Datasets:

Hendrik-a

/

arxiv

like 0

studying the stellar populations in the outskirts of spiral galaxies can provide important constraints on their structure , formation , and evolution . to that end , we present vi photometry obtained with the advanced camera for surveys for three fields located in projected distance southeast of m33 s nucleus ( corresponding to visual scale lengths or kpc in deprojected radius ) . the color - magnitude diagrams reveal a mixed stellar population whose youngest constituents have ages no greater than myr and whose oldest members have ages of at least several gyr . the presence of stars as massive as is consistent with global star formation thresholds in disk galaxies but could argue for a threshold in m33 that is on the low end of observational and theoretical expectations . the metallicity gradient as inferred by comparing the observed red giant branch ( rgb ) to the galactic globular clusters is consistent with m33 s inner disk gradient traced by several other studies . the surface density of rgb stars drops off exponentially with a radial scale length of . the scale length increases with age in a manner similar to the vertical scale height of several nearby late - type spirals . based on the metallicity gradient , density gradient , and mixed nature of the stellar populations , we conclude these fields are dominated by a disk population although we can not rule out the presence of a small halo component .

astro-ph0611868

stars are the most visible building blocks of galaxies . hence , knowledge of the ages and chemical compositions of a galaxy s stellar populations can yield insights into its formation and evolution and the astrophysical processes involved . studies seeking to understand disk galaxy evolution typically use integrated colors of large samples of galaxies at various redshifts to infer mean ages and metallicities . however , integrating the light from all the stars in an entire galaxy or any portion of a galaxy can reduce information while increasing uncertainties . that is why it is important to more fully understand nearby systems whose stellar contents we can directly resolve . such systems provide benchmarks for their more distant , unresolved counterparts . an example of such a system is m33 , the third most massive galaxy in the local group . in addition to being one of the most common types of spirals in the universe ( de vaucouleurs 1963 ) , it is also the only other known spiral in the local group besides the galaxy and m31 . therefore , understanding m33 s evolution is an important step towards understanding the evolution of spiral galaxies in general . unlike its larger , more massive spiral counterparts in the local group , m33 has a late - type morphology ( hubble type of scd ) . like many late - type spirals , it has a relatively low dark halo virialized mass of @xmath6 ( corbelli 2003 ) and high total gas mass fraction of @xmath7 ( garnett 2002 ; corbelli 2003 ) . some empirical and theoretical studies predict such galaxies to have evolutionary histories different from those of earlier morphological types ( e.g. , scannapieco & tissera 2003 ; garnett 2002 ; ferreras et al . 2004 ; dalcanton et al . 2004 ; heavens et al . hence , m33 potentially provides a contrasting view of galaxy evolution to that provided by the galaxy and m31 . the outer disks of spiral galaxies are unique environments for several reasons . they are often characterized by warps , flares , and other effects of gas infall or gravitational interactions with nearby galaxies . the disk gravitational potential is relatively shallow and the spiral features weak compared to the inner disk . moreover , hi has a low column density but dominates the total baryonic mass . what processes affect the star formation rates in such environments ? some possibilities include galactic winds , inflow of gas onto the disk , interactions with neighbors , density waves , rotational shear , and viscous radial gas flows . ferguson et al . ( 1998 ) discovered hii regions organized in spiral arms out to 2 optical radii in three late - type disk galaxies . the recent discovery of extended uv disks in several galaxies further raises the importance of spiral structure in driving star formation in the outer regions of these galaxies ( gil de paz et al . in addition , the uv and cosmic - ray backgrounds as well as feedback from massive stars could be more important regulators of star formation in the tenuous gas of outer disks than in inner disks where dense molecular cores are readily shielded from dissociation and ionization . the stellar populations in the outskirts of disk galaxies can contain important clues to the interplay between these processes . furthermore , they tell us about the galactic collapse history , conditions in the early halo , and the progression of subsequent star formation ( freeman & bland - hawthorne 2002 ) . for example , the disk truncation seen in optical surface photometry of many galaxies could be associated with a critical gas density for star formation ( e.g. , naab & ostriker 2006 ) or the maximum specific angular momentum of the baryons before the protogalactic cloud collapsed ( e.g. , van der kruit 1987 ; pohlen et al . 2000 ; de grijs et al . 2001 ) . in tiede ( 2004 ; paper i ) we used ground - based photometry reaching the horizontal branch to study the metallicity and spatial distribution of stars in m33 s outskirts . the primary conclusion was that the metallicity gradient was consistent with that of m33 s inner disk implying that the disk extends out to a deprojected radius of at least @xmath8 kpc . the present paper is a natural follow - up to paper i because we use deeper photometry obtained with the advanced camera for surveys ( acs ) on board the _ hubble space telescope _ ( hst ) that enables a more comprehensive examination of stellar ages , metallicities , and spatial distribution in m33 s outer regions . in a companion paper ( barker et al . 2006 ; paper iii ) we present a complimentary yet distinct analysis of the detailed star formation history of this region based on the acs data . this paper is organized as follows . in [ sec : obs ] we describe the observations and photometric reduction procedure . we present and discuss the resulting color - magnitude diagrams in [ sec : cmds ] . in [ sec : surf ] we investigate the stellar surface density . then in [ sec : mdf ] and [ sec : z_grad ] we derive the metallicity distribution functions and compare them to measurements reported in the literature in the context of m33 s metallicity gradient . finally , in [ sec : disc ] and [ sec : conc ] we discuss the implications and summarize the results . in this paper , age means lookback time ( i.e. , time from the present ) and the global metallicity is [ m / h ] @xmath9 $ ] where @xmath10 . for m33 s distance , inclination , and position angle we assume , respectively , 867 @xmath11 28 kpc ( paper i ; galleti et al . 2004 ) , 56@xmath12 , and @xmath13 ( corbelli & schneider 1997 ; regan & vogel 1994 ) . we make use of observations obtained with acs during cycle 11 program go-9479 . three fields were observed at projected radii of approximately @xmath14 southeast of m33 s nucleus or @xmath15 kpc . the locations of the fields are shown in figure [ fig : m33 ] . each field was observed for a total of 760 s and 1400 s in the f606w and f814w filters , respectively . the observations for each filter were divided into two cr - split exposures to allow identification and masking of cosmic rays . no dithering was carried out . henceforth , we will refer to the acs fields as a1 , a2 , and a3 in order of increasing galactocentric distance . the observations are summarized in table [ tab : obslog ] . lccrcc a1 & 2003 - 1 - 12 & f606w & @xmath16 & 01:35:18 & 30:28:58 + a1 & 2003 - 1 - 12 & f814w & @xmath17 & 01:35:18 & 30:28:58 + a2 & 2003 - 1 - 12 & f606w & @xmath16 & 01:35:32 & 30:28:06 + a2 & 2003 - 1 - 12 & f814w & @xmath17 & 01:35:32 & 30:28:06 + a3 & 2003 - 1 - 12 & f606w & @xmath16 & 01:35:46 & 30:27:02 + a3 & 2003 - 1 - 12 & f814w & @xmath17 & 01:35:46 & 30:27:02 + w1 & 2003 - 1 - 12 & f606w & @xmath18 & 01:35:42 & 30:26:16 + w1 & 2003 - 1 - 12 & f814w & @xmath19 & 01:35:42 & 30:26:16 + w2 & 2003 - 1 - 12 & f606w & @xmath18 & 01:35:56 & 30:25:25 + w2 & 2003 - 1 - 12 & f814w & @xmath19 & 01:35:56 & 30:25:24 + w3 & 2003 - 1 - 12 & f606w & @xmath18 & 01:36:10 & 30:24:20 + w3 & 2003 - 1 - 12 & f814w & @xmath19 & 01:36:10 & 30:24:20 + the data were retrieved from the space telescope science institute ( stsci ) after going through the standard calacs pipeline and `` on - the - fly '' processing . this processing generated flt images which were bias and dark subtracted and flat - fielded . because acs flat - fields are designed to flatten a source with uniform surface brightness rather than preserve total integrated counts , it was necessary to multiply the flt images by the pixel area maps provided on the stsci website . since these fields were sparsely populated , there were not enough bright stars to produce a point - spread function ( psf ) with a sufficient signal - to - noise ratio . so an empirical high signal - to - noise ratio psf was constructed from archival observations of the galactic globular cluster ( ggc ) 47 tuc . this psf varied quadratically with position on the chips . we refer the reader to sarajedini et al . ( 2006 ) for details of the psf - making procedure . we found that psf photometry resulted in color - magnitude diagrams ( cmds ) with features that were more clearly defined than those resulting from small aperture photometry . for each m33 field we first used source extractor ( bertin & arnouts 1996 ) to identify sources , daophot to measure aperture magnitudes , and allstar to measure psf magnitudes . this process was repeated on the resulting subtracted image to catch any stars that were missed . then we input the star lists into daomaster to derive precise spatial coordinate transformations between the frames in each filter . this allowed us to coadd all the frames to produce one single master image for each field . this master image was then run through two iterations of source extractor , daophot , and allstar as before . the resulting list of objects and individual cr - split exposures were input into allframe to obtain the final psf magnitudes ( stetson 1987 ; stetson & harris 1988 ; stetson 1993 ; stetson 1994 ) . we derived mean aperture corrections from @xmath20 bright , isolated stars in a1 . the standard deviations of the aperture corrections ranged from @xmath21 mag while the standard error of the mean was always @xmath22 mag . there were not enough bright stars to derive reliable aperture corrections in a2 and a3 . since the acs fields overlapped , we identified @xmath23 stars in common between a1 and a2 and @xmath24 between a2 and a3 . from these stars we calculated mean offsets to bring the psf magnitudes of a2 and a3 onto the same scale as a1 . the standard deviation of each group of offsets was @xmath25 mag while the standard error of the mean was @xmath22 mag . we applied the appropriate cte corrections following the prescription in riess & mack ( 2004 ) . these corrections were usually @xmath26 mag . to convert the aperture magnitudes to the standard @xmath27 system , we used the theoretical transformation of sirianni et al . these authors quote an uncertainty of 0.05 mag for this transformation which we adopt as the uncertainty in the photometric zero - point in the present study . to reduce spurious detections ( e.g. incompletely masked cosmic rays , background galaxies , noise , etc . ) we made cuts on the final catalog according to the following criteria . an object was kept if it was detected on all cr - split exposures , if @xmath28 of the median @xmath29 at its magnitude , @xmath30 of the median @xmath31 at its magnitude , @xmath32 , and @xmath33 times the median @xmath34 at its magnitude . the parameter @xmath29 measures the quality of the psf fit and @xmath31 measures the sharpness or spatial extent of the object . there were 22415 , 7666 , and 3337 stars in the final catalogs for a1 , a2 , and a3 , respectively . we ran a series of extensive artificial star tests to accurately estimate the photometric errors and completeness rates . for each field , we generated two catalogs of artificial stars with known magnitudes . the magnitudes in the first catalog were chosen to mimic the observed distribution of stars in the cmd . regions with no observed stars were assigned a minimum number of artificial stars to fully cover the cmd . the second catalog was compiled from the unscattered magnitudes of the model stars making up the synthetic cmds described in paper iii . this ensured a more efficient and complete sampling of the error and completeness functions because it focused on regions where the true stellar colors and magnitudes are expected to lie . in total , @xmath35 artificial stars were generated for each field . the artificial stars were inserted into the original frames using the same psfs that we used to photometer the original frames . we placed the stars on a regular grid such that the spacing between them was 2.1 psf fitting radii ( 42 pixels ) . this distance was chosen as a compromise between the requirement that the artificial stars not change the crowding conditions and the need for high computing efficiency . to fully sample the psfs , we randomly varied the starting position of the grid and allowed for sub - pixel positions . the images were then photometered in exactly the same manner as before using exactly the same detection requirements . in figure [ fig : a1_fake_err ] we show the difference between input and output magnitude as a function of input magnitude for all recovered stars in field a1 . the gray points are the individual artificial stars while the black squares and error bars represent the median and standard deviation in bins 0.4 mag wide . this figure shows how a star is likely to get redistributed in the cmd if you know where it originates before the measurement process . for most of the magnitude range covered in each band , there is no significant difference between the input and output magnitudes . however , stars with input magnitudes @xmath36 or @xmath37 tend to be recovered brighter than their true magnitudes because measuring errors or random fluctuations in the unresolved background light may scatter faint stars below the detection limit ( stetson & harris 1988 ; gallart et al . 1995 ; bellazzini et al . 2002b ) . in figure [ fig : a1_comp ] we show the variation of the completeness rate with input magnitude and color for a1 . the solid , dotted , and dashed lines correspond to the color ranges @xmath38 , @xmath39 , and @xmath40 , respectively . the completeness rises rapidly from the faint end and , due to our detection requirements and the presence of cosmic rays and bad pixels , levels off at @xmath41 at @xmath42 and @xmath43 . the completeness rate reaches 50% at @xmath44 and @xmath45 depending on the color . the errors and completeness rates for a2 and a3 are similar to within @xmath46 over most of the magnitude range ( i.e. , crowding is not a strong limiting factor ) . parallel observations were also taken with the wide field and planetary camera 2 ( wfpc2 ) . the exposure times were 280 s in f606w and 1000 s in f814w divided into two cr - split exposures per filter . [ fig : m33 ] shows the locations of the fields which will be referred to as w1 , w2 , and w3 in order of increasing galactocentric distance . table [ tab : obslog ] contains the exposure information . we elected to use hstphot ( dolphin 2000 ) rather than daophot to obtain psf photometry . our experience has been that the former is superior in computational efficiency but the latter is superior in photometric depth . since the wfpc2 exposure times are relatively short the extra photometric depth gained by using daophot was not worth the extra computational time . photometry was obtained following the cookbook procedure in the hstphot manual . in summary , bad pixels , cosmic rays , and hot pixels were masked and the two images in each filter were coadded . then hstphot was run with individual filter and total detection thresholds of @xmath47 and @xmath48 , respectively . we enabled the options to refit sky and perform a weighted psf fit which gives the central psf pixels more weight during the fitting . to reduce spurious detections in the final catalogs we required @xmath49 , @xmath50 , and @xmath51 . even after these cuts we had to remove 72 , 154 , and 30 stars from the catalogs which were artifacts from the diffraction spikes of bright foreground stars . the final catalogs then contained 648 , 429 , and 433 stars . the cmds for a1 @xmath52 a3 are presented in figures @xmath53 . qualitatively , they bear some resemblance to the cmds of m33 s inner disk which were obtained with wfpc2 and presented in sarajedini et al . the most prominent features in all three cmds are the red clump ( rc ) at i @xmath54 24.4 and red giant branch ( rgb ) whose brightest stars reach a color of @xmath55 . field a1 also contains a significant blue plume of younger main sequence ( ms ) stars at @xmath56 and a red supergiant ( rsg ) sequence ( sometimes referred to as the `` vertical clump '' ) extending from the top of the rc and adjacent to the rgb . the cmd for field a1 is reproduced in figure [ fig : cmd_a1_iso2 ] with isochrones from girardi et al . ( 2002 ) overplotted assuming a foreground reddening of @xmath57 ( mould & kristian 1986 ; sarajedini et al . 2000 ) and extinction @xmath58 ( von hippel & sarajedini 1998 ) . and @xmath59 . ] the pair of isochrones overlapping the ms have a global metallicity [ m / h ] @xmath60 0.7 and ages of 100 and 398 myr . the three horizontal lines mark the ms positions of 5 , 4 , and 3 @xmath61 from top to bottom , respectively , for the 100 myr isochrone . in a1 , the brightest ms stars have masses close to @xmath62 . in a2 and a3 there are only a handful of possible ms stars with masses above @xmath63 and @xmath64 dozen with masses in the range @xmath65 . the termination of the ms in a1 provides an upper limit of @xmath3 myr to the ages of the youngest stars . the mean color error at i @xmath54 24 is only @xmath66 mag while the width of the ms at this magnitude is @xmath67 mag . this indicates an intrinsic color spread to the ms which could be due to a spread in metallicity , age , reddening or some combination thereof . a metallicity spread alone would have to be several tenths of a dex because the ms for [ m / h ] @xmath60 0.2 is only @xmath54 0.06 mag redder at i @xmath60 24.0 than that for [ m / h ] @xmath60 0.7 there is evidence for an age spread in the ms because a single age can not simultaneously account for the bright ms turn - off and the presence of the rsg sequence extending from the top - left of the rc . the isochrones indicate the rsg sequence is probably a few hundred myr older than the youngest ms stars . sarajedini et al . ( 2006 ) estimated a typical disk reddening of @xmath68 based on m33 rr lyraes located at @xmath69 . reddening values that large in field a1 would be surprising given that it is located almost 3 times farther out . however , little is known about the distribution of dust in m33 so we can not rule out some contribution to the ms width due to dust . the 2 @xmath52 8 gyr isochrones in fig . [ fig : cmd_a1_iso2 ] indicate an age spread of several gyr could account for much of the rgb width . however , metallicity has a larger effect on the color of the rgb than age . because of the well - known age - metallicity degeneracy , multiple combinations of age and metallicity are consistent with the position of the rgb . to get a rough idea of which ages and metallicities , we fit a guassian to the color distribution of all stars in a1 with absolute magnitudes in the range @xmath70 . the peak of the distribution lies at @xmath71 which we adopt as the dereddened value of the color index , @xmath72 . figure [ fig : rgb_met ] shows the result of comparing @xmath72 , which is the dashed line , with theoretical values extracted from the rgb loci of the girardi et al . isochrones which are displayed as solid lines . the dotted lines represent the @xmath47 width of the observed distribution . interpolating between the theoretical lines by eye we estimate that the bulk of the rgb stars in a1 could have metallicities in the range @xmath73 < -0.8 $ ] at an age of 14.1 gyr and @xmath74 < -0.4 $ ] at 2 gyr . these two ages are , respectively , the approximate maximum and minimum possible ages for first ascent rgb stars and are constrained by the age of the universe and the time at which the rgb phase transition occurs ( ferraro et al . 1995 ; barker et al . 2004 ) . in a2 the consistent ranges are @xmath75 < -0.8 $ ] at 14.1 gyr and @xmath76 < -0.4 $ ] at 2 gyr . there are not enough stars in a3 to reliably fit a gaussian using the same procedure . these limits are very approximate as we have only used the central 68% of the rgb stars . in addition , the presence of agb stars could cause an underestimate of the metallicity because they lie just to the left of first - ascent rgb stars of the same age and metallicity . we have also only used a small portion of the rgb whose overall shape depends on chemical composition and age . paper iii has a more thorough analysis which includes these effects . the color and magnitude of the rc are also sensitive to the age and metallicity of its constituent stars . in figure [ fig : cmd_a1_rc ] we show a close - up of the rc region in a1 with mean theoretical values from girardi & salaris ( 2001 ) as a function of metallicity and age . the curves have metallicities of @xmath77 , @xmath78 , and @xmath79 from left to right , respectively . the symbols on each curve represent 5 different ages . the open circles correspond to 1 gyr , squares to 2 gyr , triangles to 5 gyr , stars to 8 gyr , and filled circles to 12 gyr . the cross shows the observed mean absolute magnitude ( @xmath80 ) and dereddened color ( @xmath81 ) . this figure provides evidence for the existence of stars @xmath82 gyr old with a small dependence on metallicity . if the mean metallicity is @xmath78 then an age of @xmath83 gyr is preferred but if the metallicity is @xmath79 then @xmath84 gyr is preferred . the mean rc magnitude and color for a2 are , respectively , @xmath85 and @xmath86 while for a3 they are @xmath87 and @xmath86 . this indicates that a similar intermediate - age population exists in the other fields , as well . this fact does not rule out the presence of older ages because as can be seen in the figure stars are observed throughout the entire region spanned by the models . furthermore , the rc is biased toward younger ages because the core helium burning lifetime decreases with age ( girardi & salaris 2001 ) . indeed , there is also a sequence of stars extending from the rc toward bluer colors and fainter magnitudes which lies just above the 12 gyr point on the [ m / h ] @xmath60 1.3 curve . with such few stars it is hard to say whether it is just a random grouping or the horizontal branch ( hb ) of an older and more metal - poor population . finally , this figure also nicely shows that the left half of the rc can contain stars younger than 1 gyr . the rgb ridge lines of the ggcs m92 , ngc 6752 , and 47 tuc from brown et al . ( 2005 ) are overplotted on the cmd of a1 in figure [ fig : cmd_a1_ggc ] . the ggc data was originally in the acs instrumental system but we have transformed them using the prescription and cluster parameters outlined in brown et al.and using the same transformations that we applied to the m33 data . the ggc ridge lines span the width of m33 s rgb indicating that the _ oldest _ stars in a1 most likely have metallicities between [ fe / h ] @xmath88 ( ngc 6752 ) and @xmath89 ( 47 tuc ) . this empirical comparison is in agreement with what we found above using the girardi et al . models and demonstrates that no serious errors have been introduced in the transformation to the ground - based filter system . the hb loci of the same three ggcs are also plotted in fig . [ fig : cmd_a1_ggc ] . stars in the blue hb tail at @xmath56 belong to both m92 and ngc 6752 while the red hb at @xmath90 belongs to 47 tuc . the bulk of m33 s core helium burning stars are brighter than those of 47 tuc providing further evidence for a population younger than the ggcs . however , as noted previously , the faintest portion of m33 s rc could contain stars as old as 47 tuc . because of its magnitude and color range , the young ms makes it difficult to confirm or rule out completely the presence of a blue hb tail in m33 similar to that of m92 . however , it appears that any such population is likely to be a small fraction of the total core helium burning population at the present epoch since there is no significant overdensity of stars on the ms as would be expected from such a feature superposed on the ms . there are some stars located in the hertzprung gap between the ms and rc which could be blue hb stars , sub - giant branch ( sgb ) stars @xmath91 gyr old , or foreground stars . figures @xmath92 display the cmds for the three wfpc2 fields . due to the shorter exposure time in f606w , the limiting magnitude is about 2 magnitudes brighter than in the acs fields . most of the cmd features are washed out except for the rgb . because of the shallower photometric depth and fewer number of stars , we do not use the wfpc2 cmds in any part of this study except in the analysis of stellar surface density . the spatial distribution of stars can yield insight into what type of population we are observing ( i.e. disk , thick - disk , or halo ) . we show the surface density of rgb stars as a function of deprojected radius in figure [ fig : surf_fit ] . the stars were selected to lie between the lines in figs . @xmath92 ( note that the lines are not shown in figs . @xmath53 for clarity ) . each acs field was divided into four radial bins ( represented by diamonds ) whereas each wfpc2 field was treated in its entirety ( represented by squares ) . since w1 coincides with a3 , the wfpc2 fields were brought onto the completeness scale of the acs fields by normalizing the surface density in w1 to that in a3 . this avoids making uncertain estimates of the differing completeness rates between the acs and wfpc2 fields within the rgb selection region . however , using the raw wfpc2 densities does not change the results . hence , w1 is not shown in fig . [ fig : surf_fit ] nor is it included in the fits below . the error bars reflect the poisson uncertainty normalized by the actual area observed in each radial bin . note that we have neglected error bars in deprojected radius due to the finite thickness of the disk . seth et al . ( 2005 ) observed a sample of edge - on late - type spiral galaxies similar to m33 and found the vertical scale heights of their rgb stars to range from @xmath93 pc . if we take 500 pc as representative , then this translates to an uncertainty of @xmath94 kpc for m33 . we have also assumed that the inclination and position angle are constant . corbelli & schneider ( 1997 ) fit a tilted - ring model to the hi distribution and found that the position angle changes from @xmath95 to @xmath96 over the region we have observed . whether the stellar disk follows this change is unclear . an exponential profile was fit to the observed distribution in a least - squares sense and it is shown as the dotted line in fig . [ fig : surf_fit ] . the scale length of the fit is @xmath97 . the surface density in w3 shows a significant ( @xmath98 ) deviation from the exponential profile . to assess the level of contamination from background galaxies and foreground stars we reduced a subsample of images ( 2 in each filter ) of the hubble deep field using the same technique and thresholds that were applied to m33 . the resulting cmd is shown in figure [ fig : cmd_hdf ] and contains 7 objects in the rgb selection region . therefore , we estimate the surface density of background galaxies and foreground stars to be @xmath99 . including this constant offset in the fit reduces the exponential scale length to @xmath5 . the solid line in fig . [ fig : surf_fit ] represents the sum of the exponential ( dashed ) and offset ( dot - dashed ) . the last point still shows an excess of stars but only at the @xmath100 level . if this excess is not due to poissonian fluctuations then it could represent a transition to a more extended stellar component , a point to which we return later . the k - band surface brightness scale length of m33 has been measured to be @xmath101 ( regan & vogel 1994 ; simon et al . when dealing with integrated light , the k - band is generally thought to be a better tracer of the stellar distribution than optical bands ( but see seth et al . ( 2005 ) for an alternative view ) . regan & vogel found a systematic decrease in the surface brightness scale length with wavelength . they used a simple toy model which ascribed this trend to the absorbing effects of dust . their model neglected the effects of forward scattering and variations in stellar age and abundance but predicted that the true scale length of the stellar disk is @xmath102 . our estimate of the scale length is a more direct measurement of the underlying stellar structure and is free from many of the systematic uncertainties affecting surface photometry including dust absorption and scattering . however , it should be noted that we have sampled a small region outside that studied by regan & vogel and simon et al . hence , if m33 s density profile is not a single exponential then the difference between our measurements and theirs would be expected . rowe et al . ( 2005 ) surveyed m33 s luminous stellar populations and found a clear break in the carbon star profile at @xmath103 . while they did not make any exponential fits to their profile we estimate by eye that outside this radius the scale length of their profile is remarkably close to our result . this picture is confirmed by ferguson et al . ( 2006 ) who conducted a wide field survey of m33 with the int 2.5-m telescope . they report a similar break in the rgb density profile at about the same radius found by rowe et al . and beyond which the scale length is similar to what we have found . in paper i , we found that massive ms stars were more concentrated toward m33 s nucleus than agb stars which in turn were more concentrated than rgb stars . this implied a progression in the stellar scale length with age . inspection of the cmds presented here also suggests that the density of stars in the young ms declines faster than the density of rgb stars . to investigate this in more detail we selected several regions of the cmd on the basis that each region probes different age ranges . using the results of paper iii we found that contours of constant age run roughly diagonally since the main sequence turnoff and sub - giant branch move toward fainter and redder magnitudes with age . figure [ fig : cmd_a1_boxes ] shows the regions we selected . the boundaries were chosen to roughly follow lines of constant age and the sizes were chosen as a compromise between the need for good number statistics and a small range of ages in each box . the faint limit was chosen to avoid the region where systematic magnitude errors dominate and to minimize contamination from non - stellar sources . gallart et al . ( 1999 ) used a similar approach to isolate different age ranges in studying the star formation history of leo i. table [ tab : hvage ] lists the mean age ( averaged over all three fields ) and scale length for each box and their standard deviations . rrrrr 1 & 0.35 & 0.15 & 1.92 & 0.22 + 2 & 0.62 & 0.58 & 2.98 & 0.27 + 3 & 1.90 & 1.79 & 3.55 & 0.18 + 4 & 4.31 & 3.00 & 4.09 & 0.14 + 5 & 6.02 & 3.40 & 4.56 & 0.14 + 6 & 6.79 & 3.45 & 4.86 & 0.18 + in figure [ fig : surf_boxes ] we show the relative stellar surface density of each box for fields a1 @xmath52 a3 . there is a trend of decreasing concentration as the boxes get fainter , and , hence , older . we fit an exponential profile to each box s stellar distribution spanning all three fields . figure [ fig : hvage ] displays the behavior of the scale length with mean age . the curved lines show three power - law relations of the form @xmath104 with @xmath105 . the form of these relations is somewhat arbitrary and we show the curves only for reference . the vertical error bars are the statistical uncertainties in the scale length from the least - squares fitting procedure . the horizontal error bars represent the spread of the ages in each box . the precise ages probably represent the largest source of uncertainty in this plot . the y - errors are fairly uniform but the x - errors increase dramatically for the oldest boxes . this is because the stellar isochrones get more photometrically degenerate with age . comparison of the power - law relations suggests that @xmath106 but we refrain from making a more precise estimation due to the inherent systematic uncertainties . we discuss possible interpretations in [ sec : disc ] . following the analysis in paper i , we employ the saviane et al . ( 2000 ) grid of rgb fiducials to construct the metallicity distribution function ( mdf ) of each field . those authors used a large homogeneous photometric database of ggcs to derive a function which has one parameter , [ fe / h ] , that specifies the shape of the rgb . if an rgb star s absolute magnitude and dereddened color are known , then the function can be solved for the star s metallicity . to minimize contamination from foreground stars , agb stars , and red supergiants , we restrict the present analysis to stars in the region @xmath107 and @xmath108 . the total number of stars in the selected cmd region is 207 in a1 , 84 in a2 , and 34 in a3 . the resulting mdfs for a1 @xmath52 a3 are displayed in figure [ fig : mdf_gen ] as the solid - line generalized histograms from top to bottom , respectively . these were constructed by assigning a unit gaussian to each star with a standard deviation equal to the star s metallicity uncertainty . the median metallicity is @xmath109 , @xmath110 , and @xmath111 while the interquartile range is 0.4 , 0.6 , and 0.6 dex . the dashed line represents the `` instrumental response , '' namely the recovered distribution for a test population having a single metallicity ( bellazzini et al . 2002a ) . the test population consisted of 2000 stars whose input magnitudes reproduced the observed rgb luminosity function and whose colors reproduced the saviane et al . rgb ridge line for [ fe / h ] @xmath112 . the standard deviation of the recovered distribution is 0.04 dex and represents the total intrinsic random error introduced during the entire measuring process from the photometric reduction to the measurement of metallicities . we find that when the @xmath47 uncertainties in the distance and reddening are added in quadrature they introduce a systematic error of @xmath113 dex . we are interested in tracing out the extent of m33 s disk . to do so we must place the results of the previous section in the context of other studies . we have culled from the literature a list of rgb metallicity measurements in m33 using similar techniques . figure [ fig : z_grad ] summarizes these other measurements . the open triangle is based on the results of stephens & frogel ( 2002 , sf02 ) who imaged the central 22@xmath114 of m33 with gemini north . they obtained near - ir photometry and from the slope of the rgb calculated a mean metallicity of @xmath115 . the filled circles are the results of kim et al . ( 2002 , k02 ) , who obtained vi photometry of 10 wfpc2 fields located throughout the disk ( @xmath116 kpc ) . they found median metallicities ranging from @xmath117 to @xmath118 with a typical error of 0.09 dex . it is not clear whether their quoted errors are random or systematic . the open circles correspond to the results of galleti et al . ( 2004 , g04 ) , who obtained vi photometry of two fields to the nw of m33 s nucleus . in both fields they found a median metallicity of @xmath119 where the error represents the `` instrumental response . '' the filled squares are the results of paper i ( pi ) after applying a missing factor of cos(dec ) in the original calculation of deprojected radii . the error bars demonstrate the `` instrumental response . '' a large survey covering projected radii of @xmath120 was conducted by brooks et al . ( 2004 , b04 ) who found a peak metallicity of @xmath121 which is shown as the open ( downward - pointing ) triangle . davidge ( 2003 , d03 ) imaged a field at @xmath122 and found evidence for an excess number of stars relative to a control field which he intrepreted as agb and rgb stars in m33 . he measured the rgb metallicity to be [ fe / h ] @xmath123 which is shown as an open square . we show only the random error to be more consistent with the other points . the filled triangles represent the results of the present study while the errors are the `` instrumental response '' discussed in the previous section . the star symbols in fig . [ fig : z_grad ] correspond to 9 halo globular clusters of m33 . their metallicities were estimated by sarajedini et al . ( 2000 , s02 ) using wfpc2 vi photometry and the slopes of the cluster rgbs . the errors for the clusters are random errors propagated through the relations between rgb color , slope , and metallicity . the mean metallicity of these clusters is @xmath124 . recently , sarajedini et al . ( 2006 ) studied the rr lyrae ( rrl ) population in an acs field located at @xmath125 . they found that the rrl metallicity distribution exhibited a pronounced peak at [ fe / h ] @xmath126 which they interpreted as evidence for a field halo population . the dashed line in fig . [ fig : z_grad ] therefore represents the halo of m33 ( including clusters and field stars ) . m33 s disk presumably has a different star formation history from its halo so a transition from one to the other could be observable as a change in the apparent rgb metallicity gradient . the dotted line in fig . [ fig : z_grad ] is the fit to the inner disk fields ( filled circles ) from k02 while the solid line is their fit which excluded the innermost two fields where crowding was severe . we stress that these fits were made _ independent _ of the other metallicity measurements . in paper i , we found that the metallicity gradient was consistent with the inner disk gradient . in the present study we find that the same gradient extends out to @xmath127 . past this radius there appears to be a flattening in the gradient but the situation is somewhat uncertain . b04 interpreted their field to be dominated by a halo population because of the shallow surface density profile with a power - law slope of @xmath128 and because their metallicity matched the halo globular clusters . likewise , d03 suggested the asymptotic giant branch and rgb stars he found were the field counterparts to the halo globular clusters which may have formed over a timescale of several gyr ( see also s00 ) . unfortunately , these analyses relied heavily upon the adopted level of contamination from foreground galactic stars and background galaxies . only more precise measurements of rgb metallicities at radii past @xmath129 could determine if the disk gradient continues outward or if the slope flattens out as might be expected for a halo ( or thick - disk ) population . such measurements could require large areas to obtain a statistically significant sample of rgb stars . finally , there are several important considerations to note while examining fig . [ fig : z_grad ] . first , we have used our fiducial m33 distance , inclination , and position angle and the centers of the fields studied to determine their deprojected distances . because the d03 , b04 , and g04 fields were relatively large , it would be more accurate to plot their positions according to the mean ra and dec of the rgb stars in their fields . depending on the geometry of their fields relative to m33 s nucleus , this could have the effect of moving them inward to smaller radii because of the negative stellar density gradient . most importantly , all of the measurements presented in fig . [ fig : z_grad ] compared m33 s rgb stars to the ggcs which are old ( @xmath130 gyr ) and contain enhancements in the @xmath131-element abundance ( @xmath132 \sim 0.3 $ ] ) . as pointed out by salaris & girardi ( 2005 ) , the rgbs of galaxies with composite stellar populations , such as the lmc and smc , are often dominated by stars significantly younger than the ggcs with @xmath132 \sim 0.0 $ ] . hence , the metallicities of such systems derived using the ggcs as fiducials could be underestimated by several tenths of a dex . indeed , in paper iii we model the star formation history of fields a1 @xmath52 a3 using stellar evolutionary tracks with @xmath132 = 0.0 $ ] and we find that the true metallicity is @xmath133 dex higher but the metallicity gradient through the fields is roughly unchanged . from the girardi et al . isochrones we estimate that approximately half of this offset is due to the lower @xmath131-element abundance and half is due to a younger mean age ( @xmath134 gyr ) . if the outer regions were as young as 2 gyr , the total offset would amount to @xmath135 dex . currently there is no information on the star formation history of m33 s inner disk so it is possible the rgb metallicity gradient over its _ disk is shallower or steeper . however , it is unlikely that an age gradient can mimic the apparent metallicity difference of @xmath136 dex between m33 s central and outer regions because the latter are not young enough . it has been observed that many spiral galaxies exhibit a sharp drop in star formation across their disks , as measured by the surface brightness of the @xmath137 recombination line . the location of this drop , @xmath138 , often occurs close to where the gas surface density falls below a theoretically defined critical density required for star formation . the nature and dependence of this threshold density on the local environment is not exactly clear . kennicutt ( 1989 ) and martin & kennicutt ( 2001 ) argue that it is governed by gravitational instability as parameterized by the toomre q parameter . this parameter depends on the local epicyclic frequency and gas velocity dispersion . it accurately predicts the critical radius for many high surface brightness galaxies but fails for some low - mass spirals like m33 ( @xmath139 ) where large portions of the disk are below the threshold gas density but actively forming stars . corbelli ( 2003 ) used measurements of the gas kinematics in m33 to derive an extensive rotation curve and mass model . she found that the toomre criterion could correctly predict @xmath138 in m33 if the surface density of either the stellar disk or dark matter was added to the gas density . alternatively , a stability criterion based on the shear rate with a low gas velocity dispersion could also correctly predict @xmath138 . however , such a modified criterion does not work as well for most other galaxies ( martin & kennicutt 2001 ) . what implications do our results have for the star formation threshold in m33 ? we noted in [ sec : cmds ] that field a1 contains stars @xmath140 while a2 and a3 contain stars @xmath141 . these massive stars can be no older than their ms lifetimes of @xmath142 myr and unless we have coincidentally observed them at the end of their ms phases then they are probably even younger . this fact is direct evidence for recent star formation at @xmath143 in apparent disagreement with m33 s star formation threshold radius of @xmath144 . it is possible that these massive stars actually formed at smaller radii and have since migrated outward . given their young ages , however , there probably has not been sufficient time for this to occur . could star formation be ongoing today in our fields ? boissier et al . ( 2006 ) used galex images to study the radial variation of star formation in 43 spiral galaxies . included in their sample was m33 whose far - uv profile extended out to @xmath145 , well past the threshold radius measured with @xmath137 . indeed , nearly all of the galaxies in their sample displayed similar behavior leading to the conclusion that star formation is ongoing today in the outskirts of these galaxies but at levels too low to produce any ionizing stars . the uv continuum , however , is sensitive to the star formation rate integrated over the last 100 myr . in contrast , the @xmath137 observations trace the star formation rate over the last 20 myr ( kennicutt 1998 ) . therefore , it is possible that there are no ionizing stars alive today because star formation ended @xmath146 myr ago . this latter sencario could explain our observations of young stars in m33 s outskirts beyond the @xmath137 threshold radius and , therefore , save the applicability of the toomre q parameter to m33 . however , we would still need to explain how star formation could have occurred at all so recently despite the low disk densities in these regions . was the disk density in the recent past above the threshold density ? to answer this we need to know the current densities of the gas , stars , and dark matter . according to the tilted - ring model of corbelli ( 2003 ) the azimuthally averaged hi column density is @xmath147 , @xmath148 , and @xmath149 at the central radii of fields a1 , a2 , and a3 , respectively . complexities in the hi distribution can cause systematic and random errors when performing azimuthal averages even in detailed tilted - ring models . such is the case in m33 s outer disk where residuals between the model and observed flux at different position angles within a ring can vary by a factor of @xmath84 ( corbelli & schneider 1997 ) . azimuthally averaged column densities are what are commonly reported in the literature for other galaxies so we use them here as well . the high resolution aperture synthesis map of newton ( 1980 ) shows that the hi density throughout our fields is below the lowest contour which corresponds to @xmath150 . the amount of molecular gas in our fields is difficult to ascertain but it is unlikely to contribute significantly to the total gas content . our fields lie outside the gmc survey of engargiola et al . ( 2003 ) and are at the limit of the sensitivity and coverage of the co maps presented by heyer et al . the outermost reliable point in their map is at @xmath151 where the @xmath152 column density is @xmath153 . this can be taken as a rough upper limit for our fields considering that the co to @xmath152 conversion factor may be different in the outer disk where the metallicity is lower . a crucial point to consider is whether the gas density in our fields could have been significantly greater just a few hundred myr ago . this possibility is actually ruled out by the low star formation rate required to explain the small number of young , massive stars observed . from the star formation histories calculated in paper iii , we find that the mean star formation rate in the past 400 myr has been @xmath154 . thus , no more than @xmath155 could have been converted to stars in that time . corbelli ( 2003 ) calculated the variation in the threshold density with radius in m33 . according to her plots , the toomre criterion predicts a threshold disk density that drops from @xmath156 to @xmath157 across fields a1 @xmath52 a3 whereas the shear rate criterion predicts a threshold that is approximately constant at @xmath158 . therefore , a few hundred myr ago the gas density was below either threshold . corbelli s mass model predicts the gas to dominate the baryonic mass in these outer regions so it is also unlikely that the stellar mass contributes significantly to the total disk surface density . on the other hand , the dark matter density within the disk is about equal to the gas density . hence , including dark matter in the total disk density can explain the recent star formation in field a1 but not a2 or a3 . a growing body of evidence has suggested that a _ constant _ threshold gas density of @xmath159 describes the extent of star forming disks equally well if not better than a radially varying threshold like the toomre q parameter or shear rate criterion ( skillman 1987 ; taylor et al . 1994 ; ferguson et al . 1998 ; martin & kennicutt 2001 ) . the physical basis for such a threshold could be related to the minimum pressure needed for the formation of a cold gas phase in the ism ( elmegreen & parravano 1994 , schaye 2004 ) . if such a constant threshold applies to m33 then our results imply it is @xmath160 and could be as small as @xmath161 . indeed , boissier et al . ( 2006 ) trace m33 s far - uv profile out to gas densities of @xmath162 . we now return to our finding that the stellar scale length increases with age . this result raises several interesting questions . does this reflect something fundamental about m33 s collapse history ? does it arise from the progression of star formation on galactic scales ? or is it a by - product of subsequent dynamical processes which act to redistribute stars after they are formed ? in the common picture of inside - out galaxy formation the scale length of all stars and stellar remnants increases as the disk builds up to the present day size ( e.g. naab & ostriker 2006 ) . this means that the oldest stars we observe today should have the smallest scale length in contrast to our results which show the opposite trend in m33 . this seems to suggest an outside - in formation scenario . a different interpretation is that we are observing a transition between two distinct components in m33 , like a disk / thick - disk or disk / halo . this would explain why the number of rgb stars in field w3 is larger than predicted by a simple exponential plus constant model . if this is correct , then field a3 could have a non - negligible thick - disk / halo component which might explain why its metallicity is close to that of m33 s halo . interestingly , the m - star profile measured by rowe et al . ( 2005 ) shows a flattening at approximately the same location as fields a2 @xmath52 a3 further hinting at a second , more extended component . m33 s halo was recently isolated kinematically by mcconnachie et al . ( 2006 ) who analyzed keck deimos spectra of 280 stars located @xmath163 along the major axis . these authors found evidence for three distinct kinematic components : a disk , halo , and an intermediate component which they hypothesized could be a tidal stream in m33 . it is unclear how the intermediate component would affect our results because , if it is a stream , it might not go through our fields . the halo component , however , would be present in our fields although it s precise contribution is uncertain . our fields lie at about the same deprojected distance as those observed by mcconnachie et al . but they are on the minor axis so the halo - to - disk ratio could be larger depending on the true halo density distribution . complicating the picture is the possibility of disk orbital heating mechanisms . these processes must , at some level , modify the stellar age , metallicity , and density gradients put in place by star formation regardless of whether it progresses inside - out or vice - versa . ( e.g. , l@xmath164pine et al . 2003 , sellwood & binney 2002 , wielen et al . the classical mechanism involves gravitational encounters between stars and giant molecular clouds of mass @xmath165 ( spitzer & schwarzschild 1953 ) . this process has been shown to heat the stellar disk at a rate where the vertical and radial velocity dispersions are related by @xmath166 where @xmath167 and @xmath168 ( lacey 1984 , villumsen 1985 ) . unfortunately , observations in the galaxy indicate that 1 ) there are too few gmcs with the requisite mass ( lacey 1984 ) , 2 ) @xmath169 ( wielen 1977 ) , and 3 ) @xmath170 ( h@xmath171nninen & flynn 2002 ) . in response , other mechanisms have been proposed , from heating by spiral arms to massive halo black holes passing through the disk ( lacey & ostriker 1985 ) . the former is most promising because theoretical simulations predict @xmath172 for the resulting heating rate ( de simone et al . it s biggest problem , though , is that spiral waves have no effect on the heating rate in the vertical direction and , thus , can not explain why the heating rate has the same time dependence in the radial and vertical directions . it now appears that some combination of these processes takes place with spiral waves doing most of the heating and gmcs redistributing some of the radial peculiar velocities into the vertical direction ( e.g. , carlberg 1987 ) . the precise relative contributions could depend on the characteristics of the galaxy in question , like the mass spectrum of gmcs and the strength , number , pitch angle , and lifetime of spiral arms and other irregularities in the potential like bars and accreted satellites . an important piece of the puzzle comes from seth et al . ( 2005 ) . after analyzing the _ vertical _ distribution of stars in six late - type spirals they found that the scale height increases for successively older stellar populations in a manner such that the power - law exponent @xmath173 . for an isothermal disk , the vertical scale height is proportional to the vertical velocity dispersion . seth et al . used this proportionality to show that if orbital diffusion is responsible for their results then the heating rate in the vertical direction must be significantly less than in the galaxy , namely @xmath174 . these results are intriguingly similar to what we have found in m33 for the _ radial _ direction and point to a common origin in all late - type spirals . we have presented deep vi photometry of three fields in m33 s outskirts at deprojected radii of @xmath175 visual scale lengths . the cmds reveal a mixed stellar population with ages ranging from @xmath176 myr to at least several gyr . the presence of such young stars so far out in this galaxy is consistent with a low global star formation threshold . assuming our fields are representative of all position angles at similar deprojected radii in m33 , we argue that the threshold can not be significantly more than the present - day azimuthally averaged gas surface density of @xmath177 which is roughly consistent with , but on the low end of , theoretical expectations and previous measurements in other galaxies . the metallicity gradient as inferred by comparing the observed rgbs to the ggcs is consistent with m33 s inner disk gradient traced by several other studies . the radial surface density of rgb stars declines exponentially with a scale length of @xmath5 . the scale length increases with age in a roughly power - law fashion with an exponent that is constrained to be @xmath106 . this behavior is similar to that found for the vertical scale height in a sample of late - type spiral galaxies like m33 ( seth et al . we are unable to say whether this behavior is due to the orbital diffusion of stars as they age or to intrinsic variations in star formation history with radius . given the exponential radial distribution , metallicity gradient , and mixed ages present in our fields it is likely they belong predominantly to m33 s disk . this would mean that m33 s disk extends out to at least @xmath178 kpc , or @xmath179 v - band scale lengths . however , we can not rule out the presence of a small , extended component like a thick - disk or halo particularly in the outermost field . this possibility can be tested with kinematics and by extending the metallicity measurements in m33 beyond this region to see if its gradient changes . we thank jon holtzman for helpful comments on a draft . we also thank aaron grocholski and jonathan tan for stimulating discussions and edvige corbelli and mark heyer for providing us with their data . d.g . gratefully acknowledges support from the chilean _ centro de astrofsica _ fondap no . a.s . was supported by nsf career grant ast 00 - 94048 . barker , m. k. , sarajedini , a. , geisler , d. , harding , p. , & schommer , r. 2007 , , in press barker , m. k. , sarajedini , a. , & harris , j. 2004 , , 606 , 869 bellazzini , m. , ferraro , f. r. , origlia , l. , pancino , e. , monaco , l. , & olivia , e. 2002a , , 124 , 3222 bellazzini , m. , fusi pecci , f. , montegriffo , p. , messineo , m. , monaco , l. , & rood , r. t. 2002b , , 123 , 2541 bertin , e. , & arnouts , s. 1996 , , 117 , 393 boissier , s. , gil de paz , a. , boselli , a. , madore , b. f. , buat , v. , cortese , l. , burgarella , d. , munoz mateos , j. c. m. , barlow , t. a. , forster , k. , friedman , p. g. , martin , d. c. , morrissey , p. , neff , s. g. , schiminovich , d. , seibert , m. , small , t. , wyder , t. k. , bianchi , l. , donas , j. , heckman , t. m. , lee , y. , milliard , b. , rich , r. m. , szalay , a. s. , welsh , b. y. , yi , s. k. 2006 , , in press brooks , r. s. , wilson , c. d. , & harris , w. e. 2004 , , 128 , 237 brown , t. m. , ferguson , h. c. , smith , e. , kimble , r. a. , sweigart , a. v. , renzini , a. , rich , m. r. , & vandenberg , d. a. 2005 , , 130 , 1693 carlberg , r. g. 1987 , , 322 , 59 corbelli , e. 2003 , , 342 , 199 corbelli , e. , schneider , s. e. 1997 , , 479 , 244 dalcanton , j. j. , yoachim , p. , & bernstein , r. a. 2004 , , 608 , 189 de grijs , r. , kregel , m. , & wesson , k. h. 2001 , , 324 , 1074 de simone , r. , wu , x. , & tremaine , s. 2004 , , 350 , 627 de vaucouleurs , g. 1963 , , 8 , 31 dolphin , a. e. 2000 , , 112 , 1383 elmegreen , b. g. , & parravano , a. 1994 , , 435 , l121 engargiola , g. , plambeck , r. l , rosolowsky , e. , & blitz , l. 2003 , , 149 , 343 ferguson , a. , irwin , m. , chapman , s. , ibata , r. , lewis , g. , tanvir , n. 2006 , astro - ph/0601121 ferguson , a. m. n. , wyse , r. f. g , gallagher , j. s. , & hunter , d. a. 1998 , , 506 , l19 ferraro , f. r. , fusi pecci , f. , testa , v. , greggio , l. , corsi , c. e. , buonanno , r. , terndrup , d. m. , & zinnecker , h. 1995 , , 272 , 391 ferreras , i. , silk , j. , boehm , a. , & ziegler , b. l. 2004 , , 355 , 64 freeman , k. c. , & bland - hawthorne , j. 2002 , , 40 , 487 gallart , c. , aparicio , a. , & vlchez , j. m. 1995 , , 112 , 1928 gallart , c. , freedman , w. l. , aparicio , a. , bertelli , g. , chiosi , c. 1999 , , 118 , 2245 galleti , s. , bellazzini , m. , & ferraro , f. r. 2004 , , 423 , 925 garnett , d. r. 2002 , , 581 , 1019 gil de paz , a. , madore , b. f. , boissier , s. , swaters , r. , popescu , c. c. , tuffs , r. j. , sheth , k. , kennicutt , r. c. , jr . , bianchi , l. , thilker , d. , & martin , d. c. 2005 , , 627 , l29 girardi , l. , bertelli , g. , bressan , a. , chiosi , c. , groenewegen , m. a. t. , marigo , p. , salasnich , b. , & weiss , a. 2002 , , 391 , 195 girardi , l. , & salaris , m. 2001 , , 323 , 109 h@xmath171nninen , j. , & flynn , c. 2002 , , 337 , 731 heavens , a. , panter , b. , jimenez , r. , & dunlop , j. 2004 , nature , 428 , 625 heyer , m. h. , corbelli , e. , schneider , s. e. , & young , j. s. 2004 , , 602 , 723 kennicutt , r. c. jr . 1989 , , 344 , 685 kennicutt , r. c. jr . 1998 , , 36 , 189 kim , m. , kim , e. , lee , m. g. , sarajedini , a. , & geisler , d. 2002 , , 123 , 244 lacey , c. g. 1984 , , 208 , 687 lacey , c. g. , & ostriker , j. p. 1985 , , 299 , 633 l@xmath164pine , j. r. d. , acharova , i. a. , & mishurov , y. n. 2003 , , 589 , 210 martin , c. l. , & kennicutt , r. c. jr . 2001 , , 555 , 301 mcconnachie , a. w. , chapman , s. c. , ibata , r. a. , ferguson , a. m. n. , irwin , m. j. , lewis , g. f. , & tanvir , n. r. 2006 , , 647 , l25 mould , j. & kristian , j. 1986 , , 305 , 591 naab , t. , & ostriker , j. p. 2006 , , 366 , 899 newton , k. 1980 , , 190 , 689 pohlen , m. , dettmar , r .- j . , & ltticke , r. 2000 , , 357 , l1 regan , m. w. , & vogel , s. n. 1994 , , 434 , 536 riess , a. , & mack , j. 2004 , isr acs 2004 - 006 rowe , j. f. , richer , h. b. , brewer , j. p. , & crabtree , d. r. 2005 , , 129 , 729 salaris , m. , girardi , l. 2005 , , 357 , 669 sarajedini , a. , geisler , d. , schommer , r. , & harding , p. 2000 , , 120 , 2437 sarajedini , a. , barker , m. k. , geisler , d. , harding , p. , & schommer , r. 2006 , , 132 , 1361 saviane , i. , rosenberg , a. , piotto , g. , & aparicio , a. 2000 , , 355 , 966 scannapieco , c. , & tissera , p. b. 2003 , , 338 , 880 schaye , j. 2004 , , 609 , 667 sellwood , j. a. , & binney , j. j. 2002 , , 336 , 785 seth , a. , dalcanton , j. , & de jong , r. 2005 , , in press simon , j. d. , prada , f. , vlchez , j. m. , blitz , l. , & robertson , g. 2006 , , in press sirianni , m. , jee , m. j. , bentez , n. , blakeslee , j. p. , martel , a. r. , clampin , m. , de marchi , g. , ford , h. c. , gilliland , r. , hartig , g. f. , illingworth , g. d. , mack , j. , & mccann , w. j. 2005 , , 117 , 1049 skillman , e. d. , 1987 , in star formation in galaxies , ed . c. j. londsdale persson ( nasa cp-2466 ; washington : nasa ) , 263 spitzer , l. j. , & schwarzschild , m. 1953 , , 118 , 106 stephens , a. w. , & frogel , j. a. 2002 , , 124 , 2023 stetson , p. b. 1987 , , 99 , 191 stetson , p. b. 1993 , in stellar photometry current techniques and future developments , iau coll . 136 , ed . c. j. butler and i. elliot ( cambridge , england , cambridge university ) stetson , p. b. 1994 , , 106 , 250 stetson , p. b. , & harris , w. e. 1988 , , 96 , 909 taylor , c. l. , brinks , e. , pogge , r. w. , & skillman , e. d. 1994 , , 107 971 tiede , g. p. , sarajedini , a. , & barker , m. k. 2004 , , 128 , 224 van der kruit , p. c. 1987 , , 173 , 59 villumsen , j. v. 1985 , , 290 , 75 von hippel , t. , & sarajedini , a. 1998 , , 116 , 1789 wielen , r. 1977 , , 60 , 263 wielen , r. , fuchs , b. , & dettbarn , c. 1996 , , 314 , 438

let and be subsets of such that and . for infinitely many , let be a vector of nonnegative integers whose sum is even . we find an asymptotic expression for the number of multigraphs on the vertex set with degree sequence given by , such that every loop has multiplicity in and every non - loop edge has multiplicity in . equivalently , these are symmetric integer matrices with values allowed on the diagonal and off the diagonal . our expression holds when the maximum degree satisfies . we prove this result using the switching method , building on an asymptotic enumeration of simple graphs with given degrees ( mckay and wormald , 1991 ) . our application of the switching method introduces a novel way of combining several different switching operations into a single computation .

1303.4218

multigraphs arise in many applications including modelling transportation networks @xcite , the structure of rna @xcite and in nonparametric statistics @xcite . we use terminology `` multigraphs '' inclusively , with both multiple edges and ( possibly multiple ) loops allowed . we seek an asymptotic enumeration formula for multigraphs with a given degree sequence , satisfying certain conditions . let @xmath14 be a nonnegative integer for all pairs @xmath15 of integers which satisfy @xmath16 . then for each @xmath17 , let @xmath18 . we usually write @xmath19 instead of @xmath14 . define @xmath20 . we assume that @xmath7 is even for an infinite number of values of @xmath5 , and tacitly restrict ourselves to such @xmath5 . for subsets @xmath21 of @xmath2 , define @xmath22 to be the set of all multigraphs on the vertex set @xmath8 with degree sequence given by @xmath9 such that the multiplicity of every loop belongs to @xmath1 and the multiplicity of every non - loop edge belongs to @xmath0 . loops contribute 2 to the degree of their vertex . also define @xmath23 to be the set of multigraphs on @xmath8 with degree sequence @xmath9 ( and no restrictions on the multiplicities ) . equivalently we may think of @xmath24 as the set of all @xmath25 symmetric matrices @xmath26 with all diagonal entries in @xmath1 , all off - diagonal entries in @xmath0 and with @xmath27 for @xmath28 . note that diagonal entries are weighted by 2 in the row sum . let @xmath29 . we find an asymptotic expression for @xmath30 that holds when @xmath13 . for @xmath31 let @xmath32_r\ ] ] where @xmath33_r = a(a-1)(a-2)\cdots ( a - r+1)$ ] denotes the falling factorial . then @xmath34 and @xmath35 for all @xmath36 . for @xmath37 , define @xmath38 our main result is the following . [ main ] let @xmath0 and @xmath1 be subsets of @xmath2 such that @xmath3 and @xmath4 . suppose that @xmath39 , @xmath40 and @xmath41 . then @xmath42 for convenience , we restate the formula in the regular case . [ regular ] suppose that @xmath43 with @xmath44 even , such that @xmath45 as @xmath39 . then @xmath46 where @xmath47 setting @xmath48 we recapture the asymptotic expression for sparse simple graphs with given degrees presented in ( * ? ? ? * theorem 5.2 ) . similarly , setting @xmath49 , @xmath50 we obtain the asymptotic enumeration by degree sequence of sparse graphs with loops allowed but no multiple edges : this is the second expression in ( * ? ? ? * theorem 1.5 ) . we remark that the conditions @xmath51 and @xmath52 in theorem [ main ] can be replaced by weaker conditions : namely , that @xmath10 is non - empty and that @xmath11 has at least two elements with the smallest two consecutive . let @xmath53 denote the least element of @xmath10 and let @xmath54 be the smallest two elements of @xmath11 . this case is reduced to ours by subtracting @xmath53 from each diagonal element , @xmath55 from each off - diagonal element , and @xmath56 from each @xmath19 , provided the new degree sequence satisfies the conditions of theorem [ main ] . theorem [ main ] is proved using the switching method , building on an asymptotic enumeration of simple graphs with given degrees @xcite . our application of the switching method introduces a novel way of combining several different switching operations into a single computation . the remainder of the paper is structured as follows . the history of this asymptotic enumeration problem is briefly reviewed in section [ ss : history ] , then the new switching theorem is given in section [ s : switchinglemma ] . in section [ s : multigraphs ] we describe 15 types of switchings on multigraphs , which are used to show that certain multiplicities of edges or loops are rare . these switchings are analysed together in subsection [ ss : combine ] using the new switching theorem . in section [ s : pairings ] we complete the enumeration with the help of some calculations performed in @xcite . finally , in section [ s : naive ] we show that a nave argument leads to a formula for @xmath57 which differs asymptotically by a constant factor from the result of theorem [ main ] in the regular case . the constant factor takes two different values , depending on whether @xmath58 . the earliest work on this problem was that of read @xcite , who found exact and asymptotic formulae when @xmath59 , for all four combinations of @xmath60 and @xmath61 . the best result for @xmath62 in the sparse range is that of mckay and wormald @xcite who treated @xmath63 ; see that paper for a survey of the many earlier results on the case . in addition to @xmath64 , bender and canfield @xcite found the asymptotics for @xmath65 when @xmath66 . ( although that paper allows nonzero diagonal entries , they contribute singly to the row sums , not doubly as we have it . ) in @xcite , mckay and wormald considered @xmath64 in the dense domain defined by @xmath67 for @xmath68 and @xmath69 for all @xmath70 , where @xmath71 is the average degree . mckay @xcite found a better error term under the same conditions , while barvinok and hartigan allowed a wider range of degrees @xcite . greenhill and mckay @xcite added the option of loops in both the sparse and dense ranges by considering @xmath72 , @xmath73 . finally , mcleod and mckay @xcite analysed the case of @xmath74 when @xmath75 for @xmath76 . for sparse rectangular matrices which are not necessarily symmetric , see greenhill and mckay @xcite . in order to bound the number of multigraphs having some unusual properties , we will apply the method of switchings . it will be necessary to apply several switching types , which we could analyse one at a time as in @xcite . instead we now prove a generalized switching theorem that allows us to analyse all of them at once . in order to facilitate use of the method in future work , we will present the theorem in greater generality than is required in this paper . [ switchings ] let @xmath77 be a directed multigraph , and let @xmath78 be a positive weighting of the edges of @xmath79 . fix a non - empty finite set @xmath80 , whose elements we will call _ colours _ , and let @xmath81 be an edge colouring of @xmath79 ( which need not be proper ) . for all @xmath82 and @xmath83 denote by @xmath84 the set of edges of colour @xmath85 entering @xmath86 , and by @xmath87 the set of edges of colour @xmath85 leaving @xmath86 . we introduce the set of variables @xmath88 and consider the following system of linear inequalities on these variables : @xmath89 \\ \sum_{e\in\vargamma_c^-(v ) } s(e ) \le n(v ) , & \qquad(v\in v,\,\ , c\in c ) \label{ineq3}\\ \sum_{e\in\vargamma_c^+(v ) } \alpha(e)s(e ) \ge n(v ) , & \qquad(v\in v,\,\ , c\in c,\,\ , \vargamma_c^+(v)\ne\emptyset ) . \label{ineq4}\end{aligned}\ ] ] let @xmath90 be the set of colours entering @xmath86 . for each @xmath82 which is not a source , let @xmath91 be positive numbers such that @xmath92 . for each @xmath93 , define @xmath94 . extend this function to any directed path @xmath95 by defining @xmath96 . now suppose that @xmath97 satisfy the following conditions : * @xmath98 and @xmath99 ; * if @xmath82 is a sink of @xmath79 , or if @xmath100 for some @xmath93 , then @xmath101 . for any @xmath102 , define @xmath103 to be the set of non - trivial directed paths in @xmath79 which start in @xmath104 , end in @xmath105 , and have no internal vertices in @xmath106 . then every solution of satisfies @xmath107 where the maximum over an empty set is taken to be 0 . the case of one colour appears in ( * ? ? ? * theorem 3 ) , apart from an inconsequential difference in the conditions on @xmath108 . we proceed by reducing the general problem to one in which there is only one colour , then modifying @xmath79 slightly so that ( * ? ? ? * theorem 3 ) applies . define @xmath109 and @xmath110 to be the set of all edges ( regardless of colour ) entering @xmath86 or leaving @xmath86 , respectively . for each edge @xmath93 , define @xmath111 . if we weight inequality by @xmath112 and sum over @xmath113 , we obtain @xmath114 similarly , if @xmath115 , then we can sum over those @xmath83 with @xmath116 to obtain @xmath117 together with the nonnegativity of @xmath118 and @xmath119 , we have equations of the form of with only one colour . next remove from @xmath79 all edges @xmath120 where @xmath101 . doing so can only weaken the conditions , by decreasing the left hand side of some inequalities in ( [ 1colour1 ] ) or removing some inequalities in ( [ 1colour2 ] ) . note that none of the quantities in are changed by removal of these edges . after this change to @xmath79 , @xmath108 satisfies the requirements of ( * ? ? ? * theorem 3 ) , with the variable @xmath121 defined there set equal to @xmath108 . applying that theorem to the system defined by ( [ 1colour1 ] ) , ( [ 1colour2 ] ) and the nonnegativity of @xmath118 and @xmath119 gives . we now describe how theorem [ switchings ] can be used for counting . suppose we have a finite set of `` objects '' partitioned into disjoint classes @xmath122 , where @xmath82 for some index set @xmath123 . define @xmath124 for each @xmath82 . also suppose that for each @xmath83 we have a relation @xmath125 between objects : to be precise , @xmath125 is a multiset of ordered pairs @xmath126 of objects . ( we call @xmath125 a _ switching _ and usually define it by some operation that modifies @xmath127 to make @xmath128 . ) now define an edge - coloured directed multigraph @xmath129 with vertex set @xmath123 , where @xmath79 has a directed edge @xmath120 of colour @xmath85 if and only if @xmath130 for some @xmath131 and @xmath132 . ( there is at most one edge of each colour between any pair of distinct vertices of @xmath79 . ) for each @xmath93 let @xmath133 , counting multiplicities , where @xmath85 is the colour of @xmath120 . fix @xmath82 and @xmath83 such that @xmath134 . suppose that for any @xmath131 there are at least @xmath135 objects @xmath128 with @xmath130 , counting multiplicities . then @xmath136 similarly , for fixed @xmath137 and @xmath83 , suppose that for every @xmath132 there are at most @xmath138 objects @xmath127 with @xmath130 , counting multiplicities . then @xmath139 defining @xmath140 and @xmath141 we obtain equations . theorem [ switchings ] can thus be used to bound the relative values of @xmath142 and @xmath143 if @xmath144 satisfy the requirements of the lemma . since @xmath145 , this also bounds @xmath143 relative to @xmath146 ; i.e. , it bounds the fraction of all objects that lie in @xmath147 . define @xmath148 ( we have not attempted to optimise constants . ) given a multigraph @xmath127 , let @xmath149 denote the number of loops with multiplicity @xmath150 and let @xmath151 denote the number of non - loop edges with multiplicity @xmath150 , for @xmath152 . let @xmath153 be the set of all multigraphs with degree sequence @xmath9 , and allowing all multiplicities except for those in @xmath154 on loops and those in @xmath155 on non - loops . note that @xmath156 . we also define the subsets @xmath157 , @xmath108 of @xmath158 by @xmath159 we define 15 coloured switchings which act to reduce the number of loops or edges with high multiplicities , moving in steps from @xmath157 towards @xmath108 . these switchings are defined below , together with a description of when each should be used . indeed , any given switching will only be used on multigraphs in @xmath158 for which none of the switchings with a lower - labelled colour are applicable . an important property of all the switchings is that they do not create simple loops ( with multiplicity 1 ) and they do not create non - loop edges of multiplicity 2 or 3 , as these may not be allowed for multigraphs in @xmath158 . for each switching colour @xmath85 and multigraphs @xmath127 , @xmath128 we will define the following parameters : * @xmath160 is a lower bound on the number of ways in which a switching of colour @xmath85 can be applied to @xmath127 . * @xmath161 is an upper bound on the number of ways in which a switching of colour @xmath85 can produce @xmath128 . * if @xmath128 can be obtained from @xmath127 by performing a switching of colour @xmath85 , then we let @xmath162 be an upper bound on @xmath163 . ( note that the colour @xmath85 is determined by the edge @xmath164 . ) finally in section [ ss : combine ] we will analyse all the switchings at once by applying theorem [ switchings ] . a switching of colour 1 is used to reduce the number of loops of multiplicity equal to 2 or multiplicity at least 4 . it is applied to multigraphs @xmath165 with @xmath166 , where @xmath167 is the number of loops in @xmath127 with multiplicity 2 or multiplicity at least 4 . the switching is described by the sequence @xmath168 of distinct vertices such that there is a loop of multiplicity @xmath169 at @xmath170 and a loop of multiplicity @xmath171 at @xmath172 , with @xmath173 . let @xmath174 be the multiplicity of the edge from @xmath170 to @xmath172 in @xmath127 ( which may equal zero ) . the switching reduces the multiplicity of both loops by 2 , and increases the multiplicity of the edge @xmath175 by 4 . this operation is depicted in figure [ f : switching1 ] . = 0.9 cm ( 9,1.5)(0,0 ) ( 0,0.0 ) ( 0.2,1.15)@xmath169 ( 2.6,1.15)@xmath171 ( 4.9,1.15)@xmath176 ( 7.3,1.15)@xmath177 ( 1.5,0.5)@xmath174 ( 6.3,0.4)@xmath178 suppose that @xmath128 can be produced from @xmath127 using a switching of colour 1 . the number of ways to perform a switching of colour 1 in @xmath127 is exactly @xmath179_{2 } \geq 9m\ ] ] as there is no restriction on the value of @xmath174 . the number of ways that @xmath128 can be produced using a switching of colour 1 is bounded above by @xmath180 , which is a bound on the number of ways to choose an oriented edge with multiplicity at least 4 . hence we can set @xmath181 and @xmath182 , giving @xmath183 a switching of colour 2 is used to reduce the number of loops of multiplicity three to at most @xmath184 . it is performed for multigraphs @xmath127 for which switchings of colour 1 do not apply and such that @xmath185 . the switching is described by the sequence @xmath168 of distinct vertices such that there is a loop of multiplicity 3 at @xmath170 and at @xmath172 . let @xmath174 be the multiplicity of the edge from @xmath170 to @xmath172 in @xmath127 ( which may equal zero ) . the switching removes the loop at @xmath170 and at @xmath172 , and increases the multiplicity of the edge @xmath175 by 6 , as illustrated in figure [ f : switching2 ] . = 0.9 cm ( 9,1.5)(0,0 ) ( 0,0.0 ) ( 0.3,1.1)@xmath186 ( 2.7,1.1)@xmath186 ( 1.5,0.5)@xmath174 ( 6.4,0.4)@xmath187 suppose that the multigraph @xmath128 can be produced from @xmath127 using a switching of colour 2 . the number of ways to perform a switching of colour 2 in @xmath127 is exactly @xmath188_2 \geq 9m$ ] , and the number of ways that @xmath128 can be produced using a switching of colour 2 is at most @xmath189 ( since an oriented edge of multiplicity at least 6 determines the reverse operation , from @xmath128 to @xmath127 ) . hence we can take @xmath190 and @xmath191 , leading to @xmath192 a switching of colour 3 is used to reduce the number of loops of multiplicity 1 to at most @xmath193 . it is applied to multigraphs @xmath127 for which switchings of colour 1 and 2 do not apply and such that @xmath194 . the switching is defined by the sequence @xmath195 of distinct vertices such that each of @xmath170 , @xmath172 , @xmath196 has a simple loop in @xmath127 ( of multiplicity 1 ) , and none of the edges @xmath175 , @xmath197 , @xmath198 are present in @xmath127 . the switching removes these three loops and joins the three vertices pairwise by simple edges ( of multiplicity 1 ) , as illustrated in figure [ f : switching3 ] . = 0.9 cm ( 9,3.8)(0,0 ) ( 0,-0.3 ) ( -0.2,0.0)@xmath199 ( 3.3,0.0)@xmath199 ( 1.5,3.5)@xmath199 ( 0.6,1.5)@xmath200 ( 2.5,1.5)@xmath200 ( 1.5,-0.2)@xmath200 ( 5.8,1.5)@xmath199 ( 7.7,1.5)@xmath199 ( 6.7,-0.2)@xmath199 suppose that the multigraph @xmath128 can be produced from @xmath127 using a switching of colour 3 . the number of ways to perform a switching of colour 3 in @xmath127 is at least @xmath201_3 - o\(\kmax \ell_1(q)^2\ ) = \ell_1(q)^3(1-o(1 ) ) \geq \nfrac{1}{2}\ , m^{3/2},\ ] ] while the number of ways that @xmath128 can be produced using a switching of colour 3 is at most @xmath202 . therefore we can take @xmath203 and @xmath204 , leading to @xmath205 a switching of colour 4 is used to reduce the number of non - loop edges of multiplicity greater than @xmath206 . it is applied to multigraphs @xmath127 for which switchings of colours @xmath207 do not apply and such that @xmath208 , where @xmath209 a switching of colour 4 in @xmath127 is described by a sequence @xmath210 of distinct vertices such that * the multiplicity of edge @xmath211 in @xmath127 is @xmath169 and the multiplicity of @xmath212 in @xmath127 is @xmath171 , where @xmath213 , * the edges @xmath175 and @xmath214 have multiplicity zero in @xmath127 ( they are non - edges ) . the switching reduces the multiplicity of edges @xmath211 and @xmath212 by one and gives multiplicity 1 to edges @xmath175 and @xmath215 , as shown in figure [ f : switching45 ] . = 0.9 cm ( 8.5,4.0)(0,-0.4 ) ( 0.0,0.0 ) ( 1.3,3.0)@xmath169 ( -0.1,1.3)@xmath200 ( 1.3,-0.35)@xmath171 ( 2.9,1.3)@xmath200 ( 6.1,3.0)@xmath216 ( 5.1,1.3)@xmath199 ( 6.1,-0.35)@xmath217 ( 8.2,1.3)@xmath199 now suppose that the multigraph @xmath128 can be produced from @xmath127 using a switching of colour 4 . the number of switchings of colour 4 that can be performed in @xmath127 is bounded below by @xmath218_2 - o(\kmax^2\ , e^+(q ) ) = 4e^+(q)^2(1-o(1 ) ) \geq 64\ , \kmax\ , m ( 1-o(1 ) ) > 60\kmax\ , m,\ ] ] and the number of switchings of colour 4 that can produce @xmath128 is at most @xmath219 hence we can take @xmath220 and @xmath221 , leading to @xmath222 a switching of colour 5 is used to reduce the number of non - loop edges of multiplicity at least 5 and at most @xmath223 . for a multigraph @xmath165 , let @xmath224 switchings of colour 5 are applied to multigraphs @xmath127 for which switchings of colours @xmath225 do not apply and such that @xmath226 . a switching of colour 5 is described by a sequence @xmath210 of distinct vertices such that * the multiplicity of @xmath227 in @xmath127 is @xmath169 and the multiplicity of @xmath228 in @xmath127 is @xmath171 , where @xmath229 , and * the multiplicity of @xmath175 and @xmath214 in @xmath127 is zero ( these are non - edges ) . this switching is also illustrated by figure [ f : switching45 ] , but with different conditions on @xmath230 as above . suppose that @xmath128 can be produced from @xmath127 using a switching of colour 5 . the number of ways that a switching of colour 5 can be performed in @xmath127 is bounded below by @xmath231_2 - o(\kmax^2 e^-(q ) ) = 4 e^-(q)^2(1-o(1 ) ) \geq 30 \kmax^2 m,\ ] ] and the number of ways to produce @xmath128 using a switching of colour 5 is at most @xmath232 hence we can let @xmath233 and @xmath234 , leading to @xmath235 a switching of colour @xmath236 is used to reduce the number of edges of multiplicity @xmath237 to at most @xmath238 , for @xmath239 . switchings of colour @xmath236 are applied to multigraphs @xmath127 for which switchings of colours @xmath240 do not apply and such that @xmath241 given such a multigraph @xmath127 , a switching of colour @xmath236 is defined by a sequence @xmath242 of distinct vertices such that * @xmath243 is an edge of multiplicity @xmath237 in @xmath127 for @xmath244 , * every edge @xmath245 with @xmath246 has multiplicity 0 in @xmath127 ( it is a non - edge ) . the switching deletes these @xmath237 edges of multiplicity @xmath237 , and inserts a complete bipartite graph @xmath247 on @xmath248 , with all @xmath249 new edges simple ( that is , multiplicity 1 ) . this operation is illustrated in the figure [ f : switching8 ] for the case @xmath250 . = 0.9 cm ( 9,4.4)(0,0 ) ( 0,0.0 ) ( 1.4,0.4)@xmath237 ( 1.4,2.8)@xmath237 ( 1.4,4.0)@xmath237 ( 1.4,1.6)@xmath237 suppose that the multigraph @xmath128 can be obtained from @xmath127 using a switching of colour @xmath236 , for @xmath251 . the number of ways in which a switching of colour @xmath236 can be performed in @xmath127 is bounded below by @xmath252_j - o(\kmax^2\ , e_j(q)^{j-1 } ) \geq \nfrac{1}{2 } m^{5j/6},\ ] ] while the number of switchings of colour @xmath236 which produce @xmath128 is at most @xmath253 . hence we can define @xmath254 and @xmath255 for @xmath256 . since @xmath257 , this leads to @xmath258 before describing more coloured switchings , we prove a useful fact . [ mostlyones ] suppose that @xmath259 is such that no switching of colour 1 to 8 applies to @xmath127 . then @xmath260 . since no switching of colour 1 , 2 or 3 applies we have @xmath261 and since no switching of colour 48 applies we know that @xmath262 the result follows . switchings of colour 9 reduce the number of loops of multiplicity 2 or multiplicity at least 4 , until this number is zero . they are applied to multigraphs @xmath127 for which the switchings of colours @xmath263 do not apply and such that @xmath264 . let @xmath127 be such a multigraph . a switching of colour 9 in @xmath127 is defined by a sequence @xmath265 of distinct vertices such that * there is a loop at vertex @xmath266 in @xmath127 with multiplicity @xmath150 , where @xmath267 or @xmath268 , * @xmath227 and @xmath228 are simple edges ( of multiplicity 1 ) , * @xmath269 , @xmath270 , @xmath271 , @xmath272 are all non - edges in @xmath127 ( with multiplicity zero ) . the switching reduces the multiplicity of the loop at @xmath266 to @xmath273 , removes the two simple edges @xmath274 and @xmath275 and inserts the four simple edges @xmath269 , @xmath276 , @xmath277 , @xmath272 , as shown in figure [ f : switching9 ] . = 0.9 cm ( 8,3.4)(0,0 ) ( 0,0.0 ) ( 1.4,3.3)@xmath150 ( 6.2,3.3)@xmath273 suppose that the multigraph @xmath128 can be produced from @xmath127 by a switching of colour 9 . the number of ways to perform a switching of colour 9 in @xmath127 is at least @xmath278_2 - o(\kmax^2e_1(q)){\aftergroup\egroup\originalright } ) \geq \nfrac{1}{2}\ , l(q)m^2,\ ] ] using lemma [ mostlyones ] , while the number of ways to produce @xmath128 using a switching of colour 9 is at most @xmath279 . ( we ignore the presence of the loop at @xmath266 in this upper bound , since no such loop exists when @xmath267 . ) hence we can let @xmath280 and @xmath281 , leading to @xmath282 since @xmath283 . switchings of colour 10 reduce the number of loops of multiplicity 3 , until this number is zero . they are applied to multigraphs @xmath127 such that switchings of colours @xmath284 do not apply and such that @xmath285 . let @xmath127 be such a multigraph . a switching of colour 10 is defined by a sequence of distinct vertices @xmath286 such that * there is a loop of multiplicity 3 at @xmath266 in @xmath127 , * there is a simple edge @xmath287 in @xmath127 , for @xmath288 , * the edges @xmath289 and @xmath290 all have multiplicity 0 in @xmath127 , for @xmath288 . the switching removes the loop of multiplicity 3 and the simple edges @xmath291 for @xmath288 , and adds the six simple edges @xmath292 , as shown in figure [ f : switching10 ] . = 0.9 cm ( 8,4.7)(0,0 ) ( 0,0.0 ) ( 1.4,4.5)@xmath186 now let @xmath128 be a multigraph which can be formed from @xmath127 by a switching of colour 10 . the number of ways that a switching of colour 10 can be performed in @xmath127 is at least @xmath293_3 - o(\kmax^2 e_1(q)^2){\aftergroup\egroup\originalright } ) \geq \nfrac{1}{2}\ , \ell_3(q ) m^3,\ ] ] using lemma [ mostlyones ] , and the number of ways that @xmath128 can be produced using a switching of colour 10 is bounded above by @xmath294 . hence we can take @xmath295 and @xmath296 . this leads to @xmath297 we do not tackle single loops yet . instead , the next two switchings reduce non - loop edges of high multiplicity down to zero . switchings of colour 11 reduce the number of non - loop edges with multiplicity 4 or multiplicity at least 7 , until this number is zero . switchings of colour 11 are applied to multigraphs @xmath127 for which switchings of colours @xmath298 do not apply and @xmath299 , where @xmath300 let @xmath127 be such a multigraph . a switching of colour 11 in @xmath127 is defined by a sequence @xmath301 of distinct vertices such that * the edge @xmath302 has multiplicity @xmath150 in @xmath127 , where @xmath303 or @xmath304 , * edges @xmath287 are simple edges in @xmath127 ( with multiplicity 1 ) for @xmath288 , * the edges @xmath289 and @xmath305 are non - edges in @xmath127 for @xmath288 ( with multiplicity zero ) . the switching reduces the multiplicity of @xmath302 to @xmath306 , removes the edges @xmath291 for @xmath288 and adds the simple edges @xmath289 and @xmath307 for @xmath288 . this operation is illustrated in figure [ f : switching11 ] . = 0.9 cm ( 9,4.7)(0,0 ) ( 0,0.0 ) ( 1.3,4.1)@xmath150 ( 6.2,4.1)@xmath306 suppose that the multigraph @xmath128 can be obtained from @xmath127 by performing a switching of colour 11 . the number of ways to perform a switching of colour 11 in @xmath127 is at least @xmath308_3 - o(\kmax^2 e_1(q)^2){\aftergroup\egroup\originalright } ) \geq e(q)\ , m^3,\ ] ] using lemma [ mostlyones ] , and the number of ways to produce @xmath128 using a switching of colour 11 is at most @xmath309 . hence we can let @xmath310 and @xmath311 . this leads to @xmath312 switchings of colour 12 reduce the number of non - loop edges with multiplicity 5 or 6 , until this number is zero . they are applied to multigraphs @xmath127 such that switchings of colours @xmath313 do not apply and @xmath314 . let @xmath127 be such a multigraph . then a switching of colour 12 is defined by a sequence of distinct vertices @xmath315 such that * the edge @xmath302 has multiplicity @xmath150 in @xmath127 , where @xmath316 , * each edge @xmath291 is a simple edge in @xmath127 , with multiplicity 1 , for @xmath317 , * the edges @xmath289 and @xmath307 all have multiplicity 0 in @xmath127 for @xmath317 ( that is , they are non - edges ) . the switching reduces the multiplicity of the edge @xmath302 to @xmath318 , removes the edges @xmath291 , @xmath317 , and inserts the simple edges @xmath289 , @xmath319 for @xmath317 , as shown in figure [ f : switching12 ] . = 0.9 cm ( 9,7.1)(0,0 ) ( 0,0.0 ) ( 1.3,6.5)@xmath150 ( 6.3,6.5)@xmath318 suppose that the multigraph @xmath128 can be formed from @xmath127 by a switching of colour 12 . the number of ways that a switching of colour 12 can be performed in @xmath127 is at least @xmath320_5 - o(\kmax^2 e_1(q)^4){\aftergroup\egroup\originalright } ) \geq ( e_5(q ) + e_6(q))\ , m^5,\ ] ] using lemma [ mostlyones ] , and the number of ways that @xmath128 can be produced by a switching of colour 12 is at most @xmath321 . hence we can set @xmath322 and @xmath323 . this gives @xmath324 switchings of colour 13 are used to reduce the number of simple loops until this number is at most @xmath325 . they are applied to multigraphs @xmath127 for which switchings of colours @xmath326 do not apply and @xmath327 . let @xmath127 be such a multigraph . then a switching of colour 13 is defined by a sequence of distinct vertices @xmath328 such that there is a simple loop on @xmath266 in @xmath127 , the edge @xmath329 is a simple edge in @xmath127 , and the edges @xmath269 , @xmath270 are both absent in @xmath127 ( that is , they have multiplicity zero ) . the switching removes the simple loop at @xmath266 and the simple edge @xmath175 and inserts the two simple edges @xmath269 , @xmath270 , as shown in figure [ f : switching13 ] . = 0.9 cm ( 9,2.1)(0,0 ) ( 0,0.0 ) suppose that the multigraph @xmath128 can be obtained from @xmath127 by performing a switching of colour 13 . the number of ways that a switching of colour 13 can be performed in @xmath127 is at least @xmath330 using lemma [ mostlyones ] , and the number of ways that a switching of colour 13 can produce @xmath128 is at most @xmath331 . therefore we can define @xmath332 and @xmath333 , leading to @xmath334 using the definition of @xmath335 . switchings of colour 14 reduce the number of non - loop edges with multiplicity 2 until this number is at most @xmath336 . this switching is applied to multigraphs @xmath127 such that switchings of colours @xmath337 do not apply and @xmath338 . let @xmath127 be such a multigraph . a switching of colour 14 in @xmath127 is described by a sequence @xmath339 of distinct vertices such that * @xmath302 is an edge of multiplicity 2 in @xmath127 , * @xmath227 and @xmath228 are simple edges in @xmath127 ( with multiplicity 1 ) , * none of the edges @xmath289 or @xmath305 are present in @xmath127 , for @xmath340 ( these are all non - edges , with multiplicity zero ) . the switching removes these @xmath186 edges and replaces them with two copies of @xmath341 , one on @xmath342 centred at @xmath266 and the other on @xmath343 centred at @xmath344 . this operation is shown in figure [ f : switching14 ] . = 0.9 cm ( 9,3)(0,0 ) ( 0,0.0 ) ( 1.4,2.85)@xmath345 suppose that the multigraph @xmath128 can be produced from @xmath127 using a switching of colour 14 . the number of ways to perform a switching of colour 14 in @xmath127 is at least @xmath346_2 - o(\kmax^2 e_1(q)){\aftergroup\egroup\originalright } ) \geq e_2(q ) m^2,\ ] ] using lemma [ mostlyones ] , and the number of ways that @xmath128 can be produced using a switching of colour 14 is at most @xmath347 . therefore we can take @xmath348 and @xmath349 , leading to @xmath350 by the definition of @xmath351 . switchings of colour 15 are used to reduce the number of non - loop edges with multiplicity 3 until this number is at most @xmath352 . they are applied to multigraphs @xmath127 such that switchings of colours @xmath353 do not apply and @xmath354 . let @xmath127 be such a multigraph . then a switching of colour 15 is defined by a sequence of distinct vertices @xmath355 such that * @xmath302 has multiplicity 3 in @xmath127 , * each of @xmath291 is a simple edge in @xmath127 , for @xmath288 , * the edges @xmath289 and @xmath307 are absent in @xmath127 for @xmath288 . the switching removes the edges @xmath291 for @xmath356 ( setting the multiplicity of each to zero ) and inserts the simple edges @xmath289 , @xmath307 for @xmath288 , as illustrated in figure [ f : switching11 ] for the case that @xmath357 . suppose that the multigraph @xmath128 can be produced by performing a switching of colour 15 from @xmath127 . the number of ways to perform a switching of colour 15 in @xmath127 is at least @xmath358_3 - o(\kmax^2 e_1(q)^2){\aftergroup\egroup\originalright } ) \geq e_3(q ) m^3,\ ] ] using lemma [ mostlyones ] , while the number of ways that @xmath128 could be produced using a switching of colour 15 is at most @xmath359 . therefore we can take @xmath360 and @xmath361 , leading to @xmath362 by the definition of @xmath363 . we now explain how to apply theorem [ switchings ] to analyse these switchings on @xmath158 . ( see the statement of theorem [ switchings ] for the necessary notation . ) we now define a directed graph @xmath79 with vertex set @xmath364 , where @xmath365 is associated with the set @xmath366 containing one object . these sets are certainly disjoint . by a slight abuse of notation we will identify @xmath365 and @xmath127 from now on , and write @xmath127 for both . the edge set of @xmath79 is defined as follows : there is an edge @xmath164 in @xmath79 with colour @xmath367 if and only if @xmath128 can be obtained from @xmath127 using a switching of colour @xmath85 . since a switching is used only where no switching with a lower - labelled colour applies , there is at most one edge from @xmath127 to @xmath128 in @xmath79 . hence the endvertices of an edge uniquely determine the colour of the edge . we take @xmath368 for all @xmath369 , and so for each edge @xmath164 in @xmath79 we have @xmath370 let @xmath371 be as defined in ( [ ydef ] ) , ( [ zdef ] ) . with notation as established above , we have @xmath372 [ switchinganalysis ] it follows from the analysis of the previous sections that @xmath373 , and therefore @xmath374 , for all edges @xmath164 in @xmath79 . moreover , we have provided at least one switching which can be performed from @xmath127 , for each graph @xmath127 in @xmath375 . hence @xmath157 and @xmath108 satisfy the requirements of theorem [ switchings ] and we can conclude that @xmath376 where @xmath377 for all @xmath378 . since @xmath379 by definition has only non - trivial paths , and @xmath374 for all edges in @xmath79 , we know that @xmath380 , and therefore @xmath381 now let @xmath95 be a path in @xmath79 from some @xmath382 to some element of @xmath108 , such that all internal vertices of @xmath95 belong to @xmath383 . let @xmath164 be the first edge in @xmath95 . then @xmath164 can not have colour in @xmath384 , since these switchings only produce graphs in @xmath157 . if @xmath164 is coloured with a colour in @xmath385 then @xmath386 , so @xmath387 . the remaining possibility is that @xmath164 has colour @xmath388 for some @xmath389 , in which case @xmath95 must contain at least @xmath390 edges of colour @xmath388 . our analysis showed that @xmath391 and hence that @xmath392 , for all such edges @xmath393 . by definition @xmath394 for @xmath389 . therefore , @xmath395 which implies that @xmath396 . we conclude that @xmath397 but @xmath398 , and hence @xmath399 completing the proof . it remains to obtain an asymptotic expression for @xmath400 , which we do in the next section . in this section we work in the _ pairing model _ ( also called _ configuration model _ ) , which we now describe . this model is standard for working with random graphs of fixed degrees : see for example @xcite . consider a set of @xmath7 _ points _ partitioned into _ cells _ @xmath401 , where cell @xmath402 contains @xmath19 points for @xmath28 . take a partition @xmath95 ( called a _ pairing _ ) of the @xmath7 points into @xmath403 _ pairs _ with each pair having the form @xmath404 where @xmath405 and @xmath406 for some @xmath407 . the set of all such pairings , of which there are @xmath408 , will be denoted by @xmath409 . a _ loop _ is a pair whose two points lie in the same cell , while a _ link _ is a pair involving two distinct cells . two pairs are _ parallel _ if they involve the same cells . parallel class _ is a maximal set of mutually parallel pairs . the _ multiplicity _ of a parallel class ( and of the pairs in the class ) is the cardinality of the class . as important special cases , a _ simple pair _ is a parallel class of multiplicity one , a _ double pair _ is a parallel class of multiplicity two and a _ triple pair _ is a parallel class of multiplicity three . each pairing gives rise to a multigraph in @xmath23 by replacing each cell by a vertex , and letting the multiplicity of the edge @xmath410 equal the multiplicity of the parallel class between the corresponding cells . let @xmath411 be the set of all pairings in @xmath409 with exactly @xmath412 simple loops , exactly @xmath413 double pairs and exactly @xmath55 triple pairs , but with no loops of multiplicity greater than one and no links of multiplicity greater than three . if @xmath414 can be formed from a pairing @xmath415 , then exactly @xmath416 pairings in @xmath409 give rise to @xmath417 . now defining @xmath418 we can write @xmath419 hence it suffices to obtain an asymptotic expression for the above sum . we will need the following two summation lemmas adapted from @xcite : [ sumcor ] let @xmath420 and @xmath421 be real numbers . suppose that there exist integers @xmath422 , @xmath423 with @xmath424 and @xmath425 , and a real number @xmath426 such that @xmath427 and @xmath428 for all @xmath429 $ ] and @xmath430 $ ] . further suppose that there are real numbers @xmath431 , for @xmath432 , and @xmath433 , for @xmath434 , such that @xmath435 for @xmath432 . + given @xmath436 $ ] and @xmath437 $ ] , define @xmath438 by @xmath439 and @xmath440 for @xmath432 . then @xmath441 where @xmath442 [ sumcor2 ] let @xmath424 be an integer and , for @xmath443 , let real numbers @xmath444 , @xmath445 be given such that @xmath446 and @xmath447 . define @xmath448 , @xmath449 , @xmath450 and @xmath451 . suppose that there exists a real number @xmath452 with @xmath453 such that @xmath454 for all @xmath455 $ ] , @xmath456 $ ] . define @xmath457 by @xmath439 and @xmath458 for @xmath443 . then @xmath459 where @xmath460 in the proofs of lemmas 4.34.5 , we will use several results which were proved in @xcite ( specifically , lemmas 4.14.4 of that paper and some details of their proofs ) . that paper uses values of @xmath461 that differ from ours , but only by bounded factors , and examination of the proofs in @xcite shows that the results we wish to apply remain valid when our values of @xmath461 are used . first we perform a summation over the number of edges of multiplicity three . [ tripcount ] uniformly for @xmath462 and @xmath463 , we have @xmath464 we will apply the proof of ( * ? ? ? * lemma 4.1 ) . let @xmath465 be the first value of @xmath466 for which @xmath467 , or @xmath468 if there is no such value . in ( * ? * lemma 4.1 ) , a switching is described that converts any pairing in @xmath411 to at least one in @xmath469 , for @xmath470 , so we know that @xmath471 for @xmath472 . in particular , the present lemma is true when @xmath473 , so we assume that @xmath474 . noting that @xmath475 when the denominators are nonzero , the calculation in ( * ? ? ? * lemma 4.1 ) shows that there is some uniformly bounded function @xmath476 such that @xmath477 for @xmath470 , where @xmath478 for @xmath479 and @xmath480 for @xmath481 . now we can apply lemma [ sumcor2 ] . it is clear that @xmath482 by . if @xmath483 then @xmath484 , while if @xmath485 then the definition of @xmath486 makes it evidently nonnegative . now define @xmath487 by taking the minimum and maximum of @xmath486 and @xmath488 over @xmath489 . let @xmath455 $ ] and @xmath456 $ ] , and set @xmath490 . since @xmath491 and @xmath492 , we have that @xmath493 for @xmath7 sufficiently large , by the definition of @xmath363 . therefore lemma [ sumcor2 ] applies , and gives an upper bound @xmath494 since @xmath495 and @xmath496 , @xmath497 in the case that @xmath468 , the lower bound given by lemma [ sumcor2 ] is the same within the stated error term , so we are done . this leaves the case @xmath498 . by the counts of the second switching in the proof of ( * lemma 4.1 ) , @xmath499 is only possible if @xmath500 . if this happens for @xmath501 we have @xmath502 however , this implies that @xmath503 so the upper bound matches the trivial lower bound 1 within the error term . this completes the proof . next we perform a summation over the number of simple loops . [ loopcount ] uniformly for @xmath462 , we have @xmath504 let @xmath505 be the first value of @xmath506 for which @xmath507 , or @xmath508 if there is no such value . in the proof of ( * ? ? ? * lemma 4.2 ) , a switching is described that converts any pairing in @xmath509 to at least one in @xmath510 , for @xmath511 , so we know that @xmath473 for @xmath512 . in particular , the present lemma is true when @xmath513 , so we assume that @xmath514 . noting that @xmath515 when the denominators are nonzero , the calculation in ( * ? ? ? * lemma 4.2 ) shows that there is some uniformly bounded function @xmath516 such that @xmath517 for @xmath518 , where @xmath519 for @xmath520 and @xmath521 for @xmath522 . we can now complete the proof using @xmath523 and following the argument used in the previous lemma . the treatment of the lower bound when @xmath524 needs some additional care . from the analysis of the second switching used in ( * ? ? ? * lemma 4.2 ) , @xmath524 can only happen if @xmath525 . for @xmath526 , the trivial lower bound of 1 matches the upper bound within the stated error terms . if @xmath527 then the first two terms of the summation give the lower bound @xmath528 which matches the upper bound within the stated error terms . in either case , the proof is complete . next we perform a summation over the number of edges of multiplicity two . the exponential factor in the summand corresponds to the error term from lemma [ loopcount ] which depends on @xmath413 . [ doublecount ] for any constant @xmath529 have @xmath530 let @xmath531 be the first value of @xmath532 for which @xmath533 , or @xmath534 if no such value of @xmath413 exists . as in the previous two lemmas , the switchings described in @xcite ( see also ( * ? ? ? * lemma 4 ) ) show that @xmath513 for @xmath535 . they also show that @xmath536 is only possible if @xmath537 . in particular , the lemma is true if @xmath538 , so we assume that @xmath539 . we divide the proof into two cases , following the division used in ( * ? ? ? * lemmas 4.34.4 ) . for @xmath462 , define @xmath540 first suppose that @xmath541 . from ( * ? ? ? * lemma 4.3 ) we know that for @xmath542 , @xmath543 the current lemma follows from this , arguing as in the proofs of the previous two lemmas . ( note that the term @xmath544 in the answer is absorbed into the error term , by the assumption that @xmath545 . ) in the case that @xmath536 , the condition @xmath537 implies that 1 is a sufficient lower bound . now suppose that @xmath546 . in this case @xmath537 is not possible for @xmath532 , so @xmath534 and the series contains only positive terms . by ( * ? ? ? * lemma 4.4 ) we have for @xmath547 , @xmath548 where @xmath549 b(d ) & = b = -\frac{8}{m } + \frac{16m_3}{m_2 ^ 2 } , \\[0.4ex ] \delta_d & = o\((d-1)/m_2\ ) \,\,\ , \text { uniformly for $ 1\leq d\leq n_2$}.\end{aligned}\ ] ] ( note that @xmath550 equals @xmath551 within the precision afforded by the error term , since @xmath552 . ) now we will apply lemma [ sumcor ] to @xmath553 . let @xmath554 and @xmath555 , with the minimum and maximum taken over @xmath547 , and let @xmath556 . define @xmath557 and @xmath558 . then the conditions of lemma [ sumcor ] apply with @xmath559 and some value of @xmath560 which satisfies @xmath561 . application of that lemma now gives the desired result . recall the definition of @xmath562 , @xmath563 , @xmath564 given just before the statement of theorem [ main ] . by combining the last three summations we obtain the following . [ allcount ] as @xmath40 and @xmath41 , we have @xmath565 from lemmas [ tripcount ] and [ loopcount ] , we have @xmath566 uniformly over @xmath413 . if @xmath567 , we are finished . if @xmath568 , choose two constants @xmath569 such that the actual value of the error term @xmath570 in lies in @xmath571 $ ] for all @xmath572 . applying lemma [ doublecount ] with @xmath573 and noting that the result does not depend on @xmath529 within the precision given by the error term , we are done . finally we can prove our main result . from lemma [ switchinganalysis ] and lemma [ allcount ] together with ( [ counting - pairings ] ) , we obtain @xmath574 now @xmath575 , and it follows from ( * ? ? ? * lemma 5.1 ) that @xmath576 combining these two expressions completes the proof . let @xmath577 be a probability which we will define later . abusing notation slightly , we define the functions @xmath578 and the probability generating functions @xmath579 consider a random symmetric nonnegative integer matrix @xmath580 which is created as follows : * all entries on and above the diagonal are independent ; * for @xmath581 , let @xmath582 be a randomly chosen element of @xmath10 , with + @xmath583 for all @xmath584 ; * for @xmath585 let @xmath586 be a randomly chosen element of @xmath11 , with + @xmath587 for all @xmath588 ; * finally , let @xmath589 for @xmath585 . we write @xmath590 to mean probabilities generated by the above procedure . if @xmath591 is a fixed matrix which corresponds to an element of @xmath592 then the probability that @xmath591 is produced by the above procedure is @xmath593 this follows as the probability of @xmath591 depends only on the sum of the diagonal and above - diagonal entries ( that is , on the number of edges of the corresponding multigraph ) . for @xmath581 , let @xmath594 denote the event that the sum of row @xmath70 equals @xmath19 , recalling that diagonal entries are weighted by a factor of 2 as in . now @xmath595 since @xmath590 is uniform on ( the set of matrices corresponding to ) @xmath592 . by ( incorrectly ) assuming that the events @xmath596 are independent , we obtain the following nave estimate of @xmath597 , using ( [ fred ] ) : @xmath598 \,f(z)^{n-1 } g(z^2 ) \notag \\ & = p^{-m/2}\ , j(1)^{-\binom{n}{2 } } \ , \prod_{i=1}^n \ , [ z^{k_i}]\ , j(z)^{n-1 } j^*(z^2).\label{gnaive}\end{aligned}\ ] ] let @xmath599 denote the target average row sum . the value @xmath600 of @xmath601 which makes the expected row sum equal to @xmath71 satisfies @xmath602 we write @xmath603 for @xmath604 from now on . ( as we shall see , the exact value of the @xmath605 term does not affect theorem [ reformulate ] below , within the stated error bound , even though some of the intermediate formulae we give in the proof are affected . ) we now compare the expression given in theorem [ main ] with the nave estimate @xmath603 . for future applications , it will be convenient to express the result in terms of the scaled central moments @xmath606 , @xmath607 defined by @xmath608 for @xmath609 . to calculate the product in ( [ gnaive ] ) we consider two cases , depending on the value of @xmath563 . first assume that @xmath567 ( that is , the value 2 is not permitted off the diagonal ) . then @xmath615 where @xmath616 consequently , for any integer @xmath617 with @xmath618 we have @xmath619\ , & j(z)^{n-1}j^*(z^2)\ , = \ , \sum_{j=0}^r \binom{n-1}{r - j}p^{r - j } \ , [ z^j ] h_0(z ) \notag\\ & = \binom{n-1}{r } p^r + \binom{n-1}{r-2 } y_1 p^{r-1 } + \binom{n-1}{r-3 } \ , x_3(n-1)p^r + \binom{n-1}{r } p^r \vardelta_0,\label{zrjj}\end{aligned}\ ] ] where @xmath620 h_0(z).\ ] ] now define @xmath621 notice that the coefficients of @xmath622 , when expanded as a taylor series in @xmath623 , are nonnegative and dominate those of @xmath624 . also notice that @xmath625 uniformly for @xmath626 , since @xmath627 . therefore , setting @xmath628 , @xmath629h_0^+(z)\\ & = o\ ( h_0^+(\alpha ) - 1 - p\alpha^2 - ( n-1)p^3\alpha^3\).\end{aligned}\ ] ] since @xmath630 , by taylor s theorem we can write @xmath631 substituting this into the definition of @xmath632 , we find that @xmath633 and so @xmath634 applying this bound to , we have @xmath635&\,j(z)^{n-1}j^*(z^2 ) \notag\\ & = \binom{n-1}{r } p^r \exp\biggl ( \frac{y_1r(r-1)}{kn } + \frac{x_3 r^3}{n^2 } + o(r\kmax^3/m^2 ) \biggr)\notag \\ & = \frac{n^r p^r}{r ! } \exp\biggl ( -\frac{r(r+1)}{2n } + \frac{y_1r(r-1)}{kn } + \frac { r^3(6x_3 - 1)}{6n^2 } + o(r\kmax^3/m^2 ) \biggr ) . \label{x=0case}\end{aligned}\ ] ] now we assume that @xmath568 and perform a similar calculation . we have @xmath636 where @xmath637 we find that for any @xmath617 with @xmath638 , @xmath635\,j(z)^{n-1}j^*(z^2 ) & = \binom{n+r-2}{r } p^r + \binom{n+r-4}{r-2 } y_1 p^{r-1 } \\ & { \quad}+ \binom{n+r-5}{r-3 } ( x_3 - 1 ) ( n-1)\ , p^r + \binom{n+r-2}{r}\ , p^r \vardelta_1,\end{aligned}\ ] ] where @xmath639 h_1(z).\ ] ] the coefficients of @xmath640 are dominated by those of @xmath641 , where @xmath642 arguing as before , this implies that @xmath643 , and so we have @xmath644\,j(z)^{n-1}j^*(z^2 ) & = \binom{n+r-2}{r } p^r \exp\biggl ( \frac{y_1r(r-1)}{kn } + \frac{(x_3 - 1 ) r^3}{n^2 } + o(r\kmax^3/m^2 ) \biggr)\notag \\ & { \kern-1.2cm}=\frac{n^r p^r}{r ! } \exp\biggl ( \frac{r(r-3)}{2n } + \frac{y_1r(r-1)}{kn } + \frac { r^3(6x_3 - 7)}{6n^2 } + o(r\kmax^3/m^2 ) \biggr ) . \label{x=1case } \end{aligned}\ ] ] combining the cases and we have , for @xmath581 , @xmath645\ , j(z)^{n-1}j^*(z^2 ) & = \frac{n^{k_i}\ , p^{k_i}}{{k_i } ! } \exp\biggl ( - \biggl(\frac{1 + 2x_2}{2n } + \frac{y_1}{kn}\biggr)\ , k_i - \biggl(\frac{1 - 2x_2}{2n } - \frac{y_1}{kn}\biggr)\ , k_i^2 \notag \\ & \hspace*{3.5 cm } - \frac{1 + 6x_2 - 6x_3}{6n^2}\ , k_i^3 + o(k_i\kmax^3/m^2)\biggr ) . \end{aligned}\ ] ] multiplying these @xmath5 equations together gives @xmath646&\ , j(z)^{n-1}j^*(z^2 ) \notag\\ & = \frac{n^m p^m}{\prod_{i=1}^n k_i!}\ , \exp\biggl ( \frac{m_2(2x_2 - 1 ) -2 m } { 2n } + \frac{y_1 m_2}{kn } \notag \\ & \hspace*{40 mm } { } - \frac{(6x_2 - 6x_3 + 1)m_3}{6n^2 } + o(\kmax^3/m)\biggr ) . \label{rowexp}\end{aligned}\ ] ] we now substitute ( [ pexp ] ) , ( [ jexp ] ) and ( [ rowexp ] ) into ( [ gnaive ] ) . writing the result in terms of @xmath606 , @xmath607 , @xmath7 and @xmath5 , using the identities @xmath647 we find that @xmath648 ( at this point we can observe that no remaining terms depend on @xmath85 , which verifies our earlier claim that the exact value of the @xmath605 term in ( [ pdef ] ) does not affect the statement of this theorem . ) the proof is completed by applying theorem [ main ] and using stirling s formula . [ naive - corollary ] under the conditions of theorem [ reformulate ] , if @xmath649 then @xmath650 if the stronger bound @xmath651 holds then @xmath652 finally , if @xmath653 is a regular degree sequence with @xmath44 even , then @xmath654 for the first two statements , we just need to check that all additional terms inside the exponential factor in theorem [ reformulate ] are covered by the claimed error bounds . for this it is useful to note that @xmath655 and @xmath656 . finally , when @xmath9 is a regular degree sequence we have @xmath657 and follows . the constant @xmath658 in was previously noted for simple sparse regular graphs in @xcite , and in @xcite for simple sparse regular graphs with loops allowed . interestingly , with different error terms , the same constant was observed in @xcite for dense simple regular graphs , in @xcite for dense simple regular graphs with loops allowed , and in @xcite for dense symmetric integer matrices with zero diagonal . in these three cases , we conjectured that holds ( with some vanishing error term ) for all degree sequences except for the two extreme cases of graphs with no edges and graphs with all possible edges . the corresponding conjectures may be less likely to be true in full generality for all @xmath21 . from the first expression of corollary [ naive - corollary ] , we see that @xmath597 is closely approximated by @xmath659 whenever @xmath606 is close to 1 . it appears that , in the random matrix model described at the start of this section , @xmath606 will be concentrated around 1 with high probability whenever @xmath601 tends to zero slowly enough to ensure that @xmath40 with high probability , but quickly enough to ensure that @xmath660 with high probability . if so , this will lead to a model for the degree sequences of sparse multigraphs analogous to that obtained by mckay and wormald @xcite for graphs , and by mckay and skerman @xcite for bipartite graphs and directed graphs . details will be given in a future paper .

it is possible to provide a thermodynamic interpretation for the field equations in any diffeomorphism invariant theory of gravity . this insight , in turn , leads us to the possibility of deriving the gravitational field equations from another variational principle without using the metric as a dynamical variable . i review this approach and discuss its implications . address = iucaa , pune university campus , + ganeshkhind , pune 411007 , india . + email : [email protected]

0911.1403

an unsatisfactory feature of all theories of gravity is that the field equations do not have any direct physical interpretation . the lack of an elegant principle which can lead to the dynamics of gravity ( `` how matter tells spacetime to curve '' ) is quite striking when we compare this situation with the kinematics of gravity ( `` how spacetime makes the matter move '' ) . the latter can be determined through the principle of equivalence by demanding that all freely falling observers , at all events in spacetime , must find that the equations of motion for matter reduce to their special relativistic form . our first aim will be to remedy this and provide a physical interpretation to field equations describing gravity in any diffeomorphism invariant theory . this , in turn , will lead us to the possibility of _ deriving _ the gravitational field equations from a thermodynamic variational principle without using the metric as a dynamical variable . the alternative interpretation is based on the thermodynamics of horizons . several recent investigations have shown that there is indeed a deep connection between gravitational dynamics and horizon thermodynamics ( for a review , see @xcite ) . for example , studies have shown that : * gravitational field equations in _ a wide variety of theories _ , when evaluated on a horizon , reduce to a thermodynamic identity @xmath0 . this result , first pointed out in ref.@xcite , has now been demonstrated @xcite in several cases like the stationary axisymmetric horizons and evolving spherically symmetric horizons in einstein gravity , static spherically symmetric horizons and dynamical apparent horizons in lovelock gravity , and three dimensional btz black hole horizons , frw cosmological models in various gravity theories and even @xcite in the case horava - lifshitz gravity . it is not possible to understand , in the conventional approach , why the field equations should encode information about horizon thermodynamics . * gravitational action functionals in a wide class of theories have a a surface term and a bulk term . in the conventional approach , we _ ignore _ the surface term completely ( or cancel it with a counter - term ) and obtain the field equation from the bulk term in the action . any solution to the field equation obtained by this procedure is logically independent of the nature of the surface term . but when the _ surface term _ ( which was ignored ) is evaluated at the horizon that arises in any given solution , it gives the entropy of the horizon ! ( again , this result extends far beyond einstein s theory to situations in which the entropy is not just proportional to horizon area . ) this is possible only because there is a specific holographic relationship @xcite between the surface term and the bulk term which , however , is an unexplained feature in the conventional approach to gravitational dynamics . since the surface term has the thermodynamic interpretation as the entropy of horizons , and is related holographically to the bulk term , we are again led to an indirect connection between spacetime dynamics and horizon thermodynamics . based on these features _ which have no explanation in the conventional approach _ one can argue that there is a conceptual reason why we need to relate horizon thermodynamics with gravitational dynamics ( in a wide class of theories far more general than just einstein gravity ) and revise our perspective towards spacetime . to set the stage for this future discussion , we begin by recalling the implications of the existence of temperature for horizons . in the study of normal macroscopic systems like , for example , a solid or a gas one can _ deduce _ the existence of microstructure just from the fact that the object can be heated . this was the insight of boltzmann which led him to suggest that heat is essentially a form of motion of the microscopic constituents of matter . that is , the existence of temperature is sufficient for us to infer the existence of microstructure even without any direct experimental evidence . the non - zero temperature of a horizon shows that we can actually heat up a spacetime , just as one can heat up a solid or a gas . an unorthodox way of doing this would be to take some amount of matter and arrange it to collapse and form a black hole . the hawking radiation @xcite emitted by the black hole can be used to heat up , say , a pan of water just as though the pan was kept inside a microwave oven . in fact the same result can be achieved by just accelerating through the inertial vacuum carrying the pan of water which will eventually be heated to a temperature proportional to the acceleration @xcite . these processes show that the temperatures of all horizons are as real as any other temperature @xcite . since they give rise to a class of hot spacetimes , it follows _ la _ boltzmann that the spacetimes should possess microstructure . in the case of a solid or gas , we know the nature of this microstructure from atomic and molecular physics . hence , in principle , we can work out the thermodynamics of these systems from the underlying statistical mechanics . this is not possible in the case of spacetime because we have no clue about its microstructure . however , one of the remarkable features of thermodynamics in contrast to statistical mechanics is that the thermodynamic description is fairly insensitive to the details of the microstructure and can be developed as a fairly broad frame work . for example , a thermodynamic identity like @xmath1 has a universal validity and the information about a _ given _ system is only encoded in the form of the entropy functional @xmath2 . in the case of normal materials , this entropy arises because of our coarse graining over microscopic degrees of freedom which are not tracked in the dynamical evolution . in the case of spacetime , the existence of horizons for a particular class of observers makes it mandatory that these observers integrate out degrees of freedom hidden by the horizon . to make this notion clearer , let us start from the principle of equivalence which allows one to construct local inertial frames ( lif ) with coordinates @xmath3 , around any event in an arbitrary curved spacetime . given the lif , we can now construct a local rindler frame ( lrf ) by boosting along one of the directions with an acceleration @xmath4 , thereby locally transforming the metric to a form given by : @xmath5 where @xmath6 and @xmath7 . the observers at rest ( with @xmath8 constant ) in the lrf will perceive the @xmath9 null surface as a horizon @xmath10 ( fig 1 ) . these local rindler observers and the freely falling inertial observers will attribute different thermodynamical properties to matter in the spacetime . for example , they will attribute different temperatures and entropies to the vacuum state as well as excited states of matter fields . when some matter with energy @xmath11 moves close to the horizon say , within a few planck lengths because it formally takes infinite rindler time for matter to actually cross @xmath10 the local rindler observer will consider it to have transfered an entropy @xmath12 to the horizon degrees of freedom . we will show ( in sec . 3 ) that , when the metric satisfies the field equations of any diffeomorphism invariant theory , this transfer of entropy can be given @xcite a geometrical interpretation as the change in the entropy of the horizon . this result allows us to associate an entropy functional with the null surfaces which the local rindler observers perceive as horizons . we can now demand that the sum of the horizon entropy and the entropy of matter that flows across the horizons ( both as perceived by the local rindler observers ) , should be an extremum for all observers in the spacetime . this leads to a constraint on the geometry of spacetime which can be stated , in @xmath13 , as @xmath14 for all null vectors @xmath15 in the spacetime . the general solution to is given by @xmath16 where @xmath17 has to be a constant because of the conditions @xmath18 . hence the thermodynamic principle leads uniquely to einstein s equation with a cosmological constant in 4-dimensions . notice , however , that has a new symmetry which the standard einstein s theory does not posses ; viz . , it is invariant under the transformation @xmath19 . this has important implications for the cosmological constant problem which we will discuss in sec . 4.3 . in @xmath20 , the same entropy maximization leads to a more general class of theories called lanczos - lovelock models ( see sec . 4.2 ) . the constraint on the background geometry in arises from our demand that the thermodynamic extremum principle should hold for _ all _ local rindler observers . this is identical to the manner in which freely falling observers are used to determine how gravitational field influences matter . demanding the validity of special relativistic laws for the matter variables , as determined by all the freely falling observers , allows us to determine the influence of gravity on matter . in a similar manner , demanding the maximization of entropy of horizons ( plus matter ) , as measured by all local rindler observers , leads to the dynamical equations of gravity . in this approach , both the entropy of horizons as well as the entropy of matter flowing across the horizon will be observer dependent thereby introducing a new level of observer dependence in thermodynamics . in particular , observers in different states of motion will have different regions of spacetime accessible to them ; for example , an observer falling into a black hole will not perceive its horizon in the same manner as an observer who is orbiting around it . therefore we are forced to accept that the notion of entropy is an observer dependent concept . ( at a conceptual level this is no different from the fact different freely falling observers will measure physical quantities differently ; but in this case , standard rules of special relativity allow us to translate the results between the observers . we do not yet have a similar set of rules for quantum field theory in noninertial frames . ) all these features suggest a deep relationship between quantum theory , thermodynamics and gravity which forms the main theme of this article . the rest of the article is organized as follows : in the next section , i briefly review some of the background results needed for our discussion . section 3 describes an interpretation of gravitational field equations in a general , diffeomorphism invariant , theory of gravity . using these results , it is possible to introduce an entropy maximization principle from which one can obtain the same field equations without using the metric as a dynamical variable . this is done in sec . 4 and the last section summarizes the results . we begin by summarizing some of the standard results we will need later on . in particular , we will briefly review : ( i ) the field equations in a general class of theories of gravity as well as ( ii ) the relation between noether charge and gravitational entropy of horizons . consider a theory for gravity described by the metric @xmath21 coupled to matter with some degrees of freedom generally denoted by @xmath22 . we will take the action describing such a theory in @xmath23dimensions to be @xmath24 \label{genact}\ ] ] where @xmath25 is some scalar built from metric and curvature and @xmath26 is the matter lagrangian depending on the metric and matter variables @xmath22 . ( we have assumed that @xmath27 does not involve derivatives of curvature tensor , to simplify the discussion . ) varying @xmath28 in this action with suitable boundary conditions , we will get the equations of motion ( see e.g. refs . @xcite ) : @xmath29 where @xmath30 when @xmath31 these field equations describe a class of theories , called the lanczos - lovelock theories which have very interesting geometrical features @xcite . in this case , we have @xmath32 the crucial difference between and is that , the @xmath33 in contains no derivatives of the metric higher than second order thereby leading to field equations which are second order in the metric . in contrast , can contain up to fourth order derivatives of the metric . the lagrangians which lead to the constraint @xmath34 and hence to are , of course , quite special . they can be written as a sum of terms , each involving products of curvature tensors with the @xmath35th term being a product of @xmath36 curvature tensors . the general lanczos - lovelock lagrangian has the form , @xmath37 where the @xmath38 are arbitrary constants and @xmath39 is the @xmath36-th order lanczos - lovelock lagrangian . the @xmath40 term is proportional to @xmath41 and leads to einstein s theory . it is conventional to take @xmath42 so that the @xmath43 , reduces to @xmath44 . the normalizations for @xmath45 are somewhat arbitrary for individual @xmath39 since the @xmath38s are unspecified at this stage . the @xmath46 term gives rise to what is known as gauss - bonnet theory . because of the determinant tensor , it is obvious that in any given dimension @xmath47 we can only have @xmath48 terms where @xmath49 . it follows that , if @xmath13 , then only the @xmath50 are non - zero . of these , the gauss - bonnet term ( corresponding to @xmath46 ) gives , on variation of the action , a vanishing bulk contribution in @xmath13 . ( in dimensions @xmath51 to 8 , one can have both the einstein - hilbert term and the gauss - bonnet term and so on . ) since our aim is to provide a thermodynamic interpretation of the field equations in , we first need an expression for horizon entropy in this theory . this has been already obtained by wald @xcite but we will introduce it in a manner appropriate for our purpose . in any generally covariant theory , the infinitesimal coordinate transformations @xmath52 lead to conservation of a noether current that can be obtained as follows : the variation of the gravitational lagrangian resulting from arbitrary variations of @xmath53 generically has to a surface term and hence can be expressed in the form , @xmath54 when the variations in @xmath53 arise through the diffeomorphism @xmath55 we have , @xmath56 , with @xmath57 . substituting these in and using the ( generalized ) bianchi identity @xmath58 , we obtain the conservation law @xmath59 , for the current , @xmath60 where @xmath61 represents the boundary term which arises for the specific variation of the metric in the form @xmath62 ) . it is also convenient to introduce the antisymmetric tensor @xmath63 by @xmath64 . using the known expression for @xmath65 in , it is possible to write an explicit expression for the current @xmath66 for any diffeomorphism invariant theory . for the general class of theories we are considering , the @xmath63 and @xmath66 can be expressed @xcite in the form @xmath67 @xmath68 where @xmath69 . these expressions simplify significantly at any event @xmath70 where @xmath71 behaves like an ( approximate ) killing vector and satisfies the conditions @xmath72 ( which a true killing vector will satisfy everywhere ) . then one can easily prove that @xmath73 at the event @xmath70 ; the expression for noether current simplifies considerably and is given by @xmath74 by considering physical processes involving horizons , it can be shown that the first law of black hole dynamics , for example , is consistent with the identification of the following expression as the horizon entropy : @xmath75 where @xmath76 is the temperature of the horizon . in the final expression in the integral is over any surface with @xmath77 dimension which is a spacelike cross - section of the killing horizon on which the norm of @xmath71 vanishes . as an example , consider the special case of einstein gravity for which reduces to @xmath78 if @xmath71 be the timelike killing vector in the spacetime describing a schwarzschild black hole , we can compute the noether charge @xmath79 as an integral of @xmath63 over any two surface which is a spacelike cross - section of the killing horizon on which the norm of @xmath71 vanishes . the area element on the horizon can be taken to be @xmath80 with @xmath81 being an auxiliary vector field satisfying the condition @xmath82 . then the integral in reduces to @xmath83 where @xmath84 is the horizon area . ( we have used the relations @xmath85 and the fact that @xmath71 is a killing vector . ) this result , of course , agrees with the standard one . the expression for entropy in allows one to interpret @xmath86 , where @xmath87 is the appropriately redshifted local temperature near the horizon , as an entropy density associated with the horizon . this is the interpretation which we will exploit in what follows . with this background , we are now in a position to provide a thermodynamic interpretation for the gravitational field equations in the local rindler frames ( lrfs ) around any event . we will do this in a manner analogous to the way we use the freely falling observers to determine the kinematics of gravity . at every event in spacetime , we will introduce local rindler observers and use the horizon thermodynamics perceived by these rindler observers to constrain the background geometry . we shall begin by making the notion of local rindler observers and their coordinate systems well defined . let us choose any event @xmath70 and introduce a local inertial frame ( lif ) around it with riemann normal coordinates @xmath88 such that @xmath70 has the coordinates @xmath89 in the lif . let @xmath90 be a future directed null vector at @xmath70 and we align the coordinates of lif such that @xmath90 lies in the @xmath91 plane at @xmath70 ( fig . 1 ) . we next transform from the lif to a local rindler frame ( lrf ) coordinates @xmath92 by accelerating along the x - axis with an acceleration @xmath4 by the usual transformation . the metric near the origin now reduces to the form in where ( @xmath93 ) are the coordinates of lrf . let @xmath71 be the approximate killing vector corresponding to translation in the rindler time such that the vanishing of @xmath94 characterizes the location of the local horizon @xmath10 in lrf . usually , we shall do all the computation on a timelike surface infinitesimally away from @xmath10 with @xmath95 constant , usually called a `` stretched horizon '' . let the timelike unit normal to the stretched horizon be @xmath96 . this lrf ( with metric in ) and its local horizon @xmath10 will exist within a region of size @xmath97 ( where @xmath98 is a typical component of curvature tensor of the background spacetime ) as long as @xmath99 . this condition can always be satisfied by taking a sufficiently large @xmath4 . essentially , the introduction of the lrf uses the fact that we have two length scales in the problem at any event . first is the length scale @xmath100 associated with the curvature components of the background metric over which we have no control . the second is the length scale @xmath101 associated with the accelerated trajectory which we can choose . hence we can always ensure that @xmath99 . in fact , this is clearly seen in the euclidean sector in which the horizon maps to the origin ( fig . 2 ) . the locally flat frame in the euclidean sector will exist in a region of radius @xmath100 while the trajectory of a uniformly accelerated observer will be a circle of radius @xmath101 . hence one can always keep the latter inside the former . in the euclidean sector the rindler observer s trajectory is a circle of radius @xmath101 which can be made arbitrarily close to the origin . suppose the observer s trajectory has the usual form @xmath102 which is maintained for a time interval of the order of @xmath103 . then , the trajectory will complete a full circle _ in the euclidean sector _ irrespective of what happens later . when we work in the limit of @xmath104 , this becomes arbitrarily local in both space _ and _ time . the metric in is just the metric of the locally flat region in polar coordinates . more generally , one can choose a trajectory @xmath105 such that its acceleration @xmath106 ( where @xmath107 is the time - like four velocity ) satisfies the condition @xmath108 . in a suitably chosen lif this trajectory will reduce to the standard hyperbola of a uniformly accelerated observer . this construction also defines local rindler horizons around any event . further , the local temperature on the stretched horizon will be @xmath109 so that @xmath110 with @xmath111 . the local rindler observers will view the thermodynamics of matter around them very differently from the freely falling observers ( we will say more about this later on in sec . 3.3 ) . in particular , they will attribute a loss of entropy @xmath12 when matter with an amount of energy @xmath11 gets close to the horizon ( within a few planck lengths , say ) . in the rindler frame the appropriate energy - momentum density is @xmath112 . ( it is the integral of @xmath113 that gives the rindler hamiltonian @xmath114 , which leads to evolution in rindler time @xmath115 and appears in the thermal density matrix @xmath116 . ) a local rindler observer , moving along the orbits of the killing vector field @xmath71 with four velocity @xmath117 , will associate an energy @xmath118 with a proper volume @xmath119 . if this energy gets transfered across the horizon , the corresponding entropy transfer will be @xmath120 where @xmath121 is the local ( redshifted ) temperature of the horizon and @xmath122 is the lapse function . since @xmath123 , we find that @xmath124 consider now the gravitational entropy associated with the local horizon . from the discussion of noether charge as horizon entropy [ see ] , we know that @xmath125 , associated with the ( approximate ) killing vector @xmath71 , can be thought of as local entropy current . therefore , @xmath126 can be interpreted as the gravitational entropy associated with a volume @xmath119 as measured by an observer with four - velocity @xmath127 . ( the conservation of @xmath66 ensures that there is no irreversible entropy production or dissipation in the spacetime . ) since @xmath71 satisfies locally , it follows that @xmath128 giving the current to be @xmath129 . for observers moving along the orbits of the killing vector @xmath71 with @xmath117 we get @xmath130\ , dv_{\rm prop}\ ] ] as one approaches the horizon , @xmath131 making the second term vanish and we find that @xmath132 \ , dv_{\rm prop}\ ] ] in the same limit @xmath133 will become proportional to the original null vector @xmath134 we started with ( viz . , @xmath135 goes to @xmath136 where @xmath137 is the affine parameter associated with the null vector @xmath90 we started with ) keeping everything finite . we now see that the condition @xmath138 leads to the result @xmath139k_ak_b = 0 \label{myeqn}\ ] ] since the original null vector @xmath140 was arbitrary , this equation should hold for all null vectors for all events in the spacetime . this is equivalent to @xmath141 with some constant @xmath142 . ( the constancy of @xmath142 follows from the conditions @xmath143 . ) this provides a purely thermodynamical interpretation of the field equations of any diffeomorphism invariant theory of gravity . note that is not quite the same as the standard equation in because has an extra symmetry which standard gravitational field equations do not have : these equations are invariant under the shift @xmath144 with some constant @xmath145 . ( this symmetry has important implications @xcite for cosmological constant problem which we will discuss in sec . 4.3 . ) while the properties of lrf are relevant conceptually to define the intermediate notions ( local killing vector , horizon temperature .... ) , the essential result is independent of these notions . _ just as we introduce local inertial frames to decide how gravity couples to matter , we use local rindler frames to interpret the physical content of the field equations . _ in the above interpretation we used the local , conserved , current @xmath66 . to understand why this is to be expected for consistency , recall that we have crucially used the `` democracy of all observers '' in demanding entropy balance to hold for all observers . the mathematical content of this active version of general covariance is captured by the diffeomorphism invariance of the underlying theory which determines the dynamics of the spacetime . because the diffeomorphism invariance of the theory forced us to treat all observers at an equal footing , the diffeomorphism invariance must also lead to the conserved current @xmath66 . we have already seen that this is the case . the expression for the noether current is not unique in the sense that one can add to it any term of the form @xmath146 where @xmath147 is an anti - symmetric tensor thereby ensuring @xmath148 . this ambiguity has been extensively discussed in the literature but for providing the thermodynamic interpretation to the field equations , we have ignored this ambiguity and used the expression in . there are several reasons why the ambiguity is irrelevant for our purpose . first , in a truly thermodynamic approach , one specifies the system by specifying a thermodynamic potential , say , the entropy functional . in a local description , this translates into specifying the entropy current which determines the theory . so it is perfectly acceptable to make a specific choice for @xmath66 consistent with the symmetries of the problem . second , we shall often be interested in theories in which the equations of motion are no higher than second order . in these ( so called lanczos - lovelock models ) it is not natural to add any extra term to the noether current such that it is linear in @xmath71 as we approach the horizon with a coefficient determined entirely from metric and curvature . finally , we shall obtain in sec . 4 the field equation from maximizing an entropy functional where this ambiguity will not arise . once we realize that the real physical meaning of the field equations is contained in , it is possible to re - interpret these equations in several alternative ways all of which have the same physical content . we shall mention two of them . consider an observer who sees some matter energy flux crossing the horizon . let @xmath96 be the spacelike unit normal to the stretched horizon @xmath149 , pointing in the direction of increasing @xmath150 . if @xmath71 is the approximate , timelike killing vector corresponding to rindler time , then the energy flux through a patch of stretched horizon with normal @xmath96 will be @xmath151 and the associated entropy flux will be @xmath152 where @xmath153 is the local temperature . we require this entropy flux must match the entropy change of the locally perceived horizon . the gravitational entropy current is given by @xmath154 , such that @xmath155 gives the corresponding gravitational entropy flux . so we require @xmath156 to hold at all events . the product @xmath157 for the vector @xmath158 , which satisfies @xmath159 on the stretched horizon is @xmath160 . hence we get @xmath161 as @xmath162 and the stretched horizon approaches the local horizon and @xmath163 approaches @xmath135 ( which in turn is proportional to @xmath164 ) so that @xmath165 . so , as we approach the horizon we obtain . there is another way of interpreting this result which will be useful for further generalizations . instead of allowing matter to flow across the horizon , one could have equally well considered a virtual , infinitesimal ( planck scale ) , displacement of the @xmath10 normal to itself engulfing some matter . we only need to consider infinitesimal displacements because the entropy of the matter is not ` lost ' until it crosses the horizon ; that is , until when the matter is at an infinitesimal distance ( a few planck lengths ) from the horizon . hence an infinitesimal displacement of @xmath10 normal to itself will engulf some matter . some entropy will be again lost to the outside observers unless displacing a piece of local rindler horizon itself costs some entropy . we , therefore , expect the entropy balance condition derived earlier to ensure this and indeed it does . an infinitesimal displacement of a local patch of the stretched horizon in the direction of @xmath96 , by an infinitesimal proper distance @xmath166 , will change the proper volume by @xmath167 where @xmath168 is the metric in the transverse space . the flux of energy through the surface will be @xmath169 and the corresponding entropy flux can be obtained by multiplying the energy flux by @xmath170 . hence the ` loss ' of matter entropy to the outside observer because the virtual displacement of the horizon has engulfed some matter is @xmath171 . to find the change in the gravitational entropy , we again use the noether current @xmath66 corresponding to the local killing vector @xmath71 . multiplying by @xmath158 and @xmath110 , we get @xmath172 as the stretched horizon approaches the true horizon , @xmath173 and @xmath174 making the last term vanish . so the condition @xmath175 leads to @xmath176 which is again the same as . obviously , there are many other equivalent ways of presenting this result . as an aside , we shall comment on some new conceptual issues brought about by the existence of horizons and entropy which are relevant in this context . we recall that the mathematical formulation leading to the association of temperature with any horizon is fairly universal and it does not distinguish between different horizons , like for example rindler horizon in flat space or a schwarzschild black hole event horizon or a de sitter horizon @xcite . assuming that temperature and entropy arise for fundamentally the same reason , it would be extremely unnatural _ not _ to associate entropy with all horizons . this feature , however , brings in a new layer of observer dependent thermodynamics into the theory which though it need not come as a surprise has to be tackled . we know that while an inertial observer will attribute zero temperature and zero entropy to the inertial vacuum , a rindler observer will attribute a finite temperature and non - zero ( formally divergent ; ` entanglement ' ) entropy to the same vacuum state . _ so entropy is indeed an observer dependent concept . _ when one does quantum field theory in curved spacetime , it is not only that particles become an observer dependent notion so do the temperature and entropy . this notion can be made more quantitative as follows : consider an excited state of a quantum field with energy @xmath11 above the ground state as specified in an inertial frame . when we integrate out the unobservable modes for the rindler observer in this state , we will get a density matrix @xmath177 and the corresponding entropy will be @xmath178 . the inertial vacuum state has the density matrix @xmath17 and the entropy @xmath179 . the difference @xmath180 is finite and represents the entropy attributed to the excited state by the rindler observer . ( this is finite though @xmath181 and @xmath182 can be divergent . ) in the limit of @xmath183 , which would correspond to a rindler observer who is very close to the horizon , we can show that @xmath184 to prove this , note that if we write @xmath185 , then in the limit of @xmath183 we can concentrate on states for which @xmath186 . then we have @xmath187 where we have used the facts tr @xmath188 and @xmath189 where @xmath114 is the hamiltonian for the system in the rindler frame . the last line defines the @xmath11 in terms of the difference in the expectation values of the hamiltonian in the two states . this is the amount of entropy a rindler observer would claim to be lost when the matter disappears into the horizon . ( this result can be explicitly proved for , say , one particle excited states of the field @xcite . ) one might have naively thought that the expression for entropy of matter crossing the horizon should consist of its energy @xmath11 and _ its own _ temperature @xmath190 rather than the horizon temperature @xmath191 . but the correct expression is @xmath192 ; the horizon acts as a system with some internal degrees of freedom and temperature @xmath191 _ as far as rindler observer is concerned _ so that when one adds an energy @xmath11 to it , the entropy change is @xmath193 . obviously , a rindler observer ( or an observer at rest just outside a black hole horizon ) will attribute all these entropy changes to the horizon she perceives while an inertial observer ( or an observer falling through the schwarzschild horizon ) will see none of these phenomena . this requires us to accept the fact that many thermodynamic phenomena needs to be now thought of as specifically observer dependent . for example , if we drop some hot matter into a schwarzschild black hole , then , when it gets to a few planck lengths away from the horizon it would interact with the microscopic horizon degrees of freedom _ as far as an outside observer is concerned_. a freely falling observer through the horizon will have a completely different picture . we have learnt to live with this dichotomy as far as elementary kinematics goes ; we now need to do the same as regards thermodynamics and quantum processes . as far as an outside observer is concerned , matter takes an infinite amount of coordinate time to cross the horizon . however , this is irrelevant from a practical point of view . we have , for example , considerable evidence of very different nature to suggest planck length acts as lower bound to the length scales that can be operationally defined and that no measurements can be ultra sharp at planck scales @xcite . so one can not really talk about the location of the event horizon ignoring fluctuations of this order . hence from the operational point of view , we only need to consider matter reaching within few planck lengths of the horizon to talk about entropy loss which is what we have done in our discussion . in fact , physical processes very close to the horizon must play an important role in order to provide a complete picture of the issues we are discussing . there is already some evidence @xcite that the infinite redshift induced by the horizon plays a crucial role in this though a mathematically rigorous model is lacking . the last interpretation of the field equations ( see ) given in sec . 3.2 is similar to switching from a passive point of view to an active point of view . instead of allowing matter to fall into the horizon , we made a virtual displacement of the horizon surface to engulf the matter when it is infinitesimally close to the horizon . but for the theory to be consistent , this displacement of the horizon surface degrees of freedom should cost some entropy . by determining the functional form of this entropy density , we should be able to obtain the field equations of gravity through an extremum principle . recall that thermodynamics relies entirely on the form of the entropy functional to make predictions . hence , if we can determine the form of entropy functional for gravity ( @xmath194 ) in terms of the normals to the null surfaces , then it seems natural to demand that the dynamics should follow from the extremum prescription @xmath195=0 $ ] for _ all null surfaces in the spacetime _ where @xmath196 is the relevant matter entropy . the form of @xmath196 and @xmath197 can be determined as follows . let us begin with @xmath196 which is easy to ascertain from the previous discussion . if @xmath198 is the matter energy - momentum tensor in a general @xmath199 dimensional spacetime then an expression for matter entropy _ relevant for our purpose _ can be taken to be @xmath200 where @xmath15 is a null vector field . from our we see that the entropy density associated with proper 3-volume is @xmath201 where on the horizon the vector @xmath71 becomes proportional to a null vector @xmath15 . if we now use the rindler coordinates in in which @xmath202 and interpret the factor @xmath203 as arising from an integration of @xmath204 in the range @xmath205 we find that the entropy density associated with a proper four volume is @xmath206 . this suggests treating as the matter entropy . for example , if @xmath198 is due to an ideal fluid at rest in the lif then @xmath207 will contribute @xmath208 , which by gibbs - duhem relation is just @xmath209 where @xmath210 is the entropy density and @xmath211 is the properly redshifted temperature with @xmath212 being the periodicity of the euclidean time coordinate . then @xmath213 which matches with in the appropriate limit . this argument works for any matter source , not necessarily the ones with which we conventionally associate an entropy . what is really relevant is only the _ energy _ flux close to the horizon from which one can obtain an entropy flux . we _ do _ have the notion of _ energy _ flux across a surface with normal @xmath158 being @xmath214 which holds for _ any _ source @xmath215 . given some energy flux @xmath11 in the rindler frame , there is an associated entropy flux loss @xmath216 as given by . it is _ this _ entropy which is given by and . the only non - trivial feature in is the integration range for time which is limited to @xmath205 . this is done by considering the integrals in the euclidean sector and rotating back to the lorentzian sector but the same result can be obtained working entirely in the euclidean sector . ( there is an ambiguity in the overall scaling of @xmath15 since if @xmath15 is null so is @xmath217 for all @xmath218 ; we will comment on this ambiguity , which turns out to be irrelevant , later on . ) next , let us consider the expression for @xmath197 . we will first describe the simplest possible choice and then consider a more general expression . the simplest choice is to postulate @xmath194 to be a quadratic expression @xcite in the derivatives of the normal : @xmath219 where the explicit form of @xmath220 is ascertained below . given and the expression for the total entropy becomes : @xmath221=-\int_{\ensuremath{\mathcal{v}}}{d^dx\sqrt{-g } } \left(4p_{ab}^{{\phantom{a}}{\phantom{b}}cd } { \ensuremath{\nabla}}_cn^a{\ensuremath{\nabla}}_dn^b - t_{ab}n^an^b\right ) \ , , \label{ent - func-2}\ ] ] the field equations of gravity are to be determined by extremizing this entropy functional . however , there is one crucial conceptual difference between the extremum principle introduced here and the conventional one . usually , given a set of dynamical variables @xmath222 and a functional @xmath223 $ ] , the extremum principle will give a set of equations _ for _ the dynamical variable @xmath222 . here the situation is completely different . we expect the variational principle to hold for _ all _ null vectors @xmath15 thereby leading to a constraint on the _ background metric . _ obviously , the functional in must be rather special to accomplish this and one needs to impose restrictions on @xmath220 ( and @xmath198 , though that condition turns out to be trivial ) to achieve this . it turns out as we shall see below that two conditions are sufficient to ensure this . first , the tensor @xmath224 should have the same algebraic symmetries as the riemann tensor @xmath225 of the @xmath47-dimensional spacetime . this condition can be ensured if we define @xmath226 as @xmath227 where @xmath228 is some scalar . the motivation for this choice arises from the fact that this approach leads to the same field equations as the one with @xmath27 as gravitational lagrangian in the conventional approach ( which explains the choice of the symbol @xmath27 ) . second , we will postulate the condition : @xmath229 as well as @xmath230 which is anyway satisfied by any matter energy - momentum tensor . one possible motivation for is to ensure that the field equations are no higher than second order in the metric . ( if we think of gravity as an emergent phenomenon like elasticity , then @xmath15 is like the displacement field in elasticity . the standard entropy functional @xcite used in theory of elasticity has the form in with coefficients being elastic constants . here the coefficients are @xmath231 and the condition in may be interpreted as saying the ` elastic constants of spacetime solid ' are actually ` constants ' @xcite . ) this is , however , not a crucial condition and in fact we will see below how this condition in can be relaxed . varying the normal vector field @xmath232 after adding a lagrange multiplier function @xmath233 for imposing the condition @xmath234 , we get @xmath235 \label{ent - func-3}\ ] ] where we have used the symmetries of @xmath220 and @xmath198 . an integration by parts and the condition @xmath236 , leads to @xmath237\delta n^b } + 8\int_{{\ensuremath{\partial{\ensuremath{\mathcal{v}}}}}}{d^{d-1}x\sqrt{h}\left[k_d p_{ab}^{{\phantom{a}}{\phantom{b}}cd}\left({\ensuremath{\nabla}}_c n^a\right)\right]\delta n^b } \ , , \label{ent - func-4}\ ] ] where @xmath90 is the @xmath47-vector field normal to the boundary @xmath238 and @xmath239 is the determinant of the induced metric on @xmath238 . as usual , in order for the variational principle to be well defined , we require that the variation @xmath240 of the vector field should vanish on the boundary . the second term in therefore vanishes , and the condition that @xmath241 $ ] be an extremum for arbitrary variations of @xmath232 then becomes @xmath242 where we used the antisymmetry of @xmath220 in its upper two indices to write the first term . the definition of the riemann tensor in terms of the commutator of covariant derivatives reduces the above expression to @xmath243 and we see that the equations of motion _ do not contain _ derivatives with respect to @xmath15 which is , of course , the crucial point . this peculiar feature arose because of the symmetry requirements we imposed on the tensor @xmath220 . we need the condition in holds for _ arbitrary _ null vector fields @xmath232 . one can easily show@xcite that this requires @xmath244= 8\pi t{}_b^a + \lambda\delta^a_b \label{ent - func-71}\ ] ] where @xmath245 is an arbitrary integration constant . it is also easy to see that is equivalent to with @xmath33 given by . one crucial difference between [ along with ] and is the introduction of the cosmological constant @xmath245 ; we will discuss this later on . we mentioned earlier that the expression in depends on the overall scaling of @xmath15 which is arbitrary , since @xmath217 is a null vector if @xmath15 is null . but since the arbitrary variation of @xmath15 ( with the constraint @xmath246 ) includes scaling variations of the type @xmath247 , it is clear that this ambiguity is irrelevant for determining the equations of motion . to summarize , we have proved the following . suppose we start with the lagrangian in , define a @xmath231 by ensuring that it satisfies . varying the metric with this action will lead to field equations in . we have now proved that we will get the _ same _ field equations ( but with a cosmological constant ) if we start with the expression in , maximize it with respect to @xmath15 and demand that it holds for all @xmath15 . this result might appear a little mysterious at first sight , but the following alternative description will make clear why this works . note that , using the constraints on @xmath231 we can prove the identity @xmath248 - 4n^ap_{ab}^{{\phantom{a}}{\phantom{b}}cd } { \ensuremath{\nabla}}_c{\ensuremath{\nabla}}_dn^b = 4{\ensuremath{\nabla}}_c[p_{ab}^{{\phantom{a}}{\phantom{b}}cd } n^a{\ensuremath{\nabla}}_dn^b]-2n^ap_{ab}^{{\phantom{a}}{\phantom{b}}cd } { \ensuremath{\nabla}}_{[c}{\ensuremath{\nabla}}_{d]}n^b\nonumber\\ & = & 4{\ensuremath{\nabla}}_c[p_{ab}^{{\phantom{a}}{\phantom{b}}cd } n^a{\ensuremath{\nabla}}_dn^b]-2n^ap_{ab}^{{\phantom{a}}{\phantom{b}}cd } r^b_{\phantom{b}icd}n^i = 4{\ensuremath{\nabla}}_c[p_{ab}^{{\phantom{a}}{\phantom{b}}cd } n^a{\ensuremath{\nabla}}_dn^b]+2n^ae_{ai}n^i\end{aligned}\ ] ] where the first equality uses , the second equality uses the antisymmetry of @xmath220 in c and d , the third equality uses the standard identity for commutator of covariant derivatives and the last one is based on when @xmath246 and hold . using this in the expression for @xmath249 in and integrating the four - divergence term , we can write @xmath221=-\int_{\partial{\ensuremath{\mathcal{v}}}}{d^{d-1}x k_c\sqrt{h } } ( 4p_{ab}^{{\phantom{a}}{\phantom{b}}cd } n^a{\ensuremath{\nabla}}_dn^b ) -\int_{\ensuremath{\mathcal{v}}}{d^dx\sqrt{-g}}(2e_{ab}-t_{ab})n^an^b \label{thetrick}\ ] ] so , when we vary @xmath250 $ ] ( ignoring the surface term ) we are effectively varying @xmath251 with respect to @xmath222 and demanding that it holds for all @xmath222 . there is an ambiguity of adding a term of the form @xmath252 in the integrand of the second term in leading to the final equation @xmath253 but the bianchi identity @xmath254 along with @xmath255 will make @xmath233 actually a constant . it is now clear how we can find an @xmath249 for any theory , even if does not hold . this can be achieved by starting from the expression @xmath251 as the entropy density , using for @xmath33 and integrating by parts ( see the discussion after eq . ( 14 ) in ref . @xcite ) . in this case , we get for @xmath197 the expression : @xmath256 \label{generals}\ ] ] varying this with respect to @xmath15 will then lead to the correct equations of motion and incidentally the same surface term . while one could indeed work with this more general expression , there are four reasons to prefer the imposition of the condition in . first , it is clear from that when @xmath27 depends on the curvature tensor and the metric , @xmath33 can depend up to the fourth derivative of the metric if is not satisfied . but when we impose then we are led to field equations which have , at most , second derivatives of the metric tensor which is a desirable feature . second , as we shall see below , with that condition we can actually determine the form of @xmath27 ; it turns out that in @xmath13 , it uniquely selects einstein s theory , which is probably a nice feature . in higher dimensions , it picks out a very geometrical extension of einstein s theory in the form of lanczos - lovelock theories . third , it is difficult to imagine why the terms in should occur with very specific coefficients . in fact , it is not clear why we can not have derivatives of @xmath225 in @xmath27 , if the derivatives of @xmath224 can occur in the expression for entropy . finally , if we take the idea of elastic constants being constants , then one is led to . none of these rigorously exclude the possibility in and in fact this model has been explored recently @xcite . so far we have not fixed @xmath231 so we have not fixed the theory . in a complete theory , the explicit form of @xmath231 will be determined by the long wavelength limit of the microscopic theory just as the elastic constants can in principle be determined from the microscopic theory of the lattice . in the absence of such a theory , we need to determine @xmath231 by general considerations which is possible when @xmath231 satisfies . since this condition is identically satisfied by lanczos - lovelock models which are known to be unique , our problem can be completely solved by taking the @xmath231 as a series in the powers of derivatives of the metric as : @xmath257 where @xmath258 are coupling constants with the @xmath36 th order term derived from the lanczos - lovelock lagrangian : @xmath259 where @xmath260 is the alternating tensor . the lowest order term depends only on the metric with no derivatives . the next term depends ( in addition to metric ) linearly on curvature tensor and the next one will be quadratic in curvature etc . the lowest order term in ( which leads to einstein s theory ) is @xmath261 to the lowest order , when we use for @xmath262 , the reduces to einstein s equations . the corresponding gravitational entropy functional has the @xmath263 structure . if we think of the @xmath13 spacetime being embedded in a sufficiently large k - dimensional _ flat _ spacetime we can obtain the same structure using the gauss - codazzi equations relating the ( zero ) curvature of k - dimensional space with the curvature of spacetime . ] is : @xmath264=\int_{\ensuremath{\mathcal{v}}}\frac{d^dx}{8\pi } \left({\ensuremath{\nabla}}_an^b{\ensuremath{\nabla}}_bn^a - ( { \ensuremath{\nabla}}_cn^c)^2 \right)\ ] ] the next order term ( which arises from the gauss - bonnet lagrangian ) in is : @xmath265 and similarly for all the higher orders terms . none of them can contribute in @xmath13 so we get einstein s theory as the unique choice if we assume @xmath13 . if we assume that @xmath231 is to be built _ only _ from the metric , then this choice is unique in all @xmath47 . while the matter term in the functional in has a natural interpretation in terms of entropy transfered across the horizon , the interpretation of the gravitational part needs to be made explicit . the interpretation of @xmath197 as entropy arises from the following two facts . first , we see from the identity that this term differs from @xmath266 by a total divergence . on the other hand , we have seen earlier that the term @xmath266 can be related to the gravitational entropy of the horizon through the noether current which suggests the identification . in fact , shows that when the equations of motion holds _ the total entropy of a bulk region is entirely on its boundary_. further if we evaluate this boundary term @xmath267 ( where we have manipulated a few indices using the symmetries of @xmath231 ) in the case of a _ stationary _ horizon which can be locally approximated as rindler spacetime , one gets exactly the wald entropy of the horizon @xcite . the approach outlined above has important implications for the cosmological constant problem , which we shall now briefly mention @xcite . in the conventional approach , we start with an action principle which depends on matter degrees of freedom and the metric and vary ( i ) the matter degrees of freedom to obtain the equations of motion for matter and ( ii ) the metric @xmath28 to obtain the field equations of gravity . the equations of motion for _ matter _ remain invariant if one adds a constant , say , @xmath268 to the matter lagrangian . however , gravity breaks this symmetry which the matter sector has and @xmath17 appears as a cosmological constant term in the field equations of gravity . if we interpret the evidence for dark energy in the universe ( see ref . @xcite ; for a critical look at data , see ref . @xcite and references therein ) as due to the cosmological constant , then its value has to be fine - tuned to satisfy the observational constraints . it is not clear why a particular parameter in the low energy sector has to be fine - tuned in such a manner . in the alternative perspective described here , the functional in is clearly invariant under the shift @xmath269 or equivalently , @xmath270 , since it only introduces a term @xmath271 for any null vector @xmath222 . in other words , one _ can not _ introduce the cosmological constant as a low energy parameter in the action in this approach . we saw , however , that the cosmological constant can reappear as an _ an integration constant _ when the equations are solved . the integration constants which appear in a particular solution have a completely different conceptual status compared to the parameters which appear in the action describing the theory . it is much less troublesome to choose a fine - tuned value for a particular integration constant in the theory if observations require us to do so . from this point of view , the cosmological constant problem is considerably less severe when we view gravity from the alternative perspective . this extra symmetry under the shift @xmath270 arises because we are not treating metric as a dynamical variable in an action principle . in a universal manner , in which case , one will have the graviton coupling to the cosmological constant . in our approach , the linearized field equations for the spin-2 graviton field @xmath272 , in a suitable gauge , will be @xmath273 for all null vectors this equation is still invariant under @xmath270 showing that the graviton does _ not _ couple to cosmological constant . ] in fact one can state a stronger result @xcite . consider any model of gravity satisfying the following three conditions : ( 1 ) the metric is varied in a local action to obtain the equations of motion . ( 2 ) we demand full general covariance of the equations of motion . ( 3 ) the equations of motion for matter sector are invariant under the addition of a constant to the matter lagrangian . then , we can prove a ` no - go ' theorem that the cosmological constant problem can not be solved in such model . that is , we can not solve cosmological constant problem unless we drop one of these three demands . of these , we do not want to sacrifice general covariance encoded in ( 2 ) ; neither do we have a handle on low energy matter lagrangian so we can not avoid ( 3 ) . so the only hope we have is to introduce an approach in which gravitational field equations are obtained from varying some degrees of freedom other than @xmath21 in a maximization principle . this suggests that the so called cosmological constant problem has its roots in our misunderstanding of the nature of gravity . our approach at present is not yet developed far enough to predict the value of the cosmological constant .but providing a mechanism in which the _ bulk cosmological constant decouples from gravity _ is a major step forward . it was always thought that some unknown symmetry should make the cosmological constant ( almost ) vanish and weak ( quantum gravitational ) effects which break this symmetry could lead to its small value . _ our approach provides a model which has such symmetry . _ the small value of the observed cosmological constant has to arise from non - perturbative quantum gravitational effects at the next order , for which we do not yet have a fully satisfactory model . ( see , however , ref . it is useful to distinguish clearly ( i ) the mathematical results which can be rigorously proved from ( ii ) interpretational ideas which might evolve when our understanding of these issues deepen . * consider a functional of null vector fields @xmath274 in an arbitrary spacetime given by [ or , more generally , by ] . demanding that this functional is an extremum for all null vectors @xmath15 leads to the field equations for the background geometry given by @xmath275 where @xmath33 is given by [ or , more generally , by ] . thus field equations in a wide class of theories of gravity can be obtained from an extremum principle without varying the metric as a dynamical variable . * these field equations are invariant under the transformation @xmath270 , which relates to the freedom of introducing a cosmological constant as an integration constant in the theory . further , this symmetry forbids the inclusion of a cosmological constant term in the variational principle by hand as a low energy parameter . _ that is , we have found a symmetry which makes the bulk cosmological constant decouple from the gravity . when linearized around flat spacetime , the graviton inherits this symmetry and does not couple to the cosmological constant . * on - shell , the functional in [ or , more generally , by ] contributes only on the boundary of the region . when the boundary is a horizon , this terms gives precisely the wald entropy of the theory . it is remarkable that one can derive not only einstein s theory uniquely in @xmath13 but even lanczos - lovelock theory in @xmath20 from an extremum principle involving the null normals _ without varying @xmath21 in an action functional!_. \(ii ) as regards interpretation of these results , we see from that in the case of einstein s theory , we have a lagrangian @xmath276})n^b$ ] for a vector field @xmath15 ( except for a surface term ) that becomes vacuous in flat spacetime in which covariant derivatives become partial derivatives . there is no dynamics in @xmath15 ( in the usual sense ) but they do play a crucial role . this raises the question as to the physical meaning of these null vectors and the interpretation of our approach . as described above , this interpretation is essentially thermodynamical and the physical picture is made of the following qualitative ingredients : * assume that the spacetime is endowed with certain microscopic degrees of freedom capable of exhibiting thermal phenomena . this is just the boltzmann paradigm : _ if one can heat it , it must have microstructure ! _ ; and one can heat up a spacetime . * whenever a class of observers perceive a horizon , they are `` heating up the spacetime '' and the degrees of freedom close to a horizon participate in a very _ observer dependent _ thermodynamics . matter which flows close to the horizon ( say , within a few planck lengths of the horizon ) transfers energy to these microscopic , near - horizon , degrees of freedom _ as far as the observer who sees the horizon is concerned_. just as entropy of a normal system at temperature @xmath277 will change by @xmath278 when we transfer to it an energy @xmath11 , here also an entropy change will occur . ( a freely falling observer in the same neighbourhood , of course , will deny all these ! ) * we proved that when the field equations of gravity hold , one can interpret this entropy change in a purely geometrical manner involving the noether current . from this point of view , the normals @xmath15 to local patches of null surfaces are related to the ( unknown ) degrees of freedom that can participate in the thermal phenomena involving the horizon . * just as demanding the validity of special relativistic laws with respect to all freely falling observers leads to the kinematics of gravity , demanding the local entropy balance in terms of the thermodynamic variables as perceived by local rindler observers leads to the field equations of gravity in the form @xmath275 . as stressed in earlier sections , this involves a new layer of observer dependent thermodynamics . at a conceptual level , this may be welcome when we note that every key progress in physics involved realizing that something we thought as absolute is not absolute . with special relativity it was the flow of time and with general relativity it was the concept of global inertial frames and when we brought in quantum fields in curved spacetime it was the notion of particles and temperature . we now know that the temperature attributed to even vacuum state depends on the observer . it seems necessary to integrate the entire thermodynamic machinery ( involving what we usually consider to be the ` real ' temperature ) with this notion of lrfs having their own ( observer dependent ) temperature . there is scope for further work in this direction . t. padmanabhan , _ a dialogue on the nature of gravity _ [ arxiv:0910.0839 ] ; _ gen . _ , * 40 * , 2031 - 2036 ( 2008 ) ; _ _ phys . report__s , * 406 * , 49 ( 2005 ) [ gr - qc/0311036 ] ; aip conference proceedings , * 989 * 114 ( 2007 ) [ arxiv:0706.1654 ] . for a small sample , see e.g. , d. kothawala et al . , _ phys . * b 652 * , 338 ( 2007 ) [ arxiv : gr - qc/0701002 ] ; a. paranjape et al . , _ phys . _ * d 74 * , 104015 ( 2006 ) [ arxiv : hep - th/0607240 ] ; dawood kothawala , t. padmanabhan , _ phys . _ , * d 79 * , 104020 ( 2009)[arxiv:0904.0215 ] ; r. g. cai , et al . , _ phys . * d 78 * , 124012 ( 2008 ) [ arxiv:0810.2610 ] ; _ phys . * d 75 * , 084003 ( 2007 ) [ arxiv : hep - th/0609128 ] ; m. akbar and r. g. cai , _ phys . _ , * b 635 * , 7 ( 2006 ) [ arxiv : hep - th/0602156 ] ; _ phys . lett . _ * b 648 * , 243 ( 2007 ) [ arxiv : gr - qc/0612089 ] ; y. gong and a. wang , phys . rev . lett . * 99 * , 211301 ( 2007 ) [ arxiv:0704.0793 [ hep - th ] ] ; s. f. wu , g. h. yang and p. m. zhang , [ arxiv:0710.5394 ] ; s. f. wu , b. wang and g. h. yang , nucl . phys . b * 799 * , 330 ( 2008 ) [ arxiv:0711.1209 ] ; s. f. wu , b. wang , g. h. yang and p. m. zhang , [ arxiv:0801.2688 ] . t. padmanabhan , _ gen.rel.grav . , _ * 34 * 2029 ( 2002 ) [ gr - qc/0205090 ] ; _ brazilian jour.phys._ ( special issue ) * 35 * , 362 ( 2005 ) [ gr - qc/0412068 ] ; _ mod.phys.letts . _ * a 17 * , 1147 ( 2002 ) [ hep - th/0205278 ] ; aip conference proceedings * 861 * , 858 , [ astro - ph/0603114 ] . nathalie deruelle et al . , _ class . _ * 21 * , ( 2004 ) , 1971 , [ gr - qc/0310098 ] ; g.l . cardoso et al . , [ arxiv : hep - th/9904005v2 ] ; t. padmanabhan , _ gravitation : foundations and frontiers _ , ( cambridge university press , 2010 ) , in press . h. s. snyder _ phys . _ , * 71 * , 38 ( 1947 ) ; b. s. dewitt , _ phys . rev . _ , * 13 * , 114 ( 1964 ) ; t. yoneya _ prog . _ , * 56 * , 1310 ( 1976 ) ; t. padmanabhan _ ann ( n.y . ) , * 165 * , 38 ( 1985 ) ; _ class . quantum grav . _ * 4 * , l107 ( 1987 ) ; t. padmanabhan _ phys . lett . _ * 78 * , 1854 ( 1997 ) [ hep - th/9608182 ] ; _ phys . rev . _ * d 57 * , 6206 ( 1998 ) ; k.srinivasan et al . * d 58 * 044009 ( 1998 ) [ gr - qc/9710104 ] ; x.calmet et al.,phys.rev.lett.*93*:211101,(2004 ) ; _ int.j.mod.phys . _ * d14 * ( 2005 ) 2195 ; m. fontanini _ phys.lett . _ * b 633 * , 627 ( 2006 ) [ hep - th/0509090 ] . for a review , garay , _ int . phys . _ * a10 * , 145 ( 1995 ) . perlmutter et al . , _ astrophys . j. _ * 517*,565 ( 1999 ) ; a.g . reiss et al . , _ astron . j. _ ( 1998 ) , * 116*,1009 ; j. l. tonry et al . , _ ap . j _ , ( 2003 ) , * 594 * , 1 ; b. j. barris , _ astrophys.j . _ , * 602 * ( 2004 ) , 571 ; a. g.reiss et al . , _ astrophys.j . _ * 607 * , ( 2004 ) , 665 . h.k.jassal et al . , _ phys.rev . _ * d 72 * , 103503 ( 2005 ) [ astro - ph/0506748 ] ; [ astro - ph/0601389 ] ; t. padmanabhan , t . roy choudhury , _ mnras _ * 344 * , 823 ( 2003 ) [ astro - ph/0212573 ] ; t. roy choudhury , t. padmanabhan , _ astron.astrophys . _ * 429 * , 807 ( 2005 ) , [ astro - ph/0311622 ] ; s. nesseris , l.perivolaropoulos , _ jcap _ * 0702*,025 ( 2007 ) ; y. wang , p. mukherjee , _ phys . * d 76 * , 103533 ( 2007 ) . t. padmanabhan _ class.quan.grav._ , * 22 * , l107 , ( 2005 ) [ hep - th/0406060 ] . for earlier attempts in similar spirit , see t. padmanabhan , _ class.quan.grav . _ * 19 * , l167 ( 2002 ) , [ gr - qc/0204020 ] ; d. sorkin , _ int.j.theor.phys . _ * 36 * , 2759 ( 1997 ) ; for related work , see volovik , g. e. , [ gr - qc/0405012 ] ; j. v. lindesay et al . , [ astro - ph/0412477 ] ; y. s. myung , [ hep - th/0412224 ] ; j.d.barrow , [ gr - qc/0612128 ] ; e.elizalde et al . , [ hep - th/0502082 ] .

we present _ chandra _ observations of a small sample of nearby classical double radio galaxies which have more than one radio hotspot in at least one of their lobes . the x - ray emission from the hotspots of these comparatively low - power objects is expected to be synchrotron in origin , and therefore to provide information about the locations of high - energy particle acceleration . in some models of the relationship between the jet and hotspot the hotspots that are not the current jet termination point should be detached from the energy supply from the active nucleus and therefore not capable of accelerating particles to high energies . we find that in fact some secondary hotspots are x - ray sources , and thus probably locations for high - energy particle acceleration after the initial jet termination shock . in detail , though , we show that the spatial structures seen in x - ray are not consistent with nave expectations from a simple shock model : the current locations of the acceleration of the highest - energy observable particles in powerful radio galaxies need not be coincident with the peaks of radio or even optical emission .

0707.2865

in the standard model for powerful extragalactic double radio sources ( also known as classical double or frii [ @xcite ] sources ) , hotspots , the bright compact regions at the ends of the source , are the visible manifestation of a strong shock as the relativistic beam of energetic particles is suddenly decelerated by interaction with the slow - moving or stationary plasma within the radio lobes ( e.g. , * ? ? ? the particle acceleration at these shocks determines the energy distribution of the electrons ( and , possibly , protons ) that go on to form the large - scale lobes and expand into the external medium , and so an understanding of how and where it happens is essential to an understanding of the dynamics and environmental impact of radio sources ; in addition , the strong shocks in friis are often invoked as a possible region of acceleration for the high - energy cosmic ray population , so that it is important to understand where ( and if ) high - energy particles are accelerated in these systems . the best evidence for this shock model comes from the radio through optical spectra of hotspots , which have been shown ( e.g. , * ? ? ? * ) to be commonly consistent with the predictions of a simple model for shock particle acceleration and downstream losses @xcite . however , the idea that the hotspots always trace the shock at the jet termination is challenged by the observation that the lobes of radio galaxies and quasars very frequently have more than one compact bright radio feature that meets whatever definition of a hotspot is in use ( e.g. , * ? ? ? where these appear in the jet or embedded deep in the lobe they are usually interpreted as ` jet knots ' the assumption encoded in this terminology is that they are telling us about internal dissipation in the jet rather than disruption and that they are not relevant to the particle acceleration history of the source . but in many systems there are multiple hotspots at the far end of the lobe , and in these the configuration of the hotspots relative to the jet flow often suggests that more than one is associated with the beam termination . models to explain this observation include those in which the beam end - point moves about from place to place in the lobe ( the ` dentist s drill ' model of @xcite ) or in which material flows out from the initial impact point of the beam to impact elsewhere on the lobe edge ( the ` splatter - spot ' model of @xcite or the jet - deflection model of @xcite ) . all these models predict that one of the hotspots , the one associated with the first or current termination of the jet , should be more compact than the other or others ; it is in fact observed in the radio that where jets are explicitly seen to terminate , they always do so in the most compact , ` primary ' hotspot ( e.g. , * ? ? ? * ; * ? ? ? but it is very difficult to distinguish between different models of multiple hot spot formation using radio data alone . the secondary ( less compact ) hotspots exhibit a wide variety of structures and relationships to the primary hotspots , and , while in some cases radio structure seems to favor one model rather than another , the nature of the secondaries is never unambiguously constrained by single - frequency radio data . one area in which different models _ do _ make different predictions is that of the high - energy particle acceleration in the secondary ( less compact ) hotspots . if these are relics left behind by the motion of the jet , as in the ` dentist s drill ' model , then in general we expect shock - driven particle acceleration to have ceased ( though the secondary hotspot may continue to be fed for some time if the jet is disconnected some way upstream : @xcite ) . synchrotron losses will then deplete the high - energy electrons in the secondary hotspot . if secondary hotspots continue to be fed by outflow from the primary hotspot , then there is still an energy supply and particle acceleration will continue to operate . high - resolution x - ray observations of synchrotron radiation with _ chandra _ have provided key insights into the particle acceleration in the jets of low - power , fri - type radio galaxies , allowing us to locate the sites of particle acceleration and relate them to the dynamics of individual jets ( e.g. * ? ? ? x - ray observations are vital in these cases because the synchrotron loss timescale for x - ray - emitting electrons is tens of years , assuming field strengths close to the equipartition values : thus , unlike radio and even optical observations of synchrotron radiation , x - ray synchrotron detections tell us where particle acceleration is happening _ now _ , rather than where it has happened in the past . to date , however , it has been difficult to use x - ray observations to study particle acceleration in the hotspots of the more powerful frii radio sources , because of the importance in hotspots of a second emission process , inverse - compton emission . this process , particularly important in bright , compact hotspots , traces the low - energy electrons rather than the high - energy ones , and has been the subject of much work with _ chandra _ because of its potential to measure physical conditions ( magnetic field strengths and energy densities ) in hotspots ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? some hotspots have been known since the _ rosat _ epoch to be best described by a synchrotron rather than inverse - compton model ( e.g. * ? ? ? * ) but until recently it has not been clear what controls the relative dominance of the two processes . based on new and archival observations of a large sample of friis , we recently showed @xcite that the controlling parameter is related to the overall _ luminosity _ of the hotspot : high - luminosity hotspots never show x - ray synchrotron emission , while low - luminosity hotspots often do . we argued that this is due to the higher magnetic field strengths and photon energy densities found in the more luminous hotspots : these increase the energy loss rate for high - energy electrons and prevent efficient particle acceleration to the energies needed for x - ray synchrotron radiation . in contrast , low - luminosity hotspots can readily accelerate particles to x - ray emitting energies , and the expected inverse - compton emission is negligible , so that synchrotron radiation is dominant in these systems . it is important to note that this picture , while qualitatively plausible , relies on details of the microphysics , such as the magnetic field configuration and electron diffusion coefficient in the acceleration region ( see , e.g. , * ? ? ? * ) and so it can not yet be shown quantitatively to be correct . if this model is accepted , though , x - ray synchrotron emission can be used to probe high - energy particle acceleration in low - luminosity frii radio galaxies . an example of this is provided by observations of the low - power frii 3c403 @xcite . these observations were important for two reasons . firstly , as 3c403 is a narrow - line radio galaxy , any relativistic beaming effects ( of the kind thought to be important in some core - dominated quasars ) must be minimal if unified models are correct , as 3c403 should lie close to the plane of the sky : thus , models for anomalous x - ray emission involving beaming , like those of @xcite , need not be considered . secondly , 3c403 s e lobe is a multiple - hotspot system , and the observations show a clear difference between features of the jet and the primary hotspot , on the one hand , and the secondary hotspot ( much brighter in the radio ) , on the other : we found that the upper limit on the x - ray emission from the secondary was an order of magnitude below what would have been detected if its x - ray to radio ratio had been the same as that in the primary . this strongly suggests that , in this source at least , the secondary hotspot is unable to accelerate particles to the highest observable energies . the results of our work on 3c403 motivated us to carry out further observations of nearby sources with multiple hotspots , and to examine data available in the _ chandra _ archive , with the aim of seeing whether , and in what circumstances , the different hotspot components can give rise to high - energy particle acceleration , and so constraining the nature of multiple hotspots . in this paper we report on our results . throughout the paper we use a cosmology with @xmath0 km s@xmath1 mpc@xmath1 , @xmath2 and @xmath3 . the spectral index @xmath4 is defined in the sense that flux density @xmath5 : the relationship between the spectral index @xmath4 and the photon index @xmath6 is thus @xmath7 . we selected our targets from the sample of leahy et al . ( 1997 : hereafter l97 ) . this sample consists of frii radio galaxies with @xmath8 taken from the 3cr sample ( @xcite : see l97 for details of the selection ) . we chose this parent sample because of its low redshift ( and therefore , in general , low luminosity , implying negligible x - ray inverse - compton radiation from the hotspots in the picture of @xcite ) and because of the availability of excellent radio data , with resolution matched to or exceeding that of _ chandra _ , for almost all members of the sample . from the l97 sample , we selected the sources with clear , well separated , bright multiple hotspots as seen in the radio maps . two of these , 3c390.3 and 3c403 , had already been observed with _ ( as discussed in @xcite and @xcite respectively ) . we were awarded time for two more , 3c227 and 3c327 , giving us two broad - line and two narrow - line radio galaxies in total . basic properties of the sample objects are given in table [ sample ] . we discuss the radio and x - ray data for these objects in the following two sections . .properties of the sample of frii sources [ cols= " < , > , < , > , > , > , > " , ] as detailed comparison between the radio and x - ray spatial structures of the hotspots is important to us , we aligned the radio and x - ray frames for all the sources by shifting the x - ray nuclear position to match the best available radio position . where there were significant offsets between the radio core positions at different frequencies , we shifted the low - frequency positions to match the high - frequency ones . we expect the relative astrometry of the radio and x - ray frames to be limited by the accuracy of x - ray centroiding , but in all cases it should be better than @xmath9 . 3c390.3 is a special case , as extreme pileup has removed all counts from the center of the nuclear x - ray emission : we still believe that we have been able to determine the x - ray position to the required level of accuracy , and in this case the default astrometry of the _ chandra _ data appears to be correct . for radio data , resolutions are quoted as the major @xmath10 minor axis ( fwhm ) of the elliptical restoring gaussian : where only one dimension is quoted the restoring beam used was constrained to be circular . in this subsection we briefly comment on the features of the new _ chandra _ observations of 3c227 and 3c327 unrelated to the hotspots . the _ chandra _ observations of the other sample sources have been discussed elsewhere ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? images showing the large - scale x - ray emission from the two objects are shown in figs [ 227-l ] and [ 327-l ] . the raw _ chandra _ count rate from the core of the blrg 3c227 is roughly 0.14 s@xmath1 : with a frame time of 3.1 s , this means that we might expect it to be affected by pileup at a significant level . fitting to the total spectrum of the nucleus within 15 _ chandra _ pixels ( 7 arcsec ) , using a concentric adjacent background annulus , we found a good fit ( @xmath11 for 213 d.o.f . ) to a model consisting of two power laws , one with galactic absorption only and one with additional intrinsic absorption at the redshift of the source . the power - law component without intrinsic absorption had a steep best - fitting photon index ( @xmath12 ) while the absorbed component , with @xmath13 @xmath14 , had a very flat photon index , @xmath15 . the data are almost equally well fitted ( @xmath16 ) with an ionized absorber model ( absori ) but this has clear residuals at soft energies , suggesting that in any event some of the soft x - ray emission must come from a second nuclear x - ray component , or from soft thermal x - ray emission close to the nucleus . this is unsurprising as we know that in general radio - loud agn have a soft x - ray component which is probably related to the jet ( e.g. , * ? ? ? the significant intrinsic absorption seen here , although unusual for a broad - line object , is consistent with what is seen in some other blrg or quasars ( e.g. 3c109 , @xcite , @xcite ; 3c351 , @xcite , @xcite ) . from optical spectropolarimetry @xcite estimate that the optical broad lines and continuum may be obscured by 1 - 2 magnitudes in the @xmath17 band . although our column density would imply substantially higher extinction , @xmath18 for galactic gas / dust ratios , the optical results , including the detection of polarized broad emission lines , give us some additional reason for believing that there might be intrinsic x - ray absorption in this source . the very flat photon index of the absorbed component is likely to be partly the result of pileup , and to investigate this we also extracted a spectrum for an annular region between 2 and 15 _ chandra _ pixels , with the same background region . this excludes the region where pileup is likely to be significant . we applied energy - dependent corrections to the arf using the algorithm of the arfcorr softwaretsujimot / arfcorr.html ] implemented using the funtools package . fitting the same double power - law spectrum , we find similar parameters the slightly larger intrinsic absorbing column ( @xmath19 @xmath14 ) is consistent within the errors and only a slightly steeper photon index for the absorbed component , @xmath20 . however , the 99% confidence upper limit on the photon index in this extraction region ( which obviously contains only a small fraction of the total counts ) is 2.4 , so more typical photon indices are certainly not excluded by the non - piled up data . [ 227-core ] @xcite , who discuss observations of 3c227 with the _ rosat _ pspc , found that the nuclear spectrum was well fitted by a power - law model with @xmath21 . with the limited signal to noise of their _ rosat _ data and the restricted energy range of the pspc it is not clear that they could have identified the absorption features seen in our spectrum . for 3c327 , which has a much fainter x - ray nucleus , we extracted a spectrum from a source circle of radius 4 _ chandra _ pixels ( 2 arcsec ) with an adjacent concentric background annulus . the best - fitting spectrum ( @xmath22 for 21 d.o.f . ) again requires multiple components : at soft energies the source is dominated by an unabsorbed power law with @xmath23 , while the residuals at high energy require an additional heavily absorbed power law , with the poorly constrained @xmath6 fixed at 1.7 ( following * ? ? ? * ) and @xmath24 @xmath14 , together with a strong line - like feature modeled as a gaussian with @xmath25 kev and @xmath26 kev . except for the steep power law index for the unabsorbed component , this is a typical spectrum for a narrow - line radio galaxy : iron features around 6.4 kev are often found in these objects ( @xcite ; @xcite ; @xcite ) . the lobes of 3c227 are both clearly detected in the _ chandra _ observations ( fig . [ 227-l ] ) . extended x - ray emission from lobes of radio sources is generally attributed to inverse - compton scattering from the microwave background radiation ( cmb ) @xcite . we extracted a spectrum for the w lobe , as the e lobe spans a chip boundary , using a polygonal extraction region defined by the radio emission but excluding the central part of the lobe where the nucleus and central extended x - ray emission might contribute , as well as compact sources , including the hotspots . the spectrum was well fitted ( @xmath27 for 10 d.o.f . ) with a power - law model with @xmath28 and 1-kev unabsorbed flux density of @xmath29 njy . we used the code of @xcite to calculate the predicted inverse - compton emission from the whole lobe , modeling the lobe crudely as a uniform cylinder with length @xmath30 and radius @xmath31 , assuming equipartition , and using the same electron spectral assumptions as @xcite , with a low - energy electron energy index ( ` injection index ' ) , @xmath32 , of 2 . the equipartition magnetic field strength is 0.58 nt ; the predicted emission corresponds to a 1-kev flux density of 1.4 njy , i.e. about a factor 3.5 below what is observed . reducing the field strength to 0.29 nt is required to produce all the observed x - ray emission by the inverse - compton process . for 3c327 there is a clear detection of the e lobe and a weaker but still significant detection of the w lobe . again , we extracted a spectrum for the lobe that does not span a chip boundary , in this case the e lobe , excluding point sources and the components near the hotspots . this was well fitted ( @xmath33 for 6 d.o.f . ) with a power - law model with @xmath34 and unabsorbed 1-kev flux density of @xmath35 njy . the inverse - compton prediction here on the same assumptions ( modeling the lobe as a cylinder with length @xmath36 and radius @xmath37 ) is 1.5 njy , so again the observations exceed the equipartition prediction by a factor @xmath38 . the equipartition field here is 0.73 nt and the field required to produce the observed x - rays by inverse - compton processes is 0.34 nt . both objects are thus consistent with the trend seen in many other sources for the observed inverse - compton emission to lie somewhat above the equipartition , @xmath39 prediction @xcite . we know that 3c227 s lobes are likely to make a relatively small angle to the line of sight , but this makes only a small difference to the inverse - compton prediction for @xmath40 . in this section we discuss the x - ray , radio and ( where possible ) optical properties of the hotspots in our target systems . table [ hstable ] gives a summary of the properties discussed below for ease of reference . llllrrrrlll [ hstable ] 3c227&w&&p1/2&@xmath41&@xmath42&@xmath43&@xmath44&y&primary&resolved , flattened + & & &p3&@xmath45&@xmath46&1.63&@xmath47&y&secondary&resolved + & & &p4&@xmath48&@xmath49&1.63&@xmath50&n&jet knot?&slightly resolved + & e&&f1&@xmath51&@xmath52&1.63&@xmath47&y&primary&compact + diffuse + 3c327&e&sx1&s1?&@xmath53&@xmath54&1.7&@xmath55&?&primary&compact + & & sx2&s2?&@xmath41&@xmath54&1.7&@xmath55&?&secondary&compact + 3c390.3&n&nx1&b&@xmath56&@xmath57&@xmath58&@xmath59&y&primary&2 sub - components + & & nx2&&&@xmath60&@xmath61&@xmath62&y&?&point - like + & & &f&&@xmath63&1.95&@xmath64&n&secondary&no x - ray detection + & s&sx1&g@xmath65&@xmath66&@xmath67&1.4&@xmath50&n?&primary?&compact + & s&diffuse&g , e , d?&&@xmath68&@xmath69&@xmath70&y?&secondary&diffuse + 3c403&e&&f1&&@xmath71&@xmath72&@xmath73&y&primary&resolved + & & &f1b&&@xmath74&2.0&@xmath75&n&jet knot?&compact + & & &f2&&@xmath76&2.0&@xmath77&n&secondary&no x - ray detection + & & &f6&&@xmath78&@xmath79&@xmath80&y&jet knot&elongated + optical counterparts of 3c227 s hotspots have recently been discovered using ground - based observations @xcite . the two components of the double w hotspot and the more compact e hotspot all emit in the near - infrared . in the x - ray , similarly , emission is detected from both hotspot regions , as shown in figs [ 227-hswl ] and [ 227-hsel ] . in the w hotspot x - ray emission is clearly detected from the three compact components p1/2 , p3 and p4 , while there is weak emission not just from the compact e component f1 but also from the region around it . in the standard multiple hotspot picture p4 would probably be designated a jet knot ( cf knot f6 of 3c403 , below ) , p1/2 would be the primary hotspot and p3 a secondary hotspot . thus 3c227 is an example of a system where the secondary hotspot clearly is an x - ray source . what is most striking in the w hotspot region is the offsets between the peak of the radio and the peak of the x - ray in the x - ray detected components . this is most clearly seen in fig . [ 227-hswx ] , where the x - ray emission is compared with the high - resolution radio data used by @xcite . at this resolution the brightest hotspot is resolved into two components ( p1 and p2 ) , and both of these are associated with some x - ray emission , but the brightest x - ray emission comes from a resolved structure which is comparable in size to p2 ( with a long axis of @xmath81 or 5 kpc ) and the x - ray centroid of this structure is displaced @xmath41 ( 1.3 kpc ) in the direction of the nucleus from the peak of p1 . the centroid of the x - ray emission from p3 is displaced by @xmath82 ( 3.6 kpc ) from the peak of the radio emisission in the direction of p1/2 . finally , the x - ray emission from p4 appears slightly displaced ( @xmath83 ) from the peak of the radio in the direction of p1/2 , but the statistics are not good enough to claim a definite offset . in contrast , the centroid of the emission from f1 and its surroundings is consistent with the position of the radio peak within the large errors . the x - ray counterpart of p1/2 contained enough counts ( @xmath84 in the 0.55.0 kev energy range ) for us to extract a spectrum , which we took from a circular region with radius 6 _ chandra _ pixels ( 3 arcsec ) with background from a concentric annular region . we find a good fit ( @xmath85 for 3 d.o.f . ) with a power - law model with @xmath86 and 1-kev unabsorbed flux density @xmath44 njy . the fainter hotspots p3 and p4 contain @xmath46 and @xmath87 net 0.55.0 kev counts respectively , which for the same spectrum would correspond to flux densities of @xmath47 and @xmath88 njy . the counterpart of the f1 hotspot , which lies on the s1 chip , contains @xmath52 counts in a similarly sized extraction region , which corresponds to @xmath89 njy if again we assume the same spectrum as seen in p1/2 . mack kindly provided us with electronic versions of ground - based near - infrared ( nir ) images ( @xcite , and in preparation ) . when shifted to our radio - based co - ordinate system by alignment at the peak of the optical emission from the galaxy these show that the optical emission from the p1/2 region extends over both the radio and x - ray peaks . in detail , though , the brightest regions of nir emission are not coincident with either radio or x - ray peaks . the nir emission disappears in between the peak of the x - ray and the peak of the radio emission at p2 . similarly , the peak of the nir emission coincident with p3 agrees neither with the radio nor the x - ray peak positions , though it is roughly coincident with the x - ray centroid . there is no nir counterpart of the radio / x - ray knot p4 . 3c327 is the only object in our sample to have no optical hotspot detection : as pointed out by @xcite , a nearby bright disk galaxy makes it hard to detect potentially faint counterparts to the bright double eastern radio hotspots . in the new _ chandra _ data , although there is nearby x - ray emission ( fig . [ 327-l ] ) there is no detection of a compact component corresponding to the w hotspot . there is , however , a clear detection of compact components near the double e hotspots ( fig . [ 327-hsex ] ) . neither of the two x - ray components detected bears a very obvious relation to the radio structure . the component ( denoted sx1 ) closest to the primary hotspot ( s1 ) is separated from it by @xmath90 ( 4.5 kpc ) along the line connecting s1 and the core , while the other x - ray component ( sx2 ) is at @xmath41 ( 1.5 kpc ) from the nearest peak of the radio emission in the s2 region , and considerably further from its center , in the sense of being further away from the primary hotspot than the peak of the radio emission . in dss2 and 2mass images sx2 is close to the nearby disk galaxy , but not at its nucleus . we can not rule out the possibility that sx2 is associated with this galaxy rather than with the 3c327 hotspot system . sx1 is further away from the galaxy than sx2 . the two compact x - ray components each contain @xmath54 net 0.55.0 kev counts so that spectral fitting is not possible . if we assume a power - law spectrum with @xmath91 , the observed net counts correspond to flux densities of @xmath55 njy in each component . we estimated the hardness ratios of the two components ( defining the soft band as 0.5 - 2.0 kev and the hard band as 2.0 - 5.0 kev ) and compared them to the hardness ratio expected for a @xmath92 power law with galactic absorption ( this model was chosen because the spectral index is unlikely to be flatter than @xmath92 ) . we found that sx2 is significantly harder ( at the 99% confidence level on a binomial test ) than would be expected for such a model , while sx1 is not inconsistent with it . this suggests either that sx2 is not related to the hotspot ( e.g. that it is a background type 2 agn ) or that there is some additional source of absorption , conceivably in the disk galaxy , that affects sx2 but not sx1 . we can not distinguish between these models with the available data . the northern compact hotspot of 3c390 is a well - known x - ray source @xcite . diffuse x - ray emission from the large hotspot region of the s lobe was discussed by @xcite . a detailed examination of the _ chandra _ data shows that several components of the n hotspot complex are x - ray sources ( fig . [ 390.3hsn ] ) . the radio jet , which is detectable throughout the n lobe @xcite brightens strongly in the area shown by our image , and there is faint but clear x - ray emission associated with this section of the jet . the strongest x - ray ` source ' is the known x - ray counterpart of the primary hotspot @xcite , knot b , here denoted nx1 . the centroid of this region is only slightly ( @xmath56 , 0.4 kpc ) offset to the s of the peak of the radio emission , but the resolution of chandra shows that the x - ray feature is actually resolved into two components , separated by 1.5 arcsec ( 1.6 kpc ) and placed symmetrically on either side of the radio peak . this is a much smaller distance than that of the nearby galaxy , possibly interacting with the jet , described by @xcite , which is @xmath93 to the ne . at higher radio resolutions ( e.g. , in figure 5 of @xcite ) knot b is resolved into a linear structure which lies between the two x - ray peaks there is no sign of any double structure in the radio . nor is there any evidence in the optical image of @xcite or in the archival _ data for resolution of the hotspot in the e - w direction : @xcite show that the optical hotspot is in fact extended in the same direction as the radio . another bright x - ray source in the hotspot region is the object we denote nx2 . this has no radio counterpart and is well separated from any compact structure in the radio , so it seems most likely to be a chance superposition with a background object , although it is detected in the optical and infrared and its properties at these wavelengths are consistent with it being similar to knot b ( its infrared colors mean that it is certainly not a background normal galaxy ) . radial profile analysis shows that it is consistent with being a point source . finally , the secondary hotspot ( knots f and a ) shows no significant x - ray emission , even though in 3c390.3 there is clear evidence ( in the form of the continued collimated outflow from knot b , features n14 in fig.[390.3hsn ] ) that this hotspot is connected to the jet . the radio / x - ray ratio in the secondary hotspot region is a factor @xmath94 greater than in knot b. there is no optical detection of the secondary @xcite , nor is it detected in the archival _ spitzer _ data . [ nx2 ] we extracted spectra for the jet , for the counterpart to knot b ( nx1 ) as a whole , for its two subcomponents ( denoted nx1e and nx1w ) and for the possibly unrelated source nx2 , taking background from a nearby blank - sky region . the overall spectrum of nx1 is not particularly well fitted with a single power - law model ( @xmath95 ; @xmath96 ) . the e component of the x - ray source is well fitted with a power - law model ( @xmath97 , @xmath98 ) but the w component is not ( @xmath99 , @xmath100 ) , the poor fit being the result of one high bin at soft energies . neither component is acceptably fitted with a thermal model . within the limited statistics , it seems likely that the two components of knot b have significantly different x - ray spectra . the total 1-kev unabsorbed flux density of knot b in our extraction is @xmath59 njy , with the two sub - components being roughly equal in flux . the few counts in the jet are well fitted ( @xmath101 ) with a power - law model with @xmath102 , and a flux density @xmath103 njy . finally , nx2 is well fitted ( @xmath104 ) with a power - law model with @xmath105 comparable to the spectral indices of nx1 or its components and has a 1-kev flux density of @xmath106 njy . [ 390.3hss ] shows the x - ray emission from the southern hotspot . the brighest compact feature , denoted sx1 in the figure , is not coincident with any named radio feature , but appears at the n end of the bright ` rim ' of the hotspot discussed by @xcite . the nearest discrete compact feature , @xmath66 further around the rim , is a weak radio knot ( which we denote g@xmath65 ) visible in the high - resolution radio map of @xcite . this feature may be considered the primary hotspot or may just be a jet knot . otherwise there is no particular association between the detected x - ray emission and the radio : none of the bright knots g , e , d clearly has a compact associated x - ray source , though all have x - ray emission coincident with them , and the x - rays from the ` tail ' region bear no relation to the radio structure seen here . archival _ spitzer _ images show that the s hotspot has a 24-@xmath107 m infrared counterpart , which appears to be peaked where the radio is brightest ( i.e. in hotspots e and d ) , but the low resolution of _ spitzer _ at this wavelength prevents us from examining this in detail . taking a single spectrum of all the detectable x - ray structure in fig.[390.3hss ] , we find it to be poorly fitted ( @xmath108 ) with a single power law with @xmath109 and a total 1-kev flux density of @xmath70 njy . the poor fit probably indicates that more than one spectral component is present , but there are too few counts to try to separate these spatially . using the parameters of this fit , the compact component sx1 would have a flux density @xmath110 njy . the hotspots of 3c403 were discussed extensively by @xcite , and so we do not present images or spectral fits here . to summarize the results that are most important for this paper , the hotspots f1 and f6 in the e lobe have clear x - ray detections , with measured photon indices @xmath111 . f6 may be a jet knot , consistent with the detection of jet - related x - ray emission in the same lobe , and if so f1 is the primary hotspot . the peaks of the x - ray emission from f1 and f6 are clearly coincident with the peaks in the radio , unlike several of the other hotspots considered in this paper , but both have x - ray extensions back towards the nucleus that are not present ( or at least not nearly as prominent ) in the radio . there is no detection of the radio - bright secondary hotspots f2 and f3 and there is no diffuse x - ray emission from the hotspot region . the observations of 3c227 , 327 and 390.3 almost all show differences between the structure observed in the radio and that observed in the x - ray . to summarize , the only ` simple ' hotspots seen in these three objects are the weak compact components in the e hotspot of 3c227 and ( probably ) the s hotspot of 3c390.3 , sx1 . 3c227 shows a striking 1.5-kpc offset between the radio emission from the primary w hotspot and its x - ray counterpart , in the sense that the x - ray emission is nearer the nucleus , while having similar structure in the radio and x - ray . the well - known counterpart to 3c390.3 s hotspot b turns out to have two components , neither of which is coincident with the radio or optical detection , and which are displaced from it in a direction perpendicular to the jet axis . neither of the two features most obviously related to the e hotspots of 3c327 is coincident with the peak of the radio : one is 4 kpc along the jet axis from the primary hotspot , the other at least 1 kpc away from the secondary along the primary - secondary axis . what physical processes can account for these diverse behaviors ? one possibility is of course that the compact components are nothing to do with the radio sources . this is our preferred explanation for the compact feature nx2 in 3c390.3 n ( [ nx2 ] ) : it may also explain one or , at a stretch , both of the features we see in 3c327 e , where we have no deep optical information to constrain whether a background object could be responsible . it seems very unlikely that it could account for the observations of 3c390.3 nx1 or 3c227 , where the x - ray features are clearly resolved . we therefore consider the possible emission processes that could be responsible for producing the observed x - rays from the sources themselves . as the available x - ray data suggest power - law spectra and thus non - thermal emission , we focus on inverse - compton and synchrotron processes . as discussed by @xcite in the context of the more distant double - hotspot source 3c351 ( where inverse - compton emission almost certainly plays some role in the complex x - ray structures seen ) , synchrotron self - compton emission ( ssc ) or inverse - compton scattering of the cmb ( cmbic ) can not produce offsets between the radio and x - ray unless there is very strong spatial variation in the positions of the low - energy electrons and/or important beaming effects . we first consider a model in which there are strong point - to - point variations in the number density of low - energy electrons . to produce offset x - ray emission via inverse - compton processes , we require a large population of low - energy ( @xmath112 ) electrons at the location of the x - ray emission , while either the electron spectrum or the magnetic field strength must be tuned so as to avoid significant emission from this population of electrons at radio frequencies . we used the code of @xcite to calculate the expected inverse - compton emission from components matching p1 , p2 and the x - ray source in 3c227 , using an upper limit on the radio flux density of the x - ray component of 2 mjy to normalize the radio spectrum and using the observed sizes in the radio and x - ray to choose component sizes . we modeled the three components for simplicity as uniformly filled ellipsoids in the plane of the sky . we found that for standard broken power - law spectra for the three components we require the x - ray source to have a departure from equipartition of a factor @xmath113 in magnetic field strength , and an electron energy density that is @xmath114 times greater than that in p2 , in order to produce the observed x - rays by inverse - compton processes ( where the upper limits come from the fact that we have no unambiguous detection of a radio counterpart of the x - ray source ) . in this situation we find that the dominant photon field is the cmb , and so the conclusions are robust against uncertainties in the geometry , which affect only the number density of synchrotron photons from p1/2 . such an electron distribution seems highly unlikely simply on energetic grounds , since it requires essentially all the energy in the hotspot to be concentrated in an offset , radio - invisible region . a similarly implausible electron distribution would be required for 3c327 and 3c390.3 . in addition , we show in table [ hsic ] that the parameters of all detected and non - detected hotspots predict inverse - compton flux densities ( derived using the code of @xcite ) for equipartition magnetic fields that are much less than the observed values or limits , as was shown previously in some cases by @xcite . we therefore rule out simple inverse - compton models in what follows . lllrrrrrrl [ hsic ] 3c227&w&p1&@xmath115&8.35&16&@xmath116&0.0010&0.09&offset x - ray + & & p2&@xmath117&8.35&25&@xmath116&0.0019&0.06&offset x - ray + & & p3&@xmath118&8.35&13&@xmath47&0.0014&0.02&offset x - ray + & & p4&@xmath119&8.35&1.2&@xmath50&@xmath120&0.3&jet knot ? + & e&f1&@xmath121&8.35&5.3&@xmath47&@xmath122&0.06 + 3c327&e&s1&@xmath123&8.35&15&@xmath55&@xmath124&0.02&offset x - ray + & & s2&0.32&8.35&3.2&@xmath55&@xmath125&0.09&offset x - ray + 3c390.3&n&b&@xmath126&4.99&66&@xmath59&0.003&0.08&2 sub - components + & & f&3.7&4.99&190&@xmath64&0.009&@xmath127&upper limit + & s&g@xmath65&0.5&4.99&20&@xmath50&@xmath128&0.02&offset x - ray + & s&g&0.5&4.99&36&@xmath129&0.0013&@xmath130&upper limit + & s&e&0.7&4.99&106&@xmath129&0.0056&@xmath127&upper limit + & s&d&1.2&4.99&206&@xmath64&0.011&@xmath127&upper limit + 3c403&e&f1&0.275&8.35&16&@xmath73&@xmath131&0.04 + & & f1b&0.256&8.35&7.7&@xmath75&@xmath132&0.06&jet knot ? + & & f2&@xmath133&8.35&27&@xmath77&0.0011&@xmath130&upper limit + & & f6&0.272&8.35&27&@xmath80&0.0014&0.06&jet knot + @xcite ( hereafter gk03 ) proposed to explain the x - ray properties of hotspots in objects aligned close to the line of sight using a model involving emission from the decelerating relativistic jet . at the time they were writing , there was an apparent correlation between observations of hotspots with x - ray emission too bright to be ssc in origin and the jet side of broad - line objects , such as broad - line radio galaxies and radio - loud quasars , that are expected to lie at small angles to the line of sight in unified models . since then , further observations have established that these non - ssc hotspots can occur in objects in all orientations @xcite . however , the gk03 model is still interesting because it predicts spatial offsets between the peak of the x - ray and that of the radio . they proposed that the jet decelerates on kpc scales and that x - ray emission from the fast - moving component can be produced by inverse - compton scattering of the synchrotron photons from the slower , downstream component . this external inverse - compton process is strongly directional and can only be seen if the jet is aligned close to the line of sight ( and , of course , is approaching rather than receding ) . the gk03 model should not be confused with the process of inverse - compton scattering of the cmb that has been proposed to explain the x - ray jets in core - dominated quasars ( e.g. , * ? ? ? * ) : in the gk03 model the photon energy density is dominated by synchrotron photons from the upstream hotspot , and as a consequence the beaming factors required ( for given jet properties ) are less extreme and the range of plausible angles to the line of sight can be greater . in the beamed cmb models the x - ray emission should have no particular relationship to the radio hotspot , and the angle to the line of sight of the jet is required to be small : we therefore do not consider these models further . the gk03 model might be applied to the offset between the primary hotspot of 3c227 , p1/2 , and its x - ray counterpart . 3c227 is a broad - line object , so that the lobes make a relatively small angle to the line of sight ( @xmath134 ) and , if we take p4 to be a jet knot , the only evidence that we have suggests that the w lobe is likely to contain the jet pointing towards us , though the comparatively weak radio core , the non - detection of any other components of the jet , and the presence of intrinsic absorption in the x - ray spectrum ( [ 227-core ] ) suggest that it is not at a very small angle to the line of sight ( compared to , say , 3c390.3 , with its bright radio core , unabsorbed nuclear x - ray emission and well - detected kpc - scale jet ) . the flat x - ray spectrum of the x - ray counterpart of p1/2 is consistent with an inverse - compton model . if we suppose that the jet decelerates on scales of a few kpc ( set by the observed projected size of the offset between radio and x - ray ) from relativistic to sub - relativistic speeds , is it possible that the gk03 model could explain the observed offset in 3c227 ? in principle the answer is ` yes ' since the gk03 model can explain _ any _ observation with a suitable choice of the ( observationally unconstrained ) parameters of the position - dependent bulk flow speed and electron energy spectrum of the jet . the jet would have to be quite wide at the position of p1/2 to produce the observed x - ray morphology , but not impossibly so : the distribution of emitting particles and jet velocities would also require some fine tuning in order to produce an offset x - ray peak rather than some more jet - like x - ray structure . however , we consider an explanation in terms of the gk03 model to be hard to sustain when all the observations of our sample are considered together . the secondary hotspot of 3c227 shows a similar offset , yet the bulk flow clearly can not decelerate to sub - relativistic speeds twice and the direction of any flow between the primary and secondary is not the same as the direction between p4 and p1/2 . a series of coincidences is required to explain the similarity between the primary and secondary hotspots in 3c227 , therefore . the model clearly can not explain the offsets between radio and x - ray peaks in 3c390.3 ( not in the direction of the jet ) or the offsets and variations in radio / x - ray ratio seen in 3c327 and 3c403 ( too large an angle to the line of sight ) . we conclude that the gk03 model , while not ruled out by the data for the primary hotspot of 3c227 , is of little use in providing a general explanation of the problems posed by our observations . since inverse - compton explanations seem difficult to accept , we next consider synchrotron emission . synchrotron explanations also require point - to - point electron spectrum ( or possibly magnetic field strength ) variations to account for offsets between the radio and x - ray emission . however , the magnitude of the variation is comparatively very small : the high - energy tail of the electron population responsible for the x - ray emission is energetically , and a fortiori numerically , a negligible fraction of the total , whereas for the inverse - compton process a doubling of the emissivity by adjusting the electron spectrum requires a doubling of the number density of low - energy electrons , and so effectively a doubling of the energy density of the system . energetically , therefore , it is not difficult to produce what we see via synchrotron radiation . since the electron energy loss timescale is likely to be very short ( for field strengths close to the equipartition value ) what is required to produce the observed x - ray structures is some process that can accelerate particles wherever we observe x - ray emission . as we discussed in [ intro ] , there is strong evidence from radio and optical data that hotspots are sites of particle acceleration , and their spectra are consistent with models involving a single shock followed by downstream losses . in some cases the x - ray emission from hotspots lies on an extrapolation of these models @xcite . but observations of _ diffuse _ x - ray emission , often poorly matched to the observed radio structures ( e.g. in pictor a , @xcite ; 3c33 , @xcite ; 3c390.3 s , this paper ) make it hard to sustain a model in which the particle acceleration at the hotspots is _ only _ occurring at jet termination shocks . similar conclusions have been reached by other authors based on optical data ( e.g. , * ? ? ? * ; * ? ? ? observations of compact but offset x - ray emission , as in 3c227 , present a different problem . here it seems possible that there is a discrete acceleration region that is related to the jet termination shock , but , if so , the shock is not where we would have inferred it to be from radio observations . for 3c227 we could imagine a picture in which the x - ray emission in both primary and secondary hotspots tells us where the shock is now , while the radio traces material that has passed through this shock region , expanded and decelerated . this would imply that the x - ray emission should have a radio / optical counterpart , but for a flat synchrotron spectrum ( @xmath135 ) extending between radio and x - ray the emission at other wavebands could be undetectably faint , at the 10 @xmath107jy level in the radio ( i.e. , substantially below the upper limit of @xmath136 mjy on the flux density of this component from radio maps ) . such a flat synchrotron spectrum extending all the way to the x - ray has never been observed ( precisely because of the difficulty of detecting the radio counterparts ) but might be expected in models of shock acceleration . the bulk of the optical emission in 3c227 p1/2 seems to lie in between the x - ray and radio peaks , which is qualitatively consistent with this picture . the questions to be asked are then why other hotspots seen with similar spatial resolution , such as those in 3c403 , do not show the same radio / x - ray offsets ; why the offsets in 3c327 , if they have the same origin , are so much larger ; and what the origin is of the compact structure transverse to the jet direction in the hotspot of 3c390.3 . recent numerical simulations suggest that the picture of particle acceleration in the lobes of frii sources may be less simple than in the traditional model of acceleration at strong shocks in one or more hotspots . @xcite carried out three - dimensional mhd simulations that modeled the transport of relativistic electrons and of particle acceleration at shocks . they found that the interaction of the jet and the backflowing plasma at the head of the jet produced what they called a ` shock - web complex ' , `` a region of shocks of varying strengths and sizes spread throughout the source '' . even when there was a simple terminal shock , not all the jet material necessarily passed through it , and the terminal shock was not always the strongest shock in the system . while it is not clear that their simulations are perfectly matched to real radio sources , they are capable of producing simulated synchrotron images that show apparent clear discrete multiple hotspots @xcite and in these cases the particle acceleration is not necessarily well matched to the locations of the hotspots : hotspot locations in their model can have more to do with magnetic field amplification than with particle acceleration . the notion of a ` shock - web complex ' at the head of the jet could help to explain the diffuse x - ray emission now seen in the radio - bright but non - compact source head regions of a number of objects , as discussed above , while the idea that the particle acceleration region may not always be co - spatial with the observed radio hotspot might help to explain observed offsets . it should be possible to carry out numerical simulations that allow synthetic maps of the location of high - energy synchrotron - emitting particles to be generated for qualitative comparison with the range of structures seen in x - ray observations . we can also compare the x - ray observations of hotspots with observations of systems in which the x - ray emission is almost certain to be synchrotron in origin , the fri jets ( [ intro ] ) . in the nearest fri jet , centaurus a @xcite there is direct dynamical evidence for shock - acceleration of particles , as we believe is going on in frii hotspots . there are also offsets , albeit on scales of only tens of pc rather than kpc , between the peak of the x - ray emission and the brightest radio emission , in the sense that the radio emission peaks downstream of the x - ray . and there is diffuse x - ray emission , not associated with any compact radio source or dynamical feature of the jet , which in some cases has a diffuse optical counterpart @xcite . other fri jets show similar features . at present we do not understand the nature of the radio / x - ray peak offsets in fri jets or the distributed particle acceleration process responsible for the diffuse x - ray emission , but the qualitative similarity between the cen a jet and a jet termination region like 3c390.3 s or 3c33 s means that we might hope to gain some insight into one problem by studying the other . in both cases the observational requirement is sensitive , multi - frequency observations that allow us to construct a detailed map of the synchrotron sed as a function of position . our two new targets , 3c227 and 3c327 , provide at least one clear example ( 3c227 w ) , and possibly two , of an object where the primary and secondary hotspot are both detected in the x - ray , setting aside the problem of offsets between the components . 3c33 n @xcite is another example of a source with multiple x - ray hotspots . 3c390.3 n , on the other hand , behaves more similarly to 3c403 e : the bright secondary hotspot is not an x - ray ( or optical ) synchrotron source even though there is an apparently clear connection between the primary and secondary hotspot indicative of continuing energy supply ( but cf . the discussion of this point in @xcite ) . taking sx1 and its radio counterpart g@xmath65 to be the primary hotspot of the southern hotspot complex , a similar statement can be made for this system too . in both these cases , the upper limit on the x - ray to radio flux ratio in the non - detected hotspots , which are generally brighter in the radio , is 12 orders of magnitude below the measured value for the primary hotspots ( table [ hsic ] ) . if we assume , as discussed in the previous section , that the x - ray emission mechanism is synchrotron , then this tells us that secondary hotspots can be different : some , at least , are able to accelerate particles to the highest observable energies , but others are at least an order of magnitude less efficient than the primaries in producing x - ray emission for a given radio emissivity . this conclusion would be stronger if the nature of the x - ray emission in the secondary hotspots were more obvious . if some secondary hotspots can accelerate particles to high energies and some do not , what is the difference between them ? relic hotspots left behind by a jet that has moved ( ` dentists - drill ' model ) would certainly not be expected to have high - energy particle acceleration . but in our observations one secondary that apparently is connected to the jet ( 3c390.3 n ) does not have high - energy particle acceleration , while one that has no apparent connection in sensitive radio observations ( 3c227 w ) does . radio morphology is therefore not a good guide to a hotspot s ability to accelerate particles , or to its relationship to the energy supply . nor is the radio brightness of the hotspot . one trend that is apparent in the data is that a hotspot is more likely to be an x - ray emitter ( and therefore a privileged site for high - energy particle acceleration ? ) if it is compact . the secondary hotspots in 3c227 and 3c327 are similar in size to the primaries . those in 3c390.3 and 3c403 are several times larger . ` compact ' here appears to mean less than a few kpc in size . however , though this may be a necessary condition , it is not a sufficient one , as the non - detection of relatively compact hotspots in e.g. 3c390.3 s shows . secondary hotspot compactness could thus be an indicator of relatively well - collimated continued outflow from the primary hotspot to the secondary ( or , in the case of the @xcite model , of a well - collimated disconnected jet ) ; this makes sense , since ( for a given luminosity ) a more compact hotspot is more overpressured with respect to the lobe material and will have a shorter timescale for disappearance via adiabatic expansion in the absence of the energy supply . but the lack of a one - to - one correlation reinforces what we already know from observations of single hotspots : the ability of even low - luminosity , low-@xmath137-field hotspots , even when clearly connected to the energy supply , to produce x - ray emission is very variable and must depend on details of the microphysics that are not yet accessible to us . we have looked with very high spatial resolution at the hotspot x - ray emission from a small sample of radio galaxies that show multiple radio hotspots . as in earlier work , we argue that the x - ray emission from the hotspots comes predominantly from the synchrotron process , and so traces high - energy particle acceleration . to our knowledge this paper represents the first attempt to use synchrotron emission to probe the particle acceleration properties in a sample of frii sources , though several individual objects have previously been studied in detail . * the cores and lobes of the two new sources in our sample have x - ray properties that are entirely consistent with expectations and with the sources places in unified models . there is evidence for intrinsic absorption in the spectrum of the blrg 3c227 . * all the target sources exhibit structure in the x - ray images of their hotspots that would not have been predicted in a simple model in which particle acceleration occurs only at the jet termination as traced by the bright radio hotspot . this structure ranges from small - scale offsets in the radio and x - ray peaks ( e.g. in 3c227 w or 3c390.3 n ) through diffuse x - ray emission that is not well correlated with compact radio structure ( e.g. 3c390.3 s : see also 3c33 s , @xcite ; pictor a e , @xcite ) to point - like sources in the jet termination region that bear little obvious relationship to the current radio hotspots ( 3c327 e ) . if most or all of these structures can be taken to indicate the location of particle acceleration in these sources , then our observations support models in which the particle acceleration history in friis can be complicated , non - localized , and not well traced by radio observations . * our observations were obtained to investigate the nature of multiple hotspots , and we have found some evidence that some secondary hotspots are indeed associated with acceleration of particles to the highest observable energies , while others ( as we had found previously ) are not . this implies that at least some secondary hotspots have ongoing access to a supply of energy . all x - ray - synchrotron emitting hotspots appear to be compact , but not all compact hotspots are detected in x - rays . we can not at present say whether this is because some of these compact hotspots are true relics , disconnected from the energy supply , or whether they are x - ray faint for some other reason related to the microphysics of their particle acceleration . sensitive multi - wavelength observations in radio and optical will be required to make further progress . we are very grateful to karl - heinz mack for providing us with optical images of the hotspots of 3c227 prior to publication and for helpful discussion of the radio - optical alignment . we thank an anonymous referee for constructive comments that helped us to improve the paper . we also gratefully acknowledge financial support for this work from the royal society ( research fellowship for mjh ) and nasa ( grant go6 - 7094x to rpk ) . allen , s.w . , fabian , a.c . , idesawa , e. , inoue , h. , kii , t. , & otani , c. , 1997 , , 286 , 765 black , a.r.s . , baum , s.a . , leahy , j.p . , perley , r.a . , riley , j.m . , & scheuer , p.a.g . , 1992 , , 256 , 186 blandford , r.d . , & rees , m.j . , 1974 , , 169 , 395 brunetti , g. , bondi , m. , comastri , a. , pedani , m. , varano , s. , setti , g. , & hardcastle , m.j . , 2001 , , 561 , l157 brunetti , g. , mack , k.h . , prieto , m.a . , & varano , s. , 2003 , , 345 , l40 cohen , m.h . , ogle , p.m. , tran , h.d . , goodrich , r.w . , & miller , j.s . , 1999 , , 118 , 1963 cox , c.i . , gull , s.f . , & scheuer , p.a.g . , 1991 , , 252 , 588 crawford , c.s . , & fabian , a.c . , 1995 , , 273 , 827 croston , j.h . , hardcastle , m.j . , harris , d.e . , belsole , e. , birkinshaw , m. , & worrall , d.m . , 2005 , , 626 , 733 evans , d.a . , worrall , d.m . , hardcastle , m.j . , kraft , r.p . , & birkinshaw , m. , 2006 , , 642 , 96 fanaroff , b.l . , & riley , j.m . , 1974 , , 167 , 31p georganopoulos , m. , & kazanas , d. , 2003 , , 589 , l5 [ gk03 ] hardcastle , m.j . , & croston , j.h . , 2005 , , 363 , 649 hardcastle , m.j . , alexander , p. , pooley , g.g . , & riley , j.m . , 1997 , , 288 , 859 hardcastle , m.j . , birkinshaw , m. , & worrall , d.m . , 1998 , , 294 , 615 hardcastle , m.j . , birkinshaw , m. , & worrall , d.m . , 2001 , , 323 , l17 hardcastle , m.j . , birkinshaw , m. , cameron , r. , harris , d.e . , looney , l.w . , & worrall , d.m . , 2002 , , 581 , 948 hardcastle , m.j . , evans , d.a . , & croston , j.h . , 2006 , , 370 , 1893 hardcastle , m.j . , harris , d.e . , worrall , d.m . , & birkinshaw , m. , 2004 , , 612 , 729 hardcastle , m.j . , kraft , r.p . , & worrall , d.m . , 2006 , , 368 , l15 hardcastle , m.j . , worrall , d.m . , kraft , r.p . , forman , w.r . , jones , c. , & murray , s.s . , 2003 , , 593 , 169 harris , d.e . , leighly , k.m . , & leahy , j.p . , 1998 , , 499 , l149 harris , d.e . , et al . , 2000 , , 530 , l81 heavens , a.f . , & meisenheimer , k. , 1987 , , 225 , 335 jahoda , k. , mccammon , d. , dickey , j.m . , & lockman , f.j . , 1985 , , 290 , 229 kataoka , j. , & stawarz , l. , 2005 , , 622 , 797 kraft , r.p . , hardcastle , m.j . , worrall , d.m . , & murray , s.s . , 2005 , , 622 , 149 kraft , r.p . , birkinshaw , m. , hardcastle , m.j . , evans , d.a . , croston , j.h . , worrall , d.m . , & murray , s.s . , 2007 , in press ( astro - ph/0701458 ) leahy , j.p . , black , a.r.s . , dennett - thorpe , j. , hardcastle , m.j . , komissarov , s. , perley , r.a . , riley , j.m . , & scheuer , p.a.g . , 1997 , , 291 , 20 leahy , j.p . , & perley , r.a . , 1995 , , 277 , 1097 lonsdale , c.j . , & barthel , p.d . , 1986 , , 92 , 12 mack , k .- h . , prieto , m.a . , & brunetti , g. , 2003 , newar 47 501 meisenheimer , k. , rser , h .- j . , hiltner , p.r . , yates , m.g . , longair , m.s . , chini , r. , & perley , r.a . , 1989 , , 219 , 63 murphy , e.m . , lockman , f.j . , laor , a. , & elvis , m. , 1996 , , 105 , 369 nicastro , f. , fiore , f. , perola , g.c . , & elvis , m. , 1999 , , 512 , 136 prieto , m.a . , 1997 , , 284 , 627 prieto , m.a . , & kotilainen , j.k . , 1997 , , 491 , l77 prieto , m.a . , brunetti , g. , & mack , k.h . , 2002 , sci 298 193 rser , h .- j . , & meisenheimer , k. , 1987 , , 314 , 70 sambruna , r.m . , eracleous , m. , & mushotzky , r.f . , 1999 , , 526 , 60 saslaw , w.c . , tyson , j.a . , & crane , p. , 1978 , , 222 , 435 scheuer , p.a.g . , 1982 , in heeschen , d.s . , wade c.m . , eds , extragalactic radio sources , iau symposium 97 , reidel , dordrecht , p. 163 spinrad , h. , djorgovski , s. , marr , j. , & aguilar , l. , 1985 , , 97 , 932 stark , a.a . , gammie , c.f . , wilson , r.w . , bally , j. , linke , r.a . , heiles , c. , & hurwitz , m. , 1992 , , 79 , 77 tavecchio , f. , maraschi , l. , sambruna , r.m . , & urry , c.m . , 2000 , , 544 , l23 tregillis , i.l . , jones , t.w . , & ryu , d. , 2001 , , 557 , 475 tregillis , i.l . , jones , t.w . , ryu , d. , & park , c. , 2002 , newar 46 387 williams , a.g . , & gull , s.f . , 1985 , , 313 , 34

we report experimental studies of the temperature and magnetic field dependence of resistivity and dc magnetic susceptibility and the temperature dependence of zero field heat capacity in a tiv alloy . the temperature dependence of the normal state dc magnetic susceptibility in this tiv alloy shows tlnt behavior . the temperature dependence of resistivity follows a t behaviour in the range 20 - 50 k. on the other hand , a term lnt is needed in the expression containing the electronic and lattice heat capacities to explain the temperature dependence of heat capacity at temperatures where dependence of resistivity is observed . such temperature dependence of dc magnetic susceptibility , resistivity and heat capacity are indications of the presence of spin - fluctuations in the system . further experimental evidence for the spin fluctuations is obtained in the form of a negative value of t term in the temperature dependence of resistivity . the influence of spin - fluctuations on the superconducting properties of tiv is discussed in detail . we show from our analysis of resistivity and the susceptibility in normal and superconducting states that the spin fluctuations present in tiv alloy are itinerant in nature . there is some evidence of the existence of preformed cooper - pairs in the temperature range well above the superconducting transition temperature . our study indicates that the interesting correlations between spin - fluctuations and superconductivity may actually be quite widespread amongst the superconducting materials , and not necessarily be confined only to certain classes of exotic compounds . _ keywords _ : alloy superconductor , spin fluctuations , electrical conductivity , magnetization , heat capacity

1208.0181

the role of spin - fluctuations in superconducting materials has been a subject of sustained interest for the last 50 years . in 1960s doniach @xcite and berk and schrieffer @xcite made the important observation that itinerant ferromagnetic spin - fluctuations tend to suppress superconductivity in elemental s - wave superconductors . they pointed out that the absence of superconductivity in heavy transition elements such as palladium and platinum is due to the itinerant ferromagnetic spin - fluctuations resulting from the large @xmath6-electron density of states near the fermi level . the same formalism was used to suggest that the superconducting transition temperature t@xmath7 of elements such as nb and v was limited by spin - fluctuations @xcite . hence it was conjectured that when these metals were properly engineered in the form of alloys , the t@xmath7 could increase significantly as was observed in nb@xmath8sn , and v@xmath8si , etc . , later on , the effects of localized spin - fluctuations on the properties of dilute superconducting alloys were studied by zukermann @xcite . he showed that the superconducting transition shifted to temperatures lower than those expected from the band structure calculations performed without considering the spin - fluctuations @xcite . on the other hand , many families of superconductors have been identified during last 30 years , where spin - fluctuations actually play an important role in the superconductivity itself . these include various exotic low temperature superconductors such as y@xmath9co@xmath10 @xcite , cecoin@xmath11 @xcite , ube@xmath12 @xcite , sr@xmath13ruo@xmath14 @xcite , fese @xcite , and mo@xmath8sb@xmath10 @xcite etc . this concept of spin - fluctuation influenced superconductivity , however , came into much prominence in connection with the high temperature superconductors @xcite . this has now become a subject of even more attention with the discovery of a newer class of fe based high temperature superconductors @xcite . most recently it has been reported that the longitudinal ferromagnetic spin - fluctuations induce superconductivity in ucoge @xcite . it is well known that any addition of a magnetic impurity in non - transition element based s - wave superconductor suppresses the superconductivity due to the pair breaking . it is also reported in literature that even the addition of the non magnetic transition elements also suppresses the t@xmath15 @xcite due to the formation of localized states . however , it has been observed that the t@xmath7 of a dirty limit superconductor is not effected significantly by disorder @xcite , in fact a very high level of disorder is required to change the t@xmath7 @xcite . surprisingly an enhancement in t@xmath15 is observed in spite of increased disorder when certain transition elements are alloyed , even though these elements either have t@xmath7 s lower than the host material , or are non superconducting @xcite . the present ti - v alloys are the examples of one such system @xcite . in recent times , it has been recognized that ti - v alloys with their good mechanical , thermal and superconducting properties @xcite can be promising candidates for application as superconducting magnets in the environment of heavy neutron irradiation as in a fusion reactor @xcite . in addition , higher value of lower - critical field h@xmath16(0 ) of ti - v alloys at 2 k @xcite as compared to the commercial polycrystalline nb samples @xcite makes them potential candidates for superconducting rf cavity applications . thus a complete understanding of the normal and superconducting states in these ti - v alloys can lead to the possibility of enhancing their superconducting properties ( e.g. , increasing the experimental superconducting transition temperature t@xmath7 , the lower critical field h@xmath16(0 ) and the upper critical field h@xmath17(0 ) to the theoretical limits ) and enhancing other functional properties like critical current density by suitable engineering techniques . however , several features of ti - v alloys are not understood yet , of which we list below the two significant anomalies . there is an increase in t@xmath7 when ti content is increased from zero to 40 at . % in v @xcite . the significant enhancement in debye temperature @xmath18 or the electron phonon coupling is not expected in ti - v alloys as ti and v are adjacent atoms . the density of states decreases as ti content is increased in v @xcite . hence , the increase in t@xmath7 in ti - v alloys does not follow the mcmillan formula . there is a large region in h - t phase diagram where superconducting fluctuations are observed well above t@xmath7 and well above h@xmath17 @xcite . hake and coworkers @xcite have shown that superconducting fluctuations well above t@xmath7 and well above h@xmath17 are intrinsic property of certain class of transition element binary alloys such as ti - v , ti - mo and ti - ru systems . they have also observed that these fluctuations are not dependent on the sample preparation , surface polishing , size and shape of the sample and on the current density . several models such as the occurrence of reversible @xmath19 phase in the @xmath20 matrix @xcite , weak localization @xcite , kondo ( s - d ) interaction @xcite and associated localized spin fluctuation @xcite have been considered for explaining the observed physical properties in ti - v alloys . the increase in magnetic susceptibility with temperature was attributed mainly to the occurrence of reversible @xmath19 phase in the @xmath20 matrix @xcite . according to this model , the metastable hexagonal closed pack ( hcp ) @xmath19 phase appears inside the body centered cubic ( bcc ) @xmath20 phase when the temperature is decreased below 300 k. however , this model does not explain the large difference between the experimental value of superconducting transition temperature t@xmath7 and that estimated theoretically @xcite . prekul et al @xcite on the other hand invoked the idea of the localized spin fluctuations to understand the normal state properties of ti - v alloys . however , this model could not explain the superconductivity fluctuations observed well above t@xmath7 and well above h@xmath17 @xcite . in this direction we present a detailed study of the superconducting and normal state properties of a ti@xmath0v@xmath1 alloy investigated through the measurements of temperature dependence of electrical resistivity , dc magnetic susceptibility and heat capacity in the presence of zero and externally applied magnetic fields . the observed temperature dependence of the normal state properties is explained within the realm of itinerant spin - fluctuation model . the sample was prepared by taking high purity titanium ( 99.99 % , alfa aesar ) and vanadium ( 99 + % , aldrich ) in stoichiometric proportion and melting in an arc furnace under high purity argon atmosphere . the sample was flipped and re - melted four times to ensure homogeneity . the sample was then wrapped in ta - foil and sealed in quartz ampoule in argon atmosphere , and was annealed at 1573 k for 10 hours . the sample temperature was then lowered to 1273 k before quenching into the ice water . the angle dispersive x - ray diffraction ( xrd ) measurement were performed in a powder xrd beam line bl-12 @xcite at 19 kev ( 0.65 x 10@xmath21 m ) x - rays from indus-2 synchrotron radiation source at raja ramanna centre for advanced technology , indore . the beam line is based on si ( 311 ) double crystal monochromator and adaptive focusing optics . an image plate area detector ( mar-345 dtb ) and fit2d software were used for data acquisition and data reduction , respectively . nist lab6 standard was used for wavelength calibration . the resistivity measurements as functions of temperature and magnetic field were performed in a 5 t magnet - cryostat ( american magnetics inc , usa ) using a standard four - probe geometry . the magnetization measurements were performed using a 9 t vibrating sample magnetometer ( vsm ; quantum design , usa ) and a 7 t squid magnetometer ( mpms - xl ; quantum design , usa ) . the heat capacity measurements were performed using a 9 t physical property measurement system ( ppms ; quantum design , usa ) . the ti - v alloys form in @xmath20 phase for ti composition exceeding 20 at . the space group corresponding to @xmath20 phase is body centered cubic ( bcc ) im@xmath22 m with ti or v occupying 2a site randomly . figure 1 shows the xrd pattern of ti@xmath0v@xmath1 alloy along with the fitted curve obtained using rietveld refinement . all the peaks observed in xrd pattern are correspond to the @xmath20 phase in the ti@xmath0v@xmath1 alloy . the rietveld analysis ( red solid line ) shows that the sample is formed in bcc structure with lattice parameters 0.31879(2 ) nm , which is in agreement with the literature @xcite . figure 2 presents the superconducting properties of ti@xmath0v@xmath1 . figure 2a shows the temperature dependence of magnetization measured using vsm in the temperature range 2 - 7.2 k in an applied field of 10 mt . the measurements were performed in the zero field cooled ( zfc ) , field cooled cooling ( fcc ) and field cooled warming ( fcw ) modes . in the zfc mode , the sample is first cooled from a temperature well above 10 k to 2 k in zero field and then the measurements are performed while warming up the sample in presence of 10 mt . in the fcc mode , the magnetic field ( 10 mt ) is applied at a temperature well above 10 k , and the measurement is done while cooling the sample down to 2 k. after reaching 2 k , the measurement is continued while warming up the sample in the same field and this last protocol is called fcw . the superconducting transition temperature t@xmath7 is the temperature at which the m(t ) starts to drop towards the negative value when temperature is decreased . the t@xmath7 thus estimated at 10 mt is about 7.15 k , which is in agreement with the literature @xcite . the t@xmath7 of ti@xmath0v@xmath1 is higher than that for elemental vanadium ( 5.4 k @xcite ) or titanium ( 0.9 k @xcite ) . the meissner fraction is estimated as m@xmath23(2 k)/m@xmath24(2k ) is about 7 x 10@xmath25 at 10 mt which indicates the strong flux pinning . figure 2b shows the magnetization as a function of magnetic field measured using vsm at various constant temperatures below superconducting transition temperature 7.15 k. the measurements were performed as a function of magnetic field after cooling the sample from well above the 10 k to the desired temperature in zero magnetic field . there is a considerable hysteresis at low fields in all the isothermal m - h curves . for field greater than a characteristic field h@xmath26 magnetization becomes reversible . the upper inset ( i ) of the fig . 2b shows the m v / s h data near h@xmath17 and the lower inset ( ii ) shows the low field m v / s h data at selected temperatures which are used to estimate the h@xmath16 from the magnetization data . we plot in figure 2c the temperature dependence of resistivity below 15 k at various constant magnetic fields up to 5 t. in zero magnetic field , the resistivity decreases as temperature is lowered below 15 k and drops to zero at t@xmath7 = 7.06 k with a transition width of 0.085 k. the inset in fig . 2c shows the effect of magnetic field ( up to 5 t ) on the superconducting transition temperature ; an application of 5 t suppresses the t@xmath7 to 5.66 k. we have observed a rounding - off behavior in the temperature dependence of resistivity just above the t@xmath7 . figure 2d shows the temperature dependence of heat capacity plotted as c / t as a function of t@xmath2 in selected magnetic fields . the application of magnetic field shifts the heat capacity peak to lower temperature . the superconducting transition temperature t@xmath27 is obtained as that temperature at which a deviation is observed from the normal state linearity in c / t v / s t@xmath2 plot . in zero magnetic field , t@xmath27 is estimated to be about 7.17 k. at 8 t magnetic field , the superconducting transition is observed at 4.8 k.there is no difference in normal state heat capacity between the zero field data to that in magnetic fields . the sommerfeld coefficient of electronic heat capacity @xmath28 and the debye temperature @xmath18 are obtained to be about 9.46(2 ) x 10@xmath29 j/ mol k@xmath2 and 258.5(2 ) k respectively by fitting a straight line to the normal state heat capacity data in figure 2d . the jump in heat capacity @xmath30c/@xmath28t@xmath7 across the t@xmath7 in the zero magnetic field is estimated at the onset of the transition t@xmath27 as well as at the middle of the transition t@xmath31 = 6.63 k. the @xmath30c/@xmath28t@xmath27 is about 2.65 whereas @xmath30c/@xmath28t@xmath31 is about 2.22 indicating that ti@xmath0v@xmath1 is a bulk and a strong coupling superconductor . the value of t@xmath27 is in agreement with the t@xmath7 estimated using resistivity and magnetization measurements . the upper critical field h@xmath17 at various temperatures were estimated from such isothermal m - h curves is shown in fig . the magnetic field at which a distinct deviation takes place from the magnetic field dependence of the normal state magnetization is taken as the h@xmath17 . a straight line is fitted to the m v / s h data for the normal state m just above the superconducting state . the h@xmath17 is taken as that point at which the difference between the experimental data and the fitted line exceeds the standard deviation of the fitting . this procedure has been effectively used to estimate the h@xmath17 in superconductors such as borocarbides @xcite and skutterudite @xcite where enhanced paramagnetism is observed in the normal state . the error in the estimation of h@xmath17 is found to be less than 2 % . the temperature dependence of h@xmath17 is analyzed on the basis of formalism by orlando et al @xcite ( red solid line ) which will be discussed in detail later . the h@xmath17(t = 0 ) is estimated to be about 13.68 t. the lower critical field h@xmath16 ( fig . is estimated as that field where the slope of the linear fit to the low field data ( inset ( ii ) to fig . 2b . ) deviates about 2 % from the value of 1 . the magnetization data at low fields are corrected for demagnetization effects before estimating the h@xmath16 . the t@xmath7 estimated from the experimental h@xmath16(t ) is about 7.1 k. the temperature dependence of h@xmath16 follows a t@xmath2 dependence . the experimental data deviates from the t@xmath2 fit near t@xmath15 . the h@xmath16(0 ) is estimated by the t@xmath2 fit to be about 24.38(1 ) mt . the rounding - off behavior of electrical resistivity shown in fig . 2c . is an indication of fluctuation conductivity above the t@xmath7 ( 7.06 k ) . in order to verify this , we have studied this region through the estimation of excess conductivity @xmath32 = @xmath33 in zero magnetic field ( fig . 4a . ) and the magneto resistance @xmath34 = ( @xmath35(h ) - @xmath35(0 ) ) x 100 / @xmath35(0 ) as a function of magnetic field at temperatures above the t@xmath7 ( fig . 4b . ) . figure 4a shows the log - log plot of @xmath32 as a function of ( t - t@xmath7)/t@xmath7 in the temperature range 7.15 - 15 k. the excess conductivity may be explained in terms of the additional conductivity that results from the formation of cooper pair pockets well above t@xmath7 ; these cooper pairs are yet to be in the condensed state . the @xmath36 stands for the normal state conductivity , which is generally estimated by extrapolating the temperature dependence of resistivity from a temperature well above 3 times of t@xmath7 to the region where the fluctuation conductivity is observed @xcite . the @xmath37 is the experimentally observed conductivity , and the transition temperature t@xmath7 = 7.06 k is taken as the temperature at which the temperature - derivative of resistivity shows a maximum . according to theory of aslamazov and larkin ( al ) @xcite , the excess conductivity @xmath38 is expressed as @xmath39 where , @xmath40=(t - t@xmath7)/t@xmath7 is the reduced temperature , @xmath41 is a constant ( temperature independent ) , @xmath42=2-@xmath43/2 is the critical exponent , and @xmath43 is the dimensionality of the superconducting fluctuations . for a three dimensional superconductor , @xmath44 and @xmath42= 0.5 , where @xmath45(0 ) is the coherence length at zero temperature @xcite . the straight line fit near t@xmath7 to the data in fig.4a . , shows that @xmath42= 0.5 indicating 3@xmath43 character of the superconducting fluctuations in ti@xmath0v@xmath1 . however , the temperature range over which this fluctuation is observed is limited to 7.25 to 8.03 k. the upturn in @xmath32 below 7.25 k is due to the broadening of the superconducting transition @xcite . the maki - thompson ( mt ) @xcite effect can also contribute to the fluctuation conductivity where electron -electron interactions are strong . however , the @xmath45(0 ) estimated from the temperature independent amplitude of the al contribution to fluctuation conductivity is about 4.49 nm which is in agreement with the @xmath45(0 ) estimated from h@xmath17(0 ) . this indicates that the mt contribution to fluctuation conductivity is not dominant in ti@xmath0v@xmath1 . apart from this , mt contribution is dominant only in the clean limit @xcite whereas ti@xmath0v@xmath1 is in the dirty limit . it may be noted here that the al model is valid only in limited range of temperature close to t@xmath7 . the figure 4b shows @xmath34 as a function of magnetic field at temperatures above the t@xmath7 . at 7.5 k and 5 t , the @xmath34 is about 0.45 % . the @xmath34 at 5 t decreases as the temperature increases . hence , from fig . 4b , it appears that the magneto - resistance in ti@xmath0v@xmath1 exists up to 3t@xmath7 , where as fig . 4a . , shows that fluctuation conductivity exists up to at least 2t@xmath7 . such positive magneto - resistance below 3t@xmath7 , which is linear in magnetic field , was interpreted by hake to be due to the breaking of these pre - formed cooper pairs which formed well above t@xmath7 but were not condensed yet to give rise to superconductivity @xcite . in this latter argument , the temperature and magnetic field dependence of resistivity in the temperature range t@xmath7=7.15 k to 14.3 k in ti@xmath0v@xmath1 is due to the excess conductivity or the fluctuation conductivity resulting from cooper pair fluctuations well above t@xmath7 . then the effect of the magnetic field is to break the cooper pairs and drive the sample to its normal state . lue et al have shown experimentally that the magnetic field required to drive a system into normal state is more than the h@xmath17(0 ) @xcite for temperature even above t@xmath7 . the observed large positive magnetoresistance above t@xmath7 can not be accounted by the al fluctuation conductivity in presence of magnetic field for a 3d superconductor in the dirty limit is given by @xcite . in case of niobium as well as alloy and compound superconductors such as nb@xmath8sn , it is reported that the disorder generated by the high flux neutron irradiation does not decrease the t@xmath7 significantly @xcite . as pointed out earlier @xcite , the moderate disorder does not influence the transition temperature in dirty limit superconductors . very high disorders are required to drive the t@xmath7 to lower temperatures which are accompanied by a transition from superconducting state to insulating state at t@xmath7 @xcite . since , our system is a metallic and when vanadium is alloyed with the titanium , the t@xmath7 of the alloy increases , the origin of the observed fluctuation conductivity in ti@xmath0v@xmath1 may not be due to the disorder present in the alloy . however , in our alloy , the coherence length is about twice the unit cell . hence it is necessary to consider the effect of disorder on the superconducting properties . the transition from fluctuation conductivity region to condensate region ( @xmath35 = 0 ) in resistivity for ti@xmath0v@xmath1 takes place over a temperature width @xmath46 = 0.085 k centered on t@xmath7 = 7.06 k. this transition temperature width @xmath46 is very large when compared to the @xmath46 of nb and v which is about 0.001 - 0.01 k @xcite . the @xmath46 is estimated by extrapolating the derivative of temperature dependence of resistivity to zero from both side of t@xmath7 . within the landau theory of phase transitions @xcite , the transition broadening due to disorder is given by @xmath47 = ( dt@xmath7/dx ) ( @xmath30x@xmath2)@xmath48 where , ( @xmath30x@xmath2)@xmath48 is the root mean square ( rms ) fluctuation in the composition . for a a@xmath49b@xmath50 binary alloy , the average number of a and b atoms are @xmath51n and n@xmath52 respectively , where n = v@xmath7/@xmath53 is the number of atoms in the characteristic volume v@xmath7=(4/3)@xmath54(0)@xmath55 and @xmath53 is the mean atomic volume . if the alloy is taken to be ideally random , then the rms deviation in the number of b atom @xmath56(n@xmath52)@xmath48 and the rms deviation in the compositional variable , ( @xmath30x@xmath2)@xmath57 @xmath58 . for ti@xmath0v@xmath1 , @xmath59 1.108 x 10@xmath60 m@xmath55 , and v@xmath7 @xmath56 4.78 x 10@xmath61 m@xmath55 which leads to n @xmath56 3.93 x 10@xmath62 . the dt@xmath7/dx for ti - v alloys is about 0.05 k / at . , for v rich samples @xcite . then the @xmath30t@xmath7 due to disorder is about 16 mk which is distinctly small as compared to that observed experimentally . the transition broadening due to the fluctuation of atomic density @xcite is also observed to be negligibly small . hence , we believe that the disorder is not the origin for the observed physical properties viz . , \1 . the increase in t@xmath7 when ti content is increased from zero to 40 at . % in v. \2 . the presence of fluctuation conductivity and superconducting origin of magneto resistance above t@xmath7 . in order to understand the origin of these superconducting properties , we have performed a detailed study on the normal state properties of the ti@xmath0v@xmath1 alloy , which is presented below . figure 5(a ) shows the temperature dependence of dc susceptibility @xmath63 of ti@xmath0v@xmath1 alloy in the temperature range of 10 - 300 k measured using squid magnetometer in an applied magnetic field of 1 t. the data is corrected for background signal . the dc susceptibility is defined as the ratio of measured magnetization ( m ) and applied magnetic field ( h ) . as the temperature decreases below 300 k , the @xmath63(t ) decreases down to about 30 k before showing an upturn at still lower temperature . the isothermal field dependence of magnetization at 10 k measured with background correction using squid magnetometer up to 7 t is shown in fig . 5(b ) . the isothermal field dependence of magnetization at various temperatures t = 10 , 20 , 30 , 50 , 100 , 150 , 200 , 250 and 300 k is also measured up to 8 t using vibrating sample magnetometer . for the sake of clarity , we have provided such isothermal m v / s h curves at 10 , 30 and 300 k in fig . these curves does not show any indications of saturation even at 8 t. these curves rule out any predominant contribution from ferromagnetic impurities . since , the difference between m v / s h curves at various temperatures is not clearly seen in fig . 5(b ) , we have plotted the difference m(h , t ) - m(h , 300 k ) in the inset to fig . 5(b ) which clearly demonstrate the effect of temperature on the magnetic susceptibility . the susceptibility estimated as the high field slope of the isothermal m v / s h data at various temperatures normalized to that at 30 k is shown in the inset to the figure 5(a ) . for comparison we have also plotted the temperature dependence of @xmath63(t ) at 1 t ( measured with squid magnetometer ) normalized to that at 30 k. the increase in the dc susceptibility with temperature has been observed earlier in various 4d and 5d transition metals and rare earth and actinide based paramagnetic intermetallic compounds ( ref . @xcite and ref . @xcite and the references therein ) and mo@xmath8sb@xmath10 @xcite . the temperature dependence of susceptibility of such paramagnetic metals is observed to follow @xcite @xmath64 here t@xmath65 is a characteristic temperature which is related to the peak position t@xmath66 in temperature dependence of susceptibility as t@xmath66 = t@xmath65/@xmath67 with @xmath68 as natural logarithmic base , @xmath69 is a temperature independent constant . the solid red line in fig . 5(a ) show the fit using above equation ( 2 ) to the experimental @xmath63(t ) of the ti@xmath0v@xmath1 alloy along with a curie - weiss term for the low temperature upturn . this low temperature feature appears to be due to paramagnetic impurities ( such as other transition metal elements , oxygen , nitrogen and the oxides and nitrides of vanadium as well ) as in the case of vanadium @xcite . the values of the fitting parameters are as follows : @xmath63(0 ) = 4.92 x 10@xmath21 wba@xmath70m@xmath70 , @xmath69 = 3.88 x 10@xmath71 wba@xmath70m@xmath70k@xmath72 , t@xmath65 = 512 k. the errors in the parameters estimated are less than 10 % . figure 6 presents the resistivity of the ti@xmath0v@xmath1 alloy in the temperature range 5 - 300 k in zero magnetic field . at 300 k the electrical resistivity is about 105.5 @xmath73 m . the electrical resistivity decreases linearly down to 90 k. the decrease in resistivity is sharper when the temperature is lowered below t@xmath65 = 90 k. the residual resistivity @xmath74 is about 99 x 10@xmath75 @xmath73 m . the resistivity goes to zero below 7 k as the system becomes a superconductor . the inset to the fig . 6 shows the resistivity of ti@xmath0v@xmath1 plotted as a function of t@xmath2 . the resistivity in the temperature range 20 - 50 k is observed to be linear in t@xmath2 and can be expressed as @xmath35 ( x 10@xmath75 @xmath73 m ) = 98.98 + 5.05 x 10@xmath25 t@xmath2 . the error in the coefficients of linear fit is less than 0.1 % . however , the plot of ( @xmath35-@xmath74)/t@xmath2 as a function of t@xmath55 which shows a negative slope ( inset ( b ) ) indicating coefficient of t@xmath5 term in resistivity is negative @xcite . the estimated mean free path @xmath76 of the conduction electrons is estimated in the framework of free electron model as @xmath77 , where @xmath78 is the radius of sphere whose volume is equal to the volume per conduction electron , a@xmath79 is he bohr atomic radius and @xmath80 is the residual resistivity expressed in @xmath81-cm @xcite . in ti@xmath0v@xmath1 alloy , the mean free path estimated to be about 0.3 x 10@xmath82 m which is of the order of unit cell . in case of large residual resistivity values such as 99 x 10@xmath75 @xmath73 m , the estimation of the mean free path is not true as matthiessens rule is not valid @xcite . in such cases , the superconducting property can be used to estimate the mean free path @xcite . in the dirty limit superconductor , the coherence length is limited by the mean free path and hence @xmath83 . then the mean free path in this alloy is expected to be about 4.5 x 10@xmath82 m. according to mooij , short mean free path resulting from various scattering mechanism such as @xmath84 interaction , disorder , magnetic interaction etc . , is the reason for the small temperature variation of resistivity @xcite . the figure 7 shows the temperature dependence of the heat capacity c for ti@xmath0v@xmath1 in the temperature range 20 - 220 k. the fitted line represents the expression @xmath28 t + d(t/@xmath18 ) , where @xmath28 is the sommerfeld coefficient of electronic specific heat and d(t/@xmath18 ) is the debye specific heat for lattice . a good fitting to the experimental heat capacity curve is obtained only in the high temperature region ( 150 - 220 k ) . the value of @xmath28 and @xmath18 obtained as the fitting parameters are about 8.06(1 ) x 10@xmath29 j / mol k@xmath2 and 322.5(4 ) k respectively . the fitted @xmath28 t + d(t/@xmath18 ) curve deviates from the experimental curve below 105 k ( inset to the fig . 7 . ) . the figure 8 shows the c / t plotted as a function of t@xmath2 in the temperature range 7 - 45 k. a nonlinearity is observed in c / t with respect to t@xmath2 . such nonlinearity as well as the deviation from @xmath28 t + d(t/@xmath18 ) might originate from a finite temperature dependence of the debye temperature @xmath18 in a bcc structure @xcite . normally , the @xmath18 increases with decrease in temperature below 30 k , especially in case of bcc systems the values at low temperature increases more than that at room temperature @xcite . the variation of @xmath18 for elemental bcc vanadium in the temperature range 4.2 k to 300 k estimated by measuring elastic constants is about 403.8 k to 392.1 k which is about 3 % @xcite . in ti@xmath0v@xmath1 alloy we have observed that @xmath18 estimated from the low temperature fit is about 258 k which is quite low when compared to that estimated at high temperatures . hence , we believe that the origin of the difference between the experimentally observed heat capacity and the heat capacity estimated using @xmath28 t + d(t/@xmath18 ) is not due to the temperature variation of the @xmath18 . however , we have done the thermal expansion measurement to estimate the debye temperature @xcite as function of temperature @xcite . we have observed that the variation in @xmath18 with temperature in the range 30 - 300 k is about 7.5 % . hence , the deviation observed in heat capacity of ti@xmath0v@xmath1 from @xmath28 t + d(t/@xmath18 ) below 100 k is not due to variation in @xmath18 with temperature . the difference may also arise from the acoustic phonons which show peak structures in the phonon dispersion curves . these modes act as a einstein oscillators and hence , we have checked for contributions from einstein modes obtained from the phonon dispersion curve reported in literature for vanadium @xcite . we have observed that for any value of einstein temperature @xmath85 , the value of @xmath18 diverges during fitting . hence , it appears that the contribution to the heat capacity due to the einstein modes is negligible for the present system . by comparing our results with the literature we found that @xmath86 is observed to follow a temperature dependence such as dt@xmath55ln(t ) @xcite this is verified by the presence of linearity in @xmath87 as a function of lnt ( inset a to fig 8 . ) . whenever , t@xmath2 dependence in resistivity is observed , generally the kadowaki - woods relation is checked for its validity . the inset ( b ) to the figure 8 shows the kadowaki woods plot for ti@xmath0v@xmath1 along with several other spin fluctuations and heavy fermion systems . surprisingly , the kadowaki - woods relation is observed to be valid in our system . kadowaki and woods have empirically shown that the coefficient @xmath41 of t@xmath2 term in resistivity scales with the square of the coefficient @xmath28 of electronic specific heat for heavy fermions and spin - fluctuation systems @xcite . the proportionality constant ( kadowaki - woods ratio , a/@xmath88 ) is found to be about 1 x 10@xmath89-cm ( mol k / mj)@xmath2 @xcite . the coefficient a of the t@xmath2 term in resistivity for the present ti@xmath0v@xmath1 is estimated to be about 5.00(1 ) x 10@xmath90-cm /k2 ( inset to fig . 6 . ) . the ratio ( a/@xmath88 ) for ti@xmath0v@xmath1 is about 0.60(5 ) x 10@xmath91 @xmath81-cm ( mol k / mj)@xmath2 which is close to the proportionality constant obtained by kadowaki - woods. we summarize below the important observations made in the present set of experimental results on the normal state of ti@xmath0v@xmath1 alloy . the temperature dependence of dc susceptibility shows the characteristic `` temperature induced magnetic moment '' and follows @xmath63(t ) = @xmath63(0 ) -bt@xmath2 ln(t / t@xmath65 ) . the magnetization is linear in h. \3 . resistivity shows a t@xmath2 dependence at low temperatures . the coefficient of t@xmath5 term in resistivity is negative . there is an enhancement of sommerfeld coefficient of electronic specific heat @xmath28 . the excess heat capacity ( @xmath30 c = c - @xmath28 t - d(t/@xmath18 ) ) follows a dt@xmath55ln(t ) temperature dependence . kadowaki - woods relation is valid in this system . all these features together have been observed only in pauli enhanced paramagnets such as rco@xmath13(r = y , sc , lu , etc . , ) @xcite , etc . , where the physical properties have been explained in terms of the existence of itinerant spin fluctuations . we argue that itinerant spin fluctuations may be present in the ti@xmath0v@xmath1 alloys as well . we now present a model based on the existence of itinerant spin fluctuations in the ti@xmath0v@xmath1 alloy and show that the observed superconducting properties of the alloy can be explained in this picture . in our understanding , the characteristic temperature t@xmath65 = 90 k observed in resistivity is the spin fluctuation temperature t@xmath92 @xcite . it has been shown by frings and franse that the t@xmath92 may be identified as the temperature at which the second derivative of the susceptibility with respect to temperature is zero @xcite . in the present ti@xmath0v@xmath1 alloy , the second derivative of susceptibility goes to zero at t@xmath93 120 k. the t@xmath92 estimated from the straight line fit to data @xmath87 v / s lnt is about 53 k. however , t@xmath92 estimated by resistivity , dc susceptibility and heat capacity are in the same order of magnitude . variation of t@xmath92 among various experiments to about 2t@xmath92 is not uncommon , and is often found in literature @xcite . the presence of spin fluctuations in a material can be characterized through the stoner enhancement factor @xcite . the stoner enhancement factor s is defined as @xmath63(0)/@xmath94 where @xmath63(0 ) is the experimentally observed susceptibility and @xmath94 is the susceptibility due to density of states @xcite . the @xmath94 ( dimensionless ) is given as ( 3/2)@xmath95 = 3@xmath96 where @xmath97 is the permittivity of the free space , @xmath98 is the bohr magneton , @xmath99 is the free electron density , @xmath100 is the fermi energy , and @xmath101 is the boltzmann constant @xcite . the value of s for strong spin fluctuating systems like pd or pt is about 6 @xcite where as the value of s for superconducting nb or v is about 2 @xcite . the value of @xmath94 for ti@xmath0v@xmath1 is about 1.57 x 10@xmath25 . then for ti@xmath0v@xmath1 the value of s = 2.5 , where experimentally observed @xmath63(0 ) = 3.92 x 10@xmath25 ( dimensionless ) . this value of s is similar to that of v and nb and is much less than that of metals like pt and pd @xcite . this is also similar to the value of s for mo@xmath8sb@xmath10 system @xcite . the sommerfeld coefficient of electronic heat capacity @xmath28 enhances due to the presence of spin fluctuations as @xmath28 = @xmath102 where @xmath103 is the sommerfeld coefficient of electronic heat capacity without spin fluctuations . our results suggest that the spin fluctuations are absent temperatures more than about 100 k. our fit to the heat capacity at temperatures above 100 k provided a value of @xmath28 to be about @xmath104 = 8.06 x 10@xmath29 j / molk@xmath2 whereas the low temperature fit provided a value of @xmath28 to be about @xmath1059.46 x 10@xmath29 j / molk@xmath2 . then the value of @xmath106 can be estimated as @xmath107 which turns out to be about 0.17 . the two major effects of the itinerant spin fluctuations on the superconducting properties in the ti@xmath0v@xmath1 alloy are : ( i ) the enhancement of superconducting transition temperature in ti@xmath0v@xmath1 with respect to vanadium , and ( ii ) the origin of the fluctuation conductivity above t@xmath7 and above h@xmath17 . these effects are discussed below : \(i ) the enhancement of superconducting transition temperature in ti@xmath0v@xmath1 with respect to vanadium : the itinerant spin - fluctuations are reported to be responsible for the absence of superconductivity in heavy transition elements such as palladium and platinum @xcite . the same formalism was used to suggest that the superconducting transition temperature t@xmath7 of elements such as nb and v is limited by spin - fluctuations @xcite . the spin fluctuation coupling constant @xmath106 in v is estimated to be about 0.34 @xcite . then , when non magnetic ti is alloyed with vanadium , there is a reduction in spin fluctuation , hence , t@xmath7 increases in spite of decrease in the free electron density . at very high content of ti , the t@xmath7 again starts decreasing when the effect of loss of free electron density wins over the effect of reduction of spin fluctuations . the value of 2 - 3 times t@xmath7 falls in the range of the superconducting transition temperatures of ti - v alloys calculated by estimating the electron - phonon coupling constant @xmath108 from band structure calculation ( which is about 12 - 20 k ) @xcite . hence , we can estimate the @xmath106 for ti@xmath0v@xmath1 by considering the suppression of t@xmath7 from the theoretical limit ( in absence of spin fluctuations ) of about 14.3 k to the experimentally observed value of 7.15 k ( in presence of spin fluctuations ) as described below . daams et al have shown that the properties of superconductors with spin fluctuations can be scaled to that of a superconductor without spin fluctuation by introducing the renormalized parameters @xmath109 and @xmath110 @xcite . in this formalism , the superconducting transition temperature is given by the modified mcmillan formula @xcite as @xmath111 where @xmath112(1+@xmath113 and @xmath114 . here , @xmath115 is the coulomb coupling constant and @xmath106 is the electron - spin fluctuation coupling constant . using this formalism , the @xmath108 and @xmath106 for ti@xmath0v@xmath1 are estimated to be about 0.85 and 0.081 respectively . the value of @xmath106 estimated in this way is in agreement with that estimated from the enhancement of @xmath28 at low temperatures . the value of @xmath115 is taken to be 0.1 @xcite . the other parameters @xmath109 and @xmath110 were estimated to be 0.79 and 0.17 respectively . these values are used to analyze temperature dependence of h@xmath17 on the basis of formalism by orlando et al @xcite . the solid line in the fig . 3b . , shows the temperature dependence of h@xmath17 estimated in this manner @xcite . the spin - orbit interaction parameter @xmath116 is estimated to be about 0.3 . \(ii ) the origin of the fluctuation conductivity above t@xmath7 and above h@xmath17 : when vanadium is diluted with the titanium without changing the bcc structure , the spin fluctuations become inhomogeneous in space due to the spatial disorder in vanadium sublattice . in such a case electron spin - fluctuation coupling constant @xmath106 is expected to vary in space over the sample volume . this is similar to the disordered kondo lattice with spatial distribution of kondo temperatures t@xmath117 s @xcite leading to inhomogeneous short range conduction electron polarization which in turn results in the itinerant spin fluctuations . since , the mobility of the conduction electrons are large , the life time of these spin fluctuations decreases leading to the small spin fluctuation coupling constant @xmath106 of about 0.081 in the ti@xmath0v@xmath1 alloy . our conjecture about the itinerant nature of the spin fluctuations is also supported by the fact that there is no significant magnetic field dependence of physical properties above 20 k , where the signature of the existence of spin fluctuations are observed in the ti@xmath0v@xmath1 alloy @xcite . since the itinerant spin fluctuations are weakly dependent on the magnetic field , the @xmath118 dependence of the ( negative ) magneto - resistance is absent in ti@xmath0v@xmath1 . the temperature and magnetic field dependent dc magnetic susceptibility and electrical resistivity and the zero field heat capacity of ti@xmath0v@xmath1 have been analyzed in a quantitative manner . these results provide some experimental evidences of the presence of spin - fluctuations in this alloy system . the kadowaki - woods scaling is shown to be valid for the ti@xmath0v@xmath1 alloy . we have argued that the spin fluctuations present in ti@xmath0v@xmath1 alloy are itinerant in nature . the onset of superconductivity is suppressed in ti@xmath0v@xmath1 from its expected theoretical limit due to the presence of such spin - fluctuations while the distribution of @xmath106 induces the superconducting fluctuation above t@xmath7 . there are some indications based on the present study of the existence of preformed cooper - pairs in the temperature range well above the superconducting transition temperature . the present study suggests that the interesting correlations between spin - fluctuations and superconductivity may not necessarily be the properties of only certain classes of exotic compounds . relatively simple alloy systems like ti@xmath0v@xmath1 shows such interesting correlations . a complete understanding of the normal state and superconducting properties of this technologically important alloy system ti - v would help in tuning its properties for suitable technological applications . authors would like to thank ms . parul arora for the heat capacity measurements , and mr . v. k. sharma for magnetization measurements using squid magnetometer . s. k. deb , head , isud , rrcat and dr . p. d. gupta , director , rrcat are acknowledged for their support in the experiments done at indus-2 synchrotron radiation source at rrcat , indore . doniach s 1967 _ phys . lett . _ * 18 * 554 berk n f and schrieffer j r 1966 _ phys . rev . lett . _ * 17 * 433 rietschel h and winter h 1979 _ phys . lett . _ * 43 * 1256 wierzbowska m 2005 _ eur . j. b _ * 48 * 207 mitrovic b , schachinger e and carbotte j p 1984 _ phys . b _ * 29 * 6187 mitrovic b and carbotte j p 1986 _ phys . rev . b _ * 33 * , 591 zuckermann m j 1970 _ j. phys . c : solid st . phys . _ * 3 * 2130 sarkissian b v b 1986 _ j. phys . * 16 * 755 kawasaki yu , kawasaki s , yashima m , mitoy t , zheng g -q , kitaoka y , shishido h , settai r , haga y and nuki y o 2003 _ j. phys . . japan _ * 72 * 2308 heffner r h , ott h r , schenck a , mydosh j a and maclaughlin d e 1991 _ j. appl . phys . _ * 70 * 5782 kuwabara t and ogata m 2000 _ phys . _ * 85 * 4586 qiu y , bao w , zhao y , broholm c , stanev v , tesanovic z , gasparovic y c , chang s , hu j , qian b , fang m and mao z 2009 _ phys . lett . _ * 103 * 067008 candolfi c , lenoir b , dauscher a , bellouard c , hejtmanek j , santava e and tobola j 2007 _ phys . rev . lett . _ * 99 * 037006 jin k , butch n p , kirshenbaum k , paglione j and greene r l 2011 _ nature _ * 476 * 73 dahm t , hinkov v , borisenko s v , kordyuk a a , zabolotnyy v b , fink j , buchner b , scalapino d j , hanke w and keimer b 2009 _ nature phys . _ * 5 * 217 tacon m le , ghiringhelli g , chaloupka j , moretti sala m , hinkov v , haverkort m w , minola m , bakr m , zhou k j , blanco - canosa s , monney c , song y t , sun g l , lin c t , de luca g m , salluzzo m , khaliullin g , schmitt t , braicovich l and keimer b 2011 _ nature phys . _ * 7 * 725 kamihara y , watanabe t , hirano m and hosono h 2008 _ j. am . chem . soc . _ * 130 * 3296 hirschfeld p j , korshunov m m and mazin i i 2011 _ rep . phys . _ * 74 * 124508 hattori t , ihara y , nakai y , ishida k , tada y , fujimoto s , kawakami n , osaki e , deguchi k , sato n k and satoh i 2012 _ phys lett . _ * 108 * 066403 see . e.g. , vonsovsky s v , izyumov yu a and kurmaev e z _ superconductivity of transition metals : their alloys and compounds ( translated by brandt e h and zavarnitsyn a p ) _ 1982 ( springer verlag , berlin ) and the references therein . aoki r and ohtsuka t 1967 _ j. phys . . japan _ * 23 * 955 aoki r and ohtsuka t 1969 _ j. phys . . japan _ * 26 * 651 anderson p w 1959 _ j. phys . solids _ * 11 * 26 sacepe b , dubouchet t , chapelier c , sanquer m , ovadia m , shahar d , feigelman m and ioffe l 2011 _ nature phys . _ * 7 * 239 and the references therein collings e w , upton p e and ho j c 1975 _ j. less - comm . metals _ * 42 * 285 collings e w 1986 _ applied superconductivity , metallurgy , and physics of titanium alloys _ vol . 1 & 2 , ( plenum press ) tai m , inoue k , kikuchi a , takeuchi t , kiyoshi t , hishinuma y 2007 _ ieee trans . supercond . _ * 17 * 2542 blaugher r d 1965 _ phys . * 14 * 181 roy s b , myneni g r and sahni v c , 2008 _ supercond . technol . _ * 21 * 065002 hake r r 1969 _ phys . * 23 * 1105 hake r r 1970 _ phys . _ * 32a * 143 lue j w , montgomery a g and hake r r 1975 _ phys . b _ * 11 * 3393 collings e w 1978 _ aip conf _ * 40 * 410 sasaki t , hanada s and muto y 1987 _ physica _ * 148b * 513 sasaki t and muto y 1990 _ physica b _ * 165 & 166 * 291 prekul a f , rassokhin v a and volkenshtein n v 1973 _ zhetf pis . _ * 17 * 354 rassokhin v a , volkenshtein n v , romanov e p and prekul a f 1974 _ sov . phys.- jetp _ * 39 * 166 [ 1974 _ zh . fiz . _ * 66 * 348 ] prekul a f , rassokhin v a and volkenshtein n v 1975 _ sov . .- jetp _ * 40 * 1134 [ 1974 _ zh . * 67 * 2286 ] prekul a f , rassokhin v a and volkenshtein n v 1975 _ sov . phys .- jetp lett . _ * 22 * , 209 [ 1975 _ pismazh . eksp . teor . * 22 * 433 ] pictet o , jarlborg t and peter m 1987 _ j. phys . * 17 * 221 sinha a k , sagdeo a , gupta p , kumar a , singh m n , gupta r k , kane s r and deb s k 2011 _ aip conf . proc . _ * 1349 * 503 aurelio g , guillermet a f , cuello g j and campo j 2002 _ metall . trans . a _ * 33 * 1307 roy s b , hossain z , pradhan a k , mazumdar c , chaddah p , nagarajan r , godart c and gupta l c 1994 _ physica c _ * 228 * 319 roy s b , hossain z , pradhan a k , chaddah p , nagarajan r and gupta l c 1996 _ physica c _ * 256 * , 90 sharath chandra l s , chattopadhyay m k and roy s b 2012 _ phil . mag . _ ( in press ) orlando t p , mcniff e j jr . , foner s and beasley m r 1979 _ phys . b _ * 19 * 4545 gauzzi a and pavuna d 1995 _ phys . rev . _ * 51 * 15420 aslamazov l g and larkin a i 1968 _ phys . a _ * 26 * 238 johnson w l and tsuei c c 1976 _ phys . b _ * 13 * 4827 maki k 1968 _ prog . theor . * 39 * 897 maki k 1968 _ prog . _ * 40 * 193 thompson r s 1970 _ phys . b _ * 1 * 327 samoilov a v , yeh n -c and tsuei c c 1998 _ phys . rev . b _ * 57 * 1206 theunissen m h and kes p h 1997 _ phys . rev . b _ * 55 * 15183 ullah s and dorsey a t 1991 _ phys b _ * 44 * 262 kernohan r h and sekula s t 1967 _ j. appl . * 38 * 4904 sekula s t 1978 _ j. nucl . * 72 * 91 smith t f 1970 _ phys . * 33a * , 465 radebaugh r and keesom p h 1966 _ phys . rev . _ * 149 * 209 misawa s and kanematsu k 1976 _ j. phys . * 6 * 2119 barnea g 1975 _ j. phys . c : solid state phys . _ * 8 * l216 kriessman c j and callen h b 1954 _ phys . rev . _ * 94 * 837 childs b g , gardner w e and penfold j 1959 _ phil . mag . _ * 4 * 1126 galoshina e v 1974 _ usp . * 113 * 105 hechtfischer d 1976 _ z. phy . b _ * 23 * 255 schindler a i and rice m j 1967 _ phys . rev . _ * 164 * 759 schindler a i and coles b r 1968 _ j. appl . phys . _ * 39 * 956 see , e.g. , ashcroft n w and mermin n d 2001 _ solid state physics _ ( harcourt college publishers , new york ) mooij j h 1973 _ phys . ( a)_*17 * 521 hake r r 1967 _ phys . rev . _ * 158 * 356 and the references therein fine p c 1939 _ phys . * 56 * 355 chou c , white d and johnston h l 1958 _ phys . _ * 109 * 788 see , e.g. , gopal e s r 1966 _ specific heats at low temperatures _ ( plenum press , texas ) bolef d i , smith r e and miller j g 1971 _ phys . rev . b _ * 3 * 4100 sayetat f , fertey p and kessler m 1998 _ j. appl . * 121 sharath chandra l s , matin md , sharma v k , chattopadhyay m k and roy s b 2012 ( unpublished ) colella r and batterman b w 1970 _ phys . b _ * 1 * 3913 ikeda k and gschneidner k a jr . 1982 _ phys . b _ * 25 * 4623 kohama y , kamihara y , hirano m , kawaji h , atake t and hosono h 2008 _ phys . rev . b _ * 78 * 020512(r ) kadowaki k and woods s b 1986 _ solid state commun . _ * 58 * 507 ikeda k , gschneidner k a jr . , stierman r j , tsang t -w e and mcmasters o d 1984 _ phys . * 5039 hauser r , bauer e and gratz e 1998 _ phys . rev . b _ * 57 * 2904 gratz e , resel r , burkov a t , bauer e , markosyan a s and galatanu a 1995 _ j. phys . : condens . matter _ * 7 * 6687 frings p h and franse j j m 1985 _ phys . rev . b _ * 31 * 4355 daams j m , mitrovic b and carbotte j p 1981 _ phys lett . _ * 46 * 65 stewart g r 2001 _ rev . phys . _ * 73 * 797 and the references therein . kusunose h 2012 _ j. phys . japan _ * 81 * 023704 figure captions : + figure 1 . ( color online ) the x - ray diffraction pattern of ti@xmath0v@xmath1 in the 2@xmath119 range 15 - 60 @xmath120 . the rietveld analysis ( red solid line ) shows that the sample is formed in bcc structure with lattice parameter 0.3188 nm . ( color online ) ( a ) temperature dependence of magnetization below 7.2 k for 10 mt in zfc , fcc and fcw mode . there is a large difference between zfc and fcc / fcw values indicating strong pinning of the vortices . \(b ) magnetization as a function of magnetic field at various temperatures below t@xmath7 for ti@xmath121v@xmath122 . the inset ( i ) shows the magnetization as a function of magnetic field near h@xmath17 , where h@xmath17 and h@xmath26 are marked by arrows . the inset ( ii ) shows the low field magnetization data which is used to estimate h@xmath16 . \(c ) temperature dependence of resistivity of ti@xmath0v@xmath1 in the range 6 - 15 k at various magnetic fields up to 5 t. a rounding - off behavior is observed above t@xmath7 . the inset shows the temperature dependence of resistivity across the superconducting transition temperature . \(d ) temperature dependence of heat capacity of ti@xmath0v@xmath1 in zero , 3 t and 8 t magnetic field . the symbols are the experimental data points . the red straight line is fit to the normal state data . the dotted straight line is the guide to eye which is used to estimate the jump in heat capacity at t@xmath123 = 7.17 k whereas the dashed line is the guide to eye which is used to estimate the jump in heat capacity at t@xmath124 = 6.63 k. figure 3 . ( color online ) ( a ) the temperature dependence of h@xmath17 for ti@xmath0v@xmath1 . symbols represent the experimental data points , solid line is fit to the data using formalism from orlando et . al.@xcite ( b ) the temperature dependence of h@xmath16 for ti@xmath0v@xmath1 . symbols represent the experimental data points , solid line is fit to the data using the formula h@xmath16(t ) = h@xmath16(0 ) [ 1-(t / t@xmath7)@xmath2 ] . ( color online ) ( a)the @xmath126 = @xmath37- @xmath36 as a function of ( t - t@xmath7)/t@xmath7 in a log - log scale for ti@[email protected] straight line fit to the data shows that @xmath126 follows the eq . 1 . , with @xmath127 0.5 indicating the existence of 3d fluctuations . \(b ) magneto - resistance as a function of magnetic field up to 5 t above t@xmath7 = 7.06 k for ti@xmath0v@xmath1 . the magneto - resistance is about 0.45 % at 7.5 k and 5 t. for temperatures above 20 k , the magneto resistance is zero . ( color online ) ( a)temperature dependence of dc susceptibility ( @xmath63 = m / h ) of the ti@xmath0v@xmath1 alloy in the range 10 - 300 k in 1 t magnetic field . as the temperature decreases below at 300 k , the @xmath63(t ) decreases down to 20 k before showing a low temperature upturn . the red solid line shows the fit to the experimental data using eq . 2 . the inset shows the normalized @xmath63 with respect to @xmath63(30 k ) measured at 1 t ( black open circles ) as well as @xmath63 estimated from high field m v / s h data ( fig . 5(b ) ) at various temperatures ( red filled squares ) . \(b ) the plot of magnetization m as a function of @xmath97h at 10 k , 30 k and 300 k. the solid black line is the field dependence of magnetization measured at 10 k using vibrating sample magnetometer whereas the black open squares are the m v / s h data measured using squid magnetometer . the m is linear in h up to 8 t. the inset shows the plot of difference in magnetization between 300 k and various temperatures to highlight the variation in field dependecne of magnetization with temperature . ( color online ) the temperature dependence of resistivity of ti@xmath0v@xmath1 at zero magnetic field in the temperature range 5 - 300 k. resistivity is linear ( red dotted line ) in the temperature interval of 300 - 100 k. resistivity drops sharply below t@xmath65 = 90 k before it drops to zero at 7 k. inset ( a ) shows the resistivity of ti@xmath0v@xmath1 as a function of square of temperature in the range t = 20 k to 50 k. a linear fit to the data indicates @xmath35 is proportional to at@xmath2 at low temperatures . inset ( b ) shows the plot of @xmath128/t@xmath2 as function of t@xmath55 in the temperature range 10 - 100 k. the negative slope of the curve is an indicative of negative coefficient of t@xmath5 term in resistivity . ( color online ) the temperature dependence of heat capacity in the range 20 - 220 k. the solid squares are the experimental data points . the red solid line is fit to the data in the range 150 - 220 k using @xmath28 t + d(t/@xmath18 ) , where d(t/@xmath18 ) is the debye specific heat for lattice . the deviations are observed below 105 k ( inset to the figure ) . ( color online ) the c / t for ti@xmath0v@xmath1 as function of t@xmath2 below 45 k. the deviation from linearity is observed in this temperature range . the dotted blue line is the heat capacity data estimated by considering @xmath28 = 9.46(1 ) x 10@xmath29 j / molk@xmath2 and @xmath18 = 322.5(4 ) k. the solid red line is the heat capacity estimated after considering spin fluctuations along with the lattice and electronic contributions . the inset ( a ) shows the linear dependence of @xmath30 c / t@xmath55 on lnt at low temperatures . the inset ( b ) shows the validity of kadowaki - woods plot for ti@xmath0v@xmath1 .

veech groups uniformize teichmller geodesic curves in riemann moduli space . recently , examples of infinitely generated veech groups have been given . we show that these can even have infinitely many cusps and infinitely many infinite ends . we further show that examples exist for which each direction of an infinite end is the limit of directions of inequivalent infinite ends .

math0410132

the study of billiards on euclidean polygonal tables has benefitted greatly from the use of the theory of riemann surfaces . a classical unfolding process associates to each rational - angled polygonal table a riemann surface with holomorphic 1-form , see @xcite . using this , kerkhoff - masur - smillie @xcite showed that for almost every direction on a polygonal table , the billiard flow is ergodic . soon thereafter , veech @xcite identified a setting where the only non - ergodic directions are those of saddle connections ; these examples are identified by an associated group , the veech group , being a lattice fuchsian group . in @xcite , veech asked if in general these groups can be non - finitely generated . only recently have examples of infinitely generated veech groups been given @xcite , @xcite . here we study the geometry associated to some of these groups . the riemann moduli space @xmath0 is the space of smooth compact riemann surfaces of genus @xmath1 , up to biholomorphic equivalence . the holomorphic 1-forms on a riemann surface @xmath2 of genus @xmath1 form a @xmath1-dimensional complex vector space , @xmath3 . pairs @xmath4 with @xmath5 form a bundle @xmath6 ( of course , @xmath0 is not a manifold , but rather an orbifold , thus a fiber is @xmath7 ) . each pair @xmath4 ( with @xmath8 non - null ) identifies a translation structure on the real surface @xmath2 : local coordinates are defined by integration of @xmath8 . invoking the real structure of @xmath9 , there is a @xmath10-action on these translation surfaces : post - composition with the local coordinate functions . to each @xmath4 , there is thus a map from @xmath10 to @xmath0 , given by sending group elements to corresponding orbit elements . but , rotations act so as to leave the riemann surface @xmath2 fixed . thus there is an induced a map from @xmath11 to @xmath0 . with respect to the teichmller metric on @xmath0 , this map turns out to be an isometric immersion of the hyperbolic plane @xmath12 . the resulting image is called the _ teichmller complex geodesic _ determined by @xmath4 . we will denote this by @xmath13 . the @xmath10-stabilizer of @xmath4 induces a subgroup of the oriented isometry group of the hyperbolic plane . this induced group is the _ veech group _ , denoted @xmath14 . the teichmller complex geodesic is thus the image of the hyperbolic surface ( or possibly orbifold ) uniformized by the veech group ; the map between them is generically injective . we use this to show : [ thmgeometry ] for any genus @xmath15 , there exists a complex teichmller geodesic in the riemann moduli space @xmath0 that has infinitely many cusps and infinitely many infinite ends . we remind the reader of the definition and basic classification of ends of a hyperbolic surface below . informally , an _ end _ is a direction in which geodesics can leave compact subsets . finite volume hyperbolic surfaces can only have cusps as ends ; briefly , a _ cusp _ , or puncture , corresponds to a conformal punctured unit disk in the surface ( which can not be completed in the surface ) . an end of a hyperbolic surface which corresponds to a conformal non - trivial annulus is called a hole , or a flare . these two types of ends are called finite , an _ infinite end _ is an end which is not finite . we prove our main results using the approach of translation surfaces . in particular , in terms that we make precise below , whereas cusps always form a discrete set , we give examples such that any direction giving an infinite end is the limit of directions of inequivalent infinite ends . a translation surface @xmath16 is called a _ veech surface _ if @xmath17 is a lattice in @xmath18 ; the definition of a non - periodic connection point can be found in subsection [ sadparspecpts ] [ inffatboys ] let @xmath4 be a translation surface which is a finite ramified covering of a veech surface @xmath16 , with ramification above some non - periodic connection point and possibly the singularities of @xmath16 . then @xmath19 has infinitely many cusps and infinitely many infinite ends . choose a dirichlet fundamental domain for @xmath14 acting upon the hyperbolic plane @xmath20 ; this determines an infinite set @xmath21 of points in @xmath22 , each representing a distinct infinite end . the set @xmath21 has no isolated points . that @xmath14 , for @xmath4 as in theorem [ inffatboys ] , is infinitely generated is the main result of @xcite . for the @xmath4 of theorem [ inffatboys ] , the complicated geometry at infinity of @xmath19 is projected to a small sublocus of the riemann moduli space . [ thmbigcusp ] let @xmath23 be as in theorem [ inffatboys ] , and suppose that @xmath2 is of genus @xmath1 . then , there is a two - complex dimensional sublocus @xmath24 such that all ends of the complex teichmller geodesic generated by @xmath23 are contained in finitely many ends of @xmath25 . we prove theorem [ thmgeometry ] by restricting to the case where @xmath4 satisfies the hypotheses of theorem [ inffatboys ] an application of the riemann - hurwitz formula shows that we may assume that the corresponding @xmath16 be of genus two , we give explicit examples of such @xmath16 which admit non - periodic connection points in @xcite , it is now known @xcite , @xcite that the set of such examples is large the existence of infinitely many cusps is proven in [ manycusps ] , the existence of infinitely many infinite ends is proven in [ manyends ] , that every direction of an infinite end is the limit of distinct infinite end directions is also proven in [ manyends ] . we prove theorem [ thmbigcusp ] in [ locend ] . in 2 we give background material on translation surfaces and certain @xmath10-actions , in 3 we discuss projections of horocycles to the moduli space @xmath0 . the main results rest upon the elementary tools presented in 4 . that a hyperbolic surface uniformized by an infinitely generated fuchsian group of the first kind admits infinite ends can be deduced already from @xcite , theorem 16 . thus , the examples of infinitely generated veech group given by mcmullen @xcite must also have these . each of mcmullen s examples has its associated teichmller complex geodesics lying densely on a symmetric hilbert modular surface , all ends must lie in the finitely many ends of the hilbert modular surface . thus , if here also there are infinitely many infinite ends , then they may well show the geometry of the type we discuss here . it is a pleasure to thank f. dalbo , g. forni , and j. smillie for helpful discussions . in particular , we thank g. forni for his ideas which are reflected in [ projhoro ] . we thank a. eskin and others , see @xcite , for continued interest in these matters . in order to establish notation , we review basic notions . see say @xcite for more details ; @xcite , @xcite and @xcite use related notation and notions . see also @xcite , @xcite , @xcite for recent related developments . a _ translation surface _ is a real surface with an atlas such that all transition functions are translations . as usual , we consider maximal atlases . given a riemann surface @xmath2 and a holomorphic 1-form @xmath26 , integration of @xmath8 induces local coordinates on @xmath2 away from the zeros of @xmath8 ; regarding @xmath9 with its standard structure as @xmath27 , transition functions are easily seen to be translations . in particular , directions of flow on a translation surface are well - defined , by pull - back from the plane . one easily checks that the translation surface associated to @xmath4 is well - defined up to isomorphism of translation surface ( such isomorphism is realized by local pure translations ) , see section 2.1 of @xcite for a more formal discussion . the holomorphic 1-form @xmath8 induces a flat metric on @xmath2 off of its zeros ; when the metric is completed , a zero of multiplicity @xmath28 gives a conical singular point , of cone angle @xmath29 . we consider only translation surfaces of this type ; the notation @xmath4 hence denotes both the ordered pair of riemann surface @xmath2 with holomorphic 1-form @xmath8 , and the translation surface structure , with which @xmath8 equips @xmath2 . given a translation surface @xmath4 , and an element @xmath30 , we can post - compose the coordinate functions of the charts of the atlas of @xmath4 by @xmath31 . it is easily seen that this gives again a translation surface , denoted by @xmath32 . given a translation surface @xmath4 , let @xmath33 denote the surface which arises from deleting the singularities of @xmath4 . an _ affine diffeomorphism _ @xmath34 of @xmath4 is a homeomorphism of @xmath2 that restricts to a diffeomorphism of @xmath33 whose differential at all points is a constant element of @xmath10 . in other terms , @xmath34 induces a real - affine map in the charts of the @xmath8-atlas on @xmath2 , of constant ( jacobian ) derivative . the orientation - preserving affine diffeomorphisms form a group , denoted @xmath35 . the subgroup of @xmath10 which arises as the derivatives of affine diffeomorphisms is denoted by @xmath36 . if @xmath37 have the same derivative , then the diffeomorphism @xmath38 acts by pure translations . this diffeomorphism thus preserves the conformal structure on the underlying translation surface ; that is , @xmath38 is a conformal automorphism of the riemann surface @xmath2 ( that preserves @xmath8 ) . the kernel of the group homomorphism from @xmath35 to @xmath36 is hence certainly finite whenever @xmath2 is of genus greater than one . by pull - back on charts , each @xmath39 also gives a translation structure on the underlying surface . indeed , if @xmath31 is the constant derivative of @xmath40 with respect to the @xmath8-charts , then @xmath40 is an isomorphism of the translation surface @xmath32 with @xmath23 . thus , the @xmath10-stabilizer of @xmath4 is @xmath36 . fix some @xmath4 , and let @xmath41 be given by @xmath42 ; let @xmath43 denote the standard projection . note that the underlying riemann surface @xmath44 is common to the @xmath45-orbit of any @xmath46 the image of @xmath16 under a rotation is @xmath47 , where @xmath48 is the complex number which induces the rotation . thus , @xmath49 induces a map ( from right cosets ) @xmath50 . the image of @xmath34 is @xmath13 , the _ teichmller complex geodesic _ associated to @xmath4 . right multiplication gives the natural right group action of @xmath51 on the cosets @xmath52 . the center , @xmath53 , of @xmath51 is a subgroup of @xmath45 , and one finds that @xmath54 acts faithfully . the @xmath18-stabilizer of the map @xmath34 is the _ veech group _ @xmath55 , the image in @xmath18 of @xmath56 . this is trivially verified : @xmath57 implies that there is some rotation @xmath58 such that @xmath59 . thus , the class @xmath60 is in the @xmath55-orbit of @xmath61 . that @xmath34 is single - valued on each such orbit is obvious . the map @xmath34 is generically injective ; it can only fail to be injective at some @xmath46 if @xmath62 is larger than the automorphism group of riemann surfaces corresponding to neighboring points of @xmath0 , see @xcite and @xcite . when the veech group normalizes a true surface , then @xmath34 gives a ramified cover normalizing , in the sense of algebraic curves , @xmath13 . in @xcite , veech proved that @xmath55 is a discrete subgroup of @xmath18 . that is , it is a fuchsian group . the celebrated _ veech dichotomy _ , also proven in @xcite , states that if @xmath14 is a lattice that is , if @xmath63 has finite hyperbolic area , then the linear flow in any direction of the surface @xmath4 is either ergodic or periodic . one thus says that a translation surface is a _ veech surface _ if @xmath14 is a lattice . we also say that @xmath4 is an _ arithmetic surface _ if @xmath14 is arithmetic , in that it is @xmath18-commensurable to @xmath64 . the teichmller metric on @xmath0 is found by giving the teichmller metric on the ( ramified ) cover teichmller space @xmath65 of @xmath0 ; the teichmller modular group ( equivalently , the mapping class group ) acts isometrically on @xmath65 , with quotient @xmath0 . thus , the teichmller metric descends to @xmath0 . we avoid direct discussion of teichmller space here , the reader may wish to refer to , say , @xcite . if @xmath66 is a non - zero real , then the translation surfaces @xmath4 and @xmath67 differ only by scaling . this scaling and the @xmath10-action commute ; furthermore , the projection map @xmath68 ignores such scaling . thus , we can restrict our discussion to @xmath69 , the collection of area one @xmath4 . in informal terms , whereas a conformal map of riemann surfaces takes infinitesimal circles to circles , a _ quasi - conformal _ map of a riemann surfaces takes infinitesimal circles to ellipses . beltrami coefficient _ of a quasi - conformal homeomorphism @xmath70 is computed by @xmath71 , where @xmath72 is a local chart in @xmath44 , and the symbols @xmath73 and @xmath74 denote @xmath75 and @xmath76 , respectively . now , if @xmath77 and @xmath30 , then the identity map on @xmath2 , from @xmath4 to @xmath78 has beltrami coefficient @xmath79 , where @xmath80 is the complex dilatation of the linear map given by @xmath31 on @xmath9 with its standard real structure . but , the map sending elements @xmath81 to @xmath82 is constant on right @xmath83-cosets . this can be shown by direct calculation , or see corollary 1 of @xcite . indeed , @xmath84 injectively identifies @xmath85 with the complex unit disk , @xmath86 . the teichmller metric on @xmath85 is the pull - back of the hyperbolic metric on @xmath86 . but , as @xcite show in their lemma 4.2 , the map sending @xmath87 by @xmath88 is conformal ; it is thus an oriented isometry . hence , the veech group is the stabilizer in the oriented isometry group of @xmath50 . it is reasonable to identify @xmath85 with @xmath20 , as does @xcite . in this case the stabilizer of the map @xmath89 giving the teichmller complex geodesic is again the veech group , but acting on the left by mbius transformations conjugated with the action of @xmath90 . this conjugating action corresponds to @xmath91 , the constant derivative of a coset representative of the index two subgroup @xmath92 in the full group of affine diffeomorphisms of @xmath23 . thus , @xmath93 _ is _ @xmath14 ; compare with proposition 3.2 of @xcite . in what follows , we will ignore this innocuous conjugation , and consider @xmath14 acting on @xmath20 by mbius transformation . any translation surface of genus greater than one must have singularities . a geodesic line emanating from a singularity is called a _ separatrix _ , a separatrix connecting singularities ( with no singularity on its interior ) is a _ saddle connection_. we call a point @xmath94 a _ connection point _ if every separatrix passing through @xmath95 extends to a saddle connection . we say @xmath96 is a _ parabolic direction _ for @xmath4 if there is a parabolic element of @xmath14 whose associated affine diffeomorphism preserves the direction @xmath96 . ( in fact , the full veech dichotomy states that the non - ergodic directions of a veech surface are parabolic . ) we also call these associated affine diffeomorphisms parabolic . expressing @xmath96 as a slope , the point on @xmath22 associated to this slope is @xmath97 ; this is due to the fact that @xmath10 acts on @xmath4 by way of its _ linear _ action on @xmath27 , but the @xmath10-action on @xmath20 is _ fractional linear_. thus for example , an element of @xmath10 arising from an element of @xmath35 which fixes the vertical direction ( should such an element exist ) acts on the extended hyperbolic plane so as to fix the boundary point @xmath98 , see figure 1 . a point @xmath94 of finite orbit under the group of affine diffeomorphisms is called a _ periodic point_. a point of finite orbit under the subgroup of affine diffeomorphisms generated by the two ( primitive ) parabolic diffeomorphisms for some pair of transverse parabolic directions is called a _ rational point _ for @xmath4 . in @xcite , one shows that non - periodic points have dense orbits . in @xcite , we show that every connection point is a rational point . to date , constructions of infinitely generated veech groups rely on the following basic fact of fuchsian groups : ( 1 ) a fuchsian group of the first kind ( that is , whose limit set is all of @xmath99 ) is either a lattice group or else is infinitely generated . in @xcite , we used two further facts : ( 2 ) given a ramified covering @xmath100 of riemann surfaces , upon setting @xmath101 the translation surface @xmath4 is a _ balanced translation covering _ of the marked translation surface @xmath102 if every point of the fiber above a nonsingular branch point is ramified , in particular , @xmath14 is then commensurable to @xmath103 , where this latter is the image in @xmath17 of the affine diffeomorphisms which stabilize the branch locus of @xmath34 ; and , ( 3 ) if there is an element of this branch locus that is non - periodic ( that is , such that its @xmath104-orbit is of infinite order ) , then @xmath103 is of infinite index in @xmath17 ; in particular , it is thus not a lattice . therefore , the commensurable @xmath14 also fails to be a lattice . given any point of a translation surface , the directions of separatrices through the point are dense ( see , say , lemma 1 of @xcite , in turn a slight generalization of a result of @xcite ) ; the direction of any saddle connection on a veech surface is in fact a parabolic direction ; one finds that if the branch locus contains a connection point , then the parabolic directions of @xmath103 are dense in @xmath22 , and thus the limit set this group is certainly all of @xmath22 . now , if the connection point is non - periodic , then we conclude that that @xmath103 is infinitely generated ; again , the commensurable @xmath14 is therefore also infinitely generated . but , we show in @xcite that there exist @xmath16 that admit non - periodic connection points , and thus proved the existence of infinitely generated @xmath14 . + for an explicit example , we showed in @xcite ( see in particular proposition 7 there ) that the genus two surface given by the golden cross can be taken to play the role of @xmath16 . this surface has exactly one singularity , say @xmath105 ; let us also choose one of its infinitely many non - periodic connection points , @xmath95 . if @xmath106 is even , then we can take @xmath107 copies of our genus two surface , with a slit from @xmath105 to @xmath95 ; identify these with the usual process after having chosen some transitive permutation on @xmath107 letters . this gives a degree @xmath107 ramified covering , whose branch locus consists of @xmath108 , and which is totally ramified above each of these points . the riemann - hurwitz formula then shows that the covering riemann surface does have genus @xmath1 . one can choose the odd genus @xmath109 by way , say , of double coverings : choose @xmath110 connection points of our genus two surface ; create @xmath111 slits and glue two copies in the standard manner . for more on creating ramified coverings , see say @xcite . a _ cylinder _ on a translation surface @xmath4 is a maximal connected set of homotopic simple closed geodesics . if the genus of @xmath2 is greater than one , then every cylinder is bounded by saddle connections . upon cutting the cylinder along a line perpendicular to the direction of flow , we can represent it as indicated in figure 1 : of width @xmath112 and height ( or circumference ) @xmath113 . a cylinder has its affine dehn twist @xmath114 . the _ modulus _ of the cylinder is @xmath115 . if for some direction @xmath96 the surface @xmath4 decomposes into cylinders in this direction , then we say that @xmath96 is a _ ( completely ) periodic direction_. veech @xcite shows that a completely periodic direction @xmath96 is a parabolic direction if and only if the various cylinders in the direction @xmath96 gave commensurable inverse moduli : all @xmath116 are rational . the teichmller _ horocyclic flow _ is given by @xmath117 , where @xmath118 , with @xmath119 . the orbit of @xmath4 under this flow is called a _ horocycle_. suppose that @xmath4 is completely periodic in the vertical direction , that is @xmath4 has a decomposition into vertical cylinders . the effect of @xmath120 , with @xmath121 , restricted to a cylinder of the decomposition is to change the relative positions of the singularities on the right hand side of the cylinder with respect to those on the left . it is easily seen that on a given vertical cylinder of height @xmath113 and width @xmath112 , that a standard affine twist on the cylinder has been performed when @xmath122 . ( thus , when @xmath105 is the multiplicative inverse of the modulus of the cylinder . ) see figure 1 . horocyclic flow : moving a singularity . if the various moduli of the cylinders are commensurable , then there is a least positive @xmath105 such that @xmath120 induces on each cylinder an integer power of the basic dehn twist of the cylinder . this @xmath120 is then easily seen to belong to @xmath36 . indeed , the element @xmath120 then generates a parabolic subgroup of @xmath14 , and the @xmath14-conjugacy class of this parabolic subgroup corresponds to a _ cusp_. teichmller _ geodesic flow _ is given by @xmath123 , where @xmath124 , with @xmath125 . for positive @xmath126 , the effect of @xmath127 restricted to a vertical cylinder decomposition is to decrease heights and increase widths . note that vertical saddle connection lengths tend to zero as @xmath128 . thus , if @xmath129 has saddle connections in the vertical direction , then for @xmath126 sufficiently large , the shortest saddle connections for @xmath123 are in the vertical direction . a _ horodisk _ of index @xmath130 corresponding to a direction @xmath96 of @xmath4 is the union over @xmath131 of all the horocycles @xmath132 . a direction @xmath96 of @xmath4 is a _ fat direction _ if there is a sufficiently large positive real @xmath130 such no two points of the horodisk of index @xmath130 corresponding to @xmath96 are @xmath36-equivalent . an _ end _ of a hyperbolic manifold @xmath133 is determined by a cover of @xmath133 by compact sets @xmath134 , with each @xmath134 contained in the interior of @xmath135 ; an end @xmath136 is a sequence of components @xmath137 in the complement of @xmath134 such that @xmath138 . the definition of an end is independent of choice of sequence @xmath134 , see say @xcite . an end @xmath136 of a hyperbolic surface is called a _ cusp _ if there is a subset @xmath139 that is conformally equivalent to the complex punctured unit disk @xmath140 such that for all sufficiently large @xmath141 , @xmath142 . an end is called a hole if there is a subset @xmath143 that is conformally equivalent to a complex annulus @xmath144 with @xmath145 such that for all sufficiently large @xmath141 , @xmath146 . these two types of ends are called finite , an _ infinite end _ is an end which is not finite . in a terminology introduced by haas @xcite , one differentiates between ends of the first kind and infinite ends of the second kind ; the infinite ends that we discuss are of the first kind . to be precise , we say that @xmath136 is an _ infinite end of the first kind _ if there is a subset @xmath147 that is conformally equivalent to the complex upper half - plane @xmath148 such that for all sufficiently large @xmath141 , @xmath149 . note that each of the above definitions continues to be valid upon allowing @xmath133 to be a hyperbolic orbifold , although there is then also the further possibility that the set of elliptic fixed points enters arbitrarily far into an end . + given a fuchsian group @xmath150 , and a point @xmath151 , we say that @xmath152 _ determines an infinite end _ of @xmath153 if there is some horodisk based at @xmath152 such that any @xmath150-orbit in @xmath20 meets this horodisk in at most one point . equivalently , there is an infinite end of @xmath153 which lifts to include a horodisk based at @xmath152 . analogously , we say that a parabolic fixed point @xmath152 of @xmath150 _ determines a cusp_. the following two lemmas are virtually immediate . [ niceends ] if the fuchsian group @xmath150 is of the first kind , then any end of @xmath153 is either a cusp or an infinite end as above . [ fatptsareinfenddir ] let @xmath154 , and let @xmath155 be a direction for @xmath4 . let @xmath151 be given by @xmath156 . if @xmath96 is a parabolic direction , then @xmath152 determines a cusp of @xmath157 . if @xmath96 is a fat direction , then @xmath152 determines an infinite end of @xmath157 . for ease of discussion , we introduce the following terms . given a ( possibly ramified ) covering of hyperbolic surfaces or orbifolds @xmath158 , we say @xmath159 is _ balanced with respect to ends _ if the image of any end of @xmath133 is an end of @xmath160 and also the pre - image of any end of @xmath160 is an end of @xmath133 . the terms _ balanced with respect to infinite ends _ and _ balanced with respect to cusps _ are analogously defined . the following is well - known . [ fincoveringends ] finite degree ramified coverings of hyperbolic surfaces or orbifolds are balanced with respect to each of the following : ends , infinite ends , and cusps . proposition 3 of @xcite can be reworded as the following . [ ourends ] let @xmath16 be a veech translation surface . suppose that @xmath161 is a non - periodic connection point . let @xmath162 denote the subgroup of @xmath14 arising from the stabilizer of @xmath95 in @xmath35 . then the ( possibly ramified ) covering @xmath163 is balanced with respect to ends . however , this covering is not balanced with respect to either cusps or infinite ends ; more precisely : @xmath164 has infinite ends , but @xmath165 has no infinite ends . points defining our ends are dirichlet points . recall that a _ dirichlet point _ is a point @xmath166 which is on the boundary of some dirichlet region thus , @xmath152 admits some @xmath167 such that @xmath152 is on the boundary of the dirichlet fundamental region for @xmath150 . one can show , see @xcite , that a dirichlet point @xmath151 is either a parabolic fixed point of @xmath150 , or is such that the @xmath150-orbit of @xmath168 meets the boundary of a horodisk based at @xmath152 whose interior contains no element of this orbit . recall that @xmath151 is called , after sullivan @xcite , a _ garnett point _ for @xmath150 relative to @xmath168 if the largest horodisk based at @xmath152 whose interior contains no element of the @xmath150-orbit of @xmath168 , also does not meet this orbit . a point @xmath152 is a _ rigid garnett point _ for @xmath150 if @xmath152 is garnett with respect to all @xmath169 . now , if @xmath170 is any fuchsian group , and @xmath152 is a parabolic point for @xmath170 , then for any subgroup @xmath171 admitting a fundamental domain which contains some non - trivial horodisk based at @xmath152 , one can easily show that there is some @xmath167 such that @xmath152 is a dirichlet point for @xmath150 relative to @xmath168 . thus , the points which determine our infinite ends are not rigid garnett points . we do not yet know if they can in fact be garnett points . we discuss the projection of certain horocycles of the teichmller disk of some @xmath4 to @xmath0 . we discuss the avatars in the setting of @xmath172 of certain elements of the teichmller modular group . suppose that @xmath4 admits a cylinder @xmath173 in the vertical direction . consider the basic affine dehn twist on @xmath173 , expressed in standard coordinates simply as @xmath114 , where @xmath174 are respectively the height and width of @xmath173 , see figure 1 . recall that @xmath175 is defined as the _ modulus _ of the cylinder . this dehn twist acts as the identity on the vertical boundaries of the cylinder ; we can extend by the identity map to obtain a well - defined homeomorphism , @xmath176 , of the underlying topological surface . but , @xmath176 allows us to give a differentiable structure on the receiving surface ( such that the image of the singularities are again singularities ) ; furthermore , @xmath176 has locally constant derivative . hence , using @xmath177 we pull - back the translation atlas from @xmath4 to give a new translation structure . now , each of the various cylinders @xmath178 has image under @xmath176 of equal modulus ; by standard results in complex analysis , see say @xcite , @xmath176 is conformal . therefore , the complex structure pulled - back by @xmath177 is the same as that of @xmath2 . thus , we find a mapping of translation surfaces , @xmath179 . ( of course , @xmath180 only when @xmath173 is all of @xmath2 . ) we thus have that @xmath181 , where @xmath68 is the projection . note that the horocyclic flow @xmath182 restricts to induce a map on @xmath173 which commutes with @xmath176 acting on @xmath183 ; modulo the action of @xmath176 , this map is parametrized by @xmath184 . now suppose that there are @xmath185 cylinders in the vertical direction , @xmath186 , @xmath187 , @xmath188 . let @xmath189 denote the modulus , and let @xmath190 denote the map as above , corresponding to the cylinder @xmath191 ; the various @xmath190 commute , as maps on the underlying surface @xmath2 . similarly , we can define an @xmath192-action on the underlying surface : given @xmath193 , in cylinder @xmath178 , we twist by amount @xmath194 ; outside the collection of cylinders , the action is defined to be trivial . since this action commutes with the action of the @xmath190 , we find that the projection to @xmath0 of @xmath195 factors through @xmath196 but , @xmath197 is isomorphic to the @xmath185-torus , @xmath198 . if in fact @xmath23 admits a cylinder decomposition given by the various @xmath178 , then the horocycle @xmath182 lies in @xmath195 . indeed , the projection of this horocycle to @xmath0 factors through a curve on @xmath197 . to generalize slightly , if @xmath96 denotes the direction of a cylinder decomposition of @xmath4 into @xmath185 cylinders , we have that the projection of the corresponding horocycle factors through a curve on the @xmath185-torus denoted @xmath199 . commensurability of non - zero real numbers is an equivalence relation ; for each commensurability class of the moduli @xmath189 , there is a corresponding smallest positive value of @xmath105 which induces on each of the corresponding cylinders @xmath178 an integer power of its basic dehn twist . let @xmath200 be the number of these classes ; for the @xmath201-th class , let @xmath202 be the least common positive integer multiple of the various @xmath203 in this class . we then let @xmath204 be the composition of the various @xmath205 in this class , where @xmath206 is such that @xmath207 . again , the projection of the horocycle factors through the torus corresponding to the group generated by the @xmath204 . we find that in fact our horocycle factors through a curve on the @xmath200-torus @xmath208 . of course , @xmath96 is a parabolic direction if and only if there is a single commensurability class , in which case : @xmath209 , and the horocycle factors through a circle ; this circle lies on ( a finite component of the sequence which defines the end which is ) the corresponding cusp of @xmath63 . in all other cases , there is more than one commensurability class defining @xmath210 , and just as @xmath211 is dense in @xmath212 , so is the image of the horocycle is dense on @xmath213 . for positive @xmath126 , the effect of @xmath124 on a vertical cylinder decomposition of @xmath4 is to preserve the number of cylinders , while decreasing heights and increasing widths . indeed , @xmath127 acts so as to multiply vertical cylinder moduli by @xmath214 . therefore , if @xmath96 is a direction of a cylinder decomposition of @xmath4 as above , then @xmath215 has the same dimension as @xmath216 . note that vertical saddle connection lengths tend to zero as @xmath128 ; indeed ( having begun with vertical saddle connections ) , for @xmath126 sufficiently large , the shortest saddle connections for @xmath123 are in the vertical direction . but then , @xmath123 leaves every compact subset of @xmath0 . hence , as @xmath128 , @xmath123 penetrates into an end of @xmath0 , see say @xcite . in @xcite , we studied an invariant of the geometry at a cusp of a teichmller complex geodesic the multiplicities and relative lengths of saddle connection vectors . we use the following elementary notion in a similar manner . if @xmath95 lies interior to the sides of a cylinder , then the _ splitting ratio _ of @xmath95 is the ratio of the distance from the left hand side of the cylinder to @xmath95 to the width of the cylinder . lemma 4 of @xcite can be reworded as the following . [ irrsplit ] let @xmath16 be a translation surface . if @xmath96 is a completely periodic direction and @xmath161 has irrational splitting ratio for some cylinder of the cylinder decomposition in the direction @xmath96 , then @xmath96 is a fat direction for @xmath217 . [ splitratiofixed ] splitting ratios are @xmath10 invariant . * proof * let @xmath4 be a translation surface , and suppose that @xmath218 splits some cylinder with ratio @xmath96 . for ease of discussion , we may and do assume that the cylinder , say @xmath173 , is a vertical cylinder . suppose that @xmath30 . we develop @xmath173 to @xmath27 in the standard fashion . the ratio @xmath96 is the ratio of the length of the horizontal line segment passing from the left hand side of @xmath173 to @xmath95 to the width of the cylinder . of course , this second length is the length of the horizontal line segment joining the sides of @xmath173 and passing through @xmath95 . the image of the cylinder by @xmath31 is again a cylinder , say @xmath219 , with obvious development . the horizontal line segment crossing @xmath173 upon which @xmath95 lies has as its image some segment in @xmath219 ; since @xmath31 acts linearly , the ratio of the length of the initial segment of this image from the appropriate side of the cylinder to the image of @xmath95 to the length the segment is again @xmath96 . draw the line segment perpendicular to the sides of the cylinder which passes through the image of @xmath95 . due to the equality of the opposite angles so defined , ratios of corresponding sides of the triangles so formed are equal . a trivial cancellation , see figure 2 , shows that the splitting ratio is indeed preserved . figure 2 . preserving splitting ratios . we have the immediate consequence . [ affpressplitt ] affine diffeomorphisms preserve splitting ratios of cylinders . we distinguish infinite ends by use of the following elementary well - known result in the theory of continued fractions . we use a flat notation for continued fractions ; thus , for example @xmath220 = a+ \dfrac{1}{b+ \dfrac{1}{c + \alpha}}\,$ ] . we label the partial quotients of a real number @xmath221 $ ] by @xmath222 , @xmath223 , @xmath187 . these give rise to the convergents to @xmath96 : @xmath224 $ ] . let @xmath225 denote the fractional part of a real number : @xmath226 . [ rotateirrationals ] suppose that @xmath173 and @xmath219 are cylinders lying in tranverse directions which have non - trivial intersection . if @xmath95 irrationally splits @xmath173 , then the orbit of @xmath95 under the affine dehn twist of @xmath173 meets the cylinder @xmath219 so as to give infinitely many distinct irrational splitting ratios . * proof * let @xmath96 denote the splitting ratio of @xmath95 with respect to @xmath173 . by lemma [ splitratiofixed ] , we may normalize such that @xmath173 is in the vertical direction and @xmath219 is in the horizontal direction . let @xmath227 be a rectangle in which our two cylinders meet . let us develop the cylinders such that @xmath95 lies on the @xmath228-axis , and such there are @xmath229 and @xmath230 such that @xmath227 is bounded below by the line @xmath231 and above by @xmath232 . let us denote by @xmath113 the vertical height @xmath173 . thus in these coordinates , the orbit of @xmath95 by the vertical dehn twist has heights @xmath233 with @xmath234 . but , this is exactly the set of @xmath235 . it suffices for us to show that there are infinitely many distinct ratios @xmath236 which are irrational , and lie between @xmath229 and @xmath237 . for completeness sake , we show this elementary , presumably well - known , result . let @xmath238 . now , if @xmath239 is a rational multiple of @xmath240 , then it is necessary that @xmath241 for @xmath242 integers , with @xmath168 , @xmath243 nonzero . suppose now that @xmath96 is of the form @xmath244 , and there is an @xmath185 such that @xmath245 for nonzero integers @xmath66 and @xmath105 . we find : @xmath246 since @xmath96 is irrational , it is not linearly related to a rational number . hence , in particular , we must have that @xmath247 . that is , @xmath248 ; but then @xmath249 . this can only occur for a finite number of @xmath185 . since in the worst case scenario there are only a finite number of orbit elements to avoid , and the orbit of @xmath95 is dense in the circle on the vertical cylinder , we conclude that there are always infinitely many elements of the orbit which irrationally split the horizontal cylinder . [ connptsforever ] the affine diffeomorphisms of a translation surface preserve the set of connection points . * proof * suppose that @xmath250 is an affine diffeomorphism and @xmath218 is a connection point . given a separatrix passing from a singularity to @xmath251 , we apply @xmath252 to find a separatrix passing from some singularity through @xmath95 . since @xmath95 is a connection point , this separatrix extends to a saddle connection . but , the image by @xmath40 of this extended separatrix extends our initial separatrix passing through @xmath251 . since @xmath40 sends saddle connections to saddle connections , we are done . we use the invariant studied in @xcite to distinguish cusps . [ infcusps ] let @xmath4 be a translation surface which is a finite ramified covering of a veech surface @xmath16 , with ramification above some non - periodic connection point and possibly the singularities of @xmath16 . then the teichmller complex geodesic @xmath13 has infinitely many cusps . * proof * the translation structure of @xmath4 arises as the pull - back by a ramified holomorphic map @xmath253 of riemann surfaces , with @xmath254 . let @xmath95 denote the unique non - periodic connection point in the branch locus . the veech group @xmath14 is commensurable to the subgroup of @xmath17 formed by the derivatives of the stabilizer in @xmath255 of @xmath95 . by theorem 2 of @xcite , the image of the non - periodic @xmath95 under the action of the affine diffeomorphism group is dense on @xmath46 . let @xmath256 be the length of the shortest saddle connection of @xmath46 . given @xmath257 , we can find an image of @xmath95 under @xmath258 , say @xmath259 , which is within @xmath260 of one of the singularities , say @xmath176 . there is a unique separatrix of length less than @xmath260 connecting @xmath259 and @xmath176 . by lemma [ connptsforever ] , we can extend this separatrix to a saddle connection , passing from @xmath176 to a singularity @xmath261 . ( of course , @xmath261 may equal @xmath176 . ) let @xmath40 be an affine diffeomorphism mapping @xmath259 to @xmath95 . the image by @xmath40 of the above saddle connection is a saddle connection passing through @xmath95 . since @xmath46 is a veech surface , the direction of any saddle connection is a parabolic direction . since @xmath95 lies on this saddle connection , @xmath95 is preserved by the dehn twist of either of the cylinders ( in this given direction ) upon which @xmath95 lies . thus this direction is a parabolic direction for @xmath23 . consider a single lift to @xmath23 of the saddle connection that we have just determined . this lift is the union of two saddle connections on @xmath23 : the first joins a lift of @xmath262 to a lift of @xmath95 , the second lies on another branch of the ( ramified ) covering space @xmath2 , joining the lift of @xmath95 to a lift of @xmath263 . ( the connection vectors of these saddle connections are parallel ! ) now , the lengths of these saddle connections on @xmath23 are equal to the lengths of the corresponding segments on @xmath46 . but , the affine diffeomorphism @xmath40 preserves _ ratios _ of segment lengths ; thus the ratio of the lengths of the saddle connections on @xmath23 equals the ratio of the length along the separatrix from @xmath176 to @xmath259 to the length along its continuation from @xmath259 to @xmath261 . we find that this ratio is less than @xmath264 . our construction thus allows us to find pairs of parallel connection vectors , in parabolic directions , of @xmath23 that are of arbitrarily small ratio . however , the number of connection vectors in any given direction is finite . furthermore , the ratios of the various connection vectors lying in a parabolic direction is invariant under the action of @xmath36 see @xcite . we conclude that @xmath4 has infinitely many inequivalent parabolic directions . but , each of these parabolic directions determines a distinct cusp of @xmath19 . since the teichmller complex geodesic @xmath13 is a generically injective image of this quotient , it too has infinitely many cusps . we now give the proof of theorem [ inffatboys ] . thus , we continue with @xmath23 a ramified covering of @xmath46 as above . the periodic directions on @xmath23 are exactly those given by the periodic , and hence parabolic , directions of @xmath46 . we show that @xmath23 has many directions which are periodic , but not parabolic . [ infends ] let @xmath4 be a translation surface which is a finite ramified covering of a veech surface @xmath16 , with ramification above some non - periodic connection point and possibly the singularities of @xmath16 . then the teichmller complex geodesic @xmath13 has infinitely many infinite ends . * proof * let @xmath95 be the branch point , thus @xmath95 is a non - periodic connection point of @xmath16 . there is thus some parabolic direction of @xmath46 for which @xmath95 irrationally splits a cylinder . without loss of generality , we may normalize so that @xmath95 irrationally splits a vertical cylinder and such that the horizontal is a parabolic direction for @xmath16 . consider the vertical cylinder @xmath173 in which @xmath95 lies , and any horizontal cylinder @xmath219 which meets @xmath173 . by lemma [ rotateirrationals ] , the set of splitting ratios of @xmath219 determined by the orbit of @xmath95 under the vertical affine dehn twist includes an infinite set of irrational numbers . there is some finite power of this dehn twist which is induced by an element @xmath40 of the affine diffeomorphism group of @xmath16 ; the orbit of @xmath95 under the subgroup generated by this multiply - composed dehn twist still provides an infinite set of distinct irrational splitting ratios . for each image @xmath265 of this latter type , apply the corresponding inverse @xmath266 . now , @xmath267 ; @xmath268 is a cylinder in a corresponding parabolic direction for @xmath269 ; and , @xmath95 splits @xmath270 with the ratio that @xmath259 splits @xmath219 . of course , given any direction , @xmath95 lies interior to at most one cylinder of the possible decomposition in that direction ; thus the splitting ratio defined by @xmath95 in that direction is well - defined . furthermore , the splitting ratio that @xmath95 gives is constant on directions equivalent under the stabilizer of @xmath95 amongst the affine diffeomorphisms of @xmath16 . since the @xmath104-orbit of @xmath95 gives infinitely many distinct irrational splitting ratios , we conclude that there are infinitely many fat directions that are inequivalent under the stabilizer of @xmath95 . therefore , @xmath271 has infinitely many infinite ends . but , @xmath14 is commensurable with @xmath272 ; hence , @xmath271 and @xmath19 admit a common finite degree ( possibly ramified ) covering . since such coverings are balanced with respect to infinite ends , @xmath19 also has infinitely many infinite ends . but , the teichmller complex geodesic @xmath13 is a generically injective image of @xmath19 , it too must have infinitely many infinite ends . we next observe that the above proof actually shows a bit more . [ fatlimits ] with notation as above , let @xmath96 be a fat direction of @xmath4 . then @xmath96 is the limit of @xmath14-inequivalent fat directions of @xmath4 . * proof * from construction , @xmath96 is a parabolic direction for @xmath16 and the non - periodic connection branch point @xmath95 irrationally splits a cylinder @xmath173 of @xmath16 in this direction . recall that some power of the dehn twist in this cylinder is induced by an element @xmath40 of the affine diffeomorphism group of @xmath16 . choose any parabolic direction for @xmath16 that is transverse to @xmath96 . there is some cylinder @xmath219 in this direction that meets @xmath173 . the previous proof shows that there is a subsequence of @xmath270 giving cylinders which are split by @xmath95 with distinct irrational splitting ratios . since @xmath40 is parabolic , preserving the direction @xmath96 , the direction of the cylinder @xmath270 tends to @xmath96 as @xmath185 tends to infinity . thus , @xmath96 is the limit of these inequivalent fat directions for @xmath217 . finally , commensurability of groups allows us to choose a subsequence of these directions consisting of inequivalent fat directions for @xmath4 . we now suppose @xmath273 is such that @xmath4 satisfies the hypotheses of theorem [ inffatboys ] ; thus : @xmath16 is a veech surface @xmath16 with distinguished non - periodic connection point @xmath95 , @xmath274 , and the branch locus of @xmath34 is @xmath95 and possibly singularities of @xmath16 . the case of ramification above multiple non - periodic connection points is similar ; for clarity of exposition , we content ourselves with this main case . by construction , @xmath36 is commensurable to @xmath275 . for clarity of discussion , _ suppose further _ that the marked translation surface @xmath276 has some connection vector which is uniquely realized . this is easily arranged by choice of @xmath95 . with this added hypothesis , proposition 5.6 of @xcite applies to show that @xmath277 . [ everyaffhasbldcover ] let @xmath278 be a balanced covering of translation surfaces as above . for each @xmath279 , there is some @xmath280 with @xmath281 such that @xmath282 is a balanced translation covering . * proof * we take @xmath283 such that @xmath40 gives an isomorphism of translation surfaces of @xmath284 with @xmath16 . let us now excise the singularities from @xmath285 to form @xmath286 , and excise both @xmath251 and the singularities from @xmath46 to form @xmath287 . we then find that @xmath288 is a covering of topological surfaces . letting @xmath287 have its induced translation structure from @xmath46 , we can pull this back by @xmath289 to induce a possibly new structure on @xmath290 ; we can complete this flat metric to form some @xmath291 . by construction , @xmath289 gives a balanced translation covering by @xmath280 . the translation structure of @xmath291 is compatible with a unique complex structure ; this complex structure is also ( after possibly extending over removable singularities ) the completion of the pull - back of the complex structure on @xmath287 , which in turn is simply the restriction of the complex structure on @xmath46 . but , @xmath40 does not change complex structure , hence the pulled - back complex structure is exactly that of @xmath292 . therefore , @xmath293 . [ lineupcovs ] let @xmath278 be a balanced covering of translation surfaces as above . then for all @xmath294 , there is some @xmath280 with @xmath295 and some @xmath296 such that @xmath282 is a balanced translation covering . * proof * for each @xmath297 , the function @xmath34 itself gives a balanced translation covering @xmath298 . there is some @xmath299 giving an isomorphism of translation surfaces of @xmath284 with @xmath16 . we now proceed as in the previous proof . [ ptgivescover ] with notation as above , let @xmath300 $ ] . there is a one - to-@xmath200 correspondence between elements of the @xmath301-orbit of @xmath95 and balanced coverings @xmath302 such that @xmath293 . moreover , if @xmath291 and @xmath303 correspond to @xmath304 , then @xmath305 . now , @xmath40 is unique modulo the stabilizer of @xmath95 . replacing @xmath40 by @xmath308 , with @xmath309 of derivative @xmath310 , upon proceeding as in the proof of lemma [ everyaffhasbldcover ] , we find a balanced covering of translation surfaces @xmath311 , with @xmath312 such that @xmath313 of course , if @xmath314 , then @xmath315 reduces to the original mapping @xmath34 ; thus after the further composition with @xmath40 , we then recover the same balanced covering as with @xmath40 by itself . however , if @xmath316 is a nontrivial coset representative for @xmath36 in @xmath275 , then the reduction to the original mapping can not be achieved : @xmath317 gives a distinct balanced covering of @xmath318 . since @xmath40 is invertible , @xmath319 pull - backs a translation structure from @xmath320 that is distinct from that pulled - back by @xmath321 . in any case , each balanced covering @xmath322 announces a @xmath291 in the @xmath10-orbit of the translation structure pulled - back by @xmath289 . furthermore , we certainly have that @xmath293 . now recall that the number of balanced coverings of @xmath323 is finite and @xmath324 acts so as to permute them , see say @xcite . but , the pulled - back complex structures in such an orbit are the same : @xmath305 . the ( coarse ) moduli space of genus @xmath1 curves with @xmath66 marked points is denoted by @xmath325 . the forgetful map , given by erasing the final marked point , is @xmath326 ; this is also known as the _ universal curve_. see , say , @xcite and @xcite for these notions . if @xmath280 is of genus @xmath1 and has @xmath66 singularities , then since the @xmath10-action preserves the singularities of @xmath280 , the lift of @xmath327 to @xmath325 is a biholomorphic equivalence . given a balanced covering @xmath328 there is a relationship between @xmath13 and @xmath329 ; when @xmath330 is the union of some singularities and a singleton , then the relationship can be expressed in terms of the universal curve over @xmath331 . one could discuss the current setting in terms of the generalized hurwitz spaces as presented in wewer s dissertation @xcite . we give an elementary discussion here ; one may turn to @xcite for a deeper discussion of correspondences between moduli spaces induced by ramified coverings . we suppose that the genus @xmath113 translation surface @xmath16 has @xmath66 singularities , and consider the universal curve over the lift of @xmath331 to @xmath332 ; the roving @xmath333-st point accounts for the non - singular branch point in coverings related to the initial ramified covering of @xmath44 by @xmath2 . * proof * for general @xmath30 , we can replace @xmath335 by @xmath336 , and apply proposition [ ptgivescover ] to find the @xmath259 in the @xmath337-orbit of @xmath95 that is uniquely associated to @xmath78 . now , since @xmath338 is conjugated by @xmath31 to @xmath339 , it follows that when @xmath340 are equivalent modulo @xmath341 then the same point of the underlying surface is associated to both @xmath78 and @xmath342 . of course , for each @xmath343 , the set of these points @xmath259 is dense on the complement to the @xmath66 singularies . hence , we have a map from the @xmath10-orbit of @xmath4 to a dense subset , @xmath344 , of @xmath173 the universal curve restricted to @xmath331 , see figure 3 . [ endsalign ] with notation as above , suppose that @xmath136 is an end of @xmath157 . then the image of @xmath136 in @xmath13 lies within the image of a subset @xmath347 of @xmath344 fibering above a cusp of @xmath348 . more precisely , if @xmath136 is a cusp , then @xmath347 is homeomorphic to @xmath349 . if @xmath136 is an infinite end , then @xmath347 is homeomorphic to @xmath350 , with @xmath351 a dense subset of @xmath352 . * proof * suppose first that @xmath136 is a cusp . let @xmath353 be the direction on @xmath23 corresponding to @xmath136 as in lemma [ fatptsareinfenddir ] of [ endhypsurf ] ; normalize so that this is the vertical direction . of course , @xmath96 is a parabolic direction : @xmath4 admits a cylinder composition in this direction , with all cylinders having commensurable inverse moduli . but , the balanced covering @xmath354 sends cylinders to cylinders . hence , @xmath96 is also a parabolic direction for @xmath16 , and is such that @xmath95 splits no cylinder irrationally . that is , either @xmath95 lies on the boundary of some cylinder of the decomposition of @xmath16 in this direction , or else splits some cylinder in a rational manner . if @xmath95 lies on the boundary of cylinders , then there is equality of sets of moduli of cylinders of the @xmath4 and of moduli of cylinders of the @xmath16-decomposition in this direction . hence , the minimum value , say @xmath355 , of @xmath356 such that @xmath357 belongs to @xmath341 is also the minimum value such that this is an element of @xmath36 . again since @xmath95 lies on the boundary of @xmath16-cylinders , the action of each @xmath358 fixes @xmath95 as a point on the underlying topological surface . hence , in accordance with the proof of theorem [ fiberunivcurve ] , @xmath359 with @xmath360 $ ] describes a circle in @xmath344 . there is some @xmath361 such that the horodisk of index @xmath362 projects to the cuspidal region of @xmath136 ; similarly , there is some @xmath363 such that the horodisk of index @xmath363 for @xmath16 projects to the cuspidal region of the cusp of @xmath331 corresponding to the parabolic element @xmath364 . let @xmath365 . we find that the image in @xmath13 of the horodisk of index @xmath130 of @xmath136 meets the image of the region @xmath366\,,\ , t > t\,\ } \subset \mathcal c^{0}$ ] . we indeed have that @xmath347 is homeomorphic to @xmath349 . if @xmath95 rationally splits some cylinder of @xmath16 , then it is still the case that the single commensurability class of the inverse moduli of the cylinder decompositions of @xmath4 and of @xmath16 agree . however , the parabolic element @xmath367 of @xmath341 as above is not an element of @xmath36 ; there is some finite power of this element , thus some @xmath368 with @xmath369 minimal , that does belong to @xmath36 . we find that the horocycle @xmath370 gives an @xmath185-fold cover of the corresponding horcycle of @xmath46 . we again can let @xmath366\,,\ , t > t\,\ } \subset \mathcal c^{0}$ ] . finally , suppose that @xmath136 is an infinite end . we continue with the normalization that the vertical direction on @xmath4 corresponds to @xmath136 . we have a decomposition into cylinders for @xmath4 , which projects to a decomposition of cylinders for @xmath217 . there are at least two commensurability classes of inverse moduli of cylinders for the @xmath4-decomposition ( for otherwise @xmath136 would be a cusp ) for no @xmath121 is @xmath371 . now , @xmath46 is a veech surface , hence this direction of a cylinder decomposition must be parabolic ; there is thus a sole commensurability class of inverse moduli for this cylinder decomposition , and we can define @xmath355 as above . since the balanced covering of @xmath217 has at least two such classes , @xmath95 must irrationally split some cylinder . but , this introduces at most one new commensurability class ; we find that @xmath4 has exactly two commensurability classes of inverse moduli of cylinders for its cylinder decomposition . of more importance , the orbit of @xmath95 under powers of @xmath367 is dense in the vertical circle of @xmath16 upon which @xmath95 lies . with @xmath130 as above , we now let @xmath372\,,\ , t > t\,\ } \subset \mathcal c^{0}$ ] . this is is homeomorphic to @xmath350 , with @xmath351 a dense subset of @xmath352 . theorem [ thmbigcusp ] now follows each end of @xmath13 must be covered by some end of @xmath19 , but each of these is given by one of finitely many ends : the image of the fibering over one of the finitely many ends of @xmath331 . p. dbes , _ mthodes topologiques et analytiques en thorie inverse de galois : thorme dexistence de riemann _ , pp . 2741 , in : arithmtique des revtements algbriques - actes du colloque de saint - tienne , smin . et congr . 5 , soc france , paris , 2001 . e. gutkin , p. hubert and t. a. schmidt , _ affine diffeomorphisms of translation surfaces : periodic points , fuchsian groups , and arithmeticity _ , ann . sci . cole norm . , 4e ser . , t. 36 ( 2003 ) , 847 - 866 . d. sullivan , _ on the ergodic theory at infinity of an arbitrary discrete group of hyperbolic motions _ , in : riemann surfaces and related topics : proceedings of the 1978 stony brook conference ( state univ . new york , stony brook , n.y . , 1978 ) , pp . 465496 , ann . of math . stud . , 97 , princeton univ . press , princeton , n.j . , 1981 .

mark correlations provide a systematic approach to look at objects both distributed in space and bearing intrinsic information , for instance on physical properties . the interplay of the objects properties ( marks ) with the spatial clustering is of vivid interest for many applications ; are , e.g. , galaxies with high luminosities more strongly clustered than dim ones ? do neighbored pores in a sandstone have similar sizes ? how does the shape of impact craters on a planet depend on the geological surface properties ? in this article , we give an introduction into the appropriate mathematical framework to deal with such questions , i.e. the theory of marked point processes . after having clarified the notion of segregation effects , we define universal test quantities applicable to realizations of a marked point processes . we show their power using concrete data sets in analyzing the luminosity - dependence of the galaxy clustering , the alignment of dark matter halos in gravitational-body simulations , the morphology- and diameter - dependence of the martian crater distribution and the size correlations of pores in sandstone . in order to understand our data in more detail , we discuss the boolean depletion model , the random field model and the cox random field model . the first model describes depletion effects in the distribution of martian craters and pores in sandstone , whereas the last one accounts at least qualitatively for the observed luminosity - dependence of the galaxy clustering . draft

physics0201069

observations of spatial patterns at various length scales frequently are the only point where the physical world meets theoretical models . in many cases these patterns consist of a number of comparable objects distributed in space such as pores in a sandstone , or craters on the surface of a planet . another example is given in figure [ fig : kerscher_galaxies - circles ] , where we display the galaxy distribution as traced by a recent galaxy catalogue . the galaxies are represented as circles centered at their positions , whereas the size of the circles mirrors the luminosity of a galaxy . in order to test to which extent theoretical predictions fit the empirically found structures of that type , one has to rely on quantitative measures describing the physical information . since theoretical models mostly do not try to explain the structures individually , but rather predict some of their generic properties , one has to adopt a _ statistical point of view _ and to interpret the data as a realization of a random process . in a first step one often confines oneself to the spatial distribution of the objects constituting the patterns and investigates their clustering thereby thinking of it as a realization of a _ point process_. assuming that perspective , however , one neglects a possible linkage between the spatial clustering and the intrinsic properties of the objects . for instance , there are strong indications that the clustering of galaxies depends on their luminosity as well as on their morphological type . considering figure [ fig : kerscher_galaxies - circles ] , one might infer that luminous galaxies are more strongly correlated than dim ones . effects like that are referred to as _ mark segregation _ and provide insight into the generation and interactions of , e.g. , galaxies or other objects under consideration . the appropriate statistical framework to describe the relation between the spatial distribution of physical objects and their inner properties are _ marked point processes _ , where discrete , scalar- , or vector - valued marks are attached to the random points . + in this contribution we outline how to describe marked point processes ; along that line we discuss two notions of independence ( section [ sec : kerscher_basic ] ) and define corresponding statistics that allow us to quantify possible dependencies . after having shown that some empirical data sets show significant signals of mark segregation ( section[sec : kerscher_data ] ) , we turn to analytical models , both motivated by mathematical and physical considerations ( section [ sec : kerscher_models ] ) . + contact distribution functions as presented in the contribution by d. hug et al . in this volume are an alternative technique to measure and statistically quantify distances which finally can be used to relate physical properties to spatial structures . mark correlation functions are useful to quantify molecular orientations in liquid crystals ( see the contribution by f. schmid and n. h. phuong in this volume ) or in self - assembling amphiphilic systems ( see the contribution by u. s. schwarz and g. gompper in this volume ) . but also to study anisotropies in composite or porous materials , which are essential for elastic and transport properties ( see the contributions by d. jeulin , c. arns et al . and vogel in this volume ) , mark correlations may be relevant . the empirical data the positions @xmath1 of some objects together with their intrinsic properties @xmath2 are interpreted as a realization of a marked point process @xmath3 ( stoyan , kendall and mecke , 1995 ) . for simplicity we restrict ourselves to homogeneous and isotropic processes . + the hierarchy of joint probability densities provides a suitable tool to describe the stochastic properties of a marked point process . thus , let @xmath4 denote the probability density of finding a point at @xmath5 with a mark @xmath6 . for a homogeneous process this splits into @xmath7 where @xmath8 denotes the mean number density of points in space and @xmath9 is the probability density of finding the mark @xmath6 on an arbitrary point . later on we need moments of this mark distribution ; for real - valued marks the @xmath10th - moment of the mark - distribution is defined as @xmath11 the mark variance is @xmath12 . + accordingly , @xmath13 quantifies the probability density to find two points at @xmath14 and @xmath15 with marks @xmath16 and @xmath17 , respectively ( for second - order theory of marked point processes see @xcite ) . it effectively depends only on @xmath16 , @xmath17 , and the pair separation @xmath18 for a homogeneous and isotropic process . two - point properties certainly are the simplest non - trivial quantities for homogeneous random processes , but it may be necessary to move on to higher correlations in order to discriminate between certain models . in the following we will discuss two notions of independence , which may arise for marked point patterns . for this , consider two renaissance families , call them the sforza and the gonzaga . they used to build castles spread out more or less homogeneously over italy . in order to describe this example in terms of a marked point process , we consider the locations of the castles as points on a map of italy , and treat a castle s owner as a discrete mark , @xmath19 and @xmath20 , respectively . there are many ways how the castles can be built and related to each other . [ [ independent - sub - point - processes ] ] independent sub - point processes : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for example , the sforza may build their castles regardless of the gonzaga castles . in that case the probability of finding a sforza castle at @xmath14 and a gonzaga castle at @xmath15 factorizes into two one - point probabilities and we can think of the sforza and the gonzaga castles as uncorrelated sub - point processes . in the language of marked point processes this means , e.g. , that @xmath21 for any @xmath22 . if all the joint @xmath23-point densities factorize into a product of @xmath24-point densities of one type each , then we speak of _ independent sub - point processes_. dependent sub - point processes indicate _ interactions _ between points of different marks ; for instance , the gonzaga may build their castles close to the sforza ones in order to avoid that a region becomes dominated by the other family s castles . [ [ mark - independent - clustering ] ] mark - independent clustering : + + + + + + + + + + + + + + + + + + + + + + + + + + + + a second type of independence refers to the question whether the different families have different styles to plan their castles . for instance , the gonzaga may distribute their castles in a grid - like manner over italy , whereas the sforza may incline to build a second castle close to each castle they own . rather than asking whether two sub - point processes ( namely the gonzaga and the sforza castles , respectively ) are independent ( `` independent sub - point processes '' ) , we are now discussing whether they are _ different _ as regards their statistical clustering properties . any such difference means that the clustering _ depends _ on the intrinsic mark of a point . + whenever the two - point probability density of finding two objects at @xmath14 and @xmath15 depends on the objects intrinsic properties we speak of _ mark - dependent clustering_. it is useful to rephrase this statement by using bayes theorem and the conditional mark probability density @xmath25 in case the spatial product density @xmath26 does not vanish . @xmath27 is the probability density of finding the marks @xmath16 and @xmath17 on objects located at @xmath14 and @xmath15 , given that there are objects at these points . clearly , @xmath27 depends only on the pair separation @xmath28 for homogeneous and isotropic point processes . we speak of _ mark - independent _ clustering , if @xmath29 factorizes @xmath30 and thus does not depend on the pair separation . that means that regarding their marks , pairs with a separation @xmath31 are not different from any other pairs . on the contrary , mark - dependent clustering or _ mark segregation _ implies that the marks on certain pairs show deviations from the global mark distribution . + in order to distinguish between both sorts of independencies , let us consider the case where we are given a map of italy only showing the gonzaga castles . if the distribution of castles in italy can be understood as consisting of independent sub - point processes , we can not infer anything about the sforza castles from the gonzaga ones . however , if @xmath32 , sforza castles are likely to be found close to gonzaga ones . here , @xmath33 and @xmath34 are the probabilities that a castle belongs to the sforza or gonzaga family . if , on the other hand , mark - independent clustering applies , typical clustering properties such as the spatial clustering strength are equal for both castle distributions , and the gonzaga castles are in the statistical sense already representative of the whole castle distribution in italy . that means in particular that , if the gonzaga castles are clustered , so are the sforza ones . before we turn to applications , we have to develop practical test quantities in order to test for segregation effects in real data and to describe them in more detail . to investigate correlations between sub - point processes , suitably extended nearest neighbor distribution functions or @xmath35-functions have been employed @xcite . also the ( conditional ) cross - correlation functions can be used ( see eq . [ eq : kerscher_cond - crosscorr ] ) , for a further test see @xcite , p. 302 . here we consider a multivariate extension of the @xmath36-function @xcite , as suggested by @xcite . + for this , consider the nearest neighbor s distance distribution from an object with mark @xmath2 to other objects with mark @xmath37 , @xmath38 ( `` @xmath39 to @xmath40 '' , for details see @xcite ) . let @xmath41 denote the distribution of the nearest neighbor s distance from an object of type @xmath39 to any other object ( denoted by @xmath42 ) . finally , @xmath43 is the nearest neighbor distribution of all points . similar extensions of the empty space function are possible , too . let @xmath44 denote the distribution of the nearest @xmath39-object s distance from an arbitrary position , whereas @xmath45 is the nearest object s distance distribution from a random point in space to any object in the sample . we consider the following quantities : @xmath46 they are defined whenever @xmath47 . if two sub - point processes , defined by marks @xmath48 , are independent then one gets @xcite @xmath49 note , that the @xmath50 depend on higher - order correlations functions , similar to the @xmath36-function @xcite . suitable estimators for these @xmath36-functions are derived from estimators of the @xmath51 and @xmath20-functions @xcite . in order to quantify the mark - dependent clustering or to look for the mark segregation , it proves useful to integrate the conditional probability density @xmath29 over the marks weighting with a test function @xmath52 @xcite . this procedure reduces the number of variables and leaves us with the weighted pair average : @xmath53 the choice of an appropriate weight - function depends on whether the marks are non - quantitative labels or continuous physical quantities . 1 . for labels only combinations of indicator - functions are possible , the integral degenerates into a sum over the labels . supposed the marks of our objects belong to classes labelled with @xmath54 , the conditional cross - correlation functions are given by @xmath55 with the kronecker @xmath56 for @xmath57 and zero otherwise . mark segregation is indicated by @xmath58 for @xmath59 and @xmath60 , where @xmath61 denotes the number density of points with label @xmath39 . the @xmath62 are cross - correlation functions under the _ condition _ that two points are separated by a distance of @xmath31 ( compare @xcite , p. 264 , for applications see the martian crater distribution studied in sect . [ sec : kerscher_martian ] and figure [ fig : kerscher_mars - markcorr ] in particular ) . 2 . for positive real - valued marks @xmath6 , the following pair averages prove to be powerful and distinctive @xcite : + 1 . one of the most simplest weights to be used is the mean mark : @xmath63 quantifies the deviation of the mean mark on pairs with separation @xmath31 from the overall mean mark @xmath64 . a @xmath65 indicates mark segregation for point pairs with a separation @xmath31 , specifically their mean mark is then larger than the overall mark average . + closely related is stoyan s @xmath66 function using the squared geometric mean of the marks as a weight @xcite @xmath67 2 . accordingly , higher moments of the marks may be used to quantify mark segregation , like the mark fluctuations @xmath68 or the mark - variogram @xcite : @xmath69 3 . the mark covariance @xcite is @xmath70 mark segregation can be detected by looking whether @xmath71 differs from zero . a @xmath71 larger than zero , e.g. , indicates that points with separation @xmath31 tend to have similar marks . sometimes the mark covariance is normalized by the fluctuations @xcite : @xmath72 . + these conditional mark correlation functions can be calculated from only three independent pair averages @xcite : @xmath73 , @xmath74 , and @xmath75 . thus the above mentioned characteristics are not independent , e.g.@xmath76 . + we apply these mark correlation functions to the galaxy distribution in section [ sec : kerscher_gal ] ( figure [ fig : kerscher_ssrs - lum ] ) , to martian craters in section [ sec : kerscher_martian ] ( figure [ fig : kerscher_mars - markcorr ] ) and to pores in sandstones considered in section [ sec : kerscher_sandstone ] . 3 . also vector - valued information @xmath77 , describing , e.g. , the orientation of an anisotropic object at position @xmath1 may be available . it is therefore interesting to consider vector marks such as done by @xcite who use a mark correlation function to quantify the alignment of vector marks . here we suggest three mark correlation functions quantifying geometrically different possibilities of an alignment . in order to ensure coordinate - independence of our descriptors , we focus on scalar combinations of the vector marks in using the scalar product @xmath78 and the cross product @xmath79 . different from the case of scalar marks , it is a non - trivial task to find a set of vector - mark correlation functions which contain all possible information ( at least up to a fixed order in mark space ) . we provide a systematic account of how to construct suitable vector - mark correlation functions in a complete and unique way for general dimensions in the appendix . + here we only cite the most important results . for that we need the distance vector between two points , @xmath80 , the normalized distance vector , @xmath81 , and the normalized vector mark : @xmath82 with @xmath83 . the following conditional mark correlation functions will be used to quantify alignment effects : 1 . @xmath84 quantifies the @xmath85lignment of the two vector marks @xmath86 and @xmath87 : @xmath88 it is proportional to the cosine of the angle between @xmath86 and @xmath87 . we normalize with the mean @xmath89 . for purely independent vector marks @xmath84 is zero , whereas @xmath90 means that the marks of pairs separated by @xmath31 tend to align parallel to each other . in some applications , e.g. for the orientations of ellipsoidal objects , the vector mark is only defined up to a sign , i.e. @xmath91 and @xmath92 mean actually the same . in this case the absolute value of the scalar product is useful : @xmath93 for uncorrelated random vectors we get @xmath94 . @xmath85 and @xmath95 can readily be generalized to any dimension @xmath96 , where we expect @xmath97 @xmath98 for uncorrelated random orientations . in two dimensions @xmath95 is proportional to @xmath99 as defined by @xcite . @xmath100 quantifies the @xmath101ilamentary alignment of the vectors @xmath86 and @xmath87 with respect to the line connecting both halo positions : @xmath102 @xmath100 is proportional to the cosine of the angle between @xmath86 and the distance vector @xmath103 connecting the points . for uncorrelated random vector marks , we expect again @xmath104 ; @xmath100 becomes larger than that , whenever the vector marks of the objects tend to point to objects separated by @xmath31 an example is provided by rod - like metallic grains in an electric field : they concentrate along the field lines and orient themselves parallel to the field lines . 3 . @xmath105 quantifies the @xmath106lanar alignment of the vectors and the distance vector . @xmath105 is proportional to the volume of the rhomb defined by @xmath86 , @xmath87 and @xmath103 : @xmath107 quite obviously , this quantity can not be generalized to arbitrary dimensions ; the deeper reason for that will become clear in the appendix . we get @xmath108 for randomly oriented vectors , whereas it is becoming larger for the case that @xmath87 is perpendicular to @xmath86 as well as to @xmath103 . + applications of vector marks can be found in section [ sec : kerscher_halos ] ( figure [ fig : kerscher_halos - orientation ] ) where we consider the orientation of dark matter halos in cosmological simulations . but one can think of other applications : mark correlation functions may serve as orientational order parameters in liquid crystals in order to discriminate between nemetic and smectic phases ( see the contribution by f. schmid and n. h. phuong in this volume ) . they can also quantify the local orientation and order in liquids such as the recently measured five - fold local symmetry found in liquid lead @xcite . as a further application one could try to measure the signature of hexatic phases in two - dimensional colloidal dispersions and in 2d melting scenarios occurring in experiments and simulations of hard - disk systems ( for a review on hard sphere models see @xcite . finally , the orientations of anisotropic channels in sandstone ( see the contribution by c. arns et al . in this volume ) are relevant for macroscopic transport properties , therefore their quantitative characterization in terms of mark correlation functions might be interesting . before we move on to applications a few general remarks are in order : first , the definition of these mark characteristics based on the conditional density @xmath109 leads to ambiguities at @xmath31 equal zero as discussed by @xcite , but there is no problem for @xmath110 . furthermore , suitable estimators for our test quantities are based on estimators for the usual two - point correlation function @xcite . + mark - dependent clustering can also be defined at any @xmath23-point level . mark - independent clustering at every order is called the random labelling property @xcite . mark correlation functions based on the @xmath23-point densities may be used . for discrete marks the multivariate @xmath36-functions ( see eq . ) are an interesting alternative , sensitive to higher - order correlations . the random labelling property then leads to the relation @xmath111 which may be used as a test @xcite . in many cases already the question whether one or the other type of dependence as outlined above applies to certain data sets is a controversial issue . in the following we will apply our test quantities to a couple of data sets in order to probe whether there is an interplay between some objects marks and their positions in space . other applications to biological , ecological , mineralogical , geological data can be found in @xcite . the distribution of galaxies in space shows a couple of interesting features and challenges theoretical models trying to understand cosmological structure formation ( see e.g.@xcite ) . there has been a long debate , whether and how strongly the clustering of galaxies depends on their luminosity and their morphological type ( see , e.g.@xcite ) . the methods which have been used so far to establish such claims were based on the spatial two - point correlation function ; it was estimated from different subsamples that were drawn from a catalogue and defined by morphology or luminosity . however , some authors claimed that the signal of luminosity segregation observed by others was a spurious effect , caused by inhomogeneities in the sample and an inadequate choice of the statistics @xcite . @xcite could show that methods based on the mark - correlation functions , as discussed in sect . [ sec : kerscher_mark - segregation ] , are not impaired by inhomogeneities , and found a clear signal of luminosity and morphology segregation . + in order to quantify segregation effects in the galaxy distribution we consider the southern sky redshift survey 2 ( ssrs 2 , @xcite ) , which maps a significant fraction of the sky and provides us with the angular sky positions , the distances ( determined via the redshifts ) , and some intrinsic properties of the galaxies such as their flux and their morphological type . as marks we consider either a galaxy s luminosity estimated from its distance and flux , or its morphological type . in the latter case we effectively divide our sample into early - type galaxies ( mainly elliptical galaxies ) and late - type galaxies ( mainly spirals ) . in order to analyze homogeneous samples , we focus on a volume - limited sample of @xmath112 depth equals roughly @xmath113 million light years . the number @xmath114 accounts for the uncertainty in the measured hubble constant and is about @xmath115 . volume - limited samples are defined by a limiting depth and a limiting luminosity . one considers only those galaxies which could have been observed if they were located at the limiting depth of the sample . ] @xcite . + in a first step we ask whether the early- and the late - type galaxies form independent sub - processes . in figure [ fig : kerscher_morph_ind ] we show @xmath116 as function of the distance @xmath31 being far away from the value of one . recalling eq . , we conclude that the morphological types of galaxies are not distributed independently on the sky . not surprisingly , the inequality @xmath117 indicates positive interactions between the galaxies of both morphological types ; indeed galaxies attract each other through gravity irrespective of their morphological types . + after having confirmed the presence of interactions between the different types of galaxies , we tackle the issue whether the clustering of galaxies is different for different galaxies . we consider the luminosities as marks ( see fig . [ fig : kerscher_galaxies - circles ] ) . in figure [ fig : kerscher_ssrs - lum ] we show some of the mark - weighted conditional correlation functions . already at first glance , they show evidence for luminosity segregation , relevant on scales up to @xmath118 . to strengthen our claims , we redistribute the luminosities of the galaxies within our sample randomly , holding the galaxy positions fixed . in that way we mimic a marked point process with the same spatial clustering and the same one - point distribution of the luminosities , but without luminosity segregation . comparing with the fluctuations around this null hypothesis , we see that the signal within the ssrs 2 is significant . + the details of the mark correlation functions provide some further insight into the segregation effects . the mean mark @xmath119 indicates that the luminous galaxies are more strongly clustered than the dim ones . our signal is scale - dependent and decreasing for higher pair separations . the stronger clustering of luminous galaxies is in agreement with earlier claims comparing the correlation amplitude of several volume - limited samples @xcite . + the @xmath120 being larger than the mark variance of the whole sample , @xmath121 , shows that on galaxy pairs with separations smaller than @xmath118 the luminosity fluctuations are enhanced . the fact that the mark segregation effect extends to scales of up to @xmath118 is interesting on its own . in particular , it indicates that galaxy clusters are not the only source of luminosity segregation , since typically galaxy clusters are of the size of @xmath122 . + the signal for the covariance @xmath71 , however , could be due to galaxy pairs inside clusters . it is relevant mainly on scales up to @xmath123 indicating that the luminosities on galaxy pairs with small separations tend to assume similar values . our results in part confirm claims by @xcite , who compared the correlation functions @xmath124 for different volume - limited subsamples and different luminosity classes of the ssrs 2 catalog ( see also @xcite ) . many structures found in the universe such as galaxies and galaxy clusters show anisotropic features . therefore one can assign orientations to them and ask whether these orientations are correlated and form coherent patterns . here we discuss a similar question on the base of numerical simulations of large scale structure ( e.g. , @xcite ) . + in such simulations the trajectories of massive particles are numerically integrated . these particles represent the dominant mass component in the universe , the dark matter . through gravitational instability high density peaks ( `` halos '' ) form in the distribution of the particles ; these halos are likely to be the places where galaxies originate . in the following we will report on alignment correlations between such halos @xcite , for a further application of mark correlation functions in this field see @xcite . + the halos used by @xcite stem from a @xmath0-body simulation in a periodic box with a side length of 500 . the initial and boundary conditions were fixed according to a @xmath125cdm cosmology ( for a discussion of cosmological models see @xcite ) . halos were identified using a friend - of - friends algorithm in the dark matter distribution . not all of the halos found were taken into account ; rather the mass range and the spatial number density of the selected halos were chosen to resemble the properties of observed galaxy clusters in the reflex catalogue @xcite . typically our halos show a prolate distribution of their dark matter particles . + for each halo the direction of the elongation is determined from the major axis of the mass - ellipsoid . this leads to a marked point set where the orientation @xmath77 is attached to each halo position @xmath1 as a vector mark with @xmath126 . details can be founds in @xcite . + in fig . [ fig : kerscher_halos - orientation ] the vector - mark correlation functions as defined in eqs . , , and are shown . since only the orientation of the mass ellipsoids can be determined , we use @xmath127 ( eq . [ eq : kerscher_def - vector - corr - absolute ] ) instead of @xmath84 . the signal in @xmath127 indicates that pairs of halos with a distance smaller than 30 show a tendency of parallel alignment of their orientations @xmath128 . the deviation from a pure random alignment is in the percent range but clearly outside the random fluctuations . the alignment of the halos orientations @xmath128 with the connecting vector @xmath103 quantified by @xmath100 is significantly stronger ; it is particularly interesting that this alignment effect extends to scales of about 100 . + in a qualitative picture this may be explained by halos aligned along the filaments of the large scale structure . indeed such filaments are prominent features found in the galaxy distribution @xcite and in @xmath0-body simulations @xcite , often with a length of up to 100 . the lowered @xmath105 indicates that the volume of the rhomboid given by @xmath128 and @xmath103 is reduced for halo pairs with a separation below 80 . already a preferred alignment of @xmath128 along @xmath103 leads to such a reduction , similar to a plane - like arrangement of @xmath129 . for the halo distribution the signal in @xmath105 seems to be dominated by the filamentary alignment . + the question whether there are non - trivial orientation patterns for galaxies or galaxy clusters has been discussed for a long time . @xcite reported a significant alignment of the observed galaxy clusters out to 100 . @xcite , however claimed that this effect is small and likely to be caused by systematics ; @xcite find no indication for alignment effects at all . subsequently several authors purported to have found signs of alignments in the galaxy and galaxy cluster distribution ( see e.g.@xcite ) . our fig . [ fig : kerscher_halos - orientation ] shows that from simulations significant large - scale correlations are to be expected in the orientations of galaxy clusters , in agreement with the results by @xcite . these results are also supported by a simulation study carried out by @xcite . let us now turn to another , still astrophysical , but significantly closer object : the mars ( see figure [ fig : kerscher_mars_eg ] ) . many planets surfaces display impact craters with diameters up to @xmath130 km and a broad range of inner morphologies . these craters are surrounded by ejecta forming different types of patterns . the craters and their ejecta are likely to be caused by asteroids and periodic comets crossing the planets orbits , falling down onto the planet s surface , and spreading some of the underlying surfaces material around the original impact crater . a variety of different crater morphologies and a wide range of ejecta patterns can be found . in principle , either the different impact objects ( especially their energies ) or the various surface types of the planet may explain the repertory of patterns observed . whereas the energy variations of impact objects do not cause any peculiarities in the spatial distribution of the craters ( apart from a possible latitude dependence ) , geographic inhomogeneities are expected to originate inhomogeneities in the craters morphological properties . + we try to answer the question for the ejecta patterns origin using data collected by @xcite who already found correlations between crater characteristics and the local surface type employing geologic maps of the mars . complementary to their approach , we investigate two - point properties without any reference to geologic mars maps . we restrict ourselves only to craters which have a diameter larger than @xmath131 km and whose ejecta pattern could be classified , ending up with @xmath132 craters spread out all over the martian surface . we use spherical distances for our analysis of pairs . + in a first step we divide the ejecta patterns into two broad classes consisting of either the simple patterns ( single and double lobe morphology , i.e. sl and dl in terms of the classification by @xcite ; we speak of `` simple craters '' ) or the remaining , more complex configurations ( `` complex craters '' ) . using our conditional cross correlation functions @xmath62 as defined in equation , we see a highly significant signal for mark correlations ( figure [ fig : kerscher_mars - crosscorr ] ) . at small separations , crater pairs are disproportionally built up of simple craters at the expense of cross correlations . this can be explained assuming that crater formation depends on the local surface type : if the simple craters are more frequent in certain geological environments than in others , then there are also more pairs of them to be found as far as one focuses on distances smaller than the typical scale of one geological surface type . cross pairs are suppressed , since typical pairs with small separations belong to one geological setting where the simple craters either dominate or do not . only a small , positive segregation signal occurs for the complex craters . hence our analysis indicates that the broad class of complex craters is distributed quite homogeneously over all of the geologies . on top of this there are probably simple craters , their frequency significantly depending on the surface type . + if the ejecta patterns were independent of the surface , no mark segregation could be observed ( other sources of mark segregation are unlikely , since the martian craters are a result of a long bombardment history diluting any eventual peculiar crater correlations ) . in this sense , the signal observed indicates a surface - dependence of crater formation . this result is remarkable , given that we did not use any geological information on the mars at all . the picture emerging could be described using the random field model , where a field ( here the surface type ) determines the mark of the points ( see below ) . + in a second step , we analyze the interplay between the craters diameters and their spatial clustering . now the diameter serves as a continuous mark . the results in figure [ fig : kerscher_mars - markcorr ] show a clear signal for mark segregation in @xmath133 and @xmath134 at small scales . the latter signals that pairs with separations in a broad range up to @xmath135 km tend to have similar diameters ; this is in agreement with the earlier picture : as @xcite showed , the simple craters are mostly small - sized . pairs with relatively small separations thus often stem from the same geological setting and therefore have similar diameters and similar morphological type . + also the signal of @xmath133 seems to support this picture : since the simple craters are more strongly clustered than the other ones and since they have smaller diameters , one could expect @xmath136 . as we shall see in sect . [ sec : kerscher_models ] , however , a @xmath137 contradicts the random field model ; therefore , the mark - dependence on the underlying surface type ( thought of as a random field ) can not account for the signal observed . thus , we have to look for an alternative explanation : it seems reasonable , that , whenever a crater is found somewhere , no other crater can be observed close nearby ( because an impact close to an existing crater will either destroy the old one or cover it with ejecta such that it is not likely to be observed as a crater ) . this results in a sort of effective hard - core repulsion . this repulsion should be larger for larger craters . thus , pairs with very small separations can only be formed by small craters , therefore @xmath136 for tiny @xmath31 . the scale beyond which @xmath138 should somehow be hidden within the crater diameter distribution . indeed , at about @xmath139 km the segregation vanishes , which is about twice the largest diameter in our sample . taking into account that the ejecta patterns extend beyond the crater , this seems to be a reasonable agreement . as shown in sect . [ sec : kerscher_boolean - depletion ] a model based on these consideration is able to produce such a depletion in the @xmath140 . this effect could also in turn explain part of the cross correlations observed earlier in figure [ fig : kerscher_mars - crosscorr ] . a similar effect is to be expected for the mark variance . close pairs are only accessible to craters with a smaller range of diameters ; therefore , their variance is diminished in comparison to the whole sample . however , an effect like this is barely visible in the data . + altogether , the crater distribution is dominated by two effects : the type of the ejecta pattern and the crater diameter depend on the surface , in addition , there is a sort of repulsion effect on small scales . now we turn to systems on smaller scales . sandstone is an example of a porous medium and has extensively been investigated , mainly because oil was found in the pore network of similar stones . in order to extract the oil from the stone one can try to wash it out using a second liquid , e.g. water . therefore , one tries to understand from a theoretical point of view , how the microscopic geometry of the pore network determines the macroscopic properties of such a multi - phase flow . especially the topology and connectivity of the microcaves and tunnels prove to be crucial for the flow properties at macroscopic scales . details are given , for instance , in the contributions by c. arns et al . , h .- j . vogel et al . and j. ohser in this volume . a sensible physical model , therefore , in the first place has to rely on a thorough description of the pore pattern . + one way to understand the pore network is to think of it as a union of simple geometrical bodies . following @xcite , one can identify distinct pores together with their position and their pore radius or extension . this allows us to understand the pore structure in terms of a marked point process , where the marks are the pore radii . + in the following , we consider three - dimensional data taken from one of the fontainbleau sandstone samples through synchrotron x - ray tomography . these data trace a @xmath141 mm diameter cylindrical core extracted from a block with bulk porosity @xmath142 , where the bulk porosity is the volume fraction occupied by the pores . a piece with @xmath143 mm length ( resulting in a @xmath144 mm@xmath145 volume ) of the core was imaged and tomographically reconstructed @xcite . further details of this sample are presented in the contribution by c. arns et al . in this volume . based on the reconstructed images the positions of pores and their radii were identified as described in @xcite . + in our results for the mark correlation functions a strong depletion of @xmath140 and @xmath146 is visible for @xmath147 m in fig . [ fig : kerscher_sand ] . this small - scale effect may be explained similarly to the martian craters : large pores are never found close to each others , since they have to be separated by at least the sum of their radii . the histogram of the pore radii in fig . [ fig : kerscher_sand - hist ] shows that most of the pores have radii smaller than 100@xmath148 m , and consequently this effect is confined to @xmath147 m . in sect . [ sec : kerscher_boolean - depletion ] we discuss the boolean depletion model which is based on this geometric constraints and is able to produce such a reduction in the @xmath140 . this purely geometric constraint also explains the reduced @xmath146 and increased covariance @xmath149 . for separations larger than @xmath150 m there is no signal from the covariance , but both @xmath140 and @xmath146 show a small increase out to @xmath151 m . this indicate that pairs of pores out to these separations tend to be larger in size and show slightly increased fluctuations . however , this effect is small ( of the order of @xmath152 ) and may be explained by the definition of the holes , which may lead to `` artificial small pores '' as `` bridges '' between larger ones . this hypothesis has to be tested using different hole definitions . in any case the main conclusion seems to be that apart from the depletion effect at small scales there are no other mark correlations . given the significant mark correlations found in various applications , one may ask how these signals can be understood in terms of stochastic models . a thorough understanding of course requires a physical modeling of the individual situation . there are , however , some generic models , which we will focus on in the following : in sect . [ sec : kerscher_boolean - depletion ] we introduce the boolean depletion model , which is able to explain some of the features observed in the distribution of craters and pores in sandstone . another generic model is the _ random field model _ where the marks of the points stem from an independent random field ( section [ sec : kerscher_rfm ] ) . in sect . [ sec : kerscher_cox - rf - model ] we generalize the idea behind the random field model further in order to get the _ cox random field model _ , which allows for correlations between the point set and the random field . other model classes and their applications are discussed by e.g.@xcite . in our analysis of the martian craters and the holes in sandstone , we found that for small separations only small craters , or small holes in the sandstone , could be found . we interpreted this as a pure geometric selection effect . the boolean depletion model is able to quantify this effect , but also shows further interesting features . + the starting point is the boolean model of overlapping spheres @xmath153 ( see also the contributions by c. arns et al . and d. hug in this volume as well as @xcite ) . for that , the spheres centers @xmath1 are generated randomly and independently , i.e. according to a poisson process of number density @xmath154 . the radii @xmath155 of the spheres are then chosen independently according to a distribution function @xmath156 , i.e. with probability density @xmath157 . the main idea behind the depletion is to delete spheres which are covered by other spheres . to make this procedure unique we remove only those spheres which are completely covered by a ( notably larger ) sphere . the positions and radii of the remaining spheres define a marked point process . note , that this depletion mechanism is minimal in the sense that a lot of overlapping spheres may remain . this boolean depletion model may be considered as the low - density limit of the well - known widom - rowlinson model , or ( more generally ) of non - additive hard sphere mixtures ( see @xcite ) . + the probability that a sphere of radius @xmath155 is not removed is then given by @xmath158 with the step function @xmath159 for @xmath160 and @xmath161 otherwise , and the volume of the @xmath96-dimensional unit ball @xmath162 ( @xmath163 , @xmath164 , @xmath165 ) . the limit in eq . is performed by keeping @xmath166 constant , with @xmath0 the initial number of spheres and @xmath167 the volume of the domain . + the number density of the remaining spheres reads @xmath168 where the one - point probability density @xmath169 that a sphere has radius @xmath155 is given by @xmath170 the probability that one or both of the spheres @xmath171 and @xmath172 are not removed is given by @xmath173 with @xmath174 and @xmath175 at this point we have to consider the set union of two spheres with radii @xmath176 and @xmath177 , respectively ; the volume of this geometrical configuration can be calculated ; in three dimensions , e.g. , we have : @xmath178 for @xmath179 . otherwise this volume reduces either to the volume of the larger sphere ( @xmath180 ) or to the sum of both spherical volumes ( @xmath181 ) . + similarly as in eq . the spatial two - point density turns out to be @xmath182 such that the conditional two - point mark density simply reads @xmath183 from this we can derive all of the mark correlation functions from sect . [ sec : kerscher_mark - segregation ] . in order to get an analytically tractable model we adopt a bimodal radius distribution in the original boolean model and start therefore with @xmath184 where we assume that @xmath185 . due to the depletion the number density @xmath8 of the spheres as well as the probability @xmath186 to find the smaller radius @xmath187 at a given point are then lowered ; we get @xmath188 with @xmath189 . altogether , the bimodal model can be parameterized in terms of the radii @xmath190 @xmath187 , the ratio @xmath191 $ ] and the density @xmath192 . the latter two quantities , however are not observable from the final point process , therefore we convert them into the parameters @xmath193 $ ] and @xmath194 , so that all other quantities can be expressed in terms of these , for instance , @xmath195 , and @xmath196 , + from eq . we determine the mean mark , i.e. the mean radius of the spheres @xmath197 and from eq . the spatial product density @xmath198 1+\alpha^2 \left[\exp\left(ni(x)\right)-1\right ] & 1 \leq x < 2 , \\[1ex ] 1 & 2 \leq x , \end{cases}\ ] ] with the normalized inter - sectional volume @xmath199 of two spheres and @xmath200 . finally , using eq . one can calculate the mark correlation functions , e.g. @xmath201 1-\alpha^2(1-\alpha ) \frac{r_2-r_1}{\overline{r } } \frac{\exp\left(n i(x ) \right)-1 } { 1+\alpha^2\left[\exp\left(n i(x ) \right)-1\right ] } & 1 \leq x < 2 , \\[1ex ] 1 & 2\leq x . \end{cases}\label{eq : kerscher_km_depl}\ ] ] in fig . [ fig : kerscher_depletion ] the @xmath140 function from the boolean depletion model is shown . the model with the solid line illustrates that a reduced @xmath140 for small radii can be obtained by simply removing smaller spheres . at least qualitatively this model is able to explain the depletion effects we have seen both in the distribution of martian craters ( fig . [ fig : kerscher_mars - markcorr ] ) and in the distribution of pores in sandstone ( fig . [ fig : kerscher_sand ] ) . the jump at @xmath202 is a relict of the strictly bimodal distribution with only two radii . [ fig : kerscher_depletion ] also shows that the boolean depletion model is quite flexible , allowing for a @xmath203 , but also @xmath119 is possible . + without ignoring the considerable difference of this boolean depletion model to the pore size distribution in real sandstones ( see figures [ fig : kerscher_sandstone_eg]-[fig : kerscher_sand ] ) one may still recognize some interesting similarities : this simple model explains naturally a decrease of @xmath140 if the distribution of the radii is symmetric ( @xmath204 ) . as visible in figure [ fig : kerscher_sand - hist ] this is approximately the case for the pore radii . moreover , note that even quantitative features are captured correctly indicating that the decrease of @xmath140 visible in figure [ fig : kerscher_sand ] is indeed due to a depletion effect . for instance , the decrease starts at @xmath205 where @xmath206 is the largest occurring radius ( see the histogram in figure [ fig : kerscher_sand - hist ] ) and the value of @xmath207 at @xmath208 is in accordance with equation assuming that @xmath209 and the normalized density of pores @xmath210 necessary for a connected network . of course a more detailed analysis is necessary based on eqs . and and the histogram shown in figure [ fig : kerscher_sand - hist ] . the `` random - field model '' covers a class of models motivated from fields such as geology ( see , e.g. , @xcite ) . the level of the ground water , for instance , is thought of as a realization of a random field which may be directly sampled at points ( hopefully ) independent from the value of the field or which may influence the size of a tree in a forest . + in general , a realization of the random field model is constructed from a realization of a point process and a realization of a random field @xmath211 . the mark of each object located at @xmath1 traces the accompanying random field via @xmath212 . the crucial assumption is that the point process is stochastically independent from the random field . + we denote the mean value of the homogeneous random field by @xmath213=\overline{u^1}$ ] and the moments by @xmath214 , with the one - point probability density @xmath215 of the random field and @xmath216 the expectation over realizations of the random field . the product density of the random field is @xmath217 $ ] with @xmath28 . for a general discussion of random field models , see @xcite . + in this model the one - point density of the marks is @xmath218 , and @xmath219 etc . the conditional mark density is given by @xmath220,\ ] ] where @xmath221 is the dirac delta distribution . clearly , this expression is only well - defined under a suitable integral over the marks . with eq . one obtains @xmath222 and the mark - correlation functions defined in sect . [ sec : kerscher_mark - segregation ] read @xmath223 therefore , there are some explicit predictions for the random field model : an empirically determined @xmath133 significantly differing from one not only indicates mark segregation , but also that the data is incompatible with the random field model . looking at figure [ fig : kerscher_ssrs - lum ] we see immediately that the galaxy data are not consistent with the random field model . similar tests based on the relation between @xmath66 and the mark - variogram @xmath224 were investigated by @xcite and @xcite . the failure of the random field model to describe the luminosity segregation in the galaxy distribution allows the following plausible physical interpretation : the galaxies do not merely trace an independent luminosity field ; rather the luminosities of galaxies depend on the clustering of the galaxies . we shall try to account for this with a better model in the following section . in the random field model , the field was only used to generate the points marks . in the cox random field model , on the contrary , the random field determines the spatial distribution of the points as well . as before , consider a homogeneous and isotropic random field @xmath225 . the point process is constructed as a cox - process ( see e.g. @xcite ) . the mean number of points in a set @xmath226 is given by the intensity measure @xmath227 where @xmath228 is a proportionality factor fixing the mean number density @xmath229 . the ( spatial ) product density of the point distribution is @xmath230 where again @xmath231 denotes the product density of the random field . @xmath232 is the normalized two - point cumulant of the random field ( see below ) . we will also need the @xmath23-point densities of the random field : @xmath233.\ ] ] like in the random field model , the marks trace the field , but this time rather in a probabilistic way than in a deterministic one : the mark @xmath2 on a galaxy located at @xmath1 is a random variable with the probability density @xmath234 depending on the value of the field @xmath235 at @xmath1 . this can be used as a stochastic model for the genesis of galaxies depending on the local matter density . + in order to calculate the conditional mark correlation functions we define the conditional moments of the mark distribution given the value @xmath236 of the random field : @xmath237 the spatial mark product - density is @xmath238.\ ] ] and with eq . @xmath239 , \ ] ] for @xmath240 and zero otherwise . the mark correlation functions can therefore be expressed in terms of weighted correlations of the random field : @xmath241,\nonumber\\ { \left\langle m^2 \right\rangle_{\rm p}}(r ) = \frac{1}{\rho_2^u(r ) } { \mathbb{e}}\big [ \overline{m^2}(u({\mathbf{x}}_1))\ u({\mathbf{x}}_1 ) u({\mathbf{x}}_2)\big],\\ { \left\langle m_1 m_2 \right\rangle_{\rm p}}(r ) = \frac{1}{\rho_2^u(r ) } { \mathbb{e}}\big [ \overline{m}(u({\mathbf{x}}_1))\overline{m}(u({\mathbf{x}}_2))\ u({\mathbf{x}}_1 ) u({\mathbf{x}}_2)\big].\nonumber\end{gathered}\ ] ] to proceed further , we have to specify @xmath242 . as a simple example we choose @xmath2 equal to the value of the field @xmath235 at the point @xmath1 , such as in the random field model . thinking of the random field as a mass density field and the mark of a galaxy luminosity , that means that the galaxies trace the density field and that their luminosities are directly proportional to the value of the field . with @xmath243 the conditional mark moments become @xmath244 . the moments of the unconstrained mark distribution read @xmath245 , and the three basic pair averages are @xmath246 hence , the mark correlation functions defined in sect . [ sec : kerscher_mark - segregation ] are determined by the higher - order correlations of the random field . with the cox random field model we go beyond the random field model , e.g. @xmath247 is not equal to one any more . at this point , we have to specify the correlations of the random field @xmath211 . the simplest choice , a gaussian random field , is not feasible here , since a number density ( cp . [ eq : kerscher_cox - definition ] ) has to be strictly positive , whereas the gaussian model allows for negative values . instead , we will use the hierarchical ansatz : we first express the two- and three - point correlations in terms of normalized cumulants @xmath124 and @xmath248 ( see , e.g.,@xcite ) , @xmath249 in order to eliminate @xmath250 we use the hierarchical ansatz ( see e.g. @xcite ) : @xmath251 this ansatz is in reasonable agreement with data from the galaxy distribution , provided @xmath252 is of the order of unity ( @xcite ) . several choices for @xmath253 and @xmath252 lead to well - defined cox point process models based on the random field @xmath211 @xcite . now we can express @xmath140 from eq . entirely in terms of the two - point correlation function @xmath254 of the random field : @xmath255 where we made use of the fact that @xmath256 . inserting typical parameters found from the spatial clustering of the galaxy distribution we see from fig . [ fig : kerscher_km - coxrf ] that the cox random field model allows us to qualitatively describe the observed luminosity segregation in fig.[fig : kerscher_ssrs - lum ] . but the amplitude of @xmath133 predicted by this model is too high . the cox random field model , however , is quite flexible in allowing for different choices for @xmath242 ; also different models for the higher - order correlations of the random field may be used , e.g. a log - normal random field @xcite . clearly more work is needed to turn this into viable model for the galaxy distribution . whenever objects are sampled together with their spatial positions and some of their intrinsic properties , marked point processes are the stochastic models for those data sets . combining the spatial information and the objects inner properties one can constrain their generation mechanism and their interactions . + developing the framework of marked point processes further and outlining some of their general notions is thus of interest for physical applications . let us therefore look at mark correlations again from both a statistical and a physical perspective . we focused on two kinds of dependencies . + on the one hand , one can always ask , whether objects of different types `` know '' from each other . from a statistical point of view , this is the question whether the marked point process consists of two completely independent sub - point processes . physically , this concerns the question whether the objects have been generated together and whether they interact with each other . + on the other hand , it is often interesting to know whether the spatial distribution of the objects changes with their inner properties . for the statistician , this translates into the question whether mark segregation or mark - independent clustering is present . for the physicist such a dependency is interesting since one can learn from them whether and how the interactions distinguish between different object classes or whether the formation of the objects mark depends on the environment . + we discussed statistics capable of probing to which extent mark correlations are present in a given data set , and showed how to assess the statistical significance . applying our statistics to real data , we could demonstrate , that the clustering of galaxies depends on their luminosities . large scale correlations of the orientations of dark matter halos were found . using the mars data we could validate a picture of crater generation on the martian surface : mainly , the local geological setting determines the crater type . we also could show that the sizes of pores in sandstone are correlated . + in order to understand empirical data sets in detail , we need models to compare to . as generic models the boolean depletion model , the random field model and its extension , the cox random field models are of interest . + further application of the mark correlations properties may inspire the development of further models . it seems therefore that marked point processes could spark interesting interactions between physicists and mathematicians . certainly , the distributions of physicists and mathematicians in coffee breaks at the wuppertal conference were clustered , each . but could one observe positive cross - correlations ? using mark correlations we argue , that , even more , there is lots of space for positive interactions. . we would like to thank andreas faltenbacher , stefan gottlber and volker mller for allowing us to present some results from the orientation analysis of the dark matter halos ( sect . [ sec : kerscher_halos ] ) . for providing the sandstone data ( sect . [ sec : kerscher_sandstone ] ) and discussion we thank mark knackstedt . herbert wagner provided constant support and encouragement , especially we would like to thank him for introducing us to the concepts of geometric algebra as used in the appendix . in order to form versatile test functions for describing mark segregation effects , we integrated the conditional mark probability density @xmath29 twice in mark space thereby weighting with a function of the marks @xmath52 ( see eq . [ eq : kerscher_def - paverage ] ) . such a pair - averaging reduces the full information present in @xmath29 . so one may ask , whether or in which sense the mark correlation functions give a complete picture of the present two - point mark correlations . + for scalar marks @xmath2 this task is trivial . with a polynomial weighting function @xmath257 ( @xmath258 ) we consider moments of @xmath29 , hence , we can be complete only up to a given polynomial order in the marks @xmath16 and @xmath17 . at first order there is only the mean @xmath73 . at second order we have @xmath75 and @xmath259 . all the mark correlation functions discussed in sect . [ sec : kerscher_mark - segregation ] can be constructed from these three pair averages and @xmath259 at the two - point level , however , does not imply that one should not consider linear combinations of them . for instance , it may well be the case , that only certain linear combinations yield significant results . ] . higher - order moments of the marks involve more and more cross - terms . + for vector - valued marks , however , it is not obvious that the test quantities proposed in sect . [ sec : kerscher_mark - segregation ] trace all possible correlations between the vectors up to third order . to settle this case we have to consider the framework of geometric algebra , also called clifford algebra . a detailed introduction to geometric algebra is given in @xcite , shorter introductions are @xcite . in geometric algebra one assigns a unique meaning to the geometric product ( or clifford product ) of quantities like vectors , directed areas , directed volumes , etc . the geometric product @xmath260 of two vectors @xmath261 and @xmath262 splits into its symmetric and antisymmetric part @xmath263 here @xmath264 denotes the usual scalar product ; in three dimensions , the wedge product @xmath265 is closely related to the cross product between these two vectors . however , @xmath265 is not a vector like @xmath266 , but a bivector a directed area . higher products of vectors can be simplified according to the rules of geometric algebra ( for details see @xcite ) . + let us consider the situation where objects situated at @xmath14 and @xmath15 bear vector marks @xmath86 and @xmath87 , respectively , and let the normalized distance vector be @xmath267 . note , that @xmath103 is not a mark at all , rather it can be thought of as another vector which may be useful for constructing mark correlation functions . + for many applications it is reasonable to assume isotropy in mark space , i.e. all of the mark correlation functions are invariant under common rotations of the marks . for galaxies , e.g. , there does not seem to be an a priori preferred direction for their orientation . in more detail we have then @xmath268 and so on , where @xmath155 is an arbitrary rotation in mark space . this means that the mark correlation functions depend only on rotationally invariant combinations of the vector marks . therefore , only rotationally invariant combinations of vectors are sensible building blocks for weighting functions . we thus can restrict ourselves to scalar weighting functions , which result in coordinate - independent vector - mark correlation functions . + again we proceed by considering mixed moments as basic combinations . we restrict ourselves to scalar quantities being polynomial in the vector components . one may also discuss moments in a broader sense allowing for vector moduli . in this wider sense , for example , @xmath269 or @xmath270 would be allowed . we do not consider such quantities here , because they are not polynomial in the vector components . their squares anyway appear at higher orders . furthermore , it turns out that the characterization we will provide depends on the embedding dimension . the first- and second - order moments are identical in two and three dimensions , but at the third order they start to differ . 1 . in the strict sense of scalar quantities being linear in the vector components there are no first - order moments for vectors . at second order we encounter the following products : @xmath271 , @xmath272 , @xmath273 , @xmath274 . note , that , e.g. , @xmath275 and @xmath276 do not make any difference as regards the mark correlation functions , since the pair averages implicitly render the indices symmetric ; moreover , although the geometrical product is non - commutative , @xmath277 and @xmath278 do not lead to different mark correlation functions . furthermore , @xmath279 . @xmath280 provides us with higher moments of the modulus of the vectors . to investigate these kinds of correlations already scalar marks would be sufficient . new information is encoded in the other products . + consider @xmath281 . the symmetric part @xmath282 is clearly a scalar and defines the alignment @xmath283 ( eq . [ eq : kerscher_align ] ) . the antisymmetric part @xmath284 is a bivector . its unique modulus ( see again @xcite ) , @xmath285 , may be useful , but is no longer a polynomial in the vector components . @xmath286 appears at the fourth order . in a completely analogous way we can treat @xmath287 . the symmetric part @xmath288 defines @xmath100 . hence at second order , the only possible vector - mark correlation functions are @xmath84 and @xmath100 . 3 . at third order we have to consider products of three vectors . in general the product of three vectors @xmath289 splits into @xmath290 i.e. , a vector ( consisting of the three first terms ) , and a pseudo - scalar , a directed volume . in two dimensions the pseudo - scalar @xmath291 vanishes . + now we have to form all possible products of the three vectors @xmath129 and to derive scalars . in three dimensions the only new combination is the pseudo - scalar @xmath292 giving the oriented volume @xmath293 . unfortunately , this oriented volume averages out to zero . thus , in a strict sense , there are no interesting third - order quantities . closely related , however , is the modulus of the pseudoscalar @xmath294 proportional to our @xmath105 . this expression is invariant under permutations of the vectors . 4 . at third order and in two dimensions all of the relevant combinations are products of first- and second - order combinations ; no specifically new combination appears . this is different from the case of three dimensions , where at third order an entirely new geometric object , the pseudo - scalar @xmath292 can be constructed . there is a general scheme behind this argument : since in @xmath96 dimensions any geometrical product of more than @xmath96 vectors vanishes , all relevant combinations of vectors at orders higher than @xmath96 are essentially products of combinations of lower - order factors . baddeley , a. j. ( 1999 ) : `` sampling and censoring '' . in : _ stochastic geometry , likelihood and computation _ , ed . by o. barndorff - nielsen , w. kendall , m. van lieshout , volume 80 of _ monographs on statistics and applied probability _ , chapter 2 ( chapman and hall , london ) bhringer , h. , p. schuecker , l. guzzo , c. collins , w. voges , s. schindler , d. neumann , g. chincharini , r. cruddace , a. edge , h. macgillivray , p. shaver ( 2001 ) : `` the rosta - eso flux limited x - ray ( reflex ) galaxy cluster survey i : the construction of the cluster sample '' , _ _ , p. 826 , l. n. , c. n. a. willmer , p. pellegrini , o. l. chaves , c. rite , m. a. g. maia , m. j. geller , d. w. latham , m. j. kurtz , j. p. huchra , m. ramella , a. p. fairall , c. smith , s. lipari ( 1998 ) : `` the southern sky redshift survey '' , _ _ * 116 * , pp . 17 djorgovski , s. ( 1987 ) : `` coherent orientation effects of galaxies and clusters '' . in : _ nearly normal galaxies . from the planck time to the present _ , ed . by s. m. faber ( springer , new york ) , pp . 227233 guzzo , l. , j. bartlett , a. cappi , s. maurogordato , e. zucca , g. zamorani , c. balkowski , a. blanchard , v. cayatte , g. chincarini , c. collins , d. maccagni , h. macgillivray , r. merighi , m. mignoli , d. proust , m. ramella , r. scaramella , g. stirpe , g. vettolani ( 2000 ) : `` the eso slice project ( esp ) galaxy redshift survey . the redshift and real - space correlation functions '' , _ _ * 355 * , pp . 116 hermit , s. , b. x. santiago , o. lahav , m. a. strauss , m. davis , a. dressler , j. p. huchra ( 1996 ) : `` the two point correlation function and the morphological segregation in the optical redshift survey '' , _ _ * 283 * , p. 709 isham , v. ( 1985 ) : `` marked point processes and their correlations '' . in : _ spatial processes and spatial time series analysis _ , ed . by f. droesbeke ( publications des facults universitaires sain - louis , bruxelles ) kerscher , m. ( 2000 ) : `` statistical analysis of large scale structure in the universe '' . in : _ statistical physics and spatial statistics : the art of analyzing and modeling spatial structures and pattern formation _ , ed . by k. r. mecke , d. stoyan , number 554 in lecture notes in physics ( springer , berlin ) , astro - ph/9912329 lwen , h. ( 2000 ) : `` fun with hard spheres '' . in : _ statistical physics and spatial statistics : the art of analyzing and modeling spatial structures and pattern formation _ , ed . by k. r. mecke , d. stoyan , number 554 in lecture notes in physics ( springer , berlin ) mecke , k. ( 2000 ) : `` additivity , convexity , and beyond : application of minkowski functionals in statistical physics '' . in : _ statistical physics and spatial statistics : the art of analyzing and modeling spatial structures and pattern formation _ , ed . by k. r. mecke , d. stoyan , number 554 in lecture notes in physics ( springer , berlin ) sok , r. m. , m. a. knackstedt , a. p. sheppard , w. v. pinczewski , w. b. lindquist , a. v. a , , l. paterson ( 2000 ) : `` direct and stochastic generation of network models from tomographic images ; effect of topology on two - phase flow properties '' . in : _ proc . upscaling downunder _ , ed . by l. paterson , ( kluwer academic , dordrecht ) stoyan , d. ( 2000 ) : `` basic ideas of spatial statistics '' . in : _ statistical physics and spatial statistics : the art of analyzing and modeling spatial structures and pattern formation _ , ed . by k. r. mecke , d. stoyan , number 554 in lecture notes in physics ( springer , berlin )

a new hybridization of the cuckoo search ( cs ) is developed and applied to optimize multi - cell solar systems ; namely multi - junction and split spectrum cells . the new approach consists of combining the cs with the nelder - mead method . more precisely , instead of using single solutions as nests for the cs , we use the concept of a simplex which is used in the nelder - mead algorithm . this makes it possible to use the flip operation introduces in the nelder - mead algorithm instead of the levy flight which is a standard part of the cs . in this way , the hybridized algorithm becomes more robust and less sensitive to parameter tuning which exists in cs . the goal of our work was to optimize the performance of multi - cell solar systems . although the underlying problem consists of the minimization of a function of a relatively small number of parameters , the difficulty comes from the fact that the evaluation of the function is complex and only a small number of evaluations is possible . in our test , we show that the new method has a better performance when compared to similar but more compex hybridizations of nelder - mead algorithm using genetic algorithms or particle swarm optimization on standard benchmark functions . finally , we show that the new method outperforms some standard meta - heuristics for the problem of interest . cuckoo search , nelder - mead simplex , cascaded optimization , multi - cell solar systems , multi - junction solar cells , split spectrum solar cell system

1411.0217

in the recent years , the performance of photovoltaic ( pv ) cells has been significantly improved by using multi - cell devices . in such systems , few different cells are combined to maximize the conversion efficiency by dividing the spectrum of solar radiation and then use a proper cell for each divided spectrum . using this approach , efficiency of 44.7% has been achieved where 4 junctions are used @xcite . the two most common methods for such device concept are splits spectrum and multi - junction solar cells . with the development of new technologies , it has become possible to create more complex systems consisting of a higher number of cells . it has been shown that the performance of such systems can be further improved using optical concentration@xcite . for future development of such solar - cell systems , it is of significant importance to have bounds for the optimal possible efficiency . while it is relatively simple to find them in case of two or three layers , it becomes significantly more complex in case of a higher number of layers . this is due to the fact that it is necessary to find the minimal value of a multi parameter function , which is computationally challenging . the problem becomes even more complex because it is hard to find the corresponding gradient that could simplify the calculation . there is a wide range of non - gradient based optimization methods like simulated annealing @xcite , genetic algorithms @xcite , pivot method @xcite , particle swarm optimization @xcite , nelder - mead simplex ( nms ) method @xcite that are generally used to solve this type of problems . the performance of such methods is highly dependent on the function that we wish to minimize . in the case of the problem of interest , initial test have shown that nms algorithm manages to outperform , in several tested problem instances , mentioned more complex population based methods . one of the reasons for this is the fact that due to physical properties of the problem , we have a good initial guess of the solution . our research has focused on improving the performance of this algorithm by incorporating some type of swarm intelligence . there is a wide range of meta heuristic that mimic the behavior of groups in nature , the most prominent are ant colony optimization @xcite , bee colony optimization @xcite , particle swarm optimization ( pso ) @xcite , firefly algorithm @xcite , and cuckoo search ( cs)@xcite@xcite . in the recent years , the cs @xcite algorithm has been gaining on popularity as an optimization method due to its good performance , robustness , and simplicity of implementation . research has been conducted to improve the original algorithm by changing the method of communication inside of the colony @xcite , having adaptive parameters@xcite , using quantum inspired approaches@xcite@xcite , and adapting the algorithm for parallel applications @xcite@xcite . the cs algorithm has previously been successfully applied in the field of solar cells , more precisely for parameter estimation of photovoltaic models @xcite . previously , the nms algorithm has been hybridized using genetic algorithms @xcite@xcite , ant colony optimization @xcite and particle swarm optimization @xcite resulting in very efficient global optimization algorithms . one of the main problems of hybridized methods is that although they achieve better results , they often become very complex for implementation . in this work , we introduce a cs inspired hybridization of the nms algorithm that manages to avoid this drawback but still achieves significantly better results than the original method . the general idea of our approach is to use the simplex structure from nms algorithm as a form of a nest in the cuckoo search . one of the main advantages of this approach is that we can avoid the use of the levy flight in cs . although the levy flight is a very powerful system of exploring the solution space , its performance within the cs is highly dependent on the scaling factor introduced in this method . in the presented approach , we substitute the levy flight with the flip operation of a simplex in the in the nms algorithm . in our tests , we show that the proposed method not only performs well in the case of solar cells optimization , but also achieves good results on standard benchmark functions when a lower number of parameters is considered . the cuckoo search is often a competing method to the particle swarm optimization and genetic algorithms ( ga ) ; because of this , we also give a comparison to previously published results of hybridization of nms using this method . in the comparison we have shown that the new method manages to achieve a higher speed of convergence , while at the same time being much simpler to implement . to better evaluate the effect of the proposed hybridization of cs we have also conducted a comparison to other versions of the cs previously presented in literature . this has been done on a wide range of benchmark functions , from which it is noticeable that the new approach has a very good performance . the article is organized as follows . in the next section , we give an overview of related research , divided into 3 subsections . the first one presents the basics of the cuckoo search algorithm . in the second subsection we give an outline of the nms algorithm . in the final subsection we give an overview of previously developed hybridization of nms using swarm intelligence and genetic algorithms . the third section presents a detailed specification of the new method . the forth section is dedicated to an evaluation of the new method . in the first subsection we compare its performance on standard benchmark functions with published results for other population based hybridizations of nms . the following subsection presents a comparison to other versions of the cs algorithm . in the third subsection we present the optimization problem in pv cell and analyze the performance of the method as previously mentioned the cuckoo search is one of the population based meatheuristic for function minimization . in other words the goal of this method is to find the global minimal value @xmath0 of a function defined in the following way @xmath1 function @xmath2 is called the objective function with @xmath3 dimensions . cuckoo search ( cs ) is an optimization algorithm that mimics the brood parasitism of some cuckoo species . more precisely , cuckoo birds lay their eggs in the nests of other host birds ( of other species ) . through an evolutive process cuckoos have managed to adapt the shape and color of eggs to the one of targeted bird species . this method of survival has been converted to a meta - heuristic approach called cuckoo search . in the corresponding algorithm , each egg in a nest will represent a solution and cuckoo egg represents a newly generated solutions . the idea is to create new , similar , and potentially higher quality solutions ( cuckoos ) to replace the low quality solutions in the nests . in the simplest form , each nest contains one egg . the proposed meta heuristic follows several rules . 1 . each cuckoo in the colony generates one solution ( egg ) at each step , similar to an already existing solution ( nest ) , 2 . the best solution will be carried to the next generation , 3 . the number of available solutions that are used as nests is fixed , and at each step , new solutions are generated . this has a consequence that some solutions need to be discarding , which corresponds to eggs being found by the host bird . the discarding operation is only done on some set of worst solutions ( nests ) , more precisely for each such nest @xmath4 there is a probability @xmath5 of being found . the standard cs algorithm following these rules is presented in the following pseudo code : generate an initial population of @xmath3 host nests ; get a cuckoo randomly ( say , @xmath6 ) and replace its solution by performing levy flights ; evaluate its quality / fitness @xmath7 choose a nest among @xmath3 ( say , @xmath8 ) randomly ; replace @xmath8 by the new solution ; a fraction ( @xmath5 ) of the worse nests are abandoned and new ones are built ; keep the best solutions / nests ; rank the solutions / nests and find the current best ; pass the current best solutions to the next generation ; levy flight is an essential part of the cs algorithm . it is defined by the following equation @xmath9 where @xmath10 represents a step size . [ levi1 ] basically represents a stochastic equation for a random walk which is a markov chain . more precisely , the next position or function value that shall be evaluated depends on two parameters ; namely , the current position and transition function ( @xmath11 ) . the next position ( status ) depends only on the current position ( @xmath12 ) and probability of transition ( @xmath11 ) . levy(@xmath13 ) , which is the random step length , is drawn from a levy distribution . the consecutive positions generated through steps of the levy flight create a random walk that has a heavy tail distribution . the use of levy flight instead of a simple random walk significantly improves the performance of cs . as presented in this pseudo code , the cs algorithm has several stages . the first stage is to generate the initial population . in the main loop of the algorithm a new solution @xmath7 is generated using levy flight from a random nest @xmath6 . if this solution is better than a one belonging to random nest @xmath14 , the @xmath14 substituted by @xmath7 . finally , in the goal of increasing the diversity of the search , a fraction @xmath5 of the worst nests is changed using levy flight with a larger step applied to avoid trapping in local optimal solutions . the main advantage of the cs search is its robustness and its dependence on only a few number of parameters that are needed to be fine tuned . certainly , these parameters depend on the function that is being optimized like in the case of pso and ga . the nelder mead simplex method , frequently called the downhill simplex method or amoeba method , is widely used for finding local minima solutions , for well defined problems for which a derivative is not known . the nms falls in the general class of direct search methods.the basic component of the algorithm is a simplex . in case of a two parameter function , a simplex is a triangle , and the method results in a pattern search that is dependent of the values of the three vertices of the triangle . in a similar way for a @xmath15 dimension function a simplex will have @xmath16 vertices . in general , the method replaces some of the worst points ( which have the largest values of the @xmath2 ) of the simplex with a new vertices that are acquired using a heuristic approach . in this way the simplex is transformed to a new one which is potentially closer to the minima . there are several potential transformation refection , expansion , contraction and shrinkage that define the generation of the search @xcite . in nms algorithm , the heuristic choice and the application of the transformation are called a flip . more precisely the nms algorithm is defined in the following way . for function @xmath2 with @xmath3 dimensions , a simplex will be a convex hull consisting of @xmath17 vertices . we will define a simplex with vertices @xmath18 by @xmath19 . the nms generates iteratively a sequence of simplices by performing the following procedure . first all of the vertices inside of simplex @xmath19 are sorted depending on the objective function value @xmath20 at each of iteration some of the vertices are removed from the simplex and new ones are added . using the following steps * reflection + the reflection point @xmath21 is calculated using the following formula + @xmath22 + in eq . [ reflection ] @xmath23 represents the centroid of the simplex calculated in the following way + @xmath24 + in case the value of @xmath25 satisfies the condition @xmath26 we replace @xmath27 with @xmath21 and go to next iteration . * expansion + in case @xmath28 , an additional step is done to attempt by calculating an expansion vertex @xmath29 . + @xmath30 + if @xmath31 satisfies @xmath32 replace @xmath27 with @xmath29 otherwise with @xmath21 . * outside contraction + in case @xmath33 , compute the outside contraction point . @xmath34 + if @xmath35 satisfies @xmath36 replace @xmath27 with @xmath37 otherwise do a shrink operation . * inside contraction + in case @xmath38 , compute the inside contraction point . + @xmath39 + if @xmath40 satisfies @xmath41 replace @xmath27 with @xmath42 otherwise do a shrink operation . * shrink + @xmath43 to fully define the flip operation it is necessary to specify the values of @xmath44 , @xmath45 , @xmath46 and @xmath47 . their values should satisfy the following conditions . @xmath48 in the standard implementation these parameters have the following values : @xmath49 , @xmath50 , @xmath51 and @xmath52 . in our hybridization of nms we have used these values . we wish to mention that the nms algorithm , and consequently the proposed hybridization , can be further improved by incorporating adaptive parameters @xcite . there are two main disadvantages of the cs algorithm . first , it does not incorporate any type of local search to increase the convergence speed when it is close to some local or global minima , but only uses the levy flight to generate new test positions . contrary to this , ga and pso are much faster in narrowing to optima when they are close . the second problem is that the performance of cs is highly dependant on the value of @xmath44 in eq . [ levi1 ] . it has been shown that its value does not only depend to the on scale the test function , but for high precision of results , its value needs to adapt during the execution of the algorithm . there has been a wide range of approaches to improve the performance of cs by hybridization . it has been shown that combining cs with other types of population based algorithms is a good approach . some of the most successful ones are its combination with ga @xcite , pso @xcite , differential evolution @xcite , and ant colony optimization @xcite . although this type of hybridization manages to improve the performance of the original method , the implementation becomes significantly more complex than the basic cs . on the other hand , it could be argued that the new methods , which couples two population based techniques , that are generally more suitable for globalized searches , combines to very similar approaches . it is also important to mention that very limited research has been done in the hybridization of cs with some local search methods , which has proven to be very profitable for other population based algorithms . in our work , we attempt to improve the performance of cs , by hybridization with nms algorithm while maintaining the simplicity and robustness of the original algorithm . previously , population based algorithms have been successfully combined with nms @xcite . there are two main directions to this type of hybridization ; namely , cascaded or tandem @xcite . in the cascaded approach , the two methods are used consecutively , and resulting in a more globalized search . in case of a tandem hybridization , the two methods influence each others performance and generally result in a more localized search which is more suitable for functions with a lower number of parameters . in the case of optimization of solar cells , the functions of interest would have from 3 - 10 parameters , so we have focused on developing a more localized search using the tandem approach . the focus of the method is to find solutions of good quality with a low number of functions evaluations . the incentive for this is the fact that the calculation of efficiency of multi layer solar cells is computationally very expensive . as previously mentioned , the goal is to find the minima of a function containing a relatively small number of parameters ( 3 - 10 ) . to achieve this , we have hybridized the nms using cuckoo search . the general idea is to exploit the effectiveness of nms as a local search algorithm , but combine it with the cs to get a wider search area . by combining cs with nms , the original method becomes more capable in narrowing down to good solutions . one of the problems of cs is that except for the focusing of the search by the replacement of lower quality solutions with better ones , the search is relatively random . in the new approach , we wish to search the solution space in a more structured way . this is done by incorporating the flip operation from nms to cs . the changes to the original cs method are the following : * nests in the new method are simplexes instead of parameter values that correspond to good solution . as a consequence the initial population will be a set of simplexes . * while in the original cs , levy flight is used to explore new solutions , in the new method , this is done using the flip operation of nms . more precisely , at each step of the algorithm , nest @xmath6 is transformed to nest @xmath53 by using a flip . * in the presented approach , only a part of the vertices inside of a simplex are replaced . this is because a simple replacement of lower quality nests with better ones in the case of using simplexes combined with the flip operation is not productive . this is due to the fact that nms has a deterministic search pattern , and if this was done , we would be just repeating the tests for same solutions . the new approach can be formalized by the following pseudo code . generate an initial population of @xmath3 simplexes ; test @xmath54 solutions ; get a cuckoo randomly ( say , @xmath6 ) and replace its using a nms flip ; evaluate its quality / fitness @xmath7 choose a nest among @xmath3 ( say , @xmath8 ) randomly ; replace worst @xmath55 vertices in simplex @xmath8 by the vertices from simplex @xmath6 ; a fraction ( @xmath5 ) of the worse simplex are abandoned and new ones are built ; keep the best solutions / nests ; rank the solutions / nests and find the current best ; the initial simplexes are generated around randomly selected points , with a relatively small sizes . we use a slight modification of the approach for generating simplexes proposed by l. pfeffer , which is also used in matlab implementation of nms @xcite . a simplex @xmath19 with vertices @xmath56 is generated around a vertex @xmath57 . all of the coordinates of @xmath56 are equal to the ones of @xmath4 except the one on position @xmath6 , which is calculated as @xmath58 in eq . [ randomsimplex ] @xmath59 represents the value of @xmath56 at position @xmath6 . @xmath44 is a random parameter that is used generate simplexes of different sizes . in our test the best results where achieved when @xmath44 had a uniform distribution inside of the interval @xmath60 as it has been previously mentioned , the flip operation is simply used instead of the levy flight , as it can be seen in the pseudo code . the substitution of nest of lower quality is done by coping some of the vertices of the lowest quality by some of the high quality ones from the better simplex . at this stage , there are two important details that must be pointed out . first , the best vertex from the higher quality nest ( simplex ) is not used because this would make the search focus very quickly . if the other high quality vertexes are used , the search focuses towards good solutions but still remains diverse . this stage has two possibilities , either a relatively random search of the space in the direction of the higher quality nest if the simplexes are distant or a wider search around the good solution if the simplexes are close to each other . the algorithm has two parameters @xmath61 and @xmath55 that are used for fine tuning of the method . the part of vertexes @xmath55 that should be overwritten is dependent on the number of parameters of the function that we wish to minimize . in our experiments , it has been shown that for functions with a low number of parameters , the value @xmath62 vertices is a good choice . in general , if this number is too low the search needs many steps to actually move to the region of good solutions , and contrary if it is two large this movement is too fast . the second tuning parameter @xmath61 is used to specify how randomized the search will be , or in other words , how frequently new random nests will be generated . in our test it has been shown that having @xmath61 equal to two times the number of nests is a good choice . in this section , we present results of using the new method to optimize two sets of problems . the first test are conducted on standard benchmark test functions and compared to previously published results @xcite . more precisely , we have compared our method to the hybridizations of nms with pso and nms with ga on problems of 3 - 10 parameters . since pso and ga are methods that often competing methods cs , this gives us a good evaluation of the new hybridized method . to fully evaluate the effect of the proposed hybridization of the cs , in the second section we present a comparison to other version of the cs . in the final part of our tests , we analyze the performance of the new method on a practical problem of optimization of multijunction and split spectrum solar cells . we have tested the effectiveness of our method for systems with a 3 - 10 different cells of both types . to be able to have a quantitative evaluation of the new method , it has been compared to nms , simulated annealing and ga . the method has been implemented by creating code using matlab r2013a . the calculations have been done on a machine with intel(r ) core(tm ) i7 - 2630 qm cpu 2.00 ghz , 4 gb of ddr3 - 1333 ram , running on microsoft windows 7 home premium 64-bit . in our first group of test , we compare the performance of the new method to hybridization of nms with two other population based meta heuristics , ga and pso . test has been done on 10 different test functions as presented by fan et al . @xcite and chelouah and siarry @xcite where their definitions can be found . tests for functions with two variables have been done for the branin rcoc ( @xmath63 ) , b2 , goldstein and price ( @xmath64 ) , shubert ( @xmath65 ) , zakharov ( @xmath66 ) ; with three variables hartmann ( @xmath67 ) , and with four variables on the shekel @xmath68 . the test has also been done on the rosenbrock ( @xmath69 ) function for 2 , 5 and 10 parameters . for each of the test functions , 100 runs of the nms - cs algorithm have been done with different initial simplex positions similarly to what is done in article @xcite . in the two competing algorithms the initial vertices are randomly selected inside of the specified search domain , while in the case of the proposed hybridization these vertices are used for generating the initial simplixes using the method presented in the previous section . in the results section of the mentioned article , `` successful '' runs are defined for solutions that have a certain level of precision . we have excluded this measure from the results given in table [ table : necreloc ] , since our method has had a 100 successless runs for each of the 10 test functions ; as it had been the case with nms - ga and nms - pso given in @xcite . in our test , we have used the same stopping criteria as in article @xcite or when a high enough precision is reached . the stopping criterion is given in the following equation @xmath70 where @xmath71 is the average of the best @xmath15 solutions . this criterion is based on the standard deviation of the objective function values over @xmath15 best solutions.the chosen value for @xmath72 . in the case of the previous published work @xmath15 was equivalent to one third of the population . in the case of the proposed method @xmath73 would represent the best solution inside of simpex @xmath6 , and consequently we would only observe the best third ( @xmath15 ) simplexes . the nms - cs has used 6 nests for all the test functions except @xmath74 where it was 20 . this is lower then the recommended 25 nests for the cs which is used in a wide range of articles . we believe that nms - cs has achieved the best results when using a much smaller number of nest due to fact that it uses the nms flip operation to narrow in on solutions . we can see the comparison of the three methods in table [ table : necreloc ] 350pt x x x x x x x test function & & & + & num . error & num . error + rc & 356&4.0e-5 & 230&1.0e-4&269 & 2.1e-5 + b2 & 529&4.0e-5 & 325&1.0e-5@xmath75 & & + gp & 422&2.0e-5 & 304&3.0e-5&313&2.4e-5 + sh & 1009&2.0e-5 & 753&3.0e-5 & & + @xmath76 & 738&6.0e-5 & 440&5.0e-5&473 & 2.1e-5 + @xmath77 & 339&4.0e-5 & 186&1.0e-5@xmath75 & & + @xmath67 & 688&5.0e-5 & 436&1.2e-4 & 418&5.0e-4 + @xmath78 & 2366&1.6e-4 & 850&6.0e-5 & 1125 & 2.8e-5 + @xmath79 & 3126&9.0e-5 & 2313&4.0e-5 & & + @xmath74 & 5194&2.0e-4 & 3303&1.2e-4&2621 & 2.2e-4 + it is observed that the new method manages to outperform nms - ga . this confirms the conclusion form article @xcite , that combining nms with swarm intelligence methods is more beneficial than with genetic algorithms . it is apparent that the new method has a very similar performance to nms - pso when the number of iterations is considered , but overall the new method manages to get slightly higher quality of results . more precisely the new method has managed to get clearly better results , in both the number of iterations and achieved precision , in four out of ten cases . due to the definition of benchmarks tests that have been previously published , to which we have made a comparison to , it was not possible to strictly conclude which method was the best preforming in the rest of the cases but it can be said that the performance of nms - pso and nms - cs was very similar . it is important to point out , that the new approach is much simper to implement compared to the other two hybridizations of nms . to evaluate the performance of the proposed hybridization of the cs , a comparison to other versions of the cs has also been conducted . more precisely the new method has been compared to the original cs algorithm proposed by yang ( yang ) @xcite and the modified version of the algorithm proposed by walton ( mcs ) @xcite . the comparison has been conducted on the benchmark functions used at the cec 2013 @xcite . the benchmark instances are divided into 3 groups , functions 1 - 5 are unimodal , 6 - 20 are multimodal and functions 21 - 28 are composite functions generated using the three to five of the functions of the two first groups . the three methods have been compared for test functions having 2 , 5 and 10 dimensions . the numerical experiments have been done using the code made available by the authors of the two competing methods @xcite . for each of the algorithms 50 independent runs have been performed for each of the test functions . we observed the average and best found solution find for the 50 runs , with a fixed number of fitness function evaluations . in case of 2 , 5 and 10 dimensional functions , a maximum of 2000 , 5000 and 20000 function evaluations was allowed . to better evaluate the performance of the methods the best , worst , average and standard deviation of the generated solutions is presented . to have the a fair comparison we have done extensive fine tuning of the parameters that specify the two algorithms for problems of these sizes through a large number of tests , of course this was also been done for the proposed method . the number of nests used for the yang s and the new algorithm was 6 , 10 and 25 for test functions of 2 , 5 and 10 dimensions . in case of mcs , doubling the number of the nest for each of the dimension has shown to give the best results . this is due to the fact that mcs only uses a high number of nest in the early steps of the algorithm . to evaluate the effectiveness we the results of these experiments are given in tables [ table : dim2 ] , [ table : dim5 ] , [ table : dim10 ] . 580pt x x x x x x x x x x x x x x x x & & & + + & min & max&avg & std & min & max&avg&std & min & max&avg&std + f1 & & & & 0.0e+0 & 3.6e-10 & 2.1e-4 & 2.5e-5 & 4.5e-5 & & & & 0.0e+0 + f2 & 2.9e-7 & 1.7e+0 & 8.5e-2 & 3.4e-1 & 3.2e+0 & 1.1e+4 & 1.5e+3 & 2.2e+3 & & & & 0.0e+0 + f3 & 1.4e-6 & 8.1e+3 & 1.6e+2 & 1.1e+3 & 6.5e+0 & 5.7e+3 & 1.1e+3 & 1.2e+3 & & & & 3.2e-14 + f4 & 6.0e-8 & 1.9e-2 & 1.6e-3 & 3.5e-3 & 1.1e-1 & 1.5e+4 & 2.0e+3 & 3.3e+3 & & & & 0.0e+0 + f5 & & & & 0.0e+0 & 3.7e-7 & 3.7e-3 & 4.9e-4 & 7.1e-4 & & & & 3.7e-14 + f6 & 8.0e-13 & 7.5e-2 & 2.1e-3 & 1.1e-2 & 6.5e-9 & 1.3e-1 & 1.1e-2 & 2.4e-2 & & & & 0.0e+0 + f7 & 8.6e-7 & & & 6.9e-3 & 1.0e-3 & 1.8e+0 & 2.2e-1 & 4.8e-1 & & 3.6e-1 & 2.0e-2 & 6.4e-2 + f8 & 2.1e-6 & 2.0e+1 & 5.4e+0 & 7.4e+0 & 1.0e-3 & 1.8e+1 & 1.0e+0 & 3.4e+0 & & & & 3.4e+0 + f9 & 1.3e-4 & 9.1e-1 & 1.6e-1 & 2.5e-1 & 7.5e-4 & 5.3e-1 & 8.8e-2 & 1.4e-1 & & & & 1.9e-1 + f10 & 9.6e-5 & 1.0e-1 & 2.7e-2 & 2.1e-2 & 7.4e-3 & 2.0e+0 & 4.3e-1 & 5.1e-1 & & & & 2.8e-2 + f11 & & & 1.8e-1 & 6.3e-1 & 3.5e-9 & 1.0e+0 & & 3.6e-1 & & & 1.8e-1 & 3.9e-1 + f12 & 5.7e-14 & 2.0e+0 & & 4.7e-1 & 1.7e-6 & 2.0e+0 & 9.0e-1 & 4.6e-1 & & & 2.9e-1 & 4.5e-1 + f13 & & 3.5e+0 & & 8.8e-1 & 2.5e-7 & 3.5e+0 & 1.7e+0 & 1.2e+0 & & & 5.4e-1 & 9.6e-1 + f14 & 2.4e-8 & & & 3.3e+0 & 5.3e-7 & 1.7e+1 & 1.4e+0 & 4.0e+0 & & 1.8e+2 & 1.6e+1 & 3.3e+1 + f15 & 7.3e-5 & 1.2e+2 & 1.8e+1 & 3.1e+1 & 9.9e-5 & & & 2.3e+1 & & 1.2e+2 & 1.5e+1 & 2.6e+1 + f16 & 1.2e-1 & 2.3e+0 & 9.5e-1 & 5.7e-1 & 1.3e-1 & 1.9e+0 & 8.2e-1 & 4.3e-1 & & & & 3.8e-2 + f17 & 2.4e-5 & & & 8.2e-1 & 4.9e-5 & 3.2e+0 & 1.3e+0 & 1.0e+0 & & 3.2e+0 & 1.6e+0 & 9.7e-1 + f18 & 1.4e-1 & 3.1e+0 & 1.6e+0 & 8.2e-1 & 1.3e-4 & & & 1.1e+0 & & 3.1e+0 & 1.7e+0 & 9.4e-1 + f19 & & & & 1.4e-2 & 1.3e-9 & 1.8e-1 & 4.7e-2 & 4.4e-2 & & 1.2e-1 & 1.3e-2 & 2.0e-2 + f20 & 4.3e-6 & 8.6e-1 & 3.5e-2 & 1.2e-1 & 1.9e-2 & 1.6e-1 & 4.0e-2 & 3.4e-2 & & & & 4.6e-2 + f21 & 1.3e-12 & 2.0e+2 & 3.8e+1 & 7.3e+1 & 7.0e-4 & 1.0e+2 & & 2.0e+1 & & & 5.8e+0 & 2.3e+1 + f22 & 2.9e-11 & & & 3.0e+1 & 6.6e-5 & 1.2e+2 & 1.8e+1 & 3.4e+1 & & 1.6e+2 & 4.7e+1 & 4.5e+1 + f23 & 1.6e-8 & 2.9e+2 & 7.9e+1 & 9.8e+1 & 1.8e-3 & & & 2.8e+1 & & 2.8e+2 & 6.2e+1 & 6.7e+1 + f24 & 1.3e-8 & & 1.0e+1 & 2.3e+1 & 4.0e-4 & 2.4e+1 & 8.2e+0 & 9.0e+0 & & 7.2e+1 & & 1.5e+1 + f25 & 2.7e-9 & 1.1e+2 & 4.5e+1 & 4.9e+1 & 1.7e-3 & 1.1e+2 & & 4.3e+1 & & & 4.0e+1 & 4.8e+1 + f26 & 1.6e-9 & & 5.6e+0 & 2.0e+1 & 4.8e-6 & 3.2e+1 & 1.7e+0 & 4.8e+0 & & 2.3e+1 & & 3.4e+0 + f27 & 3.6e+0 & 3.1e+2 & 1.5e+2 & 8.9e+1 & & 1.4e+2 & 8.9e+1 & 4.8e+1 & 8.5e-1 & & & 4.9e+1 + f28 & 2.1e-10 & 3.0e+2 & 4.0e+1 & 9.3e+1 & 1.0e-2 & 1.0e+2 & & 4.3e+1 & & & 3.2e+1 & 4.7e+1 + 580pt x x x x x x x x x x x x x x x x & & & + + & min & max&avg & std & min & max&avg&std & min & max&avg&std + f1 & & & & 0.0e+0 & 5.3e-6 & 1.2e-1 & 3.9e-3 & 1.7e-2 & & & & 0.0e+0 + f2 & 1.0e+1 & 1.1e+4 & 6.9e+2 & 1.6e+3 & 5.5e+2 & 6.3e+5 & 1.4e+5 & 1.6e+5 & & & & 1.4e-3 + f3 & 1.2e+0 & 3.4e+6 & 1.3e+5 & 5.4e+5 & 1.9e+2 & 7.3e+7 & 3.5e+6 & 1.2e+7 & & & & 2.2e+4 + f4 & 1.7e+0 & 8.2e+2 & 8.8e+1 & 1.5e+2 & 3.0e+2 & 9.2e+3 & 3.8e+3 & 2.4e+3 & & & & 8.2e-3 + f5 & & & & 1.6e-11 & 1.1e-4 & 4.7e-2 & 7.0e-3 & 9.5e-3 & 4.9e-9 & 2.2e-5 & 2.6e-6 & 5.1e-6 + f6 & 2.7e-5 & 2.7e+0 & & 6.6e-1 & 1.1e-3 & 5.1e+0 & 1.7e+0 & 1.9e+0 & & & 1.2e+0 & 1.8e+0 + f7 & & & & 7.0e-1 & 4.6e-1 & 5.5e+1 & 1.1e+1 & 1.2e+1 & 4.7e-2 & 2.3e+1 & 4.6e+0 & 6.2e+0 + f8 & 4.6e+0 & 2.0e+1 & 2.0e+1 & 2.2e+0 & 2.6e+0 & & 1.8e+1 & 5.3e+0 & & & 1.9e+1 & 4.8e+0 + f9 & 2.3e-1 & 2.8e+0 & 1.7e+0 & 5.8e-1 & 4.6e-1 & & 1.7e+0 & 6.3e-1 & & 3.1e+0 & & 6.6e-1 + f10 & 9.3e-2 & 4.6e-1 & 2.5e-1 & 9.0e-2 & 3.7e-1 & 3.7e+1 & 6.6e+0 & 7.3e+0 & & & & 5.5e-2 + f11 & & & & 9.3e-1 & 5.0e-4 & 5.0e+0 & 1.8e+0 & 1.0e+0 & 9.9e-1 & 1.8e+1 & 5.6e+0 & 3.7e+0 + f12 & 1.2e+0 & 1.1e+1 & 5.3e+0 & 2.1e+0 & 1.5e-1 & 2.3e+1 & 7.9e+0 & 4.8e+0 & & & & 3.7e+0 + f13 & 1.1e+0 & & & 3.7e+0 & 1.0e+0 & 2.8e+1 & 1.4e+1 & 6.6e+0 & & 2.8e+1 & 9.2e+0 & 4.8e+0 + f14 & 1.2e+1 & 3.2e+2 & 1.2e+2 & 7.7e+1 & & & & 6.1e+1 & 7.2e+0 & 6.9e+2 & 3.5e+2 & 1.6e+2 + f15 & 6.7e+1 & 6.0e+2 & 3.5e+2 & 1.1e+2 & 7.1e+0 & & & 1.1e+2 & & 5.9e+2 & 2.7e+2 & 1.6e+2 + f16 & 6.5e-1 & 2.0e+0 & 1.2e+0 & 3.1e-1 & 3.8e-1 & 1.9e+0 & 1.0e+0 & 3.6e-1 & & & & 2.1e-1 + f17 & 2.9e+0 & 1.4e+1 & 9.5e+0 & 2.0e+0 & & & & 1.9e+0 & 3.0e+0 & 1.9e+1 & 9.7e+0 & 3.1e+0 + f18 & 3.4e+0 & 1.9e+1 & 1.3e+1 & 3.0e+0 & 4.3e+0 & 1.8e+1 & 1.2e+1 & 3.2e+0 & & & & 2.8e+0 + f19 & 1.4e-1 & 9.0e-1 & 4.3e-1 & 1.5e-1 & 2.1e-2 & 2.1e+0 & 5.2e-1 & 3.9e-1 & & & & 1.6e-1 + f20 & 2.7e-1 & 1.5e+0 & 1.1e+0 & 3.0e-1 & & & & 3.4e-1 & 2.9e-1 & 2.5e+0 & 1.4e+0 & 6.2e-1 + f21 & 1.0e+2 & & & 9.8e+1 & 1.0e+2 & 3.1e+2 & 2.7e+2 & 7.4e+1 & & 4.0e+2 & 2.2e+2 & 1.1e+2 + f22 & 1.6e+2 & 5.4e+2 & 3.3e+2 & 9.8e+1 & & & & 1.3e+2 & 1.4e+2 & 9.2e+2 & 6.0e+2 & 1.8e+2 + f23 & 1.4e+2 & 8.4e+2 & 5.7e+2 & 1.3e+2 & & & & 1.7e+2 & 1.1e+2 & 8.9e+2 & 5.5e+2 & 1.9e+2 + f24 & 4.8e+1 & 2.1e+2 & 1.3e+2 & 3.9e+1 & 9.1e+1 & 1.4e+2 & 1.2e+2 & 9.7e+0 & & & & 3.0e+1 + f25 & & & & 2.3e+1 & 1.1e+2 & 1.4e+2 & 1.2e+2 & 9.3e+0 & 1.0e+2 & 2.0e+2 & 1.2e+2 & 1.6e+1 + f26 & 1.5e+1 & & & 2.4e+1 & 5.1e+1 & 1.2e+2 & 1.1e+2 & 1.3e+1 & & 2.0e+2 & 1.0e+2 & 3.2e+1 + f27 & 3.0e+2 & 3.8e+2 & 3.5e+2 & 2.0e+1 & 3.1e+2 & 4.1e+2 & 3.5e+2 & 2.4e+1 & & & & 2.5e+1 + f28 & 1.0e+2 & & & 8.4e+1 & 1.0e+2 & 4.0e+2 & 3.2e+2 & 6.4e+1 & & 4.0e+2 & 2.8e+2 & 8.7e+1 + 580pt x x x x x x x x x x x x x x x x & & & + + & min & max&avg & std & min & max&avg&std & min & max&avg&std + f1 & 1.6e-10 & 6.5e-9 & 1.3e-9 & 1.3e-9 & 2.4e-5 & 3.4e-2 & 2.1e-3 & 6.0e-3 & & & & 7.0e-10 + f2 & 1.4e+5 & 1.6e+6 & 5.4e+5 & 3.4e+5 & 5.8e+5 & 7.3e+6 & 3.4e+6 & 1.7e+6 & & & & 8.6e+2 + f3 & 7.7e+8 & 4.1e+42 & 8.1e+40 & 5.7e+41 & 4.0e+6 & 1.8e+9 & 2.6e+8 & 3.4e+8 & & & & 4.3e+5 + f4 & 2.4e+3 & 2.7e+4 & 8.3e+3 & 4.5e+3 & 4.9e+3 & 1.7e+4 & 9.8e+3 & 2.8e+3 & & & & 1.5e+0 + f5 & & & & 2.5e-6 & 5.1e-4 & 2.2e-1 & 1.4e-2 & 3.2e-2 & 5.2e-5 & 3.4e-3 & 5.9e-4 & 6.0e-4 + f6 & 2.5e-1 & 9.8e+0 & & 4.2e+0 & 5.0e-2 & 8.0e+1 & 3.2e+1 & 3.2e+1 & & & 6.4e+0 & 4.6e+0 + f7 & 7.6e+0 & & & 9.9e+0 & 1.7e+1 & 1.2e+2 & 5.0e+1 & 2.2e+1 & & 8.2e+1 & 3.4e+1 & 2.0e+1 + f8 & 2.0e+1 & 2.1e+1 & 2.0e+1 & 8.6e-2 & & 2.1e+1 & & 1.0e-1 & 2.0e+1 & & 2.0e+1 & 8.6e-2 + f9 & 4.0e+0 & 8.2e+0 & 6.6e+0 & 1.0e+0 & 3.5e+0 & 7.7e+0 & 5.7e+0 & 1.1e+0 & & & & 1.5e+0 + f10 & 2.3e-1 & 9.5e-1 & 6.1e-1 & 1.4e-1 & 1.2e+0 & 3.4e+1 & 7.2e+0 & 6.0e+0 & & & & 3.5e-2 + f11 & 4.0e+0 & 2.2e+1 & 1.4e+1 & 4.1e+0 & & & & 4.3e+0 & 4.0e+0 & 3.8e+1 & 1.6e+1 & 6.9e+0 + f12 & 9.3e+0 & 4.6e+1 & 3.0e+1 & 7.4e+0 & 8.1e+0 & 4.8e+1 & 2.5e+1 & 1.1e+1 & & & & 6.5e+0 + f13 & 1.3e+1 & 5.4e+1 & 3.2e+1 & 9.2e+0 & & 7.4e+1 & 3.6e+1 & 1.4e+1 & 1.4e+1 & & & 1.1e+1 + f14 & 3.8e+2 & 1.4e+3 & 1.0e+3 & 2.5e+2 & & & & 1.2e+2 & 3.6e+2 & 1.5e+3 & 1.0e+3 & 3.0e+2 + f15 & 7.6e+2 & 1.8e+3 & 1.4e+3 & 2.2e+2 & & & & 2.3e+2 & 2.2e+2 & 1.7e+3 & 9.0e+2 & 3.2e+2 + f16 & 3.6e-1 & 1.9e+0 & 1.4e+0 & 3.2e-1 & 5.5e-1 & 1.6e+0 & 1.1e+0 & 2.4e-1 & & & & 2.8e-1 + f17 & 2.1e+1 & 5.0e+1 & 3.7e+1 & 6.9e+0 & & & & 4.2e+0 & 1.2e+1 & 2.9e+1 & 1.9e+1 & 4.3e+0 + f18 & 3.2e+1 & 5.7e+1 & 4.8e+1 & 5.9e+0 & 2.1e+1 & 5.1e+1 & 3.6e+1 & 7.3e+0 & & & & 5.7e+0 + f19 & 9.8e-1 & 3.2e+0 & 2.2e+0 & 4.7e-1 & & 2.3e+0 & 1.1e+0 & 4.6e-1 & 3.2e-1 & & & 3.1e-1 + f20 & 3.1e+0 & & 3.5e+0 & 2.2e-1 & & 4.2e+0 & & 4.5e-1 & 2.6e+0 & 4.5e+0 & 3.6e+0 & 4.5e-1 + f21 & & & & 1.0e+2 & 3.0e+2 & 4.0e+2 & 4.0e+2 & 1.4e+1 & 2.0e+2 & 4.0e+2 & 4.0e+2 & 2.8e+1 + f22 & 3.4e+2 & 1.5e+3 & 1.0e+3 & 2.4e+2 & & & & 1.5e+2 & 6.0e+2 & 2.0e+3 & 1.3e+3 & 3.0e+2 + f23 & 7.1e+2 & 1.9e+3 & 1.5e+3 & 2.3e+2 & & & & 2.9e+2 & 6.2e+2 & 2.1e+3 & 1.3e+3 & 2.9e+2 + f24 & 1.5e+2 & & & 2.4e+1 & 1.2e+2 & 2.3e+2 & 2.0e+2 & 3.7e+1 & & 2.2e+2 & 2.2e+2 & 1.5e+1 + f25 & 1.7e+2 & 2.2e+2 & 2.2e+2 & 7.5e+0 & & & & 3.4e+1 & 1.7e+2 & 2.3e+2 & 2.1e+2 & 8.2e+0 + f26 & 1.3e+2 & 2.0e+2 & 1.6e+2 & 1.9e+1 & 1.1e+2 & & & 2.7e+1 & & 3.2e+2 & 1.8e+2 & 5.6e+1 + f27 & 4.3e+2 & 5.7e+2 & 5.3e+2 & 3.4e+1 & 3.4e+2 & 5.7e+2 & 4.4e+2 & 6.1e+1 & & & & 8.8e+1 + f28 & 1.0e+2 & 3.0e+2 & & 8.3e+1 & 1.0e+2 & 8.2e+2 & 5.7e+2 & 2.3e+2 & & & 4.1e+2 & 1.8e+2 + in these tables the value 0.0e+0 means that the maximal precision of matlab has been reached . from observing the results in tables [ table : dim2 ] , [ table : dim5 ] and [ table : dim10 ] it is noticeable that the new method has the greatest advantage when compared to the two other methods in the case of uni - modal functions . generally the hybridization using nms is most advantageous in a general slope towards the global minima . these two positive effects correspond to the use of the incorporation of the nms algorithm . the flip operation in a certain , loosely defined , sense approximates the gradient of the function . this flip operation gives two effects in the hybridized algorithm . in case of large simplexes , which are created using the mixing of nests global tendencies of the function are included in the search . on the other had , in the case of small ones , the flip operation makes it possible to quickly converge to the nearby local or global optimal solution . the method has had a better performance in the case of functions of 2 and 5 dimensions . when it is compared to the original cs algorithm it presents a significant improvement . it has managed to get lower values for average and minimal solutions , for many test functions , compared to yang s algorithm . the proposed method has lowered the performance of the original algorithm in only a few problem instances . this is contrary to the mcs , which is an overall improvement to the basic algorithm , but has for a wide range of test functions achieved worse results than the original cs algorithm . it is evident that the new method has a less consistent behavior that two competing versions of cs . this can be concluded when we observe the results presented in the tables [ table : dim2 ] , [ table : dim5 ] and [ table : dim10 ] , where the methods has a much higher number of best found minimal solutions than average ones . in this section , we applied the presented method to the practical problem of optimizing the performance of multi - cell solar systems and mainly on multi - junction and split spectrum solar cells . finding the optimal efficiency of such cells is of high complexity due to the fact that the underlining equations are highly nonlinear . previously , research has been conducted in finding limits for different mathematical models.@xcite these problem of finding the in multi - junction solar cells , few cells of different energy gaps ( @xmath80 ) are stacked in layers where in between buffer layers are used to allow transporting the photogenerated carriers between the cells . this type of cells is schematically represented in figure-[mjsc ] and it is conceptually a two - terminal system with a series stack of two - level cells . as @xmath80s differ , each layer ( or cell ) abosrbs a portion of the incident solar radiation ( i.e. for photonenergy above the layer @xmath80 ) . other photons are transimited to the next layer . this selective absortion / transmission is the mean for spectrum division . split spectrum solar cell system , follow a different approach . in this system , the idea is to split solar radiation by a pre - optical setup and then direct each of the split spectrum into a cell with matching @xmath80 . we can see an illustration of this type of cells in figure-[sssc ] . practically , there are few differences between the two concepts . the main difference is current continuity . in split - spectrum system , the photogenerated carriers are collected separately for each cell . so , there is no physical constraint in this regard . however , in multi - junction cells , the photogenerated current from ny cell should be reinjected in the next one . hence , it is essential to maintain the continuity . there have been several strategies to estimate the practical efficiency limits of cells of this type . estimating the upper limit for multi - cell devices such as multi - junction and split spectrum , can be achieved by extending the single junction model @xcite . in this work , we use the detailed balance model develop by hanna and nozik @xcite . in both multi - junction and solar cell , the spectrum is ideally split based on the @xmath80s of the different cells . for @xmath15-cell system where the gaps are ordered ascendingly , the photogenerated generated current in the i@xmath81 cell is @xmath82 @xmath83 is the electron charge , and @xmath84 is photon energy ( in @xmath85 ) , and @xmath86 is the standard am1.5 g flux . for the last cell , the upper limit is @xmath87 . in the ideal case , it is assumed that radiative recombination is the only cause for recombination . so , the recombination current is calculated based on the generalized black body radiation as follow @xmath88 hence , the resulted overall generated current in the i@xmath81 cell is @xmath89 then for each cell , the conversion efficiency becomes @xmath90 in split spectrum cells , the photogenerated current is extracted separately for each cell . so , the total efficiency is @xmath91 for multi - jucntion , it is necessary to have the photogenerated current flow from one cell to the other before extracted in the external terminals . this series connection requires that the current must be the same in all of them and it is equal to the lowest current achieved in any of the cells . this fact resulted in changing the total efficiency to @xmath92 the goal of the optimization is to maximize @xmath93 and hence for practical application , we are actually minimizing the value @xmath94 . in the experiments , the performance of the new method is compared with nms , sa and ga for split spectrum and multi - junction solar cells . the tests are conducted for solar cells that have 4 to 10 layers . for each of the problem sizes , we have compared the quality of results that can be archived using 1500 function evaluations . for nms , sa and ga , methods that are a part of the matlab toolbox are used . we have fine for the sa and the ga we have conducted a large number of tests on the smallest problem sizes to fine tune the parameters available in matlab for defining these methods . when specifying the properties of the sa , we found that the optimal initial temperature for the optimization algorithm was 100.the temperature function used was set with the matlab parameter temperatureexp , in which the temperature at any given step was 0.95 times the temperature from the previous one . the reannaling interval was set to 100 , the step has length square root of temperature , as defined by the matlab parameter value annealingboltz . in case of ga the population size was set to 15 out of which 2 where consider as elitist . the crossover fraction was set to 0.8 . the mutation was set by the matlab parameter value mutationuniform , which correspond to selection rate of 0.01 . from physical properties of the problem , it is known that @xmath80 values should belong to the domain @xmath95 . solar radiation vanishes for photon energy above 4 @xmath85 . for the problem of interest , it is known that high quality solutions can be found close to @xmath80 values that equally divide the photogenerated carriers or if the corresponding voltages are equably divided in the expected range . in case of nms , the search is started using 10 different starting points , out of which 2 correspond to the parameter values close to which it is expected to find good solutions and 8 random selection of parameters . in case of sa , only two starting points for which the highest quality of solutions is expected . we have used the same parameter sets as for nms and 5 new random ones for the initial population . in case of nms - cs , the same initial points as for ga have been used to generate the starting 15 nests ( simplexes ) . the calculation time for all of the methods was very similar due to the fact that the evaluation of the efficiency function is the most computationally expensive part . because of the long execution time , only a single run has been conducted , in the sense that 1500 function evaluations have been done , for each of the methods . in table [ table : sssc ] and [ table : mjsc ] we can see the results acquired for split spectrum and multijunction solar cell respectably . 350pt x x x x x x x number of cells & nms & sa & ga&nms cs + 3&51.348 & 50.876 & & + 4&55.387&53.562 & 55.105 & + 5&57.633&55.419 & 57.630 & + 6&59.110&58.594 & 59.006 & + 7&60.290&59.553 & 60.419 & + 8&61.247&60.228 & 61.315 & + 9&61.871&61.075 & 62.066 & + 10&62.350&61.517 & 62.451 & + in the case of calculating the efficiency of split spectrum solar sells , the use of sa has given results of lower quality than the other methods . nms manages to get better results than ga in case of smaller problem instances , which can be explained by the fact that the smaller size of the solution space , make a local search method more effective . in all of the tested cases , nms cs has managed to produce the results of the best quality . 350pt x x x x x x x number of cells & nms & sa & ga&nms cs + 3&50.961&50.548&50.719 & + 4&54.533&53.243&53.801 & + 5&56.183&55.086&52.141 & + 6&57.848&56.470&57.363 & + 7&59.714&58.556&59.423 & + 8&59.576&58.949&&60.140 + 9&60.619&59.073 & & 59.051 + 10&61.098&59.634 & & 61.231 + in case of multi junction solar cells , where the minimization problem becomes more complex due to the existences of constraints , similar results have been achieved . although results are less conclusive , since only one test run has been performed , which can make the choice of the random seed influence the performance of the different methods . an example of this are the performance of ga and nms cs in the case of 5 and 9 cells respectably , where the found efficiency is much lower than expected . we believe the reason for this is that the method has been trapped in a strong local minima . the new method has achieved better quality results in case of smaller problem instances , but has been slightly outperform by ga in case of the 3 problems of the greatest size . it is important to point out , that hybridized method has improved the performance of nms in all but the one extreme case . in this paper , we have presented a new method for the minimization of multi - parameter functions , with the focus on its application for the maximization of efficiency of multi - cell solar systems . the method is a hybridization of the cs with nms algorithm . the new approach incorporates some fundamental parts of the nms into cs . more precisely , it uses simplexes for nests in the cs , which makes it possible to use the nms flip operation instead of the levy flight . in this way , the good properties of nms , as a local search , are preserved while the new methods manages to also conduct a effective exploration of a larger solution space . in comparison to the original cs algorithm , the hybridized approach becomes more capable for finding high quality solutions since it is combined to a local search method . a second advantage of the new method is the higher level of robustness , since it avoids using some tuning parameters that exist in cs . the new approach has been tailored for the specific problem of solar cell optimization . although the function corresponding to this problem has a relatively low number of parameters 3 - 10 , the complexity comes from the fact that the its evaluation is computationally expensive . in our experiments , we have shown that the new approach manages to outperform nms , ga , and sa for this problem . to better evaluate the new method we have also compared its performance to other hybridizations of nms using population based meta heuristics like ga and pso . we have shown that the new method is very competitive , and in many cases outperforms , these algorithms on standard benchmark functions . one of the main advantages of the new method is its simplicity of implementation , especially if it is compared to other hybridizations of this type . r. babukarthik , r. raju , p. dhavachelvan , hybrid algorithm for job scheduling : combining the benefits of aco and cuckoo search , in : n. meghanathan , d. nagamalai , n. chaki ( eds . ) , advances in computing and information technology , vol . 177 of advances in intelligent systems and computing , springer berlin heidelberg , 2013 , pp . 479490 . a. s. brown , m. a. green , r. p. corkish , limiting efficiency for a multi - band solar cell containing three and four bands , physica e : low - dimensional systems and nanostructures 14 ( 12 ) ( 2002 ) 121 125 . r. chelouah , p. siarry , genetic and neldermead algorithms hybridized for a more accurate global optimization of continuous multiminima functions , european journal of operational research 148 ( 2 ) ( 2003 ) 335 348 . f. dimroth , world record solar cell with 44.7% efficiency , http://www.ise.fraunhofer.de/en/press-and-media/press-releases/presseinformationen-2013/world-record-solar-cell-with-44.7-efficiency ( oct . 2013 ) . m. i. hossain , a. bousselham , f. h. alharbi , optical concentration effects on conversion efficiency of a split - spectrum solar cell system , journal of physics d : applied physics 47 ( 7 ) ( 2014 ) 075101 . r. jovanovic , m. tuba , i. brajevic , parallelization of the cuckoo search using cuda architecture , in : proceedings of the 7th international conference on applied mathematics , simulation , modelling ( asm 13 ) , recent advances in mathematics , 2013 . p. koduru , s. das , s. welch , a particle swarm optimization - nelder mead hybrid algorithm for balanced exploration and exploitation in multidimensional search space . , in : h. r. arabnia ( ed . ) , ic - ai , csrea press , 2006 . j. liang , b. qu , p. suganthan , a. g. hernndez - daz , problem definitions and evaluation criteria for the cec 2013 special session on real - parameter optimization , computational intelligence laboratory , zhengzhou university , zhengzhou , china and nanyang technological university , singapore , technical report 201212 . n. sharma , n. arun , v. ravi , an ant colony optimisation and nelder - mead simplex hybrid algorithm for training neural networks : an application to bankruptcy prediction in banks , international journal of information and decision sciences 5 ( 2 ) ( 2013 ) 188 203 . m. subotic , i. misic , m. tuba , an object - oriented implementation of the firefly algorithm for continuous unconstrained optimization problems , in : proceedings of the 6th wseas international conference on computer engineering and applications , and proceedings of the 2012 american conference on applied mathematics , american - math12/cea12 , world scientific and engineering academy and society ( wseas ) , stevens point , wisconsin , usa , 2012 . m. subotic , m. tuba , n. bacanin , d. simian , parallelized cuckoo search algorithm for unconstrained optimization , in : proceedings of the 5th wseas congress on applied computing conference , and proceedings of the 1st international conference on biologically inspired computation , bica12 , world scientific and engineering academy and society ( wseas ) , stevens point , wisconsin , usa , 2012 . s. walton , modified cuckoo search ( mcs ) open source gradient free optimiser , https://www.mathworks.com/matlabcentral/linkexchange/links/2999-modified-cuckoo-search-mcs-open-source-gradient-free-optimiser ( apr . 2014 ) . g. wang , l. guo , h. duan , h. wang , l. liu , m. shao , a hybrid metaheuristic de / cs algorithm for ucav three - dimension path planning , the scientific world journalarticle i d 583973 , 11 pages , doi:10.1100/2012/583973 .

we report a new fabry - perot search for h emission from the intergalactic cloud hi 1225 in an attempt to measure the low redshift ionizing background radiation . we set a new upper limit on h emission of 8 mr ( ergs s arcsec ) . conversion of this limit to limits on the strength of the ionizing background requires knowledge of the ratio of the projected to total surface area of this cloud , which is uncertain . we discuss the plausible range of this ratio , and within this range find that the strength of the ionizing backround is in the lower range of , but consistent with , previous observational and theoretical estimates .

astro-ph0107534

the value of the local ionizing flux is of interest for several reasons : 1 ) it plays an important role in setting the level of ionization of the low density hydrogen gas which gives rise to the ly@xmath0 forest . recent simulations have been remarkably successful in reproducing the main features of the ly@xmath0 forest from high redshifts to zero redshift @xcite . as discussed in detail in 4 , for a given set of cosmological parameters , these simulations are now able to make predictions about the value of the local h i photoionization rate , and it would be valuable to determine empirically the value of this rate to compare it with the simulations . with a sufficiently accurate determination of its value and further forthcoming data on the low redshift ly@xmath0 forest , together with further improvements in the simulations it may be possible to set some constraints on these cosmological parameters , especially the baryon number . 2 ) a limit on the photoionization rate in locations which are free of local ionizing sources ( i.e. star - forming regions or ionization due to high energy particles in the neighborhood of galaxies ) can then be used to assess the relative importance of such local sources near a galaxy compared to the non local , global value . in particular , there is currently some controversy concerning the ionization source for some high velocity clouds in our own galaxy @xcite . 3 ) estimates of the photoionization rate due to the cumulative effect of active galactic nuclei can now be carried out with fair confidence ( * ? ? ? * hereafter srgpf ) ; but there remain significant uncertainties associated primarily with the luminosity functions of these objects and with the evaluation of the opacity associated with intervening neutral hydrogen . 4 ) similar remarks apply to the contribution from escaping radiation shortward of the lyman limit from hot stars in galaxies . however , this estimate is also dependent upon the fraction of the lyman continuum radiation assumed to escape from the galaxies , and direct observational evidence about the magnitude of this escape fraction is still somewhat scanty . for these reasons , we have previously set limits on the value of the non - local ionization rate at zero redshift by setting limits on the h@xmath0 flux from an isolated , large , h i cloud which is optically thick to the lyman continuum ( * ? ? ? * hereafter vwrh ) . the essence of the method is very simple . illumination by ionizing radiation of an h i cloud that is optically thick to the lyman continuum results in h@xmath0 emission that is well described by case b recombination theory . accordingly , approximately one h@xmath0 photon escapes for every two incident lyman continuum photons ; the exact value depends only mildly on parameters that are reasonably well known . the one significant uncertainty is the ratio of the observed projected surface area to that illuminated by the ionizing radiation ; however , the uncertainty is perhaps a factor of two , and can be better determined for objects where the geometry is known , such as galactic disks . the expected h@xmath0 radiation is extremely faint based on predictions for the low redshift metagalactic ionizing flux ( e.g. , srgpf ) , the h@xmath0 surface brightness is expected to be only a few millirayleighs ( for h@xmath0 , 1 mr @xmath7ergs @xmath4 s@xmath5 arcsec@xmath6 ) . by comparison , the observed h@xmath0 emission from objects ionized by other sources is significantly brighter : high velocity clouds known to be within a few kpc are typically 60200 mr @xcite ; most compact high velocity clouds lie between 30 to 300 mr @xcite ; the magellanic stream ranges up to more than 1000 mr @xcite ; a brightness of 90 mr was measured in the outer disk of ngc 253 @xcite ; and the diffuse extraplanar ionized gas imaged in edge - on spirals is typically brighter than 1000 mr @xcite . as a result of its expected faintness , detection of h@xmath0 emission due to ionization by the metagalactic ionization requires observation of extended clouds using specialized techniques . there exist other approaches to an empirical value for the global zero redshift ionization rate . however , several of these methods suffer from significant limitations . at higher redshifts , the proximity effect @xcite has been used to estimate the value of the metagalactic ionizing flux ; however at low redshift the sparseness of higher resolution data and the relative sparseness of the ly@xmath0 forestitself makes this method currently less useful . in addition , there are some issues about the interpretation of the proximity effect ( e.g. , the clustering properties of matter in the vicinity of qsos ; see @xcite for further discussion ) . a potentially very powerful technique involves observing the rapid transition from a mostly neutral hydrogen disk to a mostly ionized disk as one moves outward along the radius of spiral galaxies . from the value of this transition column density , one can infer the flux of the background ionizing radiation given a model for the vertical distribution of gas . the basic idea for this technique goes back to papers by @xcite and @xcite . the historical development of this idea is summarized by @xcite who applied it to realistic galaxy models . maloney gives a careful treatment of the radiative transfer equation and discusses a wide range of galaxy parameters . his results show that the derived ionizing flux is relatively insensitive to the dynamical models adopted for the galaxy . the application to ngc3198 , which has the most sensitive hi observations in this context , and whose geometry and dynamics are not very uncertain , yielded a transition column density of about @xmath8 and from this maloney deduced that the flux of ionizing photons is in the range @xmath9 to @xmath10 . shortly after this work @xcite published a similar analysis and confirmed these results . they also showed that the results are not very sensitive to the assumed spectral index of the ionizing background . application of this method to other objects has also been carried out . in particular , as noted above , measurements of h@xmath0 emission beyond the hi truncation radius were obtained by @xcite in ngc253 . in this case , however , these authors noted that the inferred surface brightness was higher than could likely arise only from the metagalactic radiation field . in addition , the ratio of [ nii ] @xmath116548 to h@xmath0 emission was unusually high . these authors favored a model in which a warp in the outer disk allowed ionizating flux from the inner young star population to dominate the ionization . in ngc253 the ratio of the hi disk radius to the optical disk radius is considerably smaller than in ngc3198 . nevertheless , @xcite do not reject the possibility that hidden local sources of ionization may contribute significantly to the ionizing flux even in ngc3198 . we note that there appear to be instances where no abrupt truncation of h i occurs to column densities which are smaller than @xmath12 @xmath13 . for example , @xcite present hi images of ngc 289 , a giant gas - rich low - surface - brightness spiral , and show hi contours down to a column density of @xmath14 @xmath4 ; they note that no truncation is evident . in ddo 154 , @xcite find no truncation , down to a level of @xmath15 @xmath4 . this might be indicative of a value for the ionizing background somewhat lower than that derived for ngc3198 , but it is also possible that the low observed hi column densities are a consequence of beam dilution ; with better angular resolution , the actual column densities might be observed to be substantially higher . in addition , however , as emphasized by maloney ( 2001 , private communication ) , the abruptness of the h i cutoff depends upon the run of the total hydrogen column density which is not _ a priori _ known , and the ratio of the neutral to total hydrogen column density is sensitive to the total hydrogen column density in the regime where the predominate ionization state switches from neutral to ionized ( cf figures 4 and 8 of maloney 1993 ) . in fact , the estimate of the background radiation in ngc3198 was carried out by using the observed h i column density to model the total hydrogen column density for assumed values of the ionizing background . it is the uncertainty of the true radial variation of the total column density of hydrogen at these large radii that appears to be the principal uncertainty in the determination of the background radiation using this technique . in any event , if the relatively large value of the metagalactic ionizing flux implied by the hi truncation of ngc 3198 is correct , then the h@xmath0 emission produced by this ionization should be detectable with achievable sensitivities in instances where the hi column density is high enough to ensure that most of the ionizing photons are absorbed . for these reasons , we continue to believe that the use of h@xmath0 surface brightness measurements as described in vwrh still provides the route to the least model - dependent estimate for the metagalactic ionizing flux , _ provided _ that ( i ) the targets are relatively isolated from star forming regions and ( ii ) that the geometry is well understood , for the reason discussed in section 3.3 consequently , when subsequent further improvements in our instrumentation were made , we again carried out basically the same series of observations as described in vwrh , which resulted in substantial further lowering of an upper limit on the zero redshift photoionization rate . the purpose of the present paper is to describe and discuss these observations . the selected target , hi 1225@xmath101 , is the same as in vwrh . this large and isolated cloud of h i was discovered serendipitously by @xcite , and subsequent mapping by @xcite showed it to consist of two distinct components a northeast component ( ne ) and a southwest component ( sw ) , with a weaker bridge of h i connecting the two . the ne component has embedded in it stars and a small region of star formation ( * ? ? ? * and references therein ) . however the sw component shows no trace of a stellar component ( * ? ? ? * and references therein ) . for this reason , we chose to observe only the sw component since the relatively bright h@xmath0 emission from the hot stars in the ne component would mask the weak diffuse signal we are attempting to detect . subsequently , a higher resolution h i study ( 40 ) using the vla was carried out by @xcite ; they showed that the sw component appears to be a flattened inclined disk with a systematic velocity @xmath16 km s@xmath5 and a maximum rotational velocity of @xmath1213 km s@xmath5 . we discuss the properties of the sw component further in 3.3.2 . the cocoon of h i in which the optical central part of the ne component resides makes it unlikely that ionizing radiation from the hot stars in this component will ionize a significant portion of the sw component , and the isolation of hi 1225@xmath101 from other galaxies makes the sw component an attractive target for this program . the plan of the remainder of the paper is as follows : in 2 we describe the instrumentation and observational procedures used for the results described below and in 3 we describe the reduction procedures . we present the results in 4 along with a discussion of their implications for some of the issues described above which motivated these observations . the observations described here were carried out at the carnegie observatories on las campanas , using the dupont 2.5 m telescope . a reducing camera was constructed for this purpose . the optical design was carried out by h. epps , with mechanical design by a. schier and associates . the design consisted of a collimator with an on - axis beam size of 70 mm and an effective focal length of 525 mm to match the f/7.5 incoming beam from the cassegrain focus of the dupont telescope . a 70 mm fabry - perot etalon ( whose properties are described in [ etalon ] ) was placed in the collimated beam . an f/2.5 camera reimages the telescope focal plane onto a site 2048 x 2048 ccd with 24 micron pixels , and a scale at the ccd of approximately 0.78 per pixel . the ccd was operated in a mode which yielded a readout noise of about 6e per pixel . however , as explained below , we were obliged to accept a tradeoff between adequate sampling of the signal and readout noise , and thus we binned the image in @xmath17 pixels . the combined collimator - camera reducing camera is achromatic over a very broad range and yields an unvignetted field of about 25 in diameter . the university of maryland et-70 etalon was used for this experiment . the etalon and cs100 controller are made by queensgate . the 70 mm diameter etalon has a gap of 44 @xmath18 , and a free spectral range ( fsr ) of 48.7 at 6590.4 , the expected center of the redshifted h@xmath0 . the fwhm of the instrumental response is 1.40 , and the effective bandwidth ( see 3.1 ) is 1.64 . the etalon was flushed with dry nitrogen during the observations and calibrations . the etalon can be tilted so as to direct reflections out of the field of view . a four - cavity , 6-inch square , order - isolating filter was located near the telescope focal plane . the filter has an intrinsic fwhm of 38 , with a transmission greater than 75% for 6581 @xmath19 . since the etalon fsr is 48.7 , in this wavelength range photometric calibration on continuum objects does not suffer from contamination from other orders . we observed hi 1225@xmath101 the nights of 1997 april 710 using the reducing camera and fabry - perot etalon described in [ instrumentation ] . the first three nights were photometric , while the atmospheric transparency varied during the course of the last night . consequently , only data from the first three nights were used in the analysis described here . to reach the very low surface brightness required for this experiment demanded careful attention to observing and reduction strategy . both shot - noise and systematic errors could contribute to the error budget . a key strategy for minimizing systematic errors is to switch as rapidly as possible between the target and sky positions . the lower limit on the switching time is set by the need to minimize the relative error contributed by ccd read noise . as discussed in [ instrumentation ] , the reducing camera was designed for this experiment with high throughput , high speed , and a large field of view . nonetheless , due to the narrow effective bandwidth ( 1.64 ) , further steps were required to make the relative contribution of ccd read - noise to the total error sufficiently small . first , as noted above we used 4@xmath204 on - chip binning of the ccd ; this yielded 3.1 pixels . hereafter we will use `` pixel '' to refer to one of the @xmath17 binned pixels . second , all exposures times were 900 seconds in length . to minimize errors introduced by variations in atmospheric and instrumental response , we adopted the following observing procedure . first , we observed blank sky at a position offset in ra to the west ( referred to as w ) of the subsequent hi 1225@xmath101 target observation ( referred to as gh1 ) , with the ra offset chosen so that each pair of w and gh1 exposures was observed through very similar azimuth and elevation ranges . in this way , we sought to minimize the effects of systematic errors due to azimuthal and elevation variations in the atmospheric , telescope , and instrument response . to eliminate linear temporal drifts , we repeated an exposure on hi 1225@xmath101(gh2 ) followed by an exposure on a sky position offset to the east ( e ) . a w - gh1-gh2-e set of four exposures , observed in the manner described , greatly reduced errors due to directional and temporal variations in the instrumental and atmospheric response . we also spatially dithered each exposure by an amount varying from 5 to 45 in order to minimize errors due to stars and other features . the monochromatic response in the focal plane is along a ring whose radius @xmath21 varies according to the usual @xmath22 where @xmath23 is the wavelength that transmits on - axis and @xmath24 is the camera focal length . as usual , @xmath23 ( and hence @xmath11 ) is adjusted by varying the etalon gap spacing , @xmath25 , according to @xmath26 , where @xmath27 is the order , @xmath28 is the refractive index of the nitrogen gas in the etalon gap , and @xmath29 is the angle with respect to the optical axis . in figure [ ring_image ] we show a typical 900 second exposure toward the target position . due to the large field , non - zero etalon tilt angle , and the quadratic nature of the wavelength response ( e.g. , equation [ ring_radius ] ) , three to four orders are present ; i.e. three to four rings are present for each night sky line . the on - axis order is 136 , with orders 135 , 134 , and 133 also present off the optical axis . with our large , binned pixels , the quadratic nature of the wavelength response results in a wavelength shift comparable to the spectral resolution across a pixel for orders 135 , 134 , and 133 . since the wavelength resolution is consequently slightly degraded for the outer orders , we extract separate spectra for each order . by the well known jacquinot relation , @xmath30 ( where @xmath31 is the angular area and @xmath32 is the resolution ) , the angle on the sky subtended by a complete monochromatic ring is independent of radius @xmath21 . clearly , we want as much as possible of the ring corresponding to the redshifted h@xmath0 line to be projected within the hi 1225@xmath101 cloud . but we also want as much spectral coverage on both sides of h@xmath0 as possible . coverage at shorter wavelengths extends to the blocking filter cutoff at 6577 and includes the galactic [ nii ] 6584 line and the atmospheric oh 6577 line . however , to observe emission at a given @xmath11 ( in particular around 6590.4 , the wavelength expected for the h@xmath0 centroid in hi 1225@xmath101 sw ; see [ reduction_details ] ) , equation ( [ ring_radius ] ) shows that the maximum wavelength @xmath23 ( i.e. the wavelength observed on axis ) increases as @xmath21 is increased . in order to keep a substantial fraction of the 6590.4 ring within the cloud boundary , we used a ring radius of 5565 pixels ( 170200 ) . this keeps roughly half of the ring within the cloud boundary but cuts off the red end of this spectral order at approximately 6593 which yields a rather short range of spectrum to the red of the redshifted h@xmath0 over which to define the continuum . the etalon and the blocking filter are normally tilted to direct reflections out of the field of view . to maximize the ring area on target , we chose an intermediate tilt angle for the etalon such that the entire ring at 6590.4 lies within the field of view . as shown by equation ( [ ring_radius ] ) , the wavelength transmitted by the etalon varies as a function of position . the usual method for constructing a spectrum is to obtain multiple exposures , stepping the etalon gap @xmath25 through a full free spectral range , or at least sufficiently to fully sample the line profile and continuum at all positions . however , the long exposure times make this impractical . instead , we know that h@xmath0 emission ionized by metagalactic radiation will be nearly uniform in brightness for regions that are optically thick to the lyman continuum ( as described in more detail in [ halpha_to_jnu ] ) . consequently , we can assemble the spectrum using the varying spectral information provided by the different pixels in the field of view . it is essential that the pixels used for the h@xmath0 portion of the spectrum be limited to those pixels that lie towards regions of hi 1225@xmath101 that are optically thick to the lyman continuum . pixels which transmit non - h@xmath0 wavelengths are not restricted to lie towards the cloud . figure [ sky_spectrum ] shows the average sky spectra obtained for one night for orders 136 and 135 , along with the measured transmission of the blocking filter in the tilted configuration used for the observations . three bright oh atmospheric lines are evident at 6577 , 6596 , and 6604 , along with a weak oh line near 6592 ; only the 6577 and 6592 lines lie within the spectral coverage of order 136 . also apparent is a bump near 6583 corresponding to galactic [ ] . continuum due to light scattered from the atmosphere , zodiacal dust , and galactic sources is also present at a level of @xmath33 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 . for flux calibration , we observed standard stars each night ; the stars selected included eg 274 , ltt 4364 , ltt 6248 , and ltt 7379 , using equation 3 in 3.1 to obtain absolute fluxes . most fabry - perot programs acquire observations of a diffuse monochromatic calibration source at intervals of 30 to 60 minutes during the night to monitor and correct the wavelength response . our program yields high signal / noise ratio spectra of the atmospheric night sky lines ; as described in [ reduction ] these can be used for wavelength calibration . since they are observed simultaneously with the target , using them provides a more accurate calibration and also minimizes the need for additional calibration observations during night time observing . daily calibrations were acquired . these included bias and dark frames , dome and twilight flats with the etalon gap set to the spacings used during the night , and neon calibration cubes . we varied the tilt angle for the etalon from night to night , and obtained calibrations appropriate to each night . due to the narrow effective bandpass , low surface brightness and relatively small solid angle over which we expect a signal , we are severely shot - noise limited and in addition must deal with a number of subtle systematic effects which further degrade the signal . therefore , in addition to the observing procedure described above , the reduction of the data must deal as well as possible with these difficulties and hence we describe in some detail various steps in the reduction of the data . we first give a simple overview of the basic reduction procedure and the approximations implicit in it for both the flux calibration observations and the target itself . the count rate in adu s@xmath5 expected from a very small patch of sky ( e.g. , a pixel corresponding to solid angle @xmath34 ) may be expressed as @xmath35 in this expression @xmath36 is the photon surface brightness per as a function of wavelength and direction . at any given point on the detector , the fabry - perot will have its maximum response at some wavelength @xmath37 , and at this wavelength and direction we denote the overall effective collecting area of the system as @xmath38 ( not including the atmospheric attenuation , @xmath39 and the ccd gain in adu e@xmath5 , @xmath40 , which are included separately in equation ( [ response ] ) ) . this effective area includes the spatial and wavelength dependence of the blocking filter and the pixel - to - pixel sensitivity of the ccd , determined via flat field images , as described below in 3.2 . however , the etalon will have some response in this direction to photons differing from @xmath37 and we denote this response function relative to its maximum response by @xmath41 . in principle , as in any system with finite spatial and wavelength resolution , we would need to invert equation ( [ response ] ) to reconstruct the actual distribution of surface brightness , @xmath36 . in practice , spectra for both the standard star and the target are constructed as follows . 1 ) we define a fine grid in wavelength with a uniform spacing , @xmath42 of 0.11 . 2 ) for each pixel within the target mask ( described below ) we find the wavelength @xmath43 associated with the center of that pixel , calibrated using the series of neon lamp observations described below . 3 ) we associate all the counts in that pixel with the wavelength bin in this fine grid closest to @xmath43 . after extraction on this 0.11 interval grid , the spectra were then binned on a coarser 0.34 grid , and all plots , fits and analyses of the spectra use the 0.34 grid . we refer hereafter to each element of this coarser grid as a `` wavelength bin '' to distinguish it from the spatial pixels . consider first the observations of an early type standard star for calibration of the throughput of the system . in this case , over the wavelength region of interest the flux is constant with wavelength and the spatial structure is simply the seeing disk . in this case , the total count rate from the star will be @xmath44_{std } \ \delta\lambda_{eb}\ ] ] in this expression , @xmath45_{std}$ ] is the ( known ) monochromatic flux ( in photons @xmath4 s@xmath5 @xmath5 ) , from which the effective area at this wavelength and this location on the detector may be calculated . the quantity @xmath46 is the effective bandwidth of the etalon , which is the response function @xmath47 ( equation 2 ) integrated over @xmath11 . the response function of the fabry - perot is quite accurately represented by a voigt profile , and we determine the effective bandwidth empirically using neon comparison lines . when the procedure described above is applied to an unresolved continuum object , the only error is simply that associated with the aliasing of a detector which is not adequately sampled if the intra - pixel response is not uniform ; however , this effect can be minimized by small dithering of the standard star observations . for the observation of the target , the opposite situation holds : the spatial scale is much larger than the pixel size , but the expected wavelength dependence in the neighborhood of the emission line is , in our case , comparable to the etalon resolution . in this case , the count rate from the source if it were finely sampled spatially can be shown to be just the convolution of the intrinsic emission line profile with the etalon response : @xmath48 where @xmath49 is the wavelength - integrated photon surface brightness , and @xmath50 is the normalized intrinsic line profile . we have omitted the angular dependence for clarity . the instrumental profile @xmath32 is a voigt function , and as discussed in [ reduction_details ] the convolution represented by the integral in equation ( [ extended ] ) results in an observed profile which is also well described by a voigt function . then the photon surface brightness @xmath49 is obtained from the count rate summed over the line profile @xmath51 as @xmath52 the effective area @xmath53 , where @xmath54 is the telescope area and @xmath55 is the efficiency of the entire optical path including telescope and instrument optics , blocking filter , etalon , and ccd . using the standard star observations , we determined that @xmath56 . photon rates were converted to surface brightness , corrected for the small amount of galactic extinction , which from @xcite amounts to about 0.05 mag in the r band . in practice , as in conventional spectrographs , if the sampling due to the pixel size is not adequate , there may be aliasing of the spectrum . for the etalon order nearest the optical axis of the etalon , the sampling is adequate in spite of the @xmath17 binning . however for the remaining orders this is no longer the case . if each wavelength bin were only sampled once , this could result in an emission line profile whose center and width were poorly defined . in fact , this situation is entirely mitigated because the format produced by the etalon causes each bin in the wavelength grid to lie at more - or - less random positions on the more than 100 pixels contributing to each wavelength bin , just as `` dithering '' does in a conventional undersampled spectrograph . the first steps in the data reduction for all calibration and data ccd frames were bias and dark frame subtraction ; bias and dark frames were constructed from the median value for each pixel derived from stacks of bias and dark frames in order to minimize the noise contribution from this step . next , spatial variations in the instrument and ccd response were removed by dividing by a normalized flat - field frame obtained from a stack of exposures of the standard flat - field lamp illuminating a screen on the side of the dome . we used a flat - field frame obtained at the same etalon gap spacing @xmath25 as used for the observation ; we masked pixels for which the flat - field response was less than 60% of the peak . the wavelength response given by equation ( [ ring_radius ] ) assumes ideal optics . in practice , there are deviations that result in slight shifts in the response ; for optimum resolution we characterize these deviations as follows . we illuminate the dome screen with a neon lamp , stepping the etalon gap over a free spectral range in steps of half the wavelength resolution . then at each pixel we determine the etalon gap that maximally transmits the neon calibration line . in this way , we characterize the deviations from the ideal wavelength relation given by equation ( [ ring_radius ] ) . as mentioned in [ observations ] , the standard method to monitor and calibrate temporal variations is to take periodic exposures of a diffuse monochromatic calibration lamp . because we have bright night sky lines of known wavelength in each exposure , it is more accurate and efficient to use these . in principle , we could determine the centroid and radius of one of the night sky rings ; however , in practice stars and cosmic rays make this difficult . instead , we used the following method . the wavelength error caused by an incorrect choice of optical axis and etalon gap varies as a function of position in the field . the night sky lines present in the field ( e.g. , fig . [ ring_image ] ) have known wavelengths . we compare the known wavelengths with those computed for a large range of assumed optical axes and etalon gap , and select the axis and gap which yield the smallest difference between the calculated and known wavelengths . we used the tukey biweight statistic as described by @xcite to estimate the count rate per pixel for each wavelength bin based on all the pixels contributing to that wavelength bin . as described by @xcite , the biweight estimator is more resistant against outliers and more robust than the mean , and more efficient than the median . it is highly effective at minimizing the effects of cosmic rays and stars . to check that variations across the frame do not cause offsets between the h@xmath0 part of the spectrum and the rest of the spectrum , we divided the frame into several regions and compared the spectra extracted from each region . differences in these spectra were small , and are dominated by scattered light from stars ; this effect is described next . even though spectral features in stars in the field are likely to be weak , these stars can nonetheless create spectral artifacts . recall that the count rate in each wavelength bin is determined from the biweight of the count rate for the ccd pixels corresponding to that wavelength bin . although stars are effectively rejected by the biweight estimator , which excludes pixels with count rates that deviate by more than four standard deviations from the central location , scattered light from brighter stars can cause difficulties . this is because portions of the extended skirts of the point spread function can be too weak to be rejected by the biweight estimator , yet extend over a sufficiently large number of pixels so as to bias the biweight . accordingly , we adopted the following procedure . first , we excluded regions for which scattered stellar light was detectable by visual inspection of the frames . since the error introduced by a star increases with the size of the region that escapes the biweight cutoff , and since the larger regions are of course associated with the brightest stars , this was straightforward , and in itself greatly reduced much of the problem associated with star light . this eliminated artifacts in individual spectra that were as large as @xmath3 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 , although making a difference of less than @xmath57 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 to the final averaged spectrum . as an added precaution , we identified the remaining stars in the field using the program sextractor @xcite . next , using a point spread profile determined on brighter stars , we subtracted the scattered light contribution from the remaining stars ; the resulting changes to the spectra were minor . as described in [ halpha_to_jnu ] , the incident ionizing flux is inferred from the h@xmath0 brightness toward regions with observed column densities @xmath58@xmath4 using the 75 h i maps provided in digital form by @xcite . therefore , we excluded regions with lower h i column densities in constructing the spectrum for the wavelength interval expected for the redshifted h@xmath0 emission , 6588.9 @xmath59 6591.9 . because pixels with wavelengths outside this range were not excluded , the number of pixels per wavelength bin over the wavelength interval defining the continuum was a factor of two to three higher than for the region of line emission . in figure [ wcce_image ] we show the image resulting from directly adding the gh1 and gh2 frames of one w - gh1-gh2-e set and subtracting the w and e frames . the oh night sky lines largely cancel , while the [ ] line appears black , corresponding to brighter [ ] emission in the gh1 and gh2 frames than the sky frames . note the patchy spatial distribution of the [ ] emission and the presence of two well defined velocity components . also note that the oh line residuals appear to have a slight offset in the subtracted image . this occurs because of drifts in the etalon . because of such drifts , spectra are extracted not from w - gh1-gh2-e combined frames such as shown here , but from the individual frames because drifts in the wavelength scale can be corrected during the extraction process , as described above . the ring corresponding to the approximate location of the expected redshifted h@xmath0 signal with centroid at 6590.4 is shown by the solid - line circle . target minus sky ( i.e. on - off ) spectra produced for orders 136 and 135 from each w - gh1-gh2-e set of four spectra are shown in figures [ all_spectra_136 ] and [ all_spectra_135 ] ; each on - off spectrum corresponds to 30 minutes integration time on the target . the error bars shown on the bottom spectrum of figures [ all_spectra_136 ] and [ all_spectra_135 ] are calculated from poisson statistics ; the errors are larger at the wavelengths of night sky lines due to the larger count rates , and in the wavelength range 6588.9 @xmath59 6591.9 due to the smaller number of unmasked pixels . significant non - zero residuals are seen at the wavelengths of the galactic [ ] line ; this is due to the difference in high galactic latitude [ ] emission between the on and off positions . also the atmospheric lines and the continuum level typically show non - zero residuals due to temporal and spatial variations in night sky brightness . figure [ all_averages]a is reproduced from figure [ sky_spectrum]b to enable comparison with figure [ all_averages]b which shows the target minus sky spectra for orders 135 and 136 averaged over the 15 w - gh1-gh2-e spectra obtained during the run . in figure [ all_averages]c , we show a magnified view of the spectra , including error bars . the error bars in figure [ all_averages]c are estimated from the scatter at each wavelength between the 15 spectra that contribute to the average and have decreased by approximately a factor of four from the individual w - gh1-gh2-e spectra , as expected . small negative residuals remain at the oh wavelengths , at a level of about 1% of the line strength , due to nonlinear temporal variations in the oh line strength . there is also a non - zero residual continuum level . we determined this residual continuum level ( assumed constant with wavelength over this narrow range ) by averaging over 6598.4 @xmath59 6601.5 and 6606.3 @xmath59 6610.8 for order 135 . the residual continuum level is @xmath60 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 , about 0.3% of the continuum brightness . as expected the percentage continuum residual is smaller than the oh residual since the oh lines are intrinsically more variable . there is a weak oh line at 6591.7 evident in figure [ all_averages]a and which also has been identified by osterbrock et al . from compilation of keck spectra @xcite . we fit each of the four oh lines in figure [ all_averages]a to determine the ratios of the oh line intensities . the 6591.7 line is about 10% of the strength of the 6596 line , similar to what can be derived from the @xcite keck data . based on the residual negative oh emission of the bright oh line at 6596 of about @xmath61 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 , we expect a peak residual of approximately @xmath62 ergs @xmath4 s@xmath5 arcsec@xmath6 @xmath5 for the 6591.7 line . we fit the residual fluxes in each of the oh lines shown in figures [ all_averages]b and [ all_averages]c using a voigt function ; for the fit , the gaussian and lorentzian line width parameters were held fixed using values determined from a fit to the neon calibration line . also , the wavelength was fixed at the known value of the oh lines , and the continuum was fixed at zero since the fits were made to continuum subtracted data . the residual [ ] peaks in figure [ all_averages]b are at @xmath6350 km s@xmath5 and @xmath180 km s@xmath5 respectively . the brightness of this [ ] feature is close to the detection threshold of the observations of known high velocity clouds made with the same instrument @xcite , but with the sensitivity of the present observations it is a highly significant detection and clearly not an artifact . the brightness of the 80 km s@xmath5 feature varies significantly on scales of roughly an arc minute , and arcs corresponding to this feature can even be seen in a few locations in figure [ ring_image ] . in attempting to subtract off the [ ] from the spectrum it is necessary to model the feature with both components due to the broad voigt wings in the etalon point spread function . unfortunately , however , the strength and extent in velocity space of this [ ] makes the locating of the continuum around the expected position of the redshifted h@xmath0 more difficult , since the [ ] emission is clearly present to within @xmath64 km s@xmath5 of the h@xmath0 and conceivably could extend to greater velocities at lower intensity . moreover , the clear differences in structure of the [ ] between the on and off positions makes the on - off sky subtraction problematic over the spectral region around the [ ] . note that the [ ] spectra even over the relatively small angles separating orders 136 and 135 are slightly different due to spatial variations in the [ ] . these can readily be seen by direct inspection of the difference images ( e.g. , figure [ wcce_image ] ) . although the presence of the [ ] adversely affects the measurement of the h@xmath0 signal which is our primary objective , it nonetheless is an interesting feature in itself . we are not aware of other observations in this general part of the sky that would have the sensitivity to detect h@xmath0 or [ ] emission at the level we detect . in h i , the leiden - dwingeloo survey @xcite clearly detects emission only in the range @xmath65 km s@xmath5 . a similar result is found in the hipass survey @xcite . thus , we can not at present associate the 80 km s@xmath5 [ ] emission with known h i emission . the velocity of the 80 km s@xmath5 feature is nearly large enough to qualify as a high velocity cloud @xcite ; however , @xciteemphasize that the deviation velocity ( the difference from the largest velocity permitted by galactic rotation ) is a better criterion , arguing for a deviation velocity greater than 50 km s@xmath5 as the cutoff . using the model presented by @xcite , positive velocities due to rotation are not expected in this direction due to its high galactic latitude ( @xmath66 = 64 ) ; consequently the deviation velocity is 80 km s@xmath5 , suggesting that the detected emission may be produced by a high - velocity cloud . with an emission measure of 25 mr , the [ ] emission is comparable to the weakest h@xmath0 emission from known high velocity clouds observed by @xcite ; however , [ ] emission was not detected by @xcite toward the faintest high velocity clouds , with upper limits somewhat below our detection . in other words the [ ] emission we detect is somewhat brighter than in some known h i high velocity clouds , in spite of the fact that no h i emission is seen in this direction at this velocity . the detected [ ] emission is clearly interesting and merits further study to determine its origin . the resulting spectra with the continuum , oh line , and [ ] line residuals subtracted are shown in figure [ all_averages]d . in this figure , the error bars have been calculated from poisson statistics . note that the order 135 and 136 spectra are generally the same within the statistical errors , except for part of the [ ] line ; this is due to spatial structure in the [ ] distribution and the fact that the difference in [ ] distribution between the target and sky positions must be slightly more complex than can be modeled with just two components . next we fit a voigt profile to this spectrum over the wavelength interval 6588 @xmath59 6593.5 ( corresponding to about four spectral resolution elements ) . the continuum is fixed to be zero since the continuum has been subtracted , as discussed above . the instrumental profile is taken from the neon calibration . the assumed intrinsic h@xmath0 line profile was obtained as follows . using the 75 resolution vla h i observations of @xcite ( which are given in the heliocentric frame ) , we determined the h i velocities of the gas at each pixel on the fabry - perot ring at the approximate redshifted wavelength of h@xmath0 and then shifted these velocities by + 5.8 km s@xmath5 , which puts them in the barycentric frame which we have used throughout this work . weighting each pixel equally as a first order approximation since the h@xmath0 emission is only weakly dependent on h i column density for the observed column density range , we constructed the spectrum shown in figure [ hi_prof]a . h i emission occurs mostly at two velocities because hi 1225@xmath101 has a rotation curve characteristic of a galactic disk , and the 6590.4 ring has relatively little overlap with the central part or minor axis ( e.g. , fig . [ wcce_image ] ) . this profile was then convolved with a gaussian with a dispersion of 6 km s@xmath5 ( corresponding to a typical dwarf dispersion and slightly less than the velocity dispersion of 8 km s@xmath5 of the ne component ) and also with a gaussian whose width corresponds to the thermal velocity of the ionized hydrogen at a temperture of 10000 k. this produces the profile shown in figure [ hi_prof]b . finally , this profile was convolved with the instrumental profile to produce the profile shown in figure [ hi_prof]c . this corresponds to the profile of the h@xmath0 line we expect to observe from hi 1225@xmath101 . this is well fit by a voigt profile centered at 6590.4 and fwhm 79 km s@xmath5 . a voigt profile with this shape is then used as the line profile in a fit to the observed spectrum , in which the only free parameter is the voigt line strength . the best fit line brightness is 4.2 mr . the spectrum and the 4.2 mr best fit for the oh , [ nii ] , and continuum subtracted spectrum for order 136 are shown in figure [ voigt_fit]b . to estimate the uncertainty in the line strength , we used a monte carlo approach @xcite , in which we constructed spectra using the errors estimated from the comparison of the individual w - gh1-gh2-e spectra ( i.e. the errors shown in figure [ all_averages]c ) . the resulting 1@xmath67 error bars are @xmath68 mr , and the 2@xmath67 upper limit is 7.4 mr . although we have attempted to accurately apply flatfield corrections to the data , differences in the spectral and spatial properties between the flatfield lamp and the sky data , as well as differences between the resolution in orders 135 and 136 mean that direct subtraction of the `` on only '' data in order 135 from 136 to obtain the net signal in the region of the expected h@xmath0 signature may well mask the signal we are attempting to detect . for this reason we have employed the on - off chopping described previously , which should accurately remove the effects of these differences in the instrumental response . however , although the sequence of on - off observations removes these differences as well as temporal variations in the night sky background which are linear , the characteristic time for the full cycle for one off - on - on - off sequence is over an hour , and variations over a shorter time may well occur . therefore the order 135 spectrum of figure [ all_averages]d was subtracted from order 136 in the same figure to produce figure [ voigt_fit]c . ( in order to reduce the noise in this subtraction the data in order 135 were first smoothed using a 5-point running mean . ) disappointingly , the signal present in figure [ voigt_fit]b appears to have largely disappeared . it is known that other weak night sky emission features are present . in particular , weak emission from o@xmath69 has been found in the night sky spectrum @xcite and one of these emission features falls at the expected wavelength of h@xmath0 ( t. slanger 2001 , private communication ) . no attempt has been made to model any residual emission from this source . however , the impression is that the main effect of the subtraction of 135 from 136 is simply to lower the level of the continuum so that it falls below the nominal zero level . this impression is confirmed by again fitting a voigt profile to the same data as shown in figure [ voigt_fit]c , but now allowing the continuum level to be a second free parameter . the resulting best fit , shown in figure [ voigt_fit]d , yields 3.9 mr , very similar to that found above . because of the very limited continuum region in order 136 , which is affected nearly up to the h@xmath0 signal at shorter wavelengths by the [ nii ] , as described above , and only extends to @xmath70 which is the termination of order 136 as explained in 2.2 , it is not clear whether such an offset really exists . it is also not clear what could give rise to such a continuum offset . spatial gradients in the background continuum arising from the zodiacal light and airglow continuum , although there is some disagreement on this point . ] would have to be present over an extent of only @xmath128 . it should be borne in mind however that the possible offset is only about 0.1% of the total continuum level . the data provide no evidence for such gradients ; first , the ccd dewar was rotated 180 after the second night , yet comparison of spectra obtained from the opposite rotations does not show mirrored spectra artifacts ; second , comparison of the order 135 and 134 spectra in figure [ two_sigma]d reveals no significant difference , even though these are extracted from regions offset by @xmath71 , nearly as large as the 710 offset between orders 136 and 135 . a more likely possible source of artifacts is scattered light from stars that has not entirely been removed or masked . this scattered light would be more likely to create spectral artifacts in the central order . partly this is because out of focus reflections between the etalon and blocking filter are not directed out of the beam for the central order due to the etalon tilt angle selected . also , due to the quadratic nature of the wavelength dispersion scattered star light affects relatively narrow spectral regions in the central order compared to the outer orders . this may explain why the order 135 and 134 spectra ( e.g. , figs . [ voigt_fit]a ) are quite similar , and more featureless than the order 136 spectrum . although there is some weak indication of the detection of a line from figures [ voigt_fit]b and [ voigt_fit]d , with strengths of @xmath12 4 mr , because of this uncertainty we can not claim a positive direct detection of the ionizing background radiation , but the formal 2@xmath67 upper limit and + 2@xmath67 of the best fit value , but we will simply refer to this as the `` 2@xmath67 '' upper limit . ] to any signal from the fits in either 8b or 8d is about 8 mr , and we shall use this limit in our subsequent discussion . the actual formal upper confidence limits to the fits in figures [ voigt_fit]b , c , and d are shown in figures [ two_sigma]b , [ two_sigma]c , and [ two_sigma]d , where figure [ voigt_fit]a is again reproduced for reference as figure [ two_sigma]a . following vwrh and @xcite , ( see also srgpf ) the one - sided incident ionizing flux is defined as @xmath72 where @xmath73 is the mean intensity . the factor @xmath74 assumes that the background intensity is isotropic . from our measurement of the h@xmath0surface brightness , @xmath75 , we may calculate @xmath76 from : @xmath77 which can be written as @xmath78 in equations ( [ phi_ex ] ) and ( [ phi_ex_calc ] ) , @xmath79 is the fraction of the one - sided incident flux which results in recombinations to excited states , @xmath80 is the fraction of those recombinations which produce h@xmath0 photons , @xmath81 is the total surface area exposed to the incident radiation and @xmath82 is the observed projected surface area . this expression is quite general , except that for complex geometries the appropriate averages over the cloud surface of the quantities @xmath79 and @xmath80 need to be taken , and the @xmath75 surface brightness is that averaged over the projected area . the h@xmath0 spectrum has been extracted from regions for which @xmath84 . however , as described in section 3.3.2 , recent work by @xcite suggests that hi 1225@xmath101 sw is a disk which is viewed far from the normal , with an aspect ratio of 4:1 , so that the ratio of the total area to the projected area is @xmath85 . if for simplicity we assume plain slab geometry , then the column density _ normal _ to the disk corresponding to the @xmath86 column density is @xmath87 , corresponding to an optical depth at the lyman limit of 15.7 . adopting the spectral index of 1.8 suggested by srgpf we find that at this optical depth about 87% of the normally incident photon flux is absorbed within the disk . however , some of the resulting recombinations occur to the ground state , and a portion of this diffuse lyman continuum radiation can escape from the slab after one ( or more ) such scatterings , and thus does not contribute to any balmer radiation . the fraction of the recombinations which go to the ground state depends upon the electron temperature , as does the energy distribution of this recombination radiation , and as discussed below , we adopt an electron temperature of 10000 k. since most of the region around the fabry - perot 6590.4 ring consists of h i column densities in excess of @xmath86 , we calculate the average value of @xmath79 by creating a histogram of column densities which contribute to the ring ( after reducing the observed values by the geometric factor of four ) and for each of the values of the column density in the histogram compute the value of @xmath79 , weighting each of these values by the number of pixels contributing to that bin . detailed calculations then yield a fraction of the normally incident photon flux which ultimately results in recombinations to excited levels of 0.88 , which is the quantity @xmath79 in equations ( 7 ) and ( 8) . ( had we used a spectral index of 1.0 , the number would be reduced to 0.75 . we also note that even for a normal column density 10 times lower than our minimum value the fraction of the continuum flux which results in excited state recombinations is still slightly greater than half of the 0.88 value , namely 0.42 . ) we also require the effective recombination rate for producing h@xmath0photons , or in the formulation used in equations ( 7 ) and ( 8) , the fraction @xmath80 of excited state recombinations which ultimately produce an h@xmath0 photon . the value of @xmath80 depends , as does @xmath79 , upon the electron temperature , and upon whether lyman line photons readily escape following the recombinations ( `` case a '' ) or scatter internally until they are ultimately converted to balmer and higher series photons ( `` case b '' ) . since nearly all of the recombinations occur at optical depths in the lyman lines which are very large , case b is appropriate . in vwrh we adopted a temperature of 20000 k on the assumption that a hard spectrum and low metallicity would lead to an electron temperature much higher than the value of @xmath88 k usually assumed . however this assumes a balance between atomic heating and cooling mechanisms . in regions of the low density intergalactic medium producing the low column density ly@xmath0 forest lines , @xcite deduce much lower temperatures ( @xmath89 k ) due to the fact that hubble expansion dominates over atomic cooling processes . however , these considerations may not be applicable to the material comprising hi 1225@xmath101 where the expansion has evidently been halted , and where the particle density is probably much higher than that in which the low column density ly@xmath0 lines arise , in which case thermal equilibrium is a much better approximation . thermal equilibrium calculations for a range of metallicities and particle densities for gas exposed to metagalactic radiation and partially shielded by neutral h i yield temperatures in the range 800010000 k ( wolfire 2001 private communication ) . however his calculations do not extend below column densities of @xmath90 so that on the outer skin of the cloud the temperatures might be slightly higher . we adopt an electron temperature of 10000 k. then , for case b and an electron temperature of 10000 k from data in @xcite and @xcite we find @xmath80 = 0.45 . as noted previously , @xcite have made a detailed study of the h i in hi 1225@xmath101 using the vla with 40 resolution . they deduce that the sw component is a thin rotating disk viewed nearly edge on , with an axial ratio of about 4:1 . this is based upon the shape of the contours of the higher column density regions , but also on the shape of the velocity displacement as a function of position as well as the fact that such a steep inclination would explain why the peak column density is higher than in the ne component . they also point out that if the aspect ratio is @xmath91 3 then the true ( normal ) surface density in the sw component is not high enough to trigger star formation ( none is observed ) , contrary to the situation in the ne component . for this proposed geometry , @xmath92 . however , the dynamical mass for hi 1225@xmath101 sw calculated by chengalur et al ( 1995 ) within a radius of 3 is @xmath93 m _ @xmath94 , slightly smaller than the total hi mass of hi 1225@xmath101 sw . an hi mass comparable to the dynamical mass is of course contrary to what is now known about the ratio of total to baryonic mass . this apparent discrepancy can be reduced in several ways . first , the hi mass refers to the total hi mass of hi 1225@xmath101 sw while the dynamical mass is only calculated for the inner 3 . second , the distance to hi 1225@xmath101 may be as much as a factor of three closer than 20 mpc , the distance assumed by chengalur et al @xcite . this would reduce the hi mass more than the dynamical mass . nonetheless , the rotational velocity of hi 1225@xmath101 sw , 13 km s@xmath5 , with just a small correction for inclination if the chengalur et al geometry is correct , is extremely low for the hi mass , even at the near distance see the tully - fisher type relation between rotational speed and hi mass derived by @xcite . indeed , a reasonable interpretation of the observed morphology of hi 1225@xmath101 sw is that it has two spiral arms elongated along the line towards hi 1225@xmath101 ne , with a more face - on inclination . accordingly , we must conclude that the geometry is uncertain . the extremes are represented by the 4/1 inclined disk , with @xmath92 , and a face - on disk , with @xmath95 . for comparison , a sphere gives @xmath96 . an infinitely long cylinder with the long axis in the plane of the sky , gives @xmath97 ; this factor is larger if the cylinder is inclined to the line of sight . an ellipsoid rotated about its major axis and with a major to minor axis ratio of 4 ( perhaps a reasonable model for a tidal feature ) gives @xmath98 . again , this factor would be slightly larger if the major axis is inclined somewhat to the line of sight . in the discussion in 4 we adopt @xmath96 as a fiducial factor , recognizing that there is an uncertainty of about two in either direction . we will refer to the @xmath99 values of 2 , 4 , and 8 as `` low '' , `` fiducial '' , and `` high '' in the following discussion . using the values for the geometrical factors just discussed , the 8 mr upper limit for the h@xmath0 flux , and values of @xmath79 and @xmath80 from equation ( [ phi_ex_calc ] ) we obtain values for the one sided ionizing flux of @xmath100 for the low , fiducial and high values of the geometrical correction factor @xmath99 . the first two of these are within the range inferred by @xcite , while the third is about a factor of two lower , though it should be remembered that our surface brightness value is an upper limit . expressing the radiation field as @xmath101 , we obtain @xmath102 , where @xmath103 is planck s constant and @xmath0 is the spectral index . assuming the srgpf value for the spectral index of @xmath104 , the corresponding upper limits for the mean intensity at the lyman limit are @xmath105 . the local unshielded metagalactic photoionization rate is calculated from this as @xmath106 , which yields @xmath107 , where @xmath108 has the dimensions s@xmath5 . had we adopted a different spectral index ( e.g. , 1.0 ) the corresponding numbers for our fiducial value of @xmath96 of @xmath76 , @xmath109 and log@xmath110 would be @xmath111 , @xmath112 and @xmath113 , respectively . if our possible detection at 4 mr should ultimately be confirmed , the actual values would of course all be lower by a factor of two . a review of recent attempts to detect , or set limits on , the intensity of the ionizing background may be found in srgpf . some of these are summarized in their table 1 . it should be noted however that in the paper by @xcite , their definition of @xmath76 is in terms of the ionizing specific intensity , not the flux . they also assume a value for @xmath79 of 1.0 , so their numbers in their table 2 can not be compared directly with ours . their upper limit on the h@xmath0 surface brightness from hi 1225@xmath101 sw is about 65mr . the h@xmath0 surface brightness limit of vwrh ( 20mr ) is the lowest surface brightness limit we are aware of prior to the current work , which is about 2.5 times higher than our new upper limit . our use of different assumptions about the values of @xmath79 and @xmath80 and the range of the projection factors considered here ( vwrh assumed a face on slab ) results in our current value for @xmath76 which ranges between about 2.5 to 10 times lower than that quoted in vwrh . [ [ comparison - with - estimates - of - agn - and - hot - star - contributions - to - the - background ] ] comparison with estimates of agn and hot star contributions to the background ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ a recent careful analysis of the contributions expected from both agns and hot stars to the ionizing background has also been carried out in srgpf as well as estimation of the uncertainties in these estimates . in principle , these contributions are directly observable independent of cosmological parameters , but of course local h i prevents radiation near the lyman limit from penetrating into the disk of the galaxy along essentially all lines of sight from the sun . in practice , according to srgpf , the two largest uncertainties in the estimate of the contributions from the agns are the uv luminosity functions for the agns and the assessment of the amount of attenuation from the cumulative absorption of intervening h i between ourselves and the agns . for the contributions from hot young stars embedded in galaxies the luminosity function of ionizing radiation is also somewhat uncertain , but the dominant uncertainty in this case is the fraction @xmath115 of ionizing photons which escape the galaxy . srgpf thus characterize this fraction as a free parameter but favor a value of order 5% , and for this contribution quote if we examine the various combinations of the `` best '' , `` upper '' and `` lower '' 1@xmath67 estimates in srgpf we find that our largest value ( for the face on projection ) is compatible for all but the largest of these combinations . for our fiducial value of the projection factor , our limit is compatible with , but slightly lower than , the sum of the two `` best '' contributions . the range of our upper limits on the inferred z @xmath12 0 photoionization rate of log @xmath118 may be compared with that inferred from very recent numerical simulations of the ly@xmath0 forest by @xcite . the recent advances in both computing power and numerical simulation techniques have produced results for the uncondensed gas giving rise to the ly@xmath0 forest which reproduce with impressive faithfulness the observed properties of the ly@xmath0 absorption lines over the entire range of observed redshifts from @xmath119 to @xmath71 . assuming a specific cosmological model , it is possible to infer , at any epoch , the value of the ionizing flux which can be compared with empirical determinations such as the one we have presented above . the basic argument is as follows : a complete specification of the cosmological parameters ( e.g. , @xmath120 , @xmath121 , @xmath122 , the h / he ratio , and , e.g. an lcdm initial perturbation model ) completely specifies the total baryon density at any redshift , and , in principle the course of the entire process of gas condensation . the simulations thus yield estimates of the actual density of hydrogen nuclei at any point in the spatial grid and at any redshift , along with the velocity field of that material , and thus the local column density of hydrogen nuclei . comparing these simulations with the observed values of the frequency of the h i absorption lines one can then infer the required fraction of hydrogen nuclei which are neutral and thus the h i photoionization rate , @xmath108 . @xcite deduce a value of @xmath123 at a mean redshift of 0.17 , and a rough extrapolation of this estimate to zero redshift yields a value of about @xmath124 , which is compatible with our range of upper limits . of course there are considerable uncertainties in the simulations themselves , in addition to the basic cosmological parameters , but these are most severe for the highest density regions where the relatively rare higher column density h i systems are formed . with further improvements in the sensitivity of the h@xmath0measurements , improved data on the density of the low redshift ly@xmath0 forestline density , and refinements in the simulations themselves , it will be possible to impose a constraint on the combination of the cosmological parameters which are involved in the @xcite estimate , especially those that yield the present epoch baryon density . as our detailed discussion of the reduction procedures and our final spectra make clear , there are certainly uncertainties in our fitted value for the h@xmath0 surface brightness and its upper limit . however , the uncertainty in the geometry of hi 1225@xmath101 sw also introduces an additional major uncertainty . there is also the possibility that our ( implicit ) assumption that the h i gas has a projected filling factor near unity is wrong . suppose in fact the h i gas in the cloud has already undergone significant fragmentation on scales much smaller than the 21 cm resolution , with each `` cloudlet '' having h i column densities substantially higher than that inferred under the assumption that no such small scale structure exists . this implies a corresponding reduction in the actual solid angle subtended by the high column density fragmented gas . in such circumstances the surface brightness of h@xmath0from each little cloudlet would be only slightly increased from that expected in the absence of fragmentation , however the total h@xmath0 flux would be reduced by the projected area filling factor , and the true value of the photoionization value would be higher than that we have inferred . we consider this scenario rather unlikely since one would expect star formation to be underway in at least some portions of the structure if such gas fragmentation had already occurred to such an extent . it is also possible that the tidal interaction between the ne and sw components has produced filamentary or clumpy hi structure as sometimes seen in higher linear resolution observations of tidally interacting galaxies . we have also considered the possibility of significant amounts of internal extinction of h@xmath0 in the sw component by dust . even for a normal dust / gas ratio , for @xmath125 @xmath4 the extinction is negligible ; however , for the peak observed column density of @xmath126 @xmath4 in the 40 resolution map of @xcite , the emission in that direction could be attenuated to 75% of the extinction - free value . however , most of the 6590.4 ring lies toward regions of the target where the column density is much lower ( see figure [ wcce_image ] ) , with a median column density of @xmath127 @xmath4 . in any case , a normal dust / gas ratio in a cloud in which no star formation seems to have occurred seems unlikely . it would be useful to reobserve the sw component of hi 1225@xmath101 with improved sensitivity in order to see if h@xmath0 at the level of @xmath124 mr is really present . however , the ambiguity in the geometry of hi 1225@xmath101 sw remains a difficuly . it would be very interesting for numerical simulations of interacting galaxies to be run to see if the morphology and velocity fields of the entire hi 1225@xmath101 complex can be understood , and this may shed some light on the correct geometry . beyond the issue of the geometry however , investigation of the initial conditions and subsequent evolution of the rare large isolated h i clouds such as hi 1225@xmath101 in which little or no star formation has taken place would seem a worthwhile program . in any event , it will be important to observe with increased sensitivity several other suitable targets , since although the sw hi 1225@xmath101component has some features which make it an attractive target , the [ ] galactic emission and the uncertain geometry make a definitive determination of the background flux problematic . the expert assistance of bill kunkel and oscar duhalde in resolving technical problems at the telescope is gratefully acknowledged . rjw thanks romeel dav for helpful discussions on the ly@xmath0 forestsimulations and don osterbrock and tom slanger for enlightening discussions and data concerning the night sky spectrum . snv and sv acknowledge support from nsf ast-9529167 . finally , we acknowledge the helpful comments of the referee , p. maloney , in particular for pointing out the dynamical objection to the @xcite geometrical model .

we present optical identifications and redshifts for seventeen new high redshift radio sources . fifteen of these sources are radio galaxies ; the remaining two are high redshift , steep spectrum , radio loud quasars . these objects were discovered as part of an ongoing study of compact ( ) , moderately steep spectrum ( ) sources from the mit green bank ( mg ) radio catalog ( mjy ) . spectra for the optical counterparts were obtained at the w.m . keck telescopes and are among the optically faintest radio galaxies thus far identified . redshifts range between 0.3 and 3.6 , with thirteen of the seventeen at redshifts greater than 1.5 . combining these new radio galaxies with two published mg radio galaxy spectra , we synthesize a composite mg radio galaxy spectrum and discuss the properties of these galaxies in comparison to other , more powerful , radio galaxies at similar redshifts . we suggest a radio power ionization state relation . 5 5 4 2 3 ] 3 ] 2 ] 4 [ ] 2 2 [ ] 2b 3 [ ] , 5007 # 1 # 1 # 1to 0pt#1 3 2 3mpch^-3mpc^3 0 0 # 1 # 110^#1 # 1to 0pt#1 a&a # 1;#2;#3;#4 # 1 , # 2 , # 3 , # 4 # 1#2#3#4 , # 3 , # 4 # 1#2#3 # 1

astro-ph9811344

the mit green bank ( mg ) survey ( bennett 1986 ; lawrence 1986 ) consists of 5974 radio sources in 1.87 steradians of sky in an equatorial strip which have flux densities @xmath9 mjy at 5 ghz . studies of higher flux density radio surveys such as the 3cr ( spinrad & djorgovski 1987 ) , molonglo ( mccarthy 1996 ) , and b2/1 jy of allington smith ( 1982 ) have yielded many interesting discoveries , such as the radio optical alignment of high redshift radio galaxies ( chambers 1987 ; mccarthy 1987 ) , very high redshift galaxies ( lacy 1995 ; spinrad , dey , & graham 1995 ; rawlings 1996 ) , and rich clusters at high redshift ( dickinson 1996 ) . the mg survey probes the parameter space between high flux density surveys such as the 3cr , and milli and micro jansky surveys such as the leiden berkeley deep survey and its extensions ( neuschaefer & windhorst 1995 ) . the mg range of radio flux density is also considered by the b3 survey ( vigotti 1989 ) and recent cambridge surveys ( 6c hales , baldwin , & warner 1993 ) , but the mg survey was conducted at a higher frequency than these other surveys . selecting an intermediate power sample allows us to study various properties as a function of radio power . in particular , a lower radio power suggests these galaxies may , on average , be less dominated by their active nuclei , and may thus be more representative of giant ellipticals at large redshift ( dunlop 1996 ; spinrad 1997 ; dey 1998 ) . fainter flux densities also suggest that a fraction of our sample may be at very high redshift , and indeed the median redshift of radio galaxies in the mg observed to this point is @xmath10 compared to @xmath11 for the 3cr ( mccarthy 1993 ) . over the past several years we have been pursuing the optical identification of a subset of 218 radio sources systematically selected from the mg catalog . we restrict our sample to radio sources of small angular size ( @xmath12 ) , simple ( unresolved , double , or triple ) radio morphology , and moderately steep radio spectral index ( @xmath1 ) . the size and morphological criteria primarily select against objects at low redshift . the angular size limit should also bias the sample to larger lookback times when the higher density of the intergalactic medium is more effective at confining radio lobes . the spectral index criterion eliminates low redshift interlopers and selects against flat spectrum quasars and in favor of radio galaxies , but is only moderately restrictive amongst the radio galaxies themselves ( blumenthal & miley 1979 ) . resulting identifications are often very distant objects and thus serve as useful probes for the study of galaxy formation and evolution . in combination with other complete samples , such as the 3cr , optical and near ir properties of mg sources can be used in correlative studies which span a range of radio power at any given redshift . similar studies suggest that weak radio sources ( @xmath13 mjy ) have weaker emission lines and the alignment effect becomes less pronounced with lower radio flux density samples ( rawlings & saunders 1991 ; dunlop & peacock 1993 ; eales & rawlings 1993 ; thompson 1994 ) . mccarthy ( 1993 ) has a recently presented a comprehensive review on the optical and associated properties of radio galaxies . preliminary results from our survey are described in spinrad ( 1993 ) , dey , spinrad , & dickinson ( 1995 ) , and stern ( 1996 ; 1997 ) . currently , we have optical identifications for @xmath14% of our sample and spectroscopic redshifts for @xmath15% of the sample . in this paper we report on seventeen new identifications and redshifts of mg sources . the spectra were obtained at the w.m . keck telescopes and are representative of some of the fainter radio identifications thus far . typical @xmath16 magnitudes are @xmath17 many of these sources are in the so called ` redshift desert , ' @xmath18 for which no strong features ( , ly@xmath19 , 2 , or h@xmath19 ) are redshifted into the optical window , thus making redshift determinations challenging . our mg sample currently includes 13 galaxies with @xmath20 ; we report on 4 such galaxies here . the paper is organized as follows : in 2 we present the observations and data reductions , followed by a discussion of individual systems in 3 . we construct a composite spectrum and compare it with models and other composite spectra of active galaxies in 4 . a concluding discussion comprises 5 . throughout this paper we adopt @xmath21 and @xmath22 unless otherwise noted . preliminary imaging observations of all sources were made over the course of several years ( 1988 @xmath23 1998 ) at the lick observatory 3 m shane telescope on mount hamilton and the mdm 2.4 m hiltner telescope at kitt peak . the kast double spectrograph of lick observatory ( miller & stone 1994 ) employs uv flooded reticon 1200x400 ccds with 27@xmath24 m pixels , corresponding to a plate scale of 078 pix@xmath25 typical seeing for the lick imaging was 12 18 . the charlotte camera of mdm employs a tek 1024@xmath26 ccd detector with a plate scale of 0275 pix@xmath25 typical seeing for these images was 10 15 . a subsample of seven of the seventeen sources were later reimaged with the low resolution imaging spectrometer ( lris ; oke 1995 ) at the 10 m w.m . keck telescopes on mauna kea between 1994 and 1998 . the detector is a tek 2048@xmath26 ccd with 24@xmath24 m pixels . due to a dewar change in july 1996 , the pixel scale changed from 0214 pix@xmath27 to 0212 pix@xmath27 at that time . the typical seeing was 08 10 . a journal of our observations is provided in table [ obsns ] . the images were corrected for overscan bias , flattened using a median sky flat , coadded , and calibrated using observations of standard stars . utilizing astrometry provided by b. burke and s. conner , we obtained optical identifications for our candidates based upon the radio coordinates . typically the optical morphologies are small , faint , and round . we present finding charts in fig . [ plate ] ( plates 1 3 ) . where available , we present keck images ; the remaining images were obtained at lick and the mdm as indicated in table [ obsns ] . astrometry for each identification is tabulated in table [ astrom ] with the coordinates of the offset star provided . the estimated uncertainty in the optical positions is 03 . optical and radio properties of the sources are in table [ prop ] , where the radio flux densities and spectral indices are from bennett ( 1986 ) and lawrence ( 1986 ) . radio luminosities have been calculated at a rest frame frequency of 4.8 ghz , implementing the spectral index derived from 1.4 and 4.8 ghz observations , a hubble constant of @xmath28 and @xmath29 . in table [ radflux ] we include more recent 4.85 ghz flux densities for the sample . many sources may vary at the 10% level over the 10 year baseline , though mg 2058 + 0542 apparently varied considerably more over the same period . we therefore indicate its spectral index as uncertain in table [ prop ] . magnitudes , measured in site dependent apertures as described in table [ prop ] , are from our ccd photometry and have errors of @xmath30 to @xmath31 magnitudes . observations obtained at lick observatory employed the spinrad night sky filter and yield red magnitudes in the @xmath32 system , which is related to commonly used photometric systems by djorgovski ( 1985 ) . typical optical magnitudes are @xmath33 probing the fainter envelope of our mg survey . we obtained spectroscopic observations with lris between 1994 and 1997 . we generally used a 300 @xmath34/mm grating ( blazed at 5000 ) to sample a wavelength range @xmath44000@xmath239000 , and a 1 arcsecond wide slit which yields an effective resolution fwhm of @xmath35 10 . the read noise and gain were typically 8.0@xmath36 and 1.6@xmath36 adu@xmath27 respectively . however , during the 1994 march run , the lris ccd was affected by pattern noise which mimicked extremely high read noise ( 25@xmath36 and 65@xmath36 on the blue and red sides ) , noticeably degrading the spectrum of mg 1251 + 1104 . observations were typically done at an airmass @xmath37 . hence , although the slit position angle was often selected to trace the radio morphology or to include nearby optical sources , we believe our relative spectrophotometry to be accurate for the wavelengths considered ( filippenko 1982 ) . the sources attempted spectroscopically at keck were primarily selected from the fainter ( @xmath38 ) identifications in the mg survey , many having been unsuccessful spectroscopic targets at lick observatory . exceptions are the relatively bright sources mg 0422 + 0816 ( @xmath39 ) and mg 0511 + 0143 ( @xmath40 ) which were first observed spectroscopically as a backup project at keck observatory on an inclement night . these observations utilized a 600 @xmath34/mm grating ( blazed at 5000 ) . the bright quasar mg 2041 + 1854 ( @xmath39 ) was observed at keck during twilight on a cirrusy night , while the relatively bright ( @xmath41 ) sources mg 0148 + 1028 and mg 0308 + 0720 were first surveyed spectroscopically as backup targets on 23 december 1997 ut . for each observation , we offset the telescope from the reference star listed in table [ astrom ] . the data were corrected for overscan bias , flat fielded using internal lamps taken after each observation , flux calibrated using observations of spectrophotometric standard stars ( massey 1988 ; massey & gronwall 1990 ) , and corrected for galactic reddening using the burstein & heiles ( 1982 ) maps ( see table [ prop ] ) with the cardelli , clayton , & mathis ( 1989 ) extinction curve . one dimensional spectra were extracted with a typical aperture size of 14 . the standard stars were observed both with and without an og570 filter in order to correct for the second order light contaminating the wavelength region @xmath42 . a journal of the observations is provided in table [ obsns ] and individual spectra are presented in fig . [ spectra ] with the prominent features indicated . in table [ lines ] we list the line flux densities measured for each source . for completeness , we include the only two published spectra of mg radio galaxies ( mg 1019 + 0534 dey , spinrad , & dickinson 1995 ; mg 2144 + 1928 maxfield 1998 ) in fig . [ spectra ] and table [ lines ] . _ mg 0018 + 0940 , at z=1.586 , _ exhibits a low ionization spectrum : unlike more powerful radio sources , the 3 line is stronger than the 4 doublet . 2 is clearly resolved . the optical ( @xmath16 ) image of mg 0018 + 0940 has an extended , chain like morphology oriented at a position angle of 137 in the optical . the radio source is unresolved in 4.85 ghz observations with the b array of the vla ( @xmath43 12 beam ; lawrence 1986 ) . _ mg 0046 + 1102 , at z=1.813 , _ also exhibits a low ionization spectrum , with similar carbon and magnesium line strengths to mg 0018 + 0940 . 2 is again a strong and clearly resolved feature . the optical identification is extended with a position angle of 130 in @xmath16 , while the source is unresolved in 4.85 ghz observations obtained with the b array configuration of the vla . _ mg 0122 + 1923 , at z=1.595 , _ also exhibits a low ionization spectrum , with carbon line strengths similar to the prior two radio galaxies . the optical identification contains a compact core and diffuse emission to the north oriented at a position angle of 41 . _ mg 0148 + 1028 , at z=2.845 , _ is an intriguing source showing both the bright emission lines typical of high redshift radio galaxies and the narrow absorption lines that characterize high redshift , star forming galaxies ( steidel 1996 , lowenthal 1997 ) . these lines , indicated with vertical dashed lines in fig . 2 , refer to @xmath31260 , @xmath31302 , @xmath31335 , @xmath41394 , 1403 , @xmath31502 , @xmath31526 , @xmath31608 , and @xmath31670 ( spinrad 1998 ) . the emission lines of this source are slightly broader than in those sources considered above , with fwhm @xmath43 1000 km s@xmath27 a value commonly seen in quasars , though the low 4/2 ratio leads us to classify this source as a broad lined radio galaxy . a detailed analysis of mg 0148 + 1028 is deferred to future publication ; note , however , that 5 , an unambiguous signature of starlight ( dey 1997 ) , is visible in absorption . the finding chart for this galaxy derives from a short ( 60@xmath44 ) exposure obtained through lris without a filter to identify the field ; this image is superior to the 1260@xmath44 lick image . slit spectroscopy was done at a position angle of 138 so that galaxies b and c ( see fig . 1 ) would also be observed . neither is associated with mg 0148 + 1028 ; galaxy b has 2 and 3 in emission at @xmath45 while galaxy c has 2 in emission with 2 in absorption at @xmath46 the optical identification is marginally resolved : fwhm@xmath47 , while the seeing was @xmath48 . _ mg 0308 + 0720 , at z=2.975 , _ is a radio loud quasar as evidenced by the broad emission lines ( fwhm ( ) @xmath43 5600 km s@xmath27 ) and unresolved morphology . the ly@xmath195 emission complex is unusual , with narrow ly@xmath19 superimposed upon a broad 5 emission line . recent statistical investigations of broad uv lines in luminous ( radio quiet ) quasars have suggested that the traditional broad line region ( blr ) consists of two components a low density , intermediate line region ( ilr ; fwhm @xmath43 2000 km s@xmath27 ) and a higher density , very broad line region ( vblr ; fwhm @xmath49 7000 km s@xmath27 ) blueshifted by @xmath49 1000 km s@xmath27 with respect to the ilr ( brotherton 1994 ) . differences in the relative contributions of these two components can explain much of the observed variation in quasar broad line profiles . for instance , ly@xmath19 will come predominantly from the ilr , while higher ionization lines such as 5 , @xmath41394 , 1403 , and 4 are dominated by vblr emission . in this simple model , mg 0308 + 0720 is a quasar whose broad lines are dominated by the vblr ; indeed , the ly@xmath19 spectral profile looks similar to that of the composite vblr presented in fig . 2a of brotherton ( 1994 ) . alternatively , we may be seeing a dusty quasar at high redshift . the ly@xmath195 complex is dominated by broad 5 emission . subtracting a gaussian fit to the 5 line , we find the residual ly@xmath19 to be extremely weak : the ly@xmath19/4 ratio is @xmath50 compared to a mean value of @xmath51 in composite quasar spectra ( boyle 1990 ; francis 1991 ) . the extremely red continuum is in concordance with this dust obscuration interpretation . three 2 absorption systems at redshifts @xmath52 and @xmath53 are also evident in the spectrum . _ mg 0311 + 1532 , at z=1.989 , _ exhibits a slightly higher ionization spectrum in comparison to the above radio galaxies . 2 is barely apparent as a faint broad bump near 7000 . the red continuum of this source , and particularly the abrupt feature at 7150 are likely artifacts due to scattered light from a bright star 35 to the ne . the optical identification consists of low surface brightness emission oriented at a position angle of 135 , while the 4.8 ghz morphology is extended by 51 at a position angle of 161 . _ mg 0422 + 0816 , at z=0.289 , _ was observed spectroscopically at keck as a backup object on a night of poor seeing and thick cirrus , thus making our primary faint targets untenable . the relatively bright optical flux ( @xmath39 ) and unfluxed spectrum are the byproducts of these meteorological conditions which prevented observations of our primary fainter targets . the 3 nebular lines are the most prominent feature of the spectrum . the optical counterpart consists of a marginally resolved core with diffuse emission to the south . _ mg 0511 + 0143 , at z=0.596 , _ was also observed as a back up target on an inclement night . this galaxy has a weak lined spectrum with 2 of low equivalent width and a prominent 4000 break ( labeled with an arrow in fig . [ spectra ] ) , indicative of an old stellar population . mg 0511 + 0143 was also imaged in the j and k bands at lick observatory on cirrusy nights . comparing the multiband images suggests the radio galaxy may be part of a cluster at @xmath54 the vignetting of the finding chart is due to the small field of view of the kast imaging spectrometer on the lick 3 m . the optical identification is oriented at a position angle of 48 , while the 4.8 ghz morphology is extended by 39 at a position angle of 171 . _ mg 1019 + 0534 , at z=2.765 , _ has an unusual spectrum which is previously reported in dey , spinrad , & dickinson ( 1995 ) . ly@xmath19 , usually the strongest line in high redshift radio galaxies , is very weak . this is likely due to dust attenuating the emission , implying dust formation at early epochs for this system . _ mg 1142 + 1338 , at z=1.279 , _ is a weak lined radio galaxy with strong 2 and high ionization [ ] @xmath33426 . 3 is barely visible as a broad feature while neither 2 nor 2 are visible . the optical counterpart is compact . _ mg 1251 + 1104 , at z=2.322 , _ was observed during the 1994 march run when the lris ccd was compromised by high pattern noise . ly@xmath19 is the dominant feature of the spectrum . the continuum slope is untrustworthy and a relic of instrumental problems , and lead to uncertain equivalent widths in table 5 . the optical morphology is diffuse and symmetric . _ mg 1401 + 0921 , at z=2.093 , _ exhibits a moderate ionization spectrum radio galaxy with strong 3 and weak 2 . the optical counterpart is slightly elongated with a shell structure to the nw , while the radio morphology has a classic double structure of separation 35 at a position angle of 138 . a similar asymmetric optical structure is observed in @xmath55band images of 4c 23.56 ( knopp & chambers 1997 ) . deep multi band images lead them to interpret the morphology of 4c 23.56 as deriving from a dusty galaxy illuminated by a beam from an active nucleus which is scattered into our line of sight . _ mg 2037@xmath230011 , at z=1.512 , _ is a weak lined source showing several faint emission lines . most strikingly , the 3 line has a narrow component superimposed upon a broad component . a comparison with the similar redshift radio galaxy mg 0018 + 0940 is interesting in particular , note that 2 is not seen in the spectrum of mg 2037@xmath230011 . _ mg 2041 + 1854 , at z=3.056 , _ is a radio loud quasar . the spectral index of this source is very close to the lower limit of our sample , which was imposed to separate the typically flat spectrum quasars from the typically steeper - spectrum radio galaxies . the relatively bright ( @xmath39 ) source is near the galactic plane . therefore , we took advantage of the good seeing conditions at mauna kea to take a short exposure spectrum of this source during twilight on a cirrusy night , though , consequently , the spectral fluxes are uncertain . the best image available for this source is from a 5@xmath44 exposure through the guide camera at keck ( fig . [ plate ] ; 0275 pix@xmath27 ) . the field of view is smaller than for the other objects in fig . [ plate ] and the slit is visible ; nevertheless , it is vastly superior to the poor seeing lick 3 m image . the spectrum shows the typical broad lines of a quasar , with intervening 2 absorption systems apparent at @xmath56 and @xmath57 this source is interesting in terms of being a moderately steep spectrum , radio loud quasar at @xmath58 few such sources are in the literature and may be interesting in terms of unified models of extragalactic radio sources . _ mg 2058 + 0542 , at z=1.331 , _ is typical of the sources discussed here . narrow lines are visible with a faint red continuum , possibly suggestive of star light . _ mg 2109 + 0326 , at z=1.636 , _ exhibits a slightly higher ionization spectrum than mg 2058 + 0542 , or possibly reflects multiple ionization states , as indicated by the relatively strong 4 emission line . an emission line galaxy at @xmath59 was also serendipitously discovered on the keck spectrogram of this target . _ mg 2121 + 1839 , at z=1.861 , _ shows neither the low ionization 2 nor the high ionization 4 lines ; the major features are 4 and 3 . ly@xmath19 has been detected for this object on the blue camera of the kast double spectrograph of lick observatory . keck / nirc @xmath60 band images of this target are presented in van breugel ( 1998 ) and reveal a relatively smooth morphology which is not aligned with the radio axis ( @xmath61 , extended by 63 ) . serendipitous emission line galaxies at @xmath62 and @xmath63 were also discovered on the lris spectrogram . _ mg 2144 + 1928 , at z=3.592 , _ is the highest redshift radio galaxy yet discovered in our mg sample . its redshift was first measured at lick observatory and is mentioned in spinrad ( 1993 ) . mg 2144 + 1928 is an aligned radio galaxy with multiple components showing interesting velocity structure . detailed discussion and analysis of the optical spectrum , as well as astrometry and a finding chart , is presented in maxfield ( 1998 ) . near infrared observations of this galaxy are also presented in armus ( 1997 ) and van breugel ( 1998 ) . both the emission line free @xmath64 image of van breugel ( 1998 ) and the narrow band 2.3@xmath24 m image of armus ( 1997 ) , selected to target the redshifted 3 doublet , show the host galaxy to be extended along the radio axis ( @xmath65 , extended by 85 ) . _ mg 2308 + 0336 , at z=2.457 , _ is a high redshift radio galaxy exhibiting a spectrogram indicative of mixed ionization state . note that 2 , the strongest of the carbon lines for this object , is quite broad . ly@xmath19 was also detected at lick observatory with the kast double spectrograph . the optical counterpart consists of higher surface brightness core oriented at a position angle of 137 , with fainter emission evident to the south . the 4.8 ghz radio axis is aligned with the fainter emission ( @xmath66 ) and is extended by 3 . in fig . [ compfig ] we present a composite keck / mg radio galaxy spectrum constructed from eleven of the spectra presented here , omitting the unfluxed spectra , the quasars , the extremely weak lined mg 1142 + 1338 , and mg 1251 + 1104 whose spectrum is heavily affected by pattern noise . we also include mg 1019 + 0534 and mg 2144 + 1928 , the only previously published keck spectra of mg radio galaxies ( dey , spinrad , & dickinson 1995 ; maxfield 1998 ) , to maintain the integrity of the selection criterion : namely , we combine all keck spectra of mg radio galaxies that we have amassed to date . this selection criteria ensures that we use only the highest signal to noise ratio data . the composite was constructed by shifting the individual spectra into their rest frame , rebinning to a common linear wavelength scale with 2 pix@xmath27 resolution , and scaling the spectra by the flux of the 3 emission line . we select the 3 feature since it is present in most of the spectra considered herein in objects at both high and low redshift , as well as objects exhibiting high and low ionization spectra . we justify this algorithm in that our primary objective is to study the emission line properties of the mg sample ; normalizing the spectra over a longer wavelength ( line free ) range would give undue weight to the galaxies with the lowest continua . the current algorithm biases the continuum to the weakest lined sources . accordingly , we deemphasize conclusions drawn regarding the composite continuum . the red ends of the spectra were then trimmed to minimize contamination from the telluric oh features , and , in the case of mg 0311 + 1532 , the effects of a nearby red star . the resultant vectors were then averaged . only the spectrum of mg 2144 + 1928 did not overlap with the 3 emission line . instead , we normalized its spectrum to a preliminary composite comprising the other 12 galaxies using the 4 doublet . the final spectrum is comparable to the composite radio galaxy spectrum constructed by mccarthy & lawrence ( 1998 ; hereinafter mcl98 ) primarily from 3c and mrc/1 jy sources , which we also present in fig . [ compfig ] . the emission line properties of the composite mg spectrum ( @xmath67 ) are presented in table [ composite ] . we note that the emission line widths have not been corrected for the resolution of the spectrograph . for typical line widths of 1100 km s@xmath27 observed at 7500 , the deconvolution correction is @xmath35 75 km s@xmath27 . comparisons to the emission line strengths determined from composite spectra of higher flux density radio galaxies ( mcl98 ) , qsos ( boyle 1990 ; francis 1991 ) , seyfert ii nuclei ( ferland & osterbrock 1986 ) , the liner nucleus of ngc 4579 ( barth 1996 ) , and models are presented in table [ compare ] . all line profiles in the mg composite were fit with a simple gaussian . this simple prescription overestimates the width of the doublet lines , most drastically for 2 whose rest frame separation is 7.2 . the typical fwhm for the remaining lines is 900 @xmath23 1100 km sec@xmath68 lower than that found by mcl98 for their composite radio galaxy spectrum . unlike mcl98 , we do not find that the fwhms increase towards shorter wavelengths . however , the constituent galaxies in the mcl98 composite stem from a much larger redshift range ( @xmath69 ) , allowing a longer baseline composite spectrum ( @xmath70 ) . no strong trend is evident in their composite for the restricted wavelength range where both composites overlap . mcl98 note that redshift correlates with 2 and 3 line width in 3cr galaxies , suggesting that the wavelength line width trend in mcl98 likely reflects a redshift line width correlation , in the sense that higher redshift radio galaxies have broader emission lines . this is unlikely to be a selection effect , as the 3cr is nearly completely identified . we measure an average restframe equivalent width of 300 for the ly@xmath19 emission line in the keck / mg radio galaxy sample . for comparison , from a sample of 28 radio galaxies with @xmath71 similar to the redshift range in the current sample , mccarthy ( 1993 ) finds a mean rest frame @xmath72 of @xmath73 . radio quiet , lyman break galaxies with ly@xmath19 in emission have a typical @xmath72 of @xmath74 ( steidel 1996 ) . mcl98 find that their composite radio galaxy spectrum has very strong ly@xmath19 relative to 5 , 4 , and 3 , as compared to other agn . the mg composite , however , has much weaker ly@xmath19 , with line ratios more typical of the other agn in table [ compare ] . most likely the radio galaxy discrepancies stem from the small number of ly@xmath19 emitters in the composite spectra . the 3c / mrc composite ly@xmath19 comprises 7 galaxies , while the mg composite ly@xmath19 comprises only 3 galaxies , one of which is known to be underluminous in ly@xmath19 , likely due to dust absorption ( mg 1019 + 0534 , dey , spinrad , & dickinson 1995 ) . however , one could also imagine a ly@xmath19radio power dependency , similar to the known 2 1.4 ghz radio power relation ( mccarthy 1993 ) , or that the mgs typically reside in dustier or more cold gas rich host galaxies . at longer wavelengths , where the spectra are less affected by small number statistics , there are several significant differences between the radio galaxy composites . most strikingly , the 2 doublet is one of the most prominent lines in the mg composite , with 2 / 3 @xmath75 and ly@xmath19 / 2 @xmath76 in the 3c / mrc composite , however , the 2 doublet is relatively weak , with 2 / 3 @xmath77 and ly@xmath19 / 2 @xmath78 the 2 / 2 line ratio also shows significant variation between the composite radio galaxy spectra : 2 / 2 @xmath79 for the mg sample , while 2 / 2 @xmath80 for the 3c / mrc composite . these strong discrepancies are difficult to interpret , though the mg line ratios are intriguingly more similar to those of liners and quasars than to the 3c / mrc composite radio galaxy or seyfert iis ( see table [ compare ] ) . we note that the 2 line in the mg composite is composed from five spectra , suggesting that small number statistics may still be the source of the these suggestive trends . the strengths of the various ionization stages of carbon are a more useful diagnostic of the excitation mechanisms in active galaxies : selecting emission lines from the same element avoids metallicity dependencies , while the small wavelength separation of the lines considered makes the conclusions relatively insensitive to reddening . in fig . [ carbon ] we plot 4 / 3 vs. 3 / 2 for individual galaxies in the mg sample , the composite spectra from table [ compare ] , and the models described below . we find that the mg composite appears to be in a lower ionization state than the 3c / mrc composite , with carbon line ratios more comparable to the liner nucleus of ngc 4579 than to the quasar composites . again , an alternative explanation is that mgs typically reside in dustier galaxies than 3c / mrc sources . unfortunately , the limited optical window and high redshift of the current sample deny us access to the line pairs commonly used for dust extinction measurements , such as h@xmath19/h@xmath81 , ly@xmath19/h@xmath81 , and 2/2b . however , the strength of the ly@xmath19 line , which is quite sensitive to dust , implies that there is not an immense amount of reddening in the mg composite , while the dramatic change in the 2 strength between the 3c / mrc composite and the mg composite suggests that there is a significant difference between the emission line regions in these two radio galaxy populations . finally , the relatively high 2 / 4 ratio indicates that the mg composite is in a lower ionization state as compared to the higher radio power 3c / mrc composite , despite the 4 line residing redwards of the the carbon lines so that dust reddening alone can not explain the line ratio difference . this suggests that the ionization state of radio galaxies correlates with radio power . in a detailed study of lower redshift ( @xmath82 ) southern 2 jy radio sources , however , tadhunter ( 1998 ) find ionization state , as measured by the 2/ [ ] @xmath35007 ratio , does not correlate with radio power . to test the hypothesis of a radio power ionization state relation in high redshift radio galaxies , we have amassed a sample of 30 radio galaxies with published 4 and 3 fluxes from the literature . in fig . 5 we plot 4 / 3 vs. rest frame 1.4 ghz radio power . there is a slight tendency for the stronger radio sources to have larger 4 / 3 ratios , implying higher ionization states . to test the significance the correlation , we have calculated the nonparametric spearman rank correlation coefficient , @xmath83 , for this sample , and find a value of @xmath84 . note that this statistic is independent of cosmology . the null hypothesis , that no correlation exists between radio power and ionization state can be marginally rejected : the probability of obtaining a rank correlation coefficient this high from a sample of 30 uncorrelated variables is 3.5% . if we restrict our analysis to the two larger data sets , , the keck / mg and ultra - steep source ( uss ) samples , then @xmath85 for a sample of 25 sources . the probability of obtaining a rank correlation coefficient this high from a sample of 25 uncorrelated variables is 1.5% , implying that the correlation is significant . we consider these statistical arguments suggestive , but not conclusive , of a radio power ionization state correlation in the uv spectra of high redshift radio galaxies . insufficient sources with both 2 and 2 well detected prevented a similar analysis to be done with the 2 / 2 ratio . we have computed simple , single zone photoionization models using cloudy ( ferland 1996 ) . the purpose of this exercise is not to fabricate a definitive model of the radio galaxy spectrum , but rather to demonstrate that a simple low density gas photoionization model can reproduce the overall character of the spectrum and compare the best fit parameters with those derived for the 3c / mrc composite . following mcl98 , our calculations were made for an ionization bounded slab of gas with a constant density of @xmath86 @xmath87 illuminated by a power law spectrum of ionizing radiation , @xmath88 we calculated models with a range of spectral index , @xmath89 and ionization parameter , @xmath90 . for reference , mcl98 found that @xmath91 and @xmath92 provided the best fit to the composite 3c / mrc spectrum . we find that the carbon line diagnostics are not well reproduced by these simple models . for example , the best fit cloudy model of mcl98 is represented by an asterisk in fig . though this model does a reasonable job at reproducing the overall character of the 3c / mrc emission line spectrum , the 3 / 2 ratio is much higher in the model than in the composite spectrum . 2d of allen ( 1998 ) is also enlightening : for a given 4 / 3 ratio , the 2 emission line is again too strong in the comparison agn spectra to be fit by simple single zone photoionization models . photoionization models which incorporate both optically thick ( ionization bounded ) clouds and optically thin ( matter bounded ) clouds , as described in binnette ( 1996 ) , are capable of reproducing the observed carbon line diagnostics , as are the shock models of dopita & sutherland ( 1996 ) , implying that ( the mg ) radio galaxies are more complicated than our basic model . investigations also reveal that the high equivalent width 2 line is also difficult to reproduce with the simple model : for @xmath93 and @xmath94 no model is capable of reproducing the 2 / 3 ratio we find in the mg composite radio galaxy spectrum , indicating that shocks and/or more complicated photoionization scenarios are necessary to explain the emission line spectra of these galaxies . we present optical identifications , finding charts , and spectra for seventeen new high redshift radio sources selected from the mg 5 ghz survey . we select targets of moderately steep spectral index to preferentially observe radio galaxies , and , indeed , fifteen of the seventeen sources discussed herein are radio galaxies . the remaining two sources are moderately steep spectrum , radio loud quasars which may be important in terms of unified models of extragalactic radio sources . the spectra were all taken at the w.m . keck telescopes and are representative of the fainter mg identifications attempted thus far , with typical @xmath16 magnitudes of @xmath17 we construct a composite mg radio galaxy spectrum and compare it with the higher radio power composite 3c / mrc radio galaxy spectrum of mcl98 . we find that the mg radio galaxies typically exhibit lower ionization state spectra than the 3c / mrc radio galaxies . the 2 emission line is extremely strong in the mg composite relative to the other rest frame uv emission lines , with 2 / 3 @xmath95 and 2 / @xmath96 . extensive modeling with single zone photoionization models are incapable of reproducing the high 2 / 3 ratio , indicating that shocks and/or more complicated photoionization scenarios are producing the emission line spectra of these distant radio galaxies . we have amassed a large sample of high redshift radio galaxies with published 4 and 3 line strengths . comparing the 4 / 3 ratio to the rest frame 1.4 ghz radio power , we find evidence for a correlation between ionization state and radio power . a likely interpretation is that the more powerful radio sources are in an active phase when the central engine is emitting more flux across the electromagnetic spectrum with the augmented uv flux leading to higher ionization state spectra . as we progress from the strongest radio sources to weaker sources , we find the emission line strengths attenuate , the ionization state of the emission line region diminishes , and the stellar populations apparently become more dominant . this last effect is seen both in the diminished alignment affect for weak radio sources and the discovery of several weak radio sources at moderate redshift whose spectra are devoid of the uv emission lines that dominate most radio galaxy spectra ( spinrad 1997 ) . an alternative explanation is to invoke multiple emission line regions whose relative contributions vary with radio power . the high redshift radio galaxies discussed herein are faint , and the radio power line strength correlation implies that long integrations with the new generation of large aperture telescopes is necessary to measure the emission line strengths and redshifts of these sources . redshift radio galaxies from a range of radio flux density will be the key to further investigations of the ionization state radio power relation . we are very grateful to b. burke and sam conner at mit for introducing us to the mg sample and for supplying much of the astrometry for our candidate subset . we thank aaron barth , wil van breugel , and pat mccarthy for valuable discussion and useful comments on the manuscript . we acknowledge chuck steidel for graciously obtaining the mdm images , marc davis and steve zepf for obtaining the spectra of mg 0422 + 0816 and mg 0511 + 0143 , marc davis and jeffrey newman for obtaining the spectra of mg 0148 + 1028 and mg 0308 + 0720 , and pat mccarthy for providing the composite 3c / mrc radio galaxy spectrum . we thank brian mcleod who has imaged a subset of the mg sample in the k band and t. bida , w. wack and j. aycock for invaluable help during our keck runs . the authors also gratefully acknowledge the referee , rogier windhorst , for useful comments . ds acknowledges support from igpp grant 99ap026 , ad acknowledges the support of nasa grant hf01089.0197a and partial support from a postdoctoral research fellowship at noao , hs acknowledges support from nsf grant ast 9528536 . ds , ad , and ds acknowledge peet for invaluable help at the observatories and for facilitating coherent discussion . allen , m.g . , dopita , m.a . , & tsvetanov , zl . 1998 , , 493 , 571 allington smith , j.r . 1982 , , 199 , 611 armus , l. , soifer , b.t . , murphy , t.w . , jr . , neugebauer , g. , evans , a.s . , & matthews , k. 1997 , , 495 , 276 barth , a.j . , reichert , g.a . , filippenko , a.v . , ho , l.c . , shields , j.c . , mushotzky , r.f . , & puchnarewicz , e.m . 1996 , , 112 , 1829 becker , r. , whiter , r. , & edwards , a. 1991 , , 75 , 1 bennett , c.l . , lawrence , c.r . , burke , b.f . , hewitt , j.n . , & mahoney , j. 1986 , , 61 , 1 binette , l , wilson , a.s . , & storchi bergmann , t. 1996 , , 312 , 365 blumenthal , g. , & miley , g.k . 1979 , , 80 , 13 boyle , b.j . 1990 , , 243 , 231 brotherton , m.s . , wills , b.j . , francis , p.j . , & steidel , c.s . 1994 , , 430 , 495 burstein , d. & heiles , c. 1982 , , 87 , 1165 cardelli , j.a . , clayton , c.c . , & mathis , j.s . 1989 , , 345 , 245 chambers , r.c . , miley , g.k . , & van breugel , w. 1987 , , 329 , 604 cimatti , a. , di serego alighieri , s. , vernet , j. , cohen , m. & fosbury , r.a.e . 1998 , , 499 , 21 dey , a. , spinrad , h. , dickinson , m. 1995 , , 440 , 515 dey , a. , van breugel , w. , vacca , w. , & antonucci , r. 1997 , , 490 , 698 dey , a. , 1998 , in preparation dickinson , m. , 1996 , in fresh views of elliptical galaxies , ed . buzzoni , renzini , & serrano ( asp conf . 86 ) , 283 djorgovski , s. 1985 , , 97 , 1119 dopita , m.a . & sutherland , r.s . 1996 , apjs , 102 , 161 dunlop , j.s . & peacock , j. 1991 , , 263 , 936 dunlop , j.s . , peacock , j. , spinrad , h. , dey , a. , jimenez , r. , stern , d. , & windhorst , r. 1996 , nature , 381 , 581 eales , s. & rawlings , s. 1993 , , 411 , 67 ferland , g.j . 1996 , hazy : a brief introduction to cloudy , univ . of kentucky dept . of phys . and astron . internal report ferland , g.j . & osterbrock , d.e . 1986 , , 300 , 658 filippenko , a.v . 1982 , , 94 , 715 francis , p. , hewett , p.c . , foltz , c.b . , chaffee , f.h . , weymann , r.j . , & morris , s.l . 1991 , , 373 , 465 gregory , p.c . & condon , j.j . 1991 , , 75 , 1011 griffiths , m. , wright , m. , burke , b. , & ekers , r. 1995 , , 97 , 347 hales , s.e.g . , baldwin , j.e . , & warner , p.j . 1993 , , 234 , 919 knopp , g.p . & chambers , k.c . 1997 , , 109 , 367 lacy , m. , 1994 , , 271 , 504 lawrence , c.r . , readhead , a.c.s . , moffet , a.t . , & birkinshaw , m. 1986 , , 61 , 105 lowenthal , j. 1997 , , 481 , 673 massey , p. , strobel , k. , barnes , j.v . , & anderson , e. 1988 , , 328 , 315 massey , p. & gronwall , c. 1990 , , 358 , 344 maxfield , l. , 1998 , in preparation mccarthy , p. , van breugel , w. , spinrad , h. , & djorgovski , s. 1987 , , 321 , l29 mccarthy , p. , spinrad , h. , dickinson , m. , van breugel , w. , liebert , j. , djorgovski , s. , & eisenhardt , p. 1990 , , 365 , 487 mccarthy , p , 1993 , , 31 , 639 mccarthy , p. , kapahi , v.k . , van breugel , w. , persson , s.e . , athreya , r. , & subrahmanya , c.r . 1996 , , 107 , 19 mccarthy , p. & lawrence , c.r . 1998 , in prepartion ( mcl98 ) miller , j.s . & stone , r.p.s . 1994 , lick obs . 66 neuschaefer , l.w . & windhorst , r.a . 1995 , , 96 , 371 oke , j.b . , 1995 , , 107 , 375 rawlings , s. & saunders , r. 1991 , , 349 , 138 rawlings , s. , lacy , m. , blundell , k. m. , eales , s. a. , bunker , a. j. , & garrington , s. t. 1996 , , 383 , 502 rttgering , h.j.a . , lacy , m. , miley , g.k . , chambers , k.c . , & saunders , r. 1994 , , 108 , 79 spinrad , h. & djorgovski , s. 1987 , in observational cosmology , ed . a. hewitt , g. burbridge , & l.z . fang ( iau symp . 124 ) , 129 spinrad , h. , dickinson , m. , schlegel , d. , & gonzalez , r. 1993 , in observational cosmology , ed . g. chincarini , a. iovino , & d. maccagni ( asp conf . ser . , 51 ) , 585 spinrad , h. , dey , a. , & graham , j.r . 1995 , , 438 , l51 spinrad , h. , dey , a. , stern , d. , dunlop , j. , peacock , j. , jimenez , r. , & windhorst , r. 1997 , , 484 , 581 spinrad , h. , dey , a. , stern , & bunker , a. 1998 , in proc . knaw colloq . , ed . h. rttgering , in press steidel , c.c . , giavalisco , m. , pettini , m. , dickinson , m. , & adelberger k.l . 1996 , , 462 , l17 stern , d. , spinrad , h. , dickinson , m. 1996 , , 111 , 102 stern , d. , spinrad , h. , dey , a. , dickinson , m. , & schlegel , d. 1997 , in the hst and the high redshift universe , ed . n. tanvir , a. aragon salamanca , & j. wall ( 37th herstmonceux ) , 413 . tadhunter , c.n . , morganti , r. , robinson , a. , dickson , r. , villar martin , m. , & fosbury , r.a.e . 1998 , , in press thompson , d. , djorgovski , s. , vigotti , m. , & grueff , g. 1994 , 108 , 828 van breugel , w. , stanford , s. , spinrad , h. , stern , d. , & graham , j.r . 1998 , , 502 , 614 van ojik , r. 1995 , phd thesis , rijksuniversiteit te leiden vigotti , m. , grueff , g. , perley , r. , clark , b. , & bridle , a. 1989 , , 98 , 419 lclcc mg 0018 + 0940 & i & 1995 sep 01 & keck & 600 & s & 1995 sep 01 & keck & 1500 mg 0046 + 1102 & i & 1995 sep 01 & keck & 1050 & s & 1995 sep 01 & keck & 3000 mg 0122 + 1923 & i & 1995 sep 01 & keck & 600 & s & 1997 sep 12 & keck & 2400 mg 0148 + 1028 & i & 1996 oct 15 & lick & 1260 & s & 1997 dec 23 & keck & 1800 mg 0308 + 0720 & i & 1996 oct 15 & lick & 600 & s & 1997 dec 23 & keck & 1800 mg 0311 + 1532 & i & 1994 jan 11 & mdm & 600 & s & 1995 aug 31 & keck & 3600 mg 0422 + 0816 & i & 1992 nov 25 & lick & 200 & s & 1995 oct 25 & keck & 1800 mg 0511 + 0143 & i & 1992 nov 25 & lick & 1500 & s & 1995 oct 25 & keck & 1800 mg 1142 + 1338 & i & 1994 apr 07 & mdm & 1800 & s & 1995 feb 03 & keck & 2400 mg 1251 + 1104 & i & 1995 mar 14 & keck & 1200 & s & 1995 mar 15 & keck & 4500 mg 1401 + 0921 & i & 1997 jul 01 & keck & 900 & s & 1997 jul 02 & keck & 2400 mg 2037@xmath230011 & i & 1994 sep 30 & mdm & 1500 & s & 1995 jul 23 & keck & 3600 & s & 1995 jul 24 & keck & 3600 & s & 1995 aug 31 & keck & 3600 mg 2041 + 1854 & s & 1996 jun 16 & keck & 300 mg 2058 + 0542 & i & 1992 aug 01 & lick & 1400 & i & 1993 aug 17 & lick & 1500 & s & 1994 jul 10 & keck & 3000 mg 2109 + 0326 & i & 1993 sep 14 & lick & 1100 & s & 1994 jul 09 & keck & 6550 mg 2121 + 1839 & i & 1992 aug 01 & lick & 1200 & s & 1994 jun 10 & keck & 4500 mg 2308 + 0336 & i & 1995 jul 25 & keck & 600 & s & 1995 jul 23 & keck & 2400 [ obsns ] lcll mg 0018 + 0940 & r & 00 18 55.23 & + 09 40 06.9 & o & 00 18 55.24 & + 09 40 06.8 & a & 00 18 53.93 & + 09 40 24.6 mg 0046 + 1102 & r & 00 46 41.40 & + 11 02 52.6 & o & 00 46 41.38 & + 11 02 52.5 & a & 00 46 43.88 & + 11 02 33.3 mg 0122 + 1923 & r & 01 22 29.95 & + 19 23 39.1 & o & 01 22 29.90 & + 19 23 38.6 & a & 01 22 31.27 & + 19 24 10.2 mg 0148 + 1028 & r & 01 48 28.85 & + 10 28 21.3 & o & 01 48 28.83 & + 10 28 22.0 & a & 01 48 26.23 & + 10 27 52.3 mg 0308 + 0720 & r & 03 08 41.90 & + 07 20 44.3 & o & 03 08 41.98 & + 07 20 44.9 & a & 03 08 40.35 & + 07 21 18.3 mg 0311 + 1532 & r & 03 11 56.89 & + 15 32 54.8 & o & 03 11 56.83 & + 15 32 55.4 & a & 03 11 54.52 & + 15 32 49.7 mg 0422 + 0816 & r & 04 22 24.00 & + 08 16 19.2 & o & 04 22 23.97 & + 08 16 18.7 & a & 04 22 23.95 & + 08 16 31.2 mg 0511 + 0143 & r & 05 11 04.77 & + 01 41 57.8 & o & 05 11 04.76 & + 01 42 00.3 & a & 05 11 02.89 & + 01 41 54.7 mg 1142 + 1338 & r & 11 42 23.6 & + 13 38 01.3 & o & 11 42 23.69 & + 13 38 01.4 & a & 11 42 22.80 & + 13 37 51.9 mg 1251 + 1104 & r & 12 51 00.02 & + 11 04 19.9 & o & 12 51 00.02 & + 11 04 21.6 & a & 12 50 58.67 & + 11 04 45.7 mg 1401 + 0921 & r & 14 01 18.3 & + 09 21 23.7 & o & 14 01 18.50 & + 09 21 21.2 & a & 14 01 16.66 & + 09 20 51.5 mg 2037@xmath230011 & r & 20 37 13.41 & @xmath2300 10 58.5 & o & 20 37 13.41 & @xmath2300 10 58.5 & a & 20 37 12.90 & @xmath2300 10 56.9 mg 2041 + 1854 & r & 20 41 24.2 & + 18 55 02.0 & o & 20 41 24.09 & + 18 55 00.9 & a & 20 41 25.20 & + 18 55 11.1 mg 2058 + 0542 & r & 20 58 28.95 & + 05 42 51.0 & o & 20 58 28.82 & + 05 42 50.7 & a & 20 58 30.13 & + 05 42 43.7 mg 2109 + 0326 & r & 21 09 21.71 & + 03 26 52.7 & o & 21 09 21.80 & + 03 26 51.6 & a & 21 09 20.93 & + 03 26 30.8 mg 2121 + 1839 & r & 21 21 25.48 & + 18 39 08.7 & o & 21 21 25.48 & + 18 39 09.0 & a & 21 21 25.35 & + 18 38 53.6 mg 2308 + 0336 & r & 23 08 25.0 & + 03 37 03.0 & o & 23 08 25.15 & + 03 37 03.6 & a & 23 08 26.07 & + 03 36 22.3 [ astrom ] lcccccccc mg 0018 + 0940 & 1.586 & 23.0 & 0.237 & 132 & 1.08 & 0.0 & & 34.70 mg 0046 + 1102 & 1.813 & 23.1 & 0.209 & 74 & 1.07 & 0.0 & & 34.61 mg 0122 + 1923 & 1.595 & 23.3 & 0.105 & 134 & 1.06 & 0.0 & & 34.70 mg 0148 + 1028 & 2.845 & 21.4 & 0.157 & 176 & 0.71 & 0.0 & & 35.39 mg 0308 + 0720 & 2.975 & 21.1 & 0.801 & 164 & 0.95 & 0.0 & & 35.56 mg 0311 + 1532 & 1.986 & 23.6 & 0.405 & 62 & 1.21 & 5.1 & 161 & 34.72 mg 0422 + 0816 & 0.294 & 20 : & 0.721 & 68 & 1.06 & 0.0 & & 32.53 mg 0511 + 0143 & 0.596 & 22 : & 0.393 & 98 & 1.06 & 3.9 & 171 & 33.4 2 mg 1019 + 0534 & 2.765 & 23.7 & 0.037 & 100 & 1.22 & 1.3 & 103 & 35.40 mg 1142 + 1338 & 1.279 & 23.9 & 0.093 & 149 & 1.03 & 0.0 & & 34.46 mg 1251 + 1104 & 2.322 & 24 : & 0.000 & 62 & 1.21 & 0.0 & & 34.94 mg 1401 + 0921 & 2.093 & 23.3 & 0.013 & 92 & 0.89 & 3.5 & 138 & 34.81 mg 2037@xmath230011 & 1.512 & 24.8 & 0.397 & 119 & 1.03 & 0.0 & & 34.57 mg 2041 + 1854 & 3.056 & 20 : & 0.441 & 217 & 0.76 & 0.0 & & 35 . 10 mg 2058 + 0542 & 1.381 & 23.7 & 0.425 & 283 & 1.19 : & 0.0 & & 34.90 mg 2109 + 0326 & 1.634 & 22.0 & 0.281 & 119 & 0.75 & 0.0 & & 34.55 mg 2121 + 1839 & 1.860 & 22.7 & 0.341 & 69 & 1.09 & 6.3 & 145 & 34.63 mg 2144 + 1928 & 3.592 & 23.5 : & 0.449 & 58 & 1.54 & 8.5 & 177 & 35.76 mg 2308 + 0336 & 2.457 & 23 : & 0.173 & 148 & 0.85 & 3.0 & 175 & 35.20 [ prop ] lcccc mg 0018 + 0940 & 132 & 156 & 159 @xmath97 22 & 190 @xmath97 14 mg 0046 + 1102 & 74 & 98 & 100 @xmath97 15 & mg 0122 + 1923 & 134 & 115 & 117 @xmath97 16 & mg 0148 + 1028 & 176 & 192 & 193 @xmath97 27 & mg 0308 + 0720 & 164 & 161 & 165 @xmath97 23 & 213 @xmath97 15 mg 0311 + 1532 & 62 & 53 & 54 @xmath97 9 & mg 0422 + 0816 & 68 & 113 & 116 @xmath97 17 & 102 @xmath97 12 mg 0511 + 0143 & 98 & 104 & 107 @xmath97 16 & 102 @xmath97 12 mg 1019 + 0534 & 115 & 115 & 132 @xmath97 19 & 100 @xmath97 12 mg 1142 + 1338 & 149 & 125 & 127 @xmath97 18 & mg 1251 + 1104 & 62 & & & mg 1401 + 0921 & 92 & 89 & 92 @xmath97 14 & 66 @xmath97 11 mg 2037@xmath230011&119 & 114 & & 179 @xmath97 14 mg 2041 + 1854 & 217 & 173 & 178 @xmath97 24 & mg 2058 + 0542 & 283 & 424 & 427 @xmath97 59 & 356 @xmath97 21 mg 2109 + 0326 & 119 & 81 & 86 @xmath97 14 & 75 @xmath97 11 mg 2121 + 1839 & 69 & 61 & 65 @xmath97 10 & mg 2144 + 1928 & 58 & 71 & 76 @xmath97 11 & mg 2308 + 0336 & 148 & 158 & 163 @xmath97 23 & 160 @xmath97 13 [ radflux ] lccccc mg 0018 + 0940 & 1549 & 4004.4 & 8.1 & 145 & 1.584 & 1640 & 4243.7 & 4.2 & 92 & 1.588 & ] 1909 & 4934.6 & 8.7 & 161 & 1.585 & ] 2326 & 6016.7 & 6.5 & 137 & 1.587 & [ ] 3426 & 6263.8 & 3.1 & 90 & 1.584 & 2798 & 7236.7 & 8.8 & 227 & 1.585 & 3203 & 8291.3 & 3.1 & 64 & 1.589 mg 0046 + 1102 & 1549 & 4357.2 & 6.5 & 99 & 1.812 & 1640 & 4615.8 & 5.5 & 83 & 1.814 & ] 1909 & 5366.5 & 7.9 & 124 & 1.811 & ] 2326 & 6542.6 & 7.4 & 120 & 1.813 & [ ] 3426 & 6813.2 & 1.9 & 29 & 1.811 & 2470 & 6947.5 & 2.0 & 30 & 1.813 & 2798 & 7875.6 & 11.8 & 176 & 1.814 mg 0122 + 1928 & 4 & 4033.9 & 3.2 & 54 & 1.604 & 2 & 4262.9 & 3.8 & 81 & 1.599 & 3 & 4958.0 & 3.2 & 60 & 1.597 & 2 & 6047.3 & 2.3 & 64 & 1.600 & 2 & 7273.3 & 3.1 & 77 & 1.598 mg 0148 + 1028 & ly@xmath19 & 4676.1 & 101.8 & 116 & 2.845 & 4 & 5965.3 & 116.3 & 128 & 2.851 & 2 & 6304.9 & 26.0 & 29 & 2.844 & 3 & 6400.8 & 8.4 & 9 & 2.849 & 3 & 7333.8 & 57.8 & 67 & 2.842 mg 0308 + 0720 & ly@xmath19 & 4858.3 & 181.8 & 71 & 2.995 & 5 & 4905.8 & 529.3 & 221 & 2.956 & & 5587.0 & 56.3 & 26 & 2.991 & 4 & 6162.9 & 155.9 & 75 & 2.979 & 3 & 7565.2 & 69.8 & 39 & 2.963 mg 0311 + 1532 & 4 & 4634.1 & 3.4 & 160 & 1.991 & 2 & 4903.5 & 2.0 & 103 & 1.990 & 3 & 5703.8 & 2.1 & 97 & 1.988 & 2 & & & & & 4 & 7239.3 & 1.1 & 50 & 1.988 mg 1019 + 0534 & & 4584.3 & 8.4 & 268 & 2.770 & 5 & 4664.8 & 2.3 & 66 & 2.762 & 4 & 5838.9 & 10.4 & 257 & 2.769 & 2 & 6174.9 & 8.5 & 170 & 2.765 & 3 & 7179.6 & 4.9 & 93 & 2.760 mg 1142 + 1338 & 3 & & @xmath983.0 & & & ] 2326 & & @xmath981.6 & & & [ ] 3426 & 7783.6 & 5.4 & 112 & 1.272 & [ ] 3727 & 8495.1 & 13.6 & 232 & 1.279 mg 1251 + 1104 & ly@xmath19 & 4040.5 & 23.1 & 148 & 2.324 & 1549 & 5148.4 & 3.0 & & 2.324 & 1640 & 5453.2 & 3.0 & & 2.325 & ] 1909 & 6326.7 & 5.2 & 298 & 2.314 & ] 2326 & & @xmath984.5 & & mg 1401 + 0921 & 1549 & 4793.9 & 4.1 & 165 & 2.095 & 1640 & 5072.8 & 5.0 & 171 & 2.093 & ] 1909 & 5901.1 & 3.4 & 83 & 2.091 & ] 2326 & 7187.2 & 1.7 & 89 & 2.090 & [ ] 2423 & 7521.8 & 2.9 & 121 & 2.104 mg 2037@xmath230011 & ] 1909 & 4782.2 & 4.4 & 123 & 1.505 & ] 2326 & 5843.5 & 1.0 & 39 & 1.512 & [ ] 2424 & 6093.9 & 1.0 & 43 & 1.514 mg 2041 + 1854 & ly@xmath19 & 4986 & 788.5 & 141 & 3.100 & ] 1400 & 5695:&130.8 & 28 & 3.05 : & 1549 & 6283 & 417.8 & 54 & 3.056 & ] 1909 & 7785 : & 68.3 & 25 & 3.08 : mg 2058 + 0542 & ] 1909 & 4541.3 & 7.1 & 152 & 1.379 & ] 2326 & 5537.6 & 4.1 & 120 & 1.381 & 2798 & 6667.0 & 6.6 & 206 & 1.381 & [ ] 2423 & 5769.6 & 1.3 & 45 & 1.380 & [ ] 3426 & & @xmath982.0 & & & [ ] 3727 & 8877.4 & 12.0 & 240 & 1.382 mg 2109 + 0326 & 1549 & & @xmath981.0 & & & 1640 & 4323.7 & 3.2 & 148 & 1.636 & ] 1909 & 5029.4 & 2.1 & 105 & 1.635 & ] 2326 & 6136.6 & 2.3 & 110 & 1.638 & [ ] 2423 & 6386.8 & 1.3 & 40 & 1.635 & 2798 & 7364.9 & 0.9 & 22 : & 1.631 mg 2121 + 1839 & 1549 & 4431.7 & 5.3 & 134 & 1.861 & 1640 & 4709.3 & 1.4 & 28 & 1.872 & ] 1909 & 5462.6 & 2.4 & 47 & 1.861 & ] 2326 & 6675 : & 1.5 : & 45 : & 1.869 & [ ] 3426 & & @xmath981.0 & & & 2798 & & @xmath980.5 & & mg 2144 + 1928 & ly@xmath19 & 5586.8 & 61.5 & 686 & 3.596 & 1549 & 7112.2 & 5.8 & 234 & 3.589 & 1640 & 7534.4 & 3.5 & 362 & 3.594 mg 2308 + 0336 & ly@xmath19 & 4209.2 & 29.3 & 284 & 2.462 & 1240 & 4288.2 & 5.7 & 118 & 2.458 & 1549 & 5357.9 & 6.3 & 91 & 2.458 & 1640 & 5669.1 & 3.9 & 78 & 2.457 & ] 1909 & 6589.1 & 4.5 & 110 & 2.452 & ] 2326 & 8036.9 & 8.3 & 412 & 2.455 [ lines ] lccccc ly@xmath19 & 1216 & 515 & 116 & 1130 & ly@xmath19 & 1216 & 606 & 106 & 1140 & & 1240 & 53 & 13 & 1420 & & 1549 & 131 & 31 & 1540 & doublet & 1640 & 94 & 23 & 1150 & ] & 1909 & 100 & 28 & 1260 & ] & 2326 & 83 & 30 & 1720 & & 2424 & 49 & 16 & 1780 & & 2470 & 17 & 5 & 930 & & 2800 & 116 & 42 & 2100 & doublet & 3727 & 188 & 142 & 1060 & [ composite ] lccccccc ly@xmath19 & 1216 & 515 & 1766 & 1000 & 239 & 485 & 253 & 1240 & 53 & 88 & & & 128 & & 1549 & 131 & 207 & 218 & 81 & 224 & 217 & 1640 & 94 & 181 & 37 & 4 & 26 & 62 ] & 1909 & 100 & 100 & 100 & 100 & 100 & 100 ] & 2326 & 83 & 52 & & 32 & 19 & 20 & 2424 & 49 & 51 & & @xmath988 & & 8 & 2470 & 17 & 23 & & 7 & & & 2800 & 116 & 43 & 33 & 113 & 113 & 117 & 3727 & 188 & 207 & 103 & & & 0.3 [ compare ]

we present high - resolution maps of stars , dust , and molecular gas in a strongly lensed submillimeter galaxy ( smg ) at . hatlas j114637.9 is selected from the _ herschel_-astrophysical terahertz large area survey ( h - atlas ) as a strong lens candidate mainly based on its unusually high 500 m flux density ( mjy ) . it is the only high - redshift _ planck _ detection in the 130 deg h - atlas phase - i area . keck adaptive optics images reveal a quadruply imaged galaxy in the-band while the submillimeter array and the jansky very large array show doubly imaged 880 m and co(1 ) sources , indicating differentiated distributions of the various components in the galaxy . in the source plane , the stars reside in three major kpc - scale clumps extended over.6 kpc , the dust in a compact ( kpc ) region kpc north of the stars , and the cold molecular gas in an extended ( kpc ) disk kpc northeast of the stars . the emission from the stars , dust , and gas are magnified by , , and times , respectively , by four lensing galaxies at . intrinsically , the lensed galaxy is a warm ( k ) , hyper - luminous ( ; sfr yr ) , gas - rich ( ) , young ( myr ) , and short - lived ( myr ) starburst . with physical properties similar to unlensed smgs , hatlas j114637.9 offers a detailed view of a typical smg through a powerful cosmic microscope .

1202.1829

bright submillimeter - selected galaxies ( smgs * ? ? ? * ) provide a powerful probe into the distant universe . thanks to the negative @xmath5-correction in the rayleigh - jeans tail of the dust thermal emission , flux limited submillimeter surveys with 850 @xmath2 m flux density @xmath18 mjy reach an almost uniform integrated infrared ( ir ) luminosity limit across a wide redshift range ( @xmath19 ) and yield a galaxy population mostly at redshifts between @xmath20 @xcite . with star formation rates ( sfrs ) of @xmath21 @xmath22 yr@xmath13 , the smgs are the most intense star - forming galaxies , despite their inevitably short - lived nature ( lifetime @xmath23 0.1 gyr ) . although such intense starburst systems are extremely rare in the local universe , smgs and the lyman break galaxies may contribute equally to the comoving sfr density at @xmath24 @xcite . in addition to their unique energetics , they also represent an important stage in massive galaxy formation . multiple lines of evidence suggest that smgs are likely the progenitors of massive elliptical galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which apparently have formed bulk of their stars rapidly at an early epoch ( e.g. , * ? ? ? our understanding of this important high - redshift galaxy population are limited by the sensitivity and spatial resolution of current facilities . gravitational lensing offers an elegant solution by effectively lifting both limiting factors . also thanks to the negative @xmath5-correction , it is relatively straightforward to identify strongly lensed smgs in large area submillimeter surveys . @xcite and @xcite predict that extragalactic sources with 500 @xmath2 m flux density @xmath25 mjy are mostly strongly lensed or blended smgs , nearby late - type galaxies , and radio active galactic nuclei ( agns ) . as demonstrated by @xcite , objects in the last two categories can be easily removed using data at other wavelengths , leading to an extremely high success rate in identifying strongly lensed smgs with this technique ( see also @xcite ) . this simple flux selection has produced a few well - studied strongly lensed smgs ( lockman01 @xmath26 : @xcite ; id141 @xmath27 : @xcite ; hls j091828.6@xmath28514223 @xmath29 : @xcite ; and hatlas12@xmath100 @xmath30 , the subject of this paper ) , all of which were discovered by the _ space observatory @xcite . more complex selection processes have been proposed ( e.g. , * ? ? ? * ) , which would allow selecting hundreds of fainter lensed galaxies with _ the brightest of the lensed smgs might also be detected by the _ planck _ mission @xcite . with @xmath34 resolutions , such sources are probably blended with fainter sources even in the highest frequency / resolution channels of _ planck _ ( 545 and 857 ghz , or 550 and 350 @xmath2 m ) . the phase - i 130 deg@xmath4 of the _ herschel_-astrophysical terahertz large area survey ( h - atlas ; * ? ? ? * ) covers 28 _ planck _ sources in the _ planck _ early release compact source catalog ( ercsc ; * ? ? ? @xcite find that sixteen of them are high galactic latitude cirrus , ten are low - redshift galaxies , and one is resolved into two similarly bright nearby spirals ( ngc 3719 and 3720 ) . only one _ planck _ source is dominated by high - redshift galaxies : plckerc857 g270.59 + 58.52 ( @xmath31 jy , @xmath32 jy ) . with 18 , 25 , and 36 angular resolutions at 250 , 350 and 500 @xmath2 m , respectively , _ herschel _ detect 16 objects within a 4.23 radius of the _ planck _ position ( _ planck _ has a full - width - half - maximum [ fwhm ] resolution of 4.23 at 857ghz ) . there are 15 faint ( @xmath33 mjy ) sources surrounding an unusually bright source ( hatlas j114637.9@xmath1001132 , hereafter hatlas12@xmath100 ; @xmath34 mjy ) . taking into account the differences in the beam size and the filter transmission , the _ herschel _ sources account for only @xmath328% and @xmath324% of the _ planck _ flux densities at 545 and 857 ghz , respectively , suggesting that the _ planck _ measurements are boosted because of either positive noise spikes ( i.e. , eddington bias ; * ? ? ? * ) or blending with an over - density of sources that are below the confusion limit of _ herschel _ @xcite . the reader is referred to @xcite for a detailed _ planck_-_herschel _ comparison . hatlas12@xmath100 peaks at 350 @xmath2 m in flux density ( `` 350 @xmath2 m peaker '' ) , implying a high photometric redshift given typical dust temperatures . subsequent detections of multiple carbon - monoxide ( co ) lines from this unusually bright object determined a redshift of @xmath35 ( zspectrometer , carma , z - spec ; * ? ? ? * van der werf et al . in prep ; riechers et al . in prep ) . the high 500 @xmath2 m flux density ( @xmath36 mjy ) , in combination with the confirmed high redshift , makes hatlas12@xmath100 an excellent strong lens candidate . it is also the only strongly lensed smg candidate associated with a _ detection in the entire 130 deg@xmath4 h - atlas phase - i region . although the _ planck _ detection is partly due to spurious factors ( i.e. , eddington bias and/or blending ) , the confirmation of the lensed nature of the dominating source demonstrates that _ planck _ can efficiently identify the brightest lensed smgs once galactic cirrus and low - redshift galaxies are removed . in this paper we present a detailed multi - wavelength analysis of this _ planck_-associated smg . we describe our high - resolution keck adaptive optics imaging , submillimeter array ( sma ) and jansky very large array ( jvla ) interferometric observations , and the panchromatic photometry in [ sec : obs ] . we then perform a joint strong lens modeling at rest - frame 0.5 @xmath2 m , 200 @xmath2 m , and co(1@xmath60 ) in [ sec : model ] . in [ sec : sed ] we derive the intrinsic physical properties of the smg from its spectral energy distribution ( sed ) . we conclude by discussing the physical properties of hatlas12@xmath100 in the context of unlensed @xmath17 smgs ( [ sec : summary ] ) . throughout we adopt a @xmath37cdm cosmology with @xmath38 , @xmath39 and @xmath40 = 70 km s@xmath13 mpc@xmath13 . we obtained a 3,440-s @xmath41-band ( hereafter @xmath5 ) image on 2011 april 13 ( ut ) and a 2,100-s @xmath42-band image on 2011 june 30 ( ut ) with the keckii laser guide - star adaptive - optics system ( lgsao ; * ? ? ? * ) . an @xmath43 magnitude star 48 sw of hatlas12@xmath100 served as the tip - tilt reference star . the estimated strehl ratios at the source position are @xmath323% and 5% in @xmath5 and @xmath42-band , respectively . we used the nirc2 camera at 004 pixel@xmath13 scale for both filters ( 40 field ) , and dithered with 2@xmath13 steps . the atmospheric seeing at 0.5 @xmath2 m was @xmath30.4 and 0.5 during the @xmath5 and @xmath42-band imaging , respectively . we used our idl ( interactive data language ) programs to reduce the images . after dark subtraction and flat - fielding , sky background and object masks are updated iteratively . for each frame , after subtracting a scaled median sky , the residual background is removed with b - spline models . in the last iteration , we discard the three frames of the poorest image quality and correct the nirc2 geometric distortion using the solution of p. b. cameron before combining the aligned frames . the resolution of the final @xmath5 and @xmath42-band images are 016 and 027 in fwhm , respectively . we measure the fwhms from the most compact source in the field located 10 se of hatlas12@xmath100 ( labeled `` psf '' in fig . [ fig : obs]@xmath44 ) ; we also use this object as the psf in the lens modeling ( [ sec : amoeba ] ) . the images are flux calibrated against ukirt infrared sky survey ( ukidss ; * ? ? ? * ) and reach depths of @xmath5 = 25.6 and @xmath45 ab for a 5@xmath46 detection with 01 and 02 radius apertures , respectively . limited by the small field of nirc2 , a deep wide - field image is required for astrometry calibration . optical imaging was obtained with the high - throughput auxiliary - port camera ( acam ) mounted at a folded - cassegrain focus of the 4.2-m william herschel telescope @xcite on 2011 april 26 ( ut ) . we obtained four images of 200 s on a @xmath32 field centered on hatlas12@xmath100 , without any filter . the seeing was @xmath309 . the images were reduced and combined following standard techniques in iraf . no accurate photometric calibration is possible because we did not use any broad - band filter . but by comparing sources extracted from the acam image and the sdss @xmath47-band catalog in the same field , we find that our image reaches an equivalent @xmath47-band 5-@xmath46 depth of 24.6 ab , or 2.3 magnitudes deeper than the sdss . we solve the astrometry of the acam image using the on - sky positions of sloan digital sky survey ( sdss ; * ? ? ? * ) dr8 sources inside the field . we use the astrometry routines in marc buie s idl librarybuie / idl/ ] to correct for offsets , rotation , and distortions , with four terms ( constant , @xmath48 , @xmath49 , and @xmath50 ) . sources that appear blended in the sdss catalog are excluded . with 35 sdss sources , we measure 1@xmath46 dispersions of @xmath51ra = 013 and @xmath51dec = 014 between the astrometry calibrated acam image and the sdss . finally , we use the same routines to match the nirc2 images to the acam image with 13 well - detected sources inside the 40 nirc2 field of view . the corrected nirc2 images show 1@xmath46 dispersions of @xmath51ra = 004 and @xmath51dec = 005 . we obtained sma interferometric imaging of hatlas12@xmath100 at 880 @xmath2 m ( 339.58 ghz ) in the compact array configuration with an on - source integration time ( @xmath52 ) of 1 hr and at 890 @xmath2 m ( 336.9 ghz ) in the subcompact array configuration with @xmath52 = 2 hr . the compact and subcompact observations took place on 2011 may 2 and 2012 january 14 , respectively . during both nights , atmospheric opacity was low ( @xmath53 ) and phase stability was good . both observations used an intermediate frequency coverage of 48 ghz and provide a total of 8 ghz bandwidth ( considering both sidebands ) . the quasars 1229@xmath28020 and 1058@xmath28015 were used for time - variable gain ( amplitude and phase ) calibration . the blazar 3c 279 served as the primary bandpass calibrator . for the compact data , we used titan as the absolute flux calibrator . for the subcompact data , we intended to use callisto as the flux calibrator , but jupiter might have fallen into one of the side lobes of the sma primary beam while we observed callisto . so we decided to use 3c 279 in lieu of callisto as the flux calibrator . it is possible to use 3c 279 because we have reliable measurements of its flux both before and after the observation of hatlas12@xmath100 . we used the invert and clean tasks in the multichannel image reconstruction , image analysis , and display ( miriad ) software @xcite to invert the _ uv _ visibilities and deconvolve the dirty map , respectively . we used natural weighting to obtain the best sensitivity . for the compact data , the cleaned image has a synthesized beam with a fwhm resolution of @xmath54 at a position angle ( pa ) of @xmath55 degrees east of north ; for the subcompact data , the beam is @xmath56 at pa = 65.7 deg . the primary beam of the sma is @xmath337 . the r.m.s . noise levels are 3.0 mjy beam@xmath13 and 3.6 mjy beam@xmath13 for the compact image and the subcompact image , respectively . hatlas12@xmath100 is resolved into two components by the sma ( fig . [ fig : obs ] ) . taking into account the 10% flux calibration uncertainty , the total flux is @xmath57 mjy and @xmath58 mjy for the compact image and the subcompact image , respectively . the latter agrees well with the laboca bolometer array flux measurement at 870 @xmath2 m ( [ sec : photometry ] ) . the compact array data did not fully capture the source flux because of the sparser array configuration , i.e. , @xmath59 of the total flux is distributed on spatial scales larger than those accessible to the sma in its compact array configuration . so we use the total flux from subcompact data for sed modeling ( [ sec : sedsmg ] ) . we chose to use the compact image for lens modeling ( [ sec : sma ] ) because of its higher spatial resolution . we find that using the subcompact image or the subcompact@xmath28compact combined image does not change the lens modeling result , but they give larger errors for the derived parameters . we exploited the recent upgrade to the national radio astronomy observatory s very large array @xcite , which includes the provision of ka - band receivers ( 26.540ghz ) , to observe the redshifted co(1@xmath60 ) emission from hatlas12@xmath100 at 27.06532ghz ( @xmath60ghz ; @xcite ) . observations were carried out dynamically during excellent weather conditions on 2012 january 6 and 8 . during this open shared risk observing period the available bandwidth from the new wideband interferometric digial architecture ( widar ) correlator consisted of two independently tunable output pairs of eight sub - bands each , with @xmath61-mhz full - polarisation channels per sub - band , giving a total bandwidth of 2,048mhz . at the redshift of hatlas12@xmath100 , however , the co(1@xmath60 ) line could be reached by only the bd output pair , giving @xmath62 km s@xmath13 coverage and @xmath63 km s@xmath13 resolution . we offset our tuning by 64mhz to avoid noisier edge channels . the 8 sub - bands of output pair ac were tuned to 32.5ghz . the bright compact calibration source , j1150@xmath10023 were observed every few minutes to determine accurate complex gain solutions and bandpass corrections . 3c286 ( @xmath64 = 2.1666 jy at 27.06 ghz ) was also observed to set the absolute flux scale , and the pointing accuracy was checked locally every hour . in total , around 2hr of data were obtained for hatlas12@xmath100 , with @xmath31hr of calibration . the data were reduced using @xmath65 ( 31dec12 ) following the procedures described by @xcite , though with a number of important changes : data were loaded using bdf2aips and fring was used to optimize the delays , based on 1min of data for 3c286 . the base bands were knitted together using the noifs task , yielding @xmath66 datasets with 512 @xmath67 2-mhz channels , which we then added together using the task dbcon . finally , the channels were imaged over a @xmath68 field , with natural weighting ( robust = 5 ) , to form a @xmath69 cube centered on hatlas12@xmath100 . integrating over those 55 channels found to contain line emission ( so a fwzi of @xmath70 km s@xmath13 ) yielded an r.m.s . noise level of 27@xmath2jybeam@xmath13 . the cleaned and velocity - integrated co map is shown in fig . [ fig : obs]@xmath71 . the beam is 25@xmath6722 at pa = @xmath72 . similar to the sma , the map resolves two components separated by @xmath35 . the co lines extracted from the two components show the same redshift and line profile , further confirming that they are lensed images of a single source . the best - fit gaussian to the area - integrated spectrum gives a line width of @xmath73 km s@xmath13 and a line flux of @xmath74jykm s@xmath13 . in comparison , the co(1@xmath60 ) measurements reported by @xcite using zpectrometer on the green bank telescope are : @xmath75 km s@xmath13 and @xmath76jykm s@xmath13(corrected for the 20% difference in the absolute flux density of 3c 286 ) . the reason for the discrepancy is unclear , but the two line flux measurements agree within the 1@xmath46 errors . so hereafter , we use the weighted mean of the two measurements , @xmath77jykm s@xmath13 , to derive the molecular gas mass . llccc sdss & @xmath78 & 0.36 & @xmath79 & + & @xmath80 & 0.47 & @xmath81 & + & @xmath82 & 0.62 & @xmath83 & + & @xmath47 & 0.75 & @xmath84 & + & @xmath85 & 0.89 & @xmath86 & + ukidss & @xmath49 & 1.03 & @xmath87 & + & @xmath42 & 1.25 & @xmath88 & @xmath89 + & @xmath90 & 1.63 & @xmath91 & + & @xmath5 & 2.20 & @xmath92 & @xmath93 + wise & @xmath94 & 3.35 & @xmath95 & @xmath96 + & @xmath97 & 4.60 & @xmath98 & @xmath99 + & @xmath100 & 11.56 & @xmath101 & @xmath102 + & @xmath103 & 22.09 & @xmath104 & + pacs & green & 100 & & @xmath105 + & red & 160 & & @xmath106 + spire & blue & 250 & & @xmath107 + & green & 350 & & @xmath108 + & red & 500 & & @xmath109 + laboca & & 870 & & @xmath110 + sma & & 890 & & @xmath111 + mambo & & 1200 & & @xmath112 + carma & & 2792 & & @xmath113 + & & 3722 & & @xmath114 + vla & & 214000 & & @xmath115 + photometry of hatlas12@xmath100 were obtained from the sdss ( @xmath116 and @xmath85 ) , the ukidss ( @xmath117 and @xmath5 ) , the wide - field infrared survey explorer ( _ wise _ , 3.6 and 4.6 @xmath2 m ; * ? ? ? * ) , the _ herschel_/pacs ( 100 and 160 @xmath2 m ; program i d : ot1_rivison_1 ; * ? ? ? * ) , the _ herschel_/spire ( 250 , 350 , and 500 @xmath2 m ; * ? ? ? * ; * ? ? ? * ) , the large apex bolometer camera ( laboca , 870 @xmath2 m ; * ? ? ? * ) , the sma ( 880 @xmath2 m ) , the max - planck millimetre bolometer ( mambo , 1.2 mm ; * ? ? ? * ) , the combined array for research in millimeter - wave astronomy ( carma , 2792 and 3722 @xmath2 m ; * ? ? ? * ) , and the vla first survey ( 21 cm ; * ? ? ? we obtained imaging at 870 @xmath2 m with the laboca bolometer array at the atacama pathfinder experiment ( apex ) telescope in november 2011 ( clements et al . in prep ) . laboca observed a 11.4 diameter field with a resolution of fwhm = 186 . the observations have a total integration time of @xmath330 hr reaching a 1@xmath46 sensitivity of @xmath32 mjy . we obtained 1.2 mm imaging with mambo at the iram 30-m telescope ( fwhm @xmath3 107 ) in january and february 2011 ( dannerbauer et al . in prep . ) . observing time in the on - off mode is 24 minutes , achieving a 1@xmath46 sensitivity of @xmath31 mjy . we obtained continuum observations at 81.2 and 108.2 ghz ( 3722 and 2792 @xmath2 m ; covering rest - frame co[3@xmath62 ] and co[4@xmath63 ] lines ) on 2011 mar 18 and sep 1 as part of our co follow - up campaign of bright , lensed h - atlas smgs with carma in d array ( riechers et al . in prep ) . observations were carried out for 0.9 and 1.4 hr on source , respectively , using the 3 mm receivers and a bandwidth of 3.7 ghz per sideband . hatlas12@xmath100 is unresolved in these observations , with angular resolutions of @xmath118 and @xmath119 at 81.2 and 108.2 ghz , respectively ( restored with natural baseline weighting ) . table [ tab : photo ] lists the photometry . we have included in the errors the absolute flux calibration uncertainties ( 3% for _ wise _ , 3@xmath15% for pacs , 7% for spire , 10% for sma , and 15% for laboca , mambo , and carma ) . because the lgsao image has the highest spatial resolution , we use it to find the best - fit lens model . we initially use the peak positions of the multiply - imaged source to constrain the lensing potentials ( [ sec : peak ] ) , then we exploit the @xmath5-band light distribution in the image plane to quantify the morphologies of the source as well as refining the lensing potentials ( [ sec : amoeba ] ) . finally , we use the best - fit lensing potentials and the sma and jvla images to constrain the sizes and locations of the dust and molecular gas in the source plane ( [ sec : sma ] & [ sec : evla ] ) . we use lenstool @xcite to find the best - fit parameters and their errors from the peak positions . lenstool implements a bayesian markov chain monte - carlo sampler to derive the posterior distribution of each parameter and an estimate of the evidence for the model . the lensing system is mainly made of two red filaments that are @xmath335 apart ( fig . [ fig : obs]@xmath44 ) . the outward curved shape of the northern arc can be explained if the source is intrinsically curved . hence we split each of the two arcs into three parts and build a simple lens model by putting two deflectors centered on g1 and g2 . we find that the predicted counter - images can explain the additional features close to g1 and g2 . guided by the predicted counter - images , we define three systems of lensed images ( fig . [ fig : obs]@xmath44 @xmath120 ) . the 11 peak positions in 3 separate systems provide a total of 16 constraints ( @xmath121 ) , allowing us to include shear from nearby galaxies g3 and g4 . for the lensing galaxies , we find photometric redshifts of @xmath122 and @xmath123 with the public photo-@xmath85 code eazy @xcite . we obtain the nine - band photometry from the sdss ( @xmath116 and @xmath85 ) and the ukidss ( @xmath117 and @xmath5 ) surveys . at these wavelengths , the flux from the lensed galaxy is negligible ( [ sec : sed ] ) . the redshift of g1@xmath28g2 is measured from the total fluxes of g1 and g2 , because they are blended in the seeing limited data . g3 is undetected in sdss but shows similar color as g1 and g2 . hence in the lens modeling we assume all four galaxies are at @xmath124 . note that although redshift errors of the lensing galaxies would lead to errors in the estimated lens masses , they would not change our conclusions on the lensed galaxy because the magnification factors would remain the same . for the lens model , we assume that the dark - matter plus baryonic mass profiles of the foreground lens galaxies g1 to g4 can be described as singular isothermal ellipsoids ( sies ; * ? ? ? the sie profile is parametrized by the velocity dispersion ( @xmath46 ) , the position ( @xmath125 , @xmath126 ) , the axis ratio ( @xmath127 ) , and the pa ( @xmath128 , e of n ) . we fix the positions to the centers of the galaxies . for g3 and g4 , we further fix their @xmath129 and @xmath128 to those from the light distribution , because they are not well constrained by the peak positions and there are significant correlations between the pa and ellipticity of the light and of the mass distribution ( e.g. * ? ? ? therefore , we have a total of eight free parameters . we find a best - fit with @xmath130 for dof = 8 ( degrees of freedom ) and an average positional error of 004 ( @xmath31 pixel ) . the parameters and their errors are summarized in table [ tab : lens ] . we also list the mass enclosed by the critical curve for each sie , @xmath131 where @xmath132 , @xmath133 , and @xmath134 are the angular diameter distances to the lens , to the source , and between the lens and the source , respectively . the radius of the area enclosed by the critical curve can be approximated by the circularized einstein radius : @xmath135 in the errors of masses and velocity dispersions , we have included the 1-@xmath46 uncertainty of the photometric redshift . the nominal model described above is the most favorable description of the lensing system because of the following : 1 . adding @xmath129 s and @xmath128 s of g3 and g4 as free parameters does not substantially improve the fit : the bayesian evidence , strong if @xmath136 , and decisive if @xmath137 @xcite . ] increases by only @xmath138 = 0.5 and the reduced @xmath139 actually increases from 1.0 to 1.6 as a result of the decreased degree of freedom . 2 . excluding the potentials of g3 and/or g4 does degrade the fit significantly . the bayesian evidence decreases by @xmath138 = 2.5 and 30 , and the reduced @xmath139 increases from 1.0 to 1.8 and 6.5 , when we exclude g4 and both g3 and g4 , respectively . including a group - scale potential with a piemd profile ( pseudo - isothermal elliptic mass distribution ; * ? ? ? * ) does not improve the fit . for the piemd profile , we adopt a cut - off radius of 500 kpc but allow the position , ellipticity , pa , core radius , and velocity dispersion to vary . to limit the number of free parameters , we fix the @xmath129 s and pa s of the sies to those measured from the lensing galaxies but allow their velocity dispersions to vary . so we have a total of 10 free parameters . the reduced @xmath139 of the best - fit is 1.9 , much higher than that of the nominal model . the bayesian evidence also decreases by @xmath138 = 6.9 when compared with the nominal model . llccr g1 & @xmath140 & @xmath141 & @xmath142 & + & @xmath143 & @xmath144 & @xmath145 & deg + & @xmath146 & @xmath147 & @xmath148 & km s@xmath13 + & @xmath149 & @xmath150 & @xmath151 & @xmath12 + g2&@xmath140 & @xmath152 & @xmath153 & + & @xmath143 & @xmath154 & @xmath155 & deg + & @xmath146 & @xmath156 & @xmath157 & km s@xmath13 + & @xmath149 & @xmath158 & @xmath158 & @xmath12 + g3&@xmath146 & @xmath159 & @xmath160 & km s@xmath13 + & @xmath149 & @xmath161 & @xmath162 & @xmath12 + g4&@xmath146 & @xmath163 & @xmath164 & km s@xmath13 + & @xmath149 & @xmath165 & @xmath166 & @xmath12 + & & & & + smg & @xmath167 & & @xmath168 & + & @xmath169 & & @xmath170 & + & @xmath171 & & @xmath172 & + although the peak positions can constrain the deflectors through ray tracing , they can not provide an accurate estimate of the magnification factor because the source - plane light distribution is not taken into account . because we want to estimate the intrinsic properties of the lensed galaxy , we are interested in the luminosity weighted magnification factor , which depends on the source morphology because the magnification factor is different at each source plane position . in this section we model the morphology of the lensed galaxy and refining the lensing potentials simultaneously with the @xmath5-band image . the psf is derived from the most compact source in the field , which is 10 se of hatlas12@xmath100 ( fig . [ fig : obs]@xmath44 ) . following [ sec : peak ] , we assume that the source consists of three clumps , each described as a srsic profile . again we use sie profiles for the lensing potentials . so we have a total of 29 parameters : seven parameters for each srsic profile , and eight parameters for the sie potentials . our fitting procedure is as follows . for an initial set of parameters describing the source and the lenses from [ sec : peak ] , we use lenstool to generate lensed images of the source , which is then convolved with the psf and compared with the observed image . we limit the comparison in a @xmath173 ( @xmath174 pixels ) rectangular region that encloses the lensing features . this process is iterated with amoeba_sa to find the parameters that minimize the residual between the observation and the model . amoeba_sa is based on the idl multidimensional minimization routine amoeba @xcite with simulated annealing added by e. rosolowsky . we allow a maximum of 1000 iterations in each call of amoeba_sa . for the simulated annealing , we adopt an initial `` temperature '' of 100 and decrease it by 20% in each subsequent call to amoeba_sa . a good fit with a reduced @xmath139 around unity is normally found after a few calls to amoeba_sa . for each iteration , we compute the total luminosity - weighted magnification factor ( @xmath175 ) by summing the pixel values in the image and the source planes with apertures matched by inverting the image plane aperture to the source plane . the 1-@xmath46 confidence interval of @xmath175 is found with @xmath176 . note that we compute the @xmath139 values on the residual image binned by @xmath177-pixel boxes ( fwhm = 016 = 4 pixel ) , so that the noise becomes uncorrelated between pixels ; or equivalently , one could divide the @xmath139 values from the original residual images by a factor of 16 . we find the luminosity - weighted magnification to be @xmath178 @xmath168 . the best - fit parameters for the deflectors are listed in table [ tab : lens ] . the results are very similar to those from fitting the peak positions , although the errors are smaller because the entire image provides more information than the peak positions alone . figure [ fig : lens ] shows the best - fit model . the lensed galaxy has a curved morphology , causing the northern arc bending in the opposite direction of the deflectors . in the source plane , the three clumps extend over only 021 or 1.6 kpc , and their effective radii are @xmath179 ( @xmath180 kpc ) , @xmath181 ( @xmath182 kpc ) , and @xmath183 ( @xmath184 kpc ) from w to e. the nature of the feature nne of g2 in the residual image is unclear , but it is unlikely to be at the same redshift as the lensed galaxy : tracing its position to the source plane and imaging it back predicts an _ unobserved _ equally bright counter - image 08 s of the southern arc . this feature could therefore be part of the galaxy g2 . precise astrometry calibration is crucial for a joint analysis of images from different wavelengths . because the only @xmath5 source detected by the sma is hatlas12@xmath100 , we have to estimate the astrometry offset between the two images in a statistical way . because the keck image is tied to the sdss astrometry and the sma image is tied to the radio reference frame , we cross - correlate the vla first catalog @xcite and the sdss catalog within @xmath185 of hatlas12@xmath100 and compute the optical - radio separation . ninety - four radio sources have optical counterparts within 3 . we then fit an elliptical gaussian to the two - dimensional distribution in @xmath186 and @xmath187 . the systematic offset from the peak position of the gaussian is consistent with zero ( @xmath188 , @xmath189 ) . the best - fit gaussian has @xmath46 s of 040 and 030 , and a pa of @xmath190 for the major axis . therefore , the 1@xmath46 ellipse of the astrometry offset has major / minor semi - axes of 061/045 . our result is consistent with that of @xcite , who found a @xmath301 systematic offset and a 1@xmath46 error circle of 047 in radius between first and sdss astrometry . we can constrain the astrometry offset further through lens modeling . as demonstrated by @xcite and @xcite , interferometric data are most naturally modeled with the @xmath66-plane visibilities , because it avoids beam deconvolution and naturally handles correlated noise . here , however , we opt to model the cleaned map directly , because ( 1 ) the images are essentially unresolved in the sma map , and ( 2 ) we already have a good lens model from the @xmath5-band image ( [ sec : amoeba ] ) . because of the limited spatial resolution of the sma 880 @xmath2 m image , the two centroid positions do not offer enough information to constrain the lens model . hence , for the deflectors , we fix all the parameters to the best - fit values from [ sec : amoeba ] ; for the source , we assume a circular gaussian profile with variable position and size . we shift the sma image relative to the @xmath5 model on a 2@xmath672 grid with 01 steps . at each offset position , we find the best - fit model using the same fitting procedure as in [ sec : amoeba ] . the modeling is performed on a @xmath191-pixel ( @xmath192 ) region enclosing the sma sources . figure [ fig : xychi2]@xmath44 shows a map of the minimum @xmath139 values at each offset position . the global best - fit , with reduced @xmath139 of unity , is reached when we shift the sma image @xmath193 e of the @xmath5 image . the middle panels of fig . [ fig : lens ] show this global best - fit model . the noise of the sma map is gaussian but is highly correlated . we compute the r.m.s . noise of the sma map after binning it by boxes of @xmath194 pixels . we find that the noise starts to decrease as @xmath195 for @xmath196 pixels ( fwhm @xmath197 2 = 20 pixel ) , indicating that the noise becomes uncorrelated on 20-pixel scales . therefore , we divide the @xmath139 values from the the residual images by a factor of 400 , which is equivalent to computing @xmath139 from residual images binned by @xmath198-pixel boxes . in combination with the 1@xmath46 error ellipse from first sdss cross - correlation , we determine that the astrometry offset between 880 @xmath2 m and @xmath5 images is @xmath199 and @xmath200 ; i.e. , the overlapping region between the ellipse and the 1@xmath46 contour of the @xmath139 map . collecting all of the solutions in this permitted offset region satisfying @xmath176 , we estimate a luminosity - weighted 880 @xmath2 m magnification of @xmath201 @xmath170 , and an 880 @xmath2 m source size of fwhm = @xmath202 = @xmath203 kpc . because we have fixed the deflectors with the best - fit parameters from @xmath5-band , the errors here do not include the uncertainties of the deflectors . higher resolution far - ir images are required to constrain the deflectors and the source simultaneously . dust emitting regions are often spatially offset from the uv / optical emitting regions in smgs @xcite . this is clearly the case for hatlas12@xmath100 , which shows distinctly different morphologies at @xmath5-band and 880 @xmath2 m , even after convolving the @xmath5-band image with the sma beam ( compare fig . [ fig : lens]@xmath71 & @xmath204 ) . from the lens model , we estimate a source - plane separation between the 880 @xmath2 m source and the central @xmath5 clump of @xmath205 or @xmath206 kpc ( fig . [ fig : lens]@xmath207 ) . if we assume zero astrometry offset between sma and keck , then we obtain a model that poorly fits the observation ( @xmath208 ; fig . [ fig : xychi2]@xmath44 ) . the lens model gives a slightly larger magnification ( @xmath209 ) and doubles the source size ( fwhm = @xmath210 kpc ) . however , the source - plane separation between the 880 @xmath2 m source and the central @xmath5 clump remains the same ( @xmath211 kpc ) . we use the same technique to model the jvla co(1@xmath60 ) map as in [ sec : sma ] . the lensed images are better resolved in the jvla image than in the sma image , so we use an elliptical gaussian instead of a circular gaussian for the source profile . the model has a total of six free parameters ( @xmath212 , fwhm , @xmath129 , pa , and flux density ) . again , we can constrain the astrometric offset between jvla and keck through lens modeling . figure [ fig : xychi2]@xmath213 shows the minimum @xmath139 values at each offset position relative to the @xmath5 image . to deal with the correlated noise , we scale the @xmath139 values from the residual map by the product of the fwhms of the major and minor axes of the beam . the global best - fit , with reduced @xmath139 of unity , is reached when we shift the jvla image 02 e of the @xmath5 image . the bottom panels of fig . [ fig : lens ] show the best - fit model . we estimate a co magnification of @xmath214 @xmath172 . similar to the sma image , we find a source - plane separation of @xmath215 kpc between the cold molecular gas and the stellar emission ( i.e. , the central @xmath5 clump ) . the co(1@xmath60 ) is emitted from a more extended region than the dust , but the two spatially overlap ( fig . [ fig : lens]@xmath207 ) . the co source has fwhm = @xmath216 = @xmath217 kpc along the major axis , with an axis ratio of @xmath218 . the molecular gas disk is massive . the velocity - area - integrated co brightness temperature of @xmath219\times10^{11}$]k km s@xmath13 pc@xmath4 indicates a molecular gas reservoir of @xmath220\times10^{10}$]@xmath12 after lensing correction , assuming a conversion factor of @xmath221 @xmath12 ( k km s@xmath13 pc@xmath4)@xmath13 , which is commonly assumed for starburst environments @xcite . note , however , that @xmath222 is uncertain by at least a factor of a few and it may depend on the metallicity , the gas temperature , and the velocity dispersion of the galaxy @xcite . the co magnification factor determined from the lens model ( @xmath172 ) is in excellent agreement with that estimated from the co luminosity@xmath1fwhm correlation . the observed @xmath223 and line width indicates a magnification factor of @xmath224 , based on its deviation from the correlation established by unlensed smgs ( * ? ? ? * bothwell et al . in prep ) . this agreement demonstrates that strongly lensed smgs may be effectively selected with co spectroscopy in the future . llcr g1@xmath28g2 & @xmath225 & @xmath226 & @xmath12 + & age & @xmath227 & gyr + & @xmath228 & @xmath229 & gyr + & @xmath230 & @xmath231 & mag + & sfr@xmath232 & @xmath233 & @xmath234 + & & & + smg@xmath235 & @xmath225 & @xmath236\times10^{10}$ ] & @xmath12 + & sfr@xmath232 & @xmath237 & @xmath234 + & & & + smg@xmath238 & @xmath239 & @xmath240 & k + & @xmath241 & @xmath242 & + & @xmath243 & @xmath244\times10 ^ 8 $ ] & @xmath12 + & @xmath245 & @xmath246 & @xmath2 m + & @xmath46 & @xmath247 & kpc@xmath4 + & @xmath248 & @xmath249\times10^{13}$ ] & @xmath10 + & sfr@xmath250 & @xmath251 & @xmath234 + & & & + smg@xmath252 & @xmath239 & @xmath253 & k + & @xmath241 & @xmath254 & + & @xmath243 & @xmath255\times10 ^ 9 $ ] & @xmath12 + & & & + smg@xmath256 & @xmath223 & @xmath257\times10^{10}$ ] & k km s@xmath13 pc@xmath4 + & @xmath258 & @xmath259 & @xmath10/k km s@xmath13 pc@xmath4 + & fwhm@xmath260 & @xmath261 & km s@xmath13 + & @xmath262 & @xmath263\times10^{10}$ ] & @xmath12 + & @xmath264 & @xmath265\times10^{11}$ ] & @xmath12 + & @xmath266 & @xmath267% & + & @xmath268 & @xmath269% & + useful physical parameters are encoded in the seds . the optical - to - nir sed of hatlas12@xmath100 is dominated by the foreground galaxies g1 and g2 , from which we can derive the photometric redshift and the stellar population of the lensing galaxies . the far - ir and submillimeter regime is dominated by the lensed smg , as evident in the sma image , so the data can tell us the dust and star formation properties of the smg . adopting the photometric redshift of 1.06 , we model the nine - band photometry ( @xmath78 , @xmath80 , @xmath82 , @xmath47 , @xmath85 , @xmath49 , @xmath42 , @xmath90 , and @xmath5 ) of g1+g2 with the stellar population synthesis models of ( * ? ? ? we assume a @xcite initial mass function ( imf ) , @xcite extinction law , and exponentially declining star formation history , with a range of e - folding times ( @xmath228 = 0.1 to 30 gyr ) and ages ( 0.01 to 12.5 gyr ) . for each template , we fit for the stellar mass ( @xmath225 ) and extinction ( @xmath230 ) . the best - fit model gives @xmath270 for dof = 7 ( fig . [ fig : sed]@xmath44 ) . the derived properties of g1@xmath28g2 are listed in table [ tab : sedfit ] . the intrinsic extinction is small ( @xmath271 ) and there is very little current star formation ( sfr = @xmath233 @xmath234 ) . the dust - absorbed uv / optical luminosity ( @xmath272\times10^{10}$ ] @xmath10 ) is less than 0.15% of the total integrated ir luminosity before lensing correction ( @xmath273 = @xmath274 @xmath10 ) . therefore , g1 and g2 do not contribute significantly to the far - ir fluxes , in agreement with their absence in the sma image . the stellar mass from sed modeling is @xmath380% of the total mass within the critical curves from lens modeling ( table [ tab : sedfit ] ) , implying that the galaxies are dominated by stellar mass within @xmath275 kpc . in the @xmath42 and @xmath5-bands , we obtain the photometry of the smg with an aperture contoured around the multiply - imaged features after subtracting the foreground lenses . we measure @xmath276 and @xmath277 in ab magnitudes , consistent with the red @xmath278 colors of unlensed smgs @xcite . careful modeling is required to extract smg photometry from the _ wise _ data , because the smg is blended with the foreground galaxies g1 through g4 ( fwhm = 6@xmath112 ) . we model the _ wise _ 3.4 @xmath2 m source with four elliptical gaussians of the same shape . with galfit @xcite , we fix their positions to those determined from the keck image , but we allow the gaussian shape to vary . then we measure the flux density of g1@xmath28g2 ( @xmath3205 @xmath2jy ) decomposed from that of g3 and g4 ( @xmath3117 @xmath2jy ) . finally , the flux density of the smg ( @xmath337 @xmath2jy ) is estimated from the excess of g1@xmath28g2 relative to their best - fit stellar population synthesis model ( [ sec : sedlens ] ) . unfortunately , we can not separate g3@xmath28g4 from g1@xmath28g2 in the longer wavelength channels of _ wise _ because of the inferior image quality , so we treat the excesses over the best - fit model of g1@xmath28g2 as upper limits for the smg . the far - ir - to - millimeter sed is dominated by the lensed smg , therefore no foreground subtraction is necessary . table [ tab : photo ] summarizes the photometry for hatlas12@xmath100 . we opt to model the rest - frame optical and far - ir emission separately , instead of fitting them together in a self - consistent way with magphys @xcite , because our lens model shows that they are emitted from physically distinctive regions , i.e. , the dust that attenuates the optical emission has little to do with the starburst - heated dust that emits in the far - ir . it is difficult to constrain the stellar population with only three photometric detections ( dof = 1 ) in rest - frame optical . however , we can limit the parameter space by excluding unphysical models , such as those that require negative extinction corrections and those whose ages exceed the cosmic age at @xmath279 ( 1.9 gyr ) . we use the same bc03 templates as in the previous section . the shaded region in fig . [ fig : sed]@xmath44 shows all of the permitted models with @xmath280 . these models give a range of extinctions , stellar masses , and sfrs : @xmath281 , @xmath282\times10^{10}$ ] @xmath12 , and sfr @xmath283 @xmath234 ( table [ tab : sedfit ] ) . the dust - absorbed uv / optical luminosity range from 0% to 140% of the observed @xmath273 ; but 90% of the models have dust - absorbed luminosity less than 50% of the observed @xmath273 . we fit the far - ir sed with a single - temperature modified blackbody , @xmath284 where @xmath46 is the total absorption cross section of dust particles at the optically thick limit ( i.e. , the size of the dust - obscured region ) , @xmath285 the _ planck _ function , @xmath286 the optical depth , @xmath2 the lensing magnification factor , and @xmath287 the luminosity distance . in the optically thin limit ( @xmath288 ) , dust mass can be derived based on the knowledge of the opacity @xmath289 ( absorption cross section per unit mass ) : @xmath290 it is generally assumed that the opacity follows a power law , @xmath291 , and has a normalization of @xmath292 m@xmath4 kg@xmath13 at 850 @xmath2 m @xcite . both the general `` optically thick '' ( @xmath293 ) and the optically thin ( @xmath294 ) models provide good fits to the observed sed ( fig . [ fig : sed]@xmath213 ) . for the general model , we use all of the nine detections between 100 @xmath2 m and 3 mm . the best - fit general model gives @xmath295 for dof = 5 , suggesting that the photometric errors have been overestimated . for the optically thin model , we exclude the pacs 100 @xmath2 m point , which is clearly on the wien tail where small grains tend to dominate the emission . the best - fit optically thin model gives @xmath296 for dof = 5 . the derived parameters are listed in table [ tab : sedfit ] . the optically thick model yields dust properties similar to those of the local ultra - luminous infrared galaxy ( ulirg ) arp 220 @xcite , with the optical depth exceeds unity below rest - frame @xmath3250 @xmath2 m . the intrinsic @xmath297 @xmath2 m luminosity of @xmath298\times10^{13}$ ] @xmath10 classifies hatlas12@xmath100 as a hyper - luminous infrared galaxy ( hy - lirg ) . the ir luminosity implies an sfr of @xmath251 @xmath12 yr@xmath13 for a @xcite imf @xcite . using the values of @xmath273 and @xmath299 k in the stefan - boltzmann law we obtain a spherical source radius of @xmath300 pc , which is three times larger than that of arp 220 ( 230 pc ) because of the ten times greater luminosity . the source radius is comparable to that we derive from the optically thick model ( @xmath301 pc ) and is consistent with the size we measure from modeling the sma image ( fwhm = @xmath203 kpc ; [ sec : sma ] ) . therefore , the optically thick model is preferred . the radio luminosity from the observed 1.4 ghz flux density is @xmath302\times10^{25}$ ] w hz@xmath13 for a radio spectral index of @xmath303 . assuming the radio emission is magnified by the same factor as the submillimeter emission , the ir - to - radio luminosity ratio of hatlas12@xmath100 , @xmath304 , is consistent with the radio far - ir correlation of high - redshift starburst galaxies : e.g. , @xcite measured @xmath305 for 15 smgs , while @xcite measured @xmath306 for 65 _ herschel _ 250 @xmath2 m selected galaxies . this suggests that the agn contribution is insignificant in hatlas12@xmath100 . we also do not see significant agn contribution in the mid - ir . in fig . [ fig : sed]@xmath213 , we fit the seds of the local ulirgs mrk 231 , arp 220 , and the @xmath307 smg `` cosmic eyelash '' @xcite to the far - ir sed . the _ wise _ upper limits lie well below the agn - dominated ulirg mrk 231 but are more consistent with arp 220 and the eyelash . therefore , we conclude that hatlas12@xmath100 is predominantly a starburst system . we have presented high - resolution @xmath5-band , 880 @xmath2 m , and co(1@xmath60 ) observations and the near - ir - to - centimeter sed of a _ herschel_-selected strongly lensed smg at @xmath279 ( hatlas12@xmath100 ) . the smg shows distinctly different morphologies in the three images , suggesting differential magnification due to stratified morphologies . a joint strong lens modeling shows that the smg is lensed by four galaxies at @xmath7 and the luminosity - weighted magnification factors are @xmath168 in @xmath5 , @xmath170 at 880 @xmath2 m , and @xmath172 at co(1@xmath60 ) . in the source plane , the smg consists of several stellar clumps extended over @xmath31.6 kpc with @xmath236\times10^{10}$ ] @xmath12 of stars , a compact ( @xmath31 kpc ) starburst enshrouded by @xmath244\times10 ^ 8 $ ] of dust at @xmath360k , and an extended ( @xmath36 kpc ) cold molecular gas reservoir with @xmath263\times10^{10}$ ] @xmath12 of gas . the starburst and its gas reservoir are located @xmath34 kpc from the stars . similar separations between optical and submillimeter / radio emission have been observed in unlensed high - redshift dusty starbursts ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? . however , these previous results could also be attributed to astrometry offsets across different facilities ; @xcite might be the only exception , whose tied the astrometry of the images to larger radio and optical fields . in hatlas12@xmath100 , the physical separations amongst stars , dust , and gas are less ambiguous because of the clear wavelength - dependent morphologies in the image plane : fortuitously , the stars and dust / gas straddle across the caustic , so the less obscured @xmath5-band region is quadruply imaged while the heavily obscured starburst and its gas reservoir are doubly imaged . because of the @xmath34 kpc separation between the stars and the gas - rich starburst and their similar masses , it is tempting to suggest that the smg is in the process of a major merger , which presumably is driving the starburst activity in @xmath17 smgs ( e.g. , * ? ? ? * ; * ? ? ? however , spatial separation of this scale can also be explained by differential dust obscuration in a single galaxy , as has been proposed for other high - redshift smgs where spatial offsets have been observed between rest - frame uv and submillimeter ( e.g. , gn20 and aztec 3 ; * ? ? ? * ; * ? ? ? differential magnification may affect the observed far - ir sed as well as the co ladder , as hotter dust and higher-@xmath42 co lines may have more compact morphologies @xcite . however , because the magnification map is smooth in the area , the six times difference in the sizes of the co(1@xmath60 ) and the dust emitting region only lead to a @xmath310% difference in magnification , which is smaller than the 1@xmath46 errors of our estimates of the magnification factors . therefore , differential magnification is unlikely to be significant enough to affect the far - ir sed and future co ladder measurements in hatlas12@xmath100 . how does the massive gaseous disk compare with the disks in other smgs ? we estimate an dynamical mass of @xmath265\times10^{11}$]@xmath12 and a gas fraction of @xmath308% for the co(1@xmath60 ) disk using the `` isotropic virial estimator '' ( e.g. , * ? ? ? * ) : @xmath309 where @xmath310 is the co line fwhm in km s@xmath13 and @xmath311 is half of the fwhm size of the disk in kpc . combined with the fwhm disk radius , we further estimate a gas surface density of @xmath312 @xmath12 pc@xmath313 . both @xmath314 and @xmath315 are similar to those of the extended co(1@xmath60 ) disks in the two @xmath316 smgs in @xcite . but both values are significantly smaller than those of the kinematically resolved co(6@xmath65 ) disk of the `` cosmic eyelash '' ( @xmath317 , @xmath318@xmath12pc@xmath313 ; * ? ? ? * ) . the discrepancies illustrate the limitations of these widely used but crude estimators _ and/or _ that high excitation co lines probe more compact and denser regions in a disk . higher resolution observations are clearly needed to resolve this issue . without spatially resolved gas kinematics , we refrain from estimating the disk stability parameter of @xcite . hatlas12@xmath100 is a gas - rich , initial starburst system similar to unlensed smgs and local ulirgs . its intrinsic ir luminosity well exceeds @xmath319 @xmath10 ( i.e. , hy - lirg ) , implying an enormous rate of star formation ( @xmath251 @xmath12 yr@xmath13 ) . although the molecular gas reservoir is massive and it constitutes @xmath267% of the _ visible _ baryonic mass ( @xmath320 ) and @xmath269% of the dynamical mass , it will exhaust in just @xmath321 myr at the current sfr ( assuming no gas accretion ) . the star formation timescale , @xmath322/sfr = @xmath323 myr , is only @xmath31% of the cosmic age at @xmath279 ( @xmath324 gyr ) , suggesting that hatlas12@xmath100 is an initial starburst system with maturity @xmath325 @xcite . the specific sfr , ssfr = sfr/@xmath225 = @xmath326 gyr@xmath13 , is consistent with the average @xmath17 smgs , but it is an order of magnitude higher than the median value of the star - forming main sequence of lyman break galaxies at the same epoch @xcite . we can also estimate the star formation efficiency : @xmath327 , where @xmath328 is the dynamical or free - fall timescale . for @xmath329 = 3 kpc and @xmath330 , we obtain @xmath331 , or [ 11@xmath3324]% per dynamical timescale , which is comparable to unlensed smgs but is an order of magnitude higher than normal starforming galaxies @xcite . the dust mass of @xmath244\times10^{8}$ ] @xmath12 is similar to the average dust mass of unlensed smgs ( e.g. , * ? ? ? * ) , and the gas - to - dust ratio , @xmath333/0.07 m@xmath4 kg@xmath334/0.8 ) , is comparable to that of the milky way . assuming that the dust emission is indicative of the size of the starburst and a starburst disk radius of @xmath335 kpc , the star formation surface density of @xmath336 @xmath234 kpc@xmath313 approaches the eddington limit of radiation pressure supported starburst disks ( @xmath337 @xmath234 kpc@xmath313 ; * * ; * ? ? ? * ) , similar to local ulirgs such as arp 220 and the host galaxy of the @xmath338 quasar sdss j114816.64 + 525150.3 @xcite . in conclusion , hatlas12@xmath100 is a _ bona fide _ smg with an intrinsic submillimeter flux density of @xmath339 mjy . the starburst disk , where most of the molecular gas and dust reside , is spatially separated from the less obscured stellar population by @xmath34 kpc , suggesting either a major merger or differentiated dust obscuration . the @xmath31 kpc radius starburst disk is presumably supported in large by radiation pressure on the dust grains . its physical properties , such as molecular gas mass , stellar mass , gas - to - dust ratio , gas fraction , sfr , star formation efficiency , and radio - to - far - ir luminosity ratio , are all very similar to unlensed @xmath17 smgs @xcite . the lensing boost of the effective angular resolution and sensitivity has allowed us to examine in unprecedented details the properties of a typical starburst galaxy when the universe is only 1/7 of its current age . hatlas12@xmath100 provides a prelude to a golden age of smg research , as _ herschel _ is unveiling hundreds of strongly lensed smgs before the mission completes ( e.g. , * ? ? ? * wardlow et al . in prep ) . we thank the anonymous referee for comments that helped improve the paper . hf , ac , jlw and sk acknowledge support from nsf career ast-0645427 . ipf is supported by the spanish grants esp2007 - 65812-c02 - 02 and aya2010 - 21697-c05 - 04 . sgd acknowledges a partial support from the nsf grant ast-0909182 . gdz and mn acknowledge support from asi / inaf agreement i/072/09/0 ( `` _ _ planck _ _ lfi activity of phase e2 '' ) and from miur through the prin 2009 . some of the data presented herein were obtained at the w.m . keck observatory , which is operated as a scientific partnership among the california institute of technology , the university of california and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w.m . keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . herschel_-atlas is a project with _ herschel _ , which is an esa space observatory with science instruments provided by european - led principal investigator consortia and with important participation from nasa . the h - atlas website is http://www.h - atlas.org/. the us participants acknowledge support from the nasa _ herschel_science center / jpl . partly based on observations obtained with _ planck _ ( http://www.esa.int/planck ) , an esa science mission with instruments and contributions directly funded by esa member states , nasa , and canada . support for carma construction was derived from the gordon and betty moore foundation , the kenneth t. and eileen l. norris foundation , the james s. mcdonnell foundation , the associates of the california institute of technology , the university of chicago , the states of california , illinois , and maryland , and the national science foundation . ongoing carma development and operations are supported by the national science foundation under a cooperative agreement , and by the carma partner universities .

this paper studies batyrev s notion of primitive collection . we use primitive collections to characterize the nef cone of a quasi - projective toric variety whose fan has convex support , a result stated without proof by batyrev in the smooth projective case . when the fan is non - simplicial , we modify the definition of primitive collection and explain how our definition relates to primitive collections of simplicial subdivisons . the paper ends with an open problem .

0808.1836

let @xmath0 be the the toric variety of a fan @xmath1 . when @xmath0 is smooth and projective , batyrev @xcite defines a collection @xmath2 of 1-dimensional cones of @xmath1 to be a _ primitive collection _ provided it does not span a cone of @xmath1 but every proper subset does . each primitive collection gives a _ primitive inequality _ , and one of the nice results of @xcite states that the nef cone of @xmath0 is defined by the primitive inequalities . for a proof , batyrev cited the work of oda and park @xcite and reid @xcite , without giving details . the survey article @xcite by the first author notes that batyrev s theorem applies to simplicial projective toric varieties . casagrande @xcite and sato @xcite explain how primitive collections relate to reid s paper @xcite , and kresch @xcite gives a proof in the smooth case . however , a complete proof of batyrev s result in the simplicial case has never appeared in print . in this paper , we give two proofs of batyrev s theorem , one based on @xcite and the other on @xcite . we also extend the definition of primitive collection to the non - simplicial case and show that primitive collections still have the required properties . our results apply to all quasi - projective toric varieties whose fans have convex support of maximal dimension . we use standard notation and terminology for toric varieties . let @xmath3 and @xmath4 be dual lattices of rank @xmath5 with associated real vector spaces @xmath6 and @xmath7 . let @xmath8 be a toric variety of a fan @xmath1 in @xmath9 . we always assume that the support @xmath10 of @xmath1 is convex of dimension @xmath5 . hence all maximal cones have dimension @xmath5 . given @xmath1 , @xmath11 denotes the set of @xmath12-dimensional cones of @xmath1 , and @xmath13 is the subset of @xmath11 consisting of @xmath12-dimensional cones not lying on the boundary of @xmath10 . an _ interior wall _ is an element of @xmath14 . we use the convention that @xmath15 will denote both an element of @xmath16 and its primitive generator in @xmath3 . the torus - invariant divisor associated to @xmath17 is denoted @xmath18 . also recall that a piecewise - linear function @xmath19 can be represented by giving @xmath20 for each @xmath21 , i.e. , @xmath22 if @xmath23 . we define @xmath24 as the vector space of all piecewise - linear functions on @xmath1 . the function @xmath19 is well - defined in @xmath24 if and only if the following statement holds : if @xmath25 is an interior wall and @xmath26 , @xmath27 are the @xmath5-dimensional cones on each side of @xmath25 , then @xmath28 . the support function @xmath19 of a torus - invariant cartier divisor @xmath29 satisfies @xmath30 . note the absence of minus signs . for us , @xmath19 is _ convex _ if and only @xmath31 for all @xmath32 . we also define @xmath33 to be the cone consisting of all convex piecewise - linear functions on @xmath1 . a function @xmath34 is _ strictly convex _ when @xmath35 for all @xmath32 not lying in the same cone of @xmath1 . the toric variety @xmath0 is quasi - projective if and only if there exists a strictly convex @xmath34 . when this happens , the interior of @xmath36 is nonempty and consists of all strictly convex piecewise - linear functions in @xmath24 . in section 1 we give a new definition of primitive collection and state our main theorem . we also recall the nef and mori cones and review the description of the mori cone in terms of the wall relations coming from interior walls . in section 2 we prove batyrev s theorem in the simplicial case , and then in section 3 we treat the non - simplicial case . this section also studies how primitive collections for @xmath1 relate to primitive collections for a simplicial subdivision @xmath37 of @xmath1 . the final section of the paper explores an open problem dealing with the quasi - projective hypothesis . the nef cone @xmath38 of @xmath0 is the quotient of the cone @xmath39 by all linear functions on @xmath1 . thus @xmath40 here is the central definition of our paper . [ pcdef ] a subset @xmath41 is called a _ primitive collection _ for @xmath1 if @xmath42 is not contained in a single cone of @xmath1 but every proper subset is . in the smooth projective case , batyrev defined @xmath43 to be a primitive collection when @xmath44 does not generate a cone of @xmath1 but every proper subset does . when @xmath1 is smooth or more generally simplicial , this is clearly equivalent to definition [ pcdef ] . let @xmath45 be a primitive collection . we say that @xmath34 satisfies the _ primitive inequality for _ @xmath46 if @xmath47 if @xmath34 is convex , i.e. , @xmath48 , then @xmath19 clearly satisfies the primitive inequality for every primitive collection . in other words , @xmath49 the main result of our paper is that the inclusion is in fact an equality , i.e. , the nef cone is defined by the primitive inequalities . here is the precise statement . [ mainthm ] let @xmath0 be a quasi - projective toric variety coming from the fan @xmath1 in @xmath50 . if @xmath10 is convex of dimension @xmath5 , then @xmath51 section 2 will prove theorem [ mainthm ] when @xmath0 is simplicial and section 3 will treat the non - simplicial case . the proof of the main theorem will use extremal rays . hence we need recall the mori cone of a toric variety . although this material is well - known to experts , we include many details since the results we need do not appear explicitly in the literature . we begin with the exact sequence @xmath52 which dualizes to @xmath53 where @xmath54 denotes the dual of @xmath24 and @xmath55 is the dual of @xmath56 . the cone @xmath57 contains the image of @xmath58 and hence dualizes to a cone @xmath59 we call @xmath60 the _ mori cone _ of @xmath0 . when @xmath0 is quasi - projective , @xmath38 has maximal dimension in @xmath56 , so that the mori cone @xmath61 is strongly convex . the unique minimal generators of the mori cone are called _ extremal rays_. we now review the combinatorial description of @xmath60 in terms of the interior walls of @xmath1 . the basic observation is that relations among elements of @xmath16 give elements of @xmath55 . the map @xmath62 gives a commutative diagram @xmath63 & m_{\mathbb{r}}\ar[r ] & { \mathbb{r}}^{\fan(1 ) } \ar[r ] & a_{n-1}(x)_{\mathbb{r}}\ar[r ] & 0\\ 0 \ar[r ] & m_{\mathbb{r}}\ar[r ] \ar[u ] & { \mathrm{pl}}(\fan ) \ar[r ] \ar[u ] & { \mathrm{pic}}(x)_{\mathbb{r}}\ar[r ] \ar[u ] & 0 } \ ] ] where @xmath64 is the chow group of @xmath65-cycles modulo rational equivalence . this dualizes to @xmath66 & a_{n-1}(x)_{\mathbb{r}}^ * \ar[r ] \ar[d ] & { \mathbb{r}}^{\fan(1 ) * } \ar[r ] \ar[d ] & n_{\mathbb{r}}\ar[r ] \ar[d ] & 0\\ 0 \ar[r ] & n_1(x ) \ar[r ] & { \mathrm{pl}}(\fan)^ * \ar[r ] & n_{\mathbb{r}}\ar[r ] \ar[u ] & 0 } \ ] ] in the top row , the map @xmath67 sends the standard basis element @xmath68 to @xmath69 . thus @xmath70 can be interpreted as all linear relations among the @xmath71 , and the surjective map @xmath72 shows that all elements of @xmath55 come from linear relations among the @xmath73 . interior walls of @xmath1 give the following linear relations . given an interior wall @xmath25 , let @xmath26 and @xmath27 be the @xmath5-dimensional cones on each side of @xmath25 , i.e. , @xmath74 . pick @xmath75 linearly independent vectors @xmath76 in @xmath77 and pick vectors @xmath78 , @xmath79 . then there is a nontrivial relation @xmath80 where the final condition holds since @xmath81 and @xmath82 lie on opposite sides of the wall . hence the coefficients @xmath83 are unique up to multiplication by a positive constant . let @xmath84 have components @xmath85 for @xmath86 and @xmath87 otherwise . using the above diagram , we see that @xmath88 . [ ltau_def ] depending on the context , we use the term _ wall relation _ to refer to the equation , the vector @xmath89 , or its image @xmath90 . notice that in the non - simplicial case , a given wall can have many choices for the @xmath91 in the wall relation , while in the simplicial case , @xmath91 are uniquely determined by the wall . [ thm_walls_mori ] let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 . 1 . for @xmath92 , the different choices of the wall relation all give the same @xmath93 up to a positive constant . the mori cone in @xmath55 is given by @xmath94 this is the kleiman - nakai criterion from oda and park ( * ? ? ? * thm.2.3 ) . we give the details since their definition of @xmath95 differs from ours . let @xmath96 and pick a wall relation @xmath97 , @xmath98 as in . rescaling by a positive constant , we may assume @xmath99 , so that @xmath100 then @xmath101 is the linear functional on @xmath24 given by @xmath102 since @xmath103 for @xmath104 and @xmath105 since @xmath106 . however , implies that @xmath107 hence the above formula for @xmath108 simplifies to @xmath109 note that @xmath110 . since @xmath111 and @xmath112 is 1-dimensional , it follows that up to a positive constant , @xmath113 for _ any _ @xmath114 . this proves the first part of the theorem and also shows that our @xmath95 agrees with the @xmath95 appearing in the statement of ( * ? ? ? then the second part follows immediately from ( * ? ? ? * thm.2.3 ) . let @xmath115 be a primitive collection for @xmath1 . then @xmath116 lies in some unique minimal cone @xmath26 of @xmath117 . pick a subset @xmath118 satisfying @xmath119 the equation gives the vector @xmath120 defined by @xmath121 from , it follows that @xmath122 . depending on the context , we use the term _ primitive relation _ to refer to the equation , the vector @xmath123 , or its image @xmath124 under the map @xmath72 . when @xmath0 is smooth , the case @xmath125 can not occur ( * ? ? ? 3.1 ) , but happens frequently in the simplicial case . in this case , we bound @xmath126 as follows . [ poscoeffs ] the coefficients in the primitive relation satisfy @xmath127 when @xmath125 . hence @xmath128 is determined by the positive entries of @xmath129 . suppose @xmath130 and @xmath131 . subtracting @xmath132 from each side of gives @xmath133 since @xmath134 lie in a cone of @xmath117 , this equation implies that @xmath134 and @xmath135 lie in the same cone of @xmath117 . adding in @xmath132 shows that @xmath128 lies in a cone of @xmath117 , which is impossible . the minimal cone @xmath26 containing @xmath136 need not be simplicial , so there may be many subsets @xmath137 satisfying . but when there are many choices for @xmath129 , they all give the same element @xmath124 , as shown by the following proposition . [ primrel_unique ] let @xmath138 be a primitive collection for @xmath1 and let @xmath139 be defined as above . then : 1 . when regarded as an element of @xmath54 , @xmath140 is the linear functional on @xmath24 defined by @xmath141 2 . @xmath142 . let @xmath143 be the smallest cone containing @xmath116 . since @xmath19 is linear on @xmath26 and @xmath118 , we obtain @xmath144 this proves the first part of the proposition , and then the second part follows immediately from the first part and since @xmath145 . we can formulate theorem [ mainthm ] in terms of primitive relations as follows . [ equivalent ] let @xmath0 be the toric variety of the fan @xmath1 in @xmath50 . if @xmath10 is convex of dimension @xmath5 , then the following are equivalent : 1 . @xmath146 for all primitive collections @xmath147 . @xmath148 , where the sum is over all primitive collections for @xmath1 . this follows easily from proposition [ primrel_unique ] and @xmath145 . the strategy for proving theorem [ mainthm ] in the simplicial case is the observation , implicit in @xcite , that every minimial generator of @xmath60 is a primitive relation @xmath140 for some primitive collection @xmath128 . then theorem [ mainthm ] for simplicial fans follows immediately from proposition [ equivalent ] . we give the details of this argument in section 2 . an interior wall @xmath25 gives a complete torus - invariant curve @xmath149 in @xmath0 . let @xmath150 denote the linear functional that sends an @xmath151-cartier divisor @xmath29 to the intersection product @xmath152 . thus @xmath153 up to a positive multiple , this gives the same class as the wall relation @xmath90 from definition [ ltau_def ] . although this result is well - known to experts , we include a proof for completeness . [ ctaultau ] let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 . for each @xmath92 , @xmath154 is a positive multiple of the wall relation @xmath95 appearing in theorem [ thm_walls_mori ] . when @xmath1 is simplicial , we have @xmath155 and as in the proof of theorem [ thm_walls_mori ] , we have the wall relation @xmath156 since @xmath1 is simplicial , the divisors @xmath157 corresponding to @xmath73 are @xmath158-cartier , so that @xmath159 is defined . by ( * ( 2.7 ) ) , we have @xmath160 the proof in @xcite assumes @xmath1 is simplicial and complete and @xmath25 is any wall ; the argument applies without change when @xmath1 is simplicial and @xmath25 is an interior wall . for the general case , we use the well - known fact that @xmath1 has a simplicial refinement @xmath37 such that @xmath161 ( see corollary [ cor_simpl_ref_general ] and remark [ rem_simpl_ref_general ] ) . if @xmath162 is the toric variety of @xmath37 , then we have a proper map @xmath163 let @xmath164 be an interior wall of @xmath37 contained in @xmath25 , and let @xmath165 and @xmath166 be the corresponding curves in @xmath162 and @xmath0 . the induced map @xmath167 has degree @xmath168 $ ] , which implies that @xmath169 . let @xmath29 be a cartier divisor on @xmath0 . by the projection formula , @xmath170 if we write @xmath171 on @xmath0 , then @xmath172 on @xmath162 since @xmath161 . if @xmath173 is the wall relation of @xmath164 coming from , then up to a positive constant , @xmath174 since @xmath37 is simplicial . however , the wall relation for @xmath164 is one of the ( possibly many ) wall relations for @xmath25 , i.e. , @xmath175 is one of the possible choices for @xmath176 . then the formula @xmath177 shows ( again up to a positive constant ) that the class of @xmath166 in @xmath55 is the image of @xmath178 in @xmath55 . in other words , @xmath179 equals @xmath95 up to a positive constant , as claimed . we conclude our discussion of the mori cone explaining how our definition of @xmath60 relates to the standard geometric approach . since @xmath0 need not be complete , we work in the relative context . let @xmath180 be the affine toric variety of the strongly convex cone @xmath181 . this gives a proper toric morphism @xmath182 . following @xcite or @xcite , the mori cone of @xmath182 is defined as follows . let @xmath183 be the free group generated by irreducible curves in @xmath0 that map to a point in @xmath180 . then we have a natural pairing @xmath184 by restricting to torus - invariant curves coming from interior walls , one sees easily that this pairing is nondegenerate with respect to @xmath185 , i.e. , if a cartier divisor satisfies @xmath186 for all torus - invariant curves @xmath187 coming from interior walls , then @xmath188 = 0 $ ] in @xmath185 . it follows that the above pairing induces a perfect pairing @xmath189 thus @xmath190 is what we call @xmath55 . dropping the @xmath180 from the notation is reasonable since in our situation @xmath180 is determined functorially by @xmath0 . finally , @xmath191 is the cone generated by irreducible curves in @xmath0 that map to a point in @xmath180 , and the _ mori cone _ is its closure @xmath192 in @xmath55 . then theorem [ thm_walls_mori ] and proposition [ ctaultau ] easily imply that @xmath193 where @xmath194 is the class of the torus - invariant curve @xmath166 associated to @xmath25 . this is the _ relative toric cone theorem_. the relative toric cone theorem is stated for toric morphisms @xmath195 by matsuki ( * ? ? ? . 14 - 1 - 4 ) or reid ( * ? ? ? * ( 1.7 ) ) . as pointed out by fujino and sato in ( 4.3 ) , this fails when the torus action on @xmath137 has no fixed points . they give the easy example of the projection map @xmath196 . the fibers of this map are never torus - invariant , so that torus - invariant curves can not generate @xmath197 . fortunately , the relative toric cone theorem holds for our map @xmath182 because @xmath198 is convex of dimension @xmath5 . a nice feature of the simplicial case is that @xmath199 . hence an interior wall @xmath25 gives @xmath200 , and a primitive collection @xmath128 gives @xmath201 . let @xmath25 be an interior wall of a simplicial fan @xmath1 with @xmath155 , and let @xmath81 and @xmath82 be the generators that are needed to span the cones on each side of the wall . the uniquely determined wall relation is @xmath202 by the discussion following . [ reid_primcoll ] let @xmath1 be a quasi - projective simplicial fan with convex support of dimension @xmath5 . let @xmath25 be an extremal wall , meaning that the wall relation generates an extremal ray @xmath203 . then : 1 . @xmath204 is a primitive collection for @xmath1 . 2 . in @xmath205 , the primitive relation @xmath129 of @xmath128 and wall relation @xmath178 of @xmath25 are equal up to a positive constant . we give two proofs . the first is based on ( * ? ? ? 2.4 ) , which assumes that @xmath0 is smooth and complete . we adapt the argument to the simplicial case as follows . we first make a useful observation about convex support functions . if @xmath19 is convex and @xmath206 , we can change @xmath19 by a linear function so that @xmath207 now take an extremal wall @xmath25 with wall relation . consider the set @xmath208 we will prove that @xmath128 is a primitive collection whose primitive relation @xmath140 equals @xmath95 up to a positive constant . recall that @xmath201 and @xmath209 since @xmath117 is simplicial . we first prove by contradiction that @xmath210 for all @xmath143 . suppose @xmath211 and take a strictly convex support function @xmath19 . we may assume that @xmath19 is of the form . since @xmath212 for @xmath213 , we have @xmath214 however , @xmath215 by , and @xmath216 implies @xmath217 for @xmath218 . it follows that @xmath219 , which is impossible since @xmath19 is strictly convex and @xmath220 . thus no cone of @xmath117 contains all rays in @xmath128 . it follows that some subset @xmath221 is a primitive collection . this gives the primitive relation @xmath222 , and we also have @xmath223 . let @xmath224 where @xmath225 . we claim that if @xmath226 is sufficiently small , then @xmath227 to prove this , suppose that @xmath228 and @xmath229 . then the definition of @xmath230 forces @xmath231 , so that @xmath232 by lemma [ poscoeffs ] . combined with @xmath221 , we see that @xmath233 by the definition of @xmath128 . but we can clearly choose @xmath226 sufficiently small so that @xmath234 this inequality and the definition of @xmath230 imply @xmath235 , a contradiction . we next claim that @xmath236 . by , we have @xmath237 where the second inclusion follows from and the definition of @xmath178 . now let @xmath19 be convex . by with @xmath238 , we may assume @xmath239 then @xmath240 where the final inequality follows since @xmath215 and @xmath228 can only happen only when @xmath241 . this proves that @xmath236 . since @xmath242 by proposition [ primrel_unique ] , the equation @xmath243 expresses @xmath209 as a sum of elements of @xmath60 . but @xmath95 is extremal . this forces @xmath244 and @xmath230 to lie in the ray generated by @xmath178 . since @xmath244 is nonzero , @xmath178 is a positive multiple of @xmath244 . in particular , they have the same positive entries . then @xmath245 follows from the definition of @xmath128 and lemma [ poscoeffs ] . this completes the first proof . the second proof begins with the extremal wall relation . for @xmath246 set @xmath247 in @xcite , reid proves that @xmath248 ( see the lemma on @xcite ) and @xmath249 ( see and ( * ? ? ? reid assumes that @xmath1 is simplicial and complete . his proofs generalize to our situation without change see @xcite . let @xmath250 , so that @xmath251 . in order to prove that @xmath128 is a primitive collection , we first show that @xmath252 is not a cone in @xmath1 . so assume @xmath253 and consider the relation @xmath254 where the coefficients on the left are positive . then @xmath255 lies in the relative interior of the cone @xmath256 , but @xmath257 lies in the wall @xmath258 since @xmath259 and @xmath260 for @xmath261 . it follows that @xmath262 , which is a contradiction since @xmath81 and @xmath82 do not lie in the wall . now we show that every proper subset of @xmath128 generates a cone of @xmath1 . let @xmath263 be any proper subset of @xmath264 . then @xmath265 is a face of @xmath266 for any @xmath267 . but @xmath268 by . hence @xmath269 is a primitive collection . finally , we consider the primitive relation of @xmath128 , which can be written @xmath270 where @xmath26 is the minimal cone of @xmath1 containing @xmath271 . since @xmath272 and @xmath273 by and , it follows that is a relation among @xmath91 . this relation is unique up to a nonzero constant since @xmath117 is simplicial . thus @xmath129 is a nonzero multiple of @xmath178 , necessarily positive by lemma [ poscoeffs ] and the definition of @xmath128 . batyrev clearly knew this proposition , though it is not stated explicitly in @xcite . proposition [ reid_primcoll ] is closely related to theorem 1.5 in casagrande s paper @xcite and appears implicitly in the remarks preceeding proposition 2.2 in sato s paper @xcite . we can now prove the simplicial case of our main theorem . [ theorem_mori_primitive ] let @xmath1 be a simplicial quasi - projective fan in @xmath50 with convex support of dimension @xmath5 . then the cone @xmath36 is defined by the primitive inequalities , i.e. , @xmath274 by proposition [ equivalent ] it suffices to show that the primitive relations @xmath140 generate the mori cone . we already know that @xmath275 ( proposition [ primrel_unique ] ) and that @xmath60 is generated by the extremal wall relations @xmath95 ( theorem [ thm_walls_mori ] ) . furthermore , @xmath60 is generated by extremal wall relations since @xmath0 is quasi - projective . hence it suffices to show that every extremal wall relation is a primitive relation . this is what we proved in corollary [ reid_primcoll ] , and the theorem follows . here is an example of theorem [ theorem_mori_primitive ] . [ ex_simplicial ] figure [ figure : counterex_def_h ] shows the complete simplicial fan @xmath1 in @xmath276 with five minimal cone generators : @xmath277 and six maximal cones : @xmath278 ( 0,0),title="fig : " ] # 1#2#3#4#5 @font ( 945,1527)(1801,-2098 ) ( 2746,-691)(0,0)[lb ] ( 2476,-2041)(0,0)[lb ] ( 2656,-1006)(0,0)[lb ] ( 1801,-1006)(0,0)[lb ] ( 1936,-736)(0,0)[lb ] the primitive collections for this fan are : @xmath279 so that @xmath34 is convex if and only if @xmath280 to get a more concrete characterization , we use the associated primitive relations : @xmath281 let @xmath282 be the torus - invariant divisor associated to @xmath283 . then the divisor @xmath284 corresponds to the support function @xmath19 satisfying @xmath285 . it follows that @xmath29 is nef if and only if @xmath286 in contrast , @xmath1 has 9 walls , so theorem [ thm_walls_mori ] describes @xmath60 using 9 generators , corresponding to 9 wall inequalities defining @xmath57 . fortunately , these can be simplified considerably . we denote by @xmath287 the wall that is spanned by @xmath283 and @xmath288 . by abuse of notation we will also call @xmath287 the corresponding class in @xmath60 . one can compute that the 9 walls fall into three groups : @xmath289 hence @xmath290 and @xmath291 give the extremal rays of the mori cone , while @xmath292 do not give an extremal ray . one can check that the primitive collection @xmath293 generates the same ray as @xmath290 and @xmath294 generates the same ray as @xmath295 . example [ ex_simplicial ] is nice because there were few primitive collections . however , there are examples where the primitive collections vastly outnumber the interior walls . [ polygon_example ] let @xmath1 be a complete fan in @xmath296 with @xmath297 minimal generators , say @xmath298 , arranged counterclockwise around the origin . then there are @xmath299 walls , all interior . one easily checks that the primitive collections are given by @xmath300 for @xmath301 and @xmath302 not adjacent . hence the fan @xmath1 has @xmath303 primitive collections . this is greater than the number of walls provided @xmath304 . in order to prove our main theorem in the non - simplicial case , we need to consider simplicial refinements . here we present a general theorem about the existence of simplicial refinements with special properties . [ thrm_simpl_ref_general ] let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 and fix @xmath305 such that @xmath306 is linearly independent for all @xmath143 . then : 1 . there exists a simplicial refinement @xmath37 of @xmath1 satisfying @xmath161 such that @xmath307 generates a cone in @xmath37 for all @xmath143 . if in addition @xmath1 is quasi - projective , then the refinement @xmath37 in part ( 1 ) can be chosen to be quasi - projective . to create @xmath37 , assign a weight @xmath308 to each @xmath73 as follows : * for @xmath309 , set @xmath310 . * for @xmath311 , pick @xmath312 generic . the exact meaning of generic will be explained in the course of the proof . for each cone @xmath21 , fix @xmath313 in the interior of the dual cone @xmath314 . the affine hyperplane @xmath315 defined by @xmath316 intersects @xmath26 in a convex polytope @xmath317 with vertices @xmath318 , where @xmath319 for @xmath320 . the idea is to triangulate @xmath321 using a variant of the method used in example 1.1 of @xcite , p. 215 . consider @xmath322 since the vectors @xmath323 lie on the 1-dimensional rays of @xmath26 , it is easy to see that the vertices of @xmath324 consist of the origin and the points @xmath325 for @xmath326 . furthermore , the faces of @xmath324 not containing the origin project to a polyhedral subdivision of @xmath321 . projecting from the origin in @xmath327 , we get a refinement @xmath328 of @xmath26 that satisfies @xmath329 . figure [ figure : simplicial_proof_weights ] shows two 3-dimensional cones @xmath26 , each with a set @xmath330 and a choice of weights giving the polytope @xmath324 inside @xmath26 . ( 0,0 ) , @xmath330 , and a choice of weights.,title="fig : " ] # 1#2#3#4#5 @font ( 3914,3619)(1059,-3008 ) ( 1756 , 29)(0,0)[lb ] ( 1351,-1006)(0,0)[lb ] ( 1296,479)(0,0)[lb ] ( 3736 , 29)(0,0)[lb ] ( 3421,-1006)(0,0)[lb ] ( 4231,-1096)(0,0)[lb ] ( 3296,479)(0,0)[lb ] we claim that the fans @xmath328 have the following three properties : * @xmath331 generates a cone of @xmath328 for all @xmath143 . * if a face @xmath25 lies in maximal cones @xmath26 and @xmath27 , then @xmath328 and @xmath332 induce the same refinement of @xmath25 . * if the @xmath308 are sufficiently generic for @xmath333 , then @xmath328 is simplicial for all @xmath143 . assuming a c , the set @xmath334 is a fan that refines @xmath1 by b and satisfies @xmath335 furthermore , @xmath37 is simplicial by c. finally , given @xmath143 , @xmath331 generates a cone of @xmath37 by a. hence the proof of part ( 1 ) of the theorem will be complete once we prove a c . _ proof of a_. consider the hyperplane @xmath315 . since @xmath336 for all @xmath15 , the polytope @xmath324 lies on the side of @xmath337 containing the origin , and the intersection @xmath338 is clearly the convex hull of the points @xmath318 for @xmath339 by the choice of the weights @xmath308 . it follows that @xmath330 generates a cone of @xmath328 . _ proof of b_. suppose @xmath25 is a face of @xmath26 . set @xmath340 and observe that @xmath341 this tells us that the refinement of @xmath25 induced by @xmath342 is determined entirely by the vectors @xmath323 for @xmath343 . since we are working with fixed weights in this part of the proof , the only choice involved is this refinement of @xmath25 comes from the vectors @xmath318 for @xmath344 . recall that @xmath345 , where @xmath337 is defined by @xmath346 and @xmath347 . now suppose that @xmath25 is a face of @xmath27 for a maximal cone @xmath348 . the refinement of @xmath25 induced by @xmath332 is determined by @xmath349 where @xmath350 determines the affine hyperplane @xmath351 whose intersection with @xmath343 determines @xmath352 . now consider the map @xmath353 defined as follows : * if @xmath354 , then @xmath355 . * if @xmath354 and @xmath356 , then @xmath357 . the map @xmath358 sends line segments to line segments and is homogeneous of degree @xmath359 for non - negative scalars . furthermore , @xmath360 , and thus @xmath361 for all @xmath343 . it follows that @xmath362 this map preserves faces , so that @xmath363 and @xmath364 induce the same refinement of @xmath25 . this completes the proof of b. _ proof of c_. we may assume @xmath365 . for @xmath366 , write @xmath367 when we have @xmath368 , , we write instead @xmath369 and we set @xmath370 . now suppose that @xmath342 is non - simplicial for some choice of weights @xmath308 . this implies that @xmath324 has a face @xmath371 of dimension @xmath75 not containing the origin that is not an @xmath65-simplex . it follows that @xmath371 has at least @xmath372 vertices . pick @xmath372 vertices of @xmath371 as follows . we first pick those vertices of @xmath371 of the form @xmath373 for @xmath374 . there are at most @xmath5 such vertices since @xmath307 is linearly independent . since @xmath375 , we can extend them to affinely independent vertices @xmath376 . since @xmath371 is non - simplicial , we can pick one more , @xmath377 . now consider the @xmath378 matrix @xmath379 by construction , the vectors @xmath380 are affinely independent , but once we add the last vector , @xmath381 are affinely dependent since they lie in a @xmath65-dimensional face . it follows that this matrix has rank exactly @xmath5 . since the weights @xmath382 are nonzero , the same is true for the matrix @xmath383 the determinant of @xmath384 must vanish . the resulting linear equation in the @xmath385 gives a necessary condition for @xmath342 to be non - simplicial . furthermore , our careful choice of @xmath386 guarantees that @xmath387 actually appears in the determinant and that @xmath388 for some @xmath389 . since there are only finitely many maximal cones in @xmath1 and for each @xmath26 we have a fixed choice of the @xmath390 ( determined by the fixed choice of @xmath347 ) , we get a finite system of non - trivial linear equations in the @xmath391 for @xmath389 that give necessary conditions for @xmath342 to be non - simplicial . if we pick the weights @xmath308 to avoid these finitely many subspaces , the resulting subdivisions @xmath342 will be all simplicial . this completes the proof of c , and part ( 1 ) follows . turning to part ( 2 ) , we assume that @xmath1 is quasi - projective . it suffices to find a simplicial refinement @xmath37 of @xmath1 satisfying @xmath161 such that the induced map @xmath392 is projective . the latter happens when @xmath37 has a piecewise - linear function @xmath393 which is strictly convex relative to @xmath1 , meaning that for all @xmath143 , @xmath394 is strictly convex with respect to the subfan @xmath395 ( see ( * ) on page 27 and theorem 10 on pages 3132 of @xcite ) . since @xmath1 is quasi - projective , we can find @xmath48 which is strictly convex . we first modify @xmath19 so that it takes positive values on @xmath16 . to see why this is possible , consider the cone @xmath396 since @xmath19 is strictly convex for @xmath1 , it follows that @xmath397 is a strongly convex cone with minimal generators given by @xmath398 for @xmath71 . hence we can find @xmath399 such that @xmath400 replacing @xmath19 with @xmath401 , we may assume @xmath402 for all @xmath71 , as claimed . the proof of part ( 1 ) used a hyperplane @xmath315 for each @xmath21 . more precisely , we picked @xmath403 such that @xmath404 , and then for @xmath326 , @xmath318 was the unique vector in @xmath405 satisfying @xmath406 . using @xmath19 , we get a consistent set of hyperplanes since @xmath407 is linear , i.e. , @xmath408 for some @xmath409 . our hypothesis that @xmath402 for all @xmath71 guarantees that @xmath313 is in the interior of @xmath410 . hence we can use these @xmath313 s to give the hyperplanes @xmath337 . then the point @xmath318 is the unique vector in @xmath405 satisfying @xmath411 . now pick generic weights @xmath308 for @xmath311 . then the properties a c are satisfied ( note b is now trivial because of our consistent choice of the @xmath337 ) . thus we get a simiplicial refinement @xmath37 of @xmath1 that satisifes part ( 1 ) of the theorem . now define @xmath412 by setting @xmath413 and extending linearly on each cone @xmath414 . this gives a well - defined function in @xmath415 since @xmath37 is simplicial . assuming @xmath19 is rational , we can also assume that the @xmath308 are rational . hence we can assume that @xmath358 is rational as well . we claim that @xmath358 is strictly convex with respect to @xmath416 for each @xmath143 . to see this , first observe that @xmath417 since @xmath411 . it follows that inside @xmath26 , the inequality @xmath418 defines @xmath322 then the convexity of @xmath324 implies that if @xmath419 , then @xmath420 with equality if and only if @xmath421 lie in the same cone of @xmath342 . to prove this , we may assume @xmath422 , so that @xmath423 then @xmath424 , so that @xmath425 it is equally easy to show that equality occurs exactly when @xmath421 lie in the same cone of @xmath342 . hence @xmath358 has the required properties , which completes the proof of part ( 2 ) of the theorem . [ cor_simpl_ref_general ] if @xmath1 is a fan in @xmath50 with convex support of dimension @xmath5 , then there exists a simplicial refinement @xmath37 with the same 1-dimensional generators . furthermore , if @xmath1 is quasi - projective , then we can assume that @xmath37 is also quasi - projective . apply theorem [ thrm_simpl_ref_general ] with @xmath426 . [ rem_simpl_ref_general ] this corollary guarantees the existence of simplicial refinements that introduce no new generators and preserve quasi - projectivity . this result has other proofs , including fujino @xcite ( via the toric mori program ) and thompson @xcite ( via stellar subdivision ) . we can now prove the non - simplicial case of our main theorem . [ theorem_mori_nonsimp ] let @xmath1 be a non - simplicial quasi - projective fan in @xmath9 with convex support of dimension @xmath5 . then the cone @xmath36 is defined by the primitive inequalities , i.e. , @xmath427 by corollary [ cor_simpl_ref_general ] , @xmath1 has a quasi - projective simplicial refinement @xmath37 satisfying @xmath335 . then observe that @xmath428 and that @xmath429 since @xmath37 is simplicial and quasi - projective . hence @xmath430 we divide primitive collections @xmath431 for @xmath37 into two types : @xmath432 note that if @xmath34 , then @xmath433 when @xmath138 is a type a primitive collection for @xmath37 . hence these can be omitted in , so that @xmath434 however , a type b primitive collection @xmath128 for @xmath37 is a primitive collection for @xmath1 . this is easy to prove . first , @xmath128 is not contained in any cone of @xmath1 by the definition of type b , and second , every proper subset of @xmath128 is contained in a cone of @xmath37 and hence lies in a cone of @xmath1 since @xmath37 refines @xmath1 . it follows that @xmath435 is a primitive inequality for @xmath1 whenver @xmath436 is a type b primitive collection for @xmath37 . hence shows that a subset of the primitive inequalities for @xmath1 define @xmath36 inside @xmath24 . using the inclusion , the theorem now follows immediately . here is an example to illustrate theorem [ theorem_mori_nonsimp ] and its proof . [ nonsimpl_ex ] figure [ figure : counterex_def ] shows the complete non - simplicial fan @xmath1 in @xmath276 with five minimal generators : @xmath437 and five maximal cones : @xmath438 ( 0,0),title="fig : " ] # 1#2#3#4#5 @font ( 945,1527)(1801,-2098 ) ( 2746,-691)(0,0)[lb ] ( 2476,-2041)(0,0)[lb ] ( 2656,-1006)(0,0)[lb ] ( 1801,-1006)(0,0)[lb ] ( 1936,-736)(0,0)[lb ] the primitive collections for this fan are : @xmath439 a first observation is that if we used batyrev s definition of primitive collection in this case , we would want every proper subset of @xmath293 and @xmath294 to generate a cone of @xmath1 . this clearly is nt true , and in fact this example has _ no _ primitive collections if we use batyrev s definition . this explains why definition [ pcdef ] is the correct definition in the non - simplicial case . theorem [ theorem_mori_nonsimp ] states that @xmath440 is defined by the primitive inequalities coming from the primitive collections @xmath293 and @xmath294 . however , the proof of the theorem shows that we need only one . to see why , consider the simplicial refinement @xmath37 of @xmath1 given by subdividing non - simplicial cone @xmath441 along @xmath442 . this gives the fan pictured in example [ ex_simplicial ] . the fan @xmath37 has the same generators @xmath443 as @xmath1 , and the primitive collections for @xmath37 are @xmath444 one easily checks that @xmath445 is of type a and @xmath294 is of type b and hence is a primitive collection for @xmath1 . by , @xmath36 is defined by @xmath294 , so that @xmath34 is convex if and only if @xmath446 it is interesting to note that the type a primitive collection @xmath447 also plays an important role . the primitive relation of @xmath445 is @xmath448 now take @xmath34 . as noted in the proof of theorem [ theorem_mori_nonsimp ] , this type a primitive collection gives the _ equality _ @xmath449 which by the above primitive relation implies @xmath450 it is easy to see that this equality defines @xmath24 inside of @xmath415 . in other words , @xmath451 lies in @xmath24 if and only if it satisfies coming from the type a primitive collection for @xmath37 . if we turn our attention to the other primitive collection @xmath452 for @xmath1 , then one easily sees that @xmath453 is convex if and only if @xmath454 this follows by considering the other simplicial refinement of @xmath1 obtained by subdividing @xmath441 along @xmath455 . example [ nonsimpl_ex ] has some interesting features : * every primitive collection for @xmath1 comes from a type b primitive collection for a simplicial refinement @xmath37 of @xmath1 satisfying @xmath161 . * for each such refinement @xmath37 of @xmath1 , the type a primitive collections for @xmath37 define @xmath456 . we will see below that these properties hold in general . we begin with the following useful property of primitive collections . [ thm_strict ] let @xmath1 be a fan in @xmath50 such that @xmath1 has convex support of dimension @xmath5 . if @xmath128 is a primitive collection for @xmath1 , then every proper subset @xmath457 of @xmath128 is linearly independent . we use induction on @xmath458 . if @xmath459 there is nothing to show . now assume that @xmath460 , @xmath461 , and that every @xmath12-element subset of @xmath457 is linearly independent . we show that @xmath457 is linearly independent by contradiction . hence suppose @xmath457 is linearly dependent . then our induction hypothesis implies that the subspace @xmath462 has dimension @xmath12 . define @xmath463 . we omit the straightforward proof that @xmath464 is a fan in @xmath462 . now fix @xmath465 and let @xmath466 be the minimal cone of @xmath1 containing @xmath467 . notice that @xmath466 does not contain @xmath15 since @xmath128 is a primitive collection . also let @xmath468 be the minimal cone of @xmath1 containing @xmath457 . the cones @xmath469 and @xmath470 are in the fan @xmath464 and @xmath471 since @xmath15 is contained in @xmath472 but not in @xmath473 . therefore , their intersection is at most @xmath474-dimensional . on the other hand , the intersection contains @xmath12 linearly independent vectors @xmath475 which is a contradiction . [ cor_strict ] let @xmath1 be a fan in @xmath50 such that @xmath10 is convex support of dimension @xmath5 . then every primitive collection for @xmath1 has at most @xmath372 elements . this follows immediately from proposition [ thm_strict ] since any maximal proper subset @xmath476 is linearly independent and hence has at most @xmath5 elements . therefore @xmath477 has at most @xmath372 elements . proposition [ thm_strict ] and corollary [ cor_strict ] are trivial in the simplicial case . let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 , and let @xmath37 be a simplicial refinement with @xmath478 . given @xmath143 , let @xmath479 . the following convexity result will be useful . [ lemma_subfab_convex ] let @xmath26 be a non - simplicial cone in @xmath1 and take an interior wall @xmath164 of @xmath328 with @xmath480 , @xmath481 . then @xmath482 is convex . since @xmath483 , @xmath164 divides @xmath26 into two convex subcones @xmath484 with @xmath485 . we may assume @xmath486 . given @xmath487 , it follows easily that the line segment @xmath488 lies in @xmath489 . [ corollary_subfab_convex ] in the situation of lemma [ lemma_subfab_convex ] , let @xmath128 be the two element set @xmath490 thus @xmath128 consists of the generators of @xmath491 not lying in the wall @xmath492 . then @xmath128 is a primitive collection for @xmath37 . first note that @xmath128 is contained in neither @xmath493 nor @xmath494 . since @xmath128 is contained in the convex set @xmath482 , it follows that @xmath128 is contained in no cone of @xmath37 . thus @xmath128 is a primitive collection for @xmath37 since has it only has two elements . as in the proof of theorem [ theorem_mori_nonsimp ] , a primitive collection for @xmath37 has type a when it is contained in a cone of @xmath1 . hence the primitive collection for @xmath37 constructed in corollary [ corollary_subfab_convex ] has type a. the idea is that these two - element primitive collections define @xmath24 inside @xmath415 . let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 and let @xmath37 be a simplicial refinement with @xmath495 . then @xmath496 the inclusion @xmath497 is obvious since elements of @xmath24 are linear on cones of @xmath1 and a type a primitive collection is contained in such a cone . for the opposite inclusion , take @xmath451 such that @xmath498 for all two element type a primitive collections for @xmath37 . for each @xmath499 , there is @xmath500 such that @xmath501 for @xmath104 . it suffices to show that @xmath502 for cones @xmath503 that lie in the same cone @xmath26 of @xmath1 and intersect in a wall @xmath504 . this is the situation of corollary [ corollary_subfab_convex ] , where @xmath505 and @xmath506 is a two element type a primitive collection for @xmath37 . we label the elements of @xmath128 so that @xmath507 and @xmath508 . since @xmath489 is convex by lemma [ lemma_subfab_convex ] , it contains @xmath509 . we may assume @xmath510 without loss of generality . then @xmath511 since @xmath512 , it follows that @xmath502 . this completes the proof . in the fan @xmath1 pictured in figure [ figure : counterex_def ] in example [ nonsimpl_ex ] , we saw that every primitive collection for @xmath1 came from a primitive collection for a simplicial subdivision of @xmath1 . in general , if @xmath37 is a simplicial subdivision of @xmath1 with @xmath161 , we say that a primitive collection @xmath128 for @xmath1 is _ supported _ on @xmath37 if @xmath128 is also a primitive collection for @xmath37 . we now prove that all primitive collections for @xmath1 are supported on such simplicial subdivisions . here is the precise result . [ prim_lemma ] let @xmath1 be a fan in @xmath50 with convex support of dimension @xmath5 and let @xmath128 be a primitive collection for @xmath1 . then there exists a simplicial refinement @xmath37 with @xmath161 such that @xmath128 is a primitive collection for @xmath37 . furthermore , if @xmath1 is quasi - projective , then @xmath37 can be chosen to be quasi - projective . by proposition [ thm_strict ] , every proper subset @xmath128 is linearly independent . in particular , if @xmath143 , then @xmath330 is a proper subset of @xmath128 ( since @xmath128 is a primitive collection ) and hence is linearly independent . thus we can apply theorem [ thrm_simpl_ref_general ] to obtain a simplicial refinement @xmath37 such that @xmath330 generates a cone of @xmath37 for all @xmath143 . the theorem also allows us to assume @xmath37 is quasi - projective whenever @xmath1 is . we claim that @xmath128 is a primitive collection for @xmath37 . first note that if @xmath128 were contained in a cone of @xmath37 , then it would be contained in a cone of @xmath1 , which we know to be false . now let @xmath457 be a proper subset of @xmath128 . then @xmath457 is contained in a cone @xmath513 , so that @xmath514 . since @xmath330 generates a cone of @xmath37 , it follows that @xmath457 is contained in a cone of @xmath37 . hence @xmath128 is a primitive collection for @xmath37 . when @xmath1 is non - simiplicial , it may be impossible to find a _ single _ simplicial refinement @xmath37 such that _ every _ primitive collection for @xmath1 is also primitive for @xmath37 . in figure [ figure : counterex_def ] from example [ nonsimpl_ex ] , we see two primitive collections @xmath515 and @xmath516 , but there is no simplicial refinement @xmath37 of @xmath1 with @xmath161 that supports both @xmath293 and @xmath294 . in this section we explore an open question about primitive collections . in @xcite , casagrande raises the question of whether @xmath36 is defined by primitive inequalities when @xmath1 is not quasi - projective . here is a classic example . [ ex_simplicial_nonqp ] the following example of a non - projective smooth complete fan is taken from fulton @xcite . consider the fan @xmath1 in @xmath276 with seven minimal generators : @xmath517 the cones of @xmath1 are obtained by projecting from the origin through the triangulated polytope shown in figure [ figure : ex_simplicial_nonqp ] . the fan @xmath1 has 15 walls and 10 maximal cones . ( 0,0 ) # 1#2#3#4#5 @font ( 2519,2700)(1149,-2543 ) ( 3668,-1463)(0,0)[lb ] ( 2888,-2498)(0,0)[lb ] ( 3203,-892)(0,0)[lb ] ( 1149,-1703)(0,0)[lb ] ( 2408 , 37)(0,0)[lb ] ( 1651,-938)(0,0)[lb ] ( 2520,-1432)(0,0)[lb ] the seven primitive collections for @xmath1 and their associated primitive relations are : @xmath518 by , a convex function @xmath48 satisfies the primitive inequalities : @xmath519 notice that adding up the third , fourth and sixth inequalities yields an equality , hence we have 3 equalities : @xmath520 to see what this says about the nef cone @xmath38 , note that @xmath521 assume @xmath522 . then the three equalities give @xmath523 . define @xmath524 and @xmath525 . then inequalities imply @xmath526 and @xmath527 . it follows that @xmath38 is contained in the 2-dimensional cone pictured in figure [ figure : ex_simpl_nqp_nef ] . ( 0,0 ) # 1#2#3#4#5 @font ( 2163,2010)(1789,-1873 ) ( 3691,-1658)(0,0)[lb ] ( 1921 , 29)(0,0)[lb ] ( 3541,-458)(0,0)[lb ] since @xmath56 has dimension 4 and @xmath38 has dimension at most two , we see that @xmath0 is non - projective since the nef cone does not have maximal dimension . it is also easy to see that the cone in figure [ figure : ex_simpl_nqp_nef ] actually equals the nef cone @xmath38just show that the generators of this cone are nef . for example , when @xmath528 , note that @xmath1 is a refinement of the complete fan @xmath529 with 1-dimensional generators @xmath530 . the toric variety of @xmath529 is @xmath531 , and the class corresponding to @xmath532 is the pullback of an ample divisor on @xmath531 , hence nef on @xmath0 . for @xmath533 , one proceeds similarly by noting that @xmath1 is a refinement of the projective non - simplical fan @xmath534 with 1-dimensional generators @xmath535 . other more substantial examples can be found in chapter 7 of scaramuzza s thesis @xcite . based on this , we make the following conjecture , which we credit to casagrande . [ casagrande ] let @xmath0 be a simplicial toric variety coming from the fan @xmath1 in @xmath50 . if @xmath10 is convex of dimension @xmath5 , then @xmath51 besides the evidence provided by examples , we also have the theoretical result of casagrande ( * ? ? ? 5.6 ) , which states that if a smooth complete non - projective toric variety @xmath0 has a toric blow - up @xmath536 with @xmath537 projective , then conjecture [ casagrande ] holds for @xmath0 . we stated conjecture [ casagrande ] for the simplicial case because of the following result . [ if_conjecture ] assume that conjecture [ casagrande ] is true . let @xmath0 be a non - simplicial toric variety of a fan @xmath1 in @xmath50 such that @xmath10 is convex of dimension @xmath5 . then : @xmath51 furthemore , every primitive collection for @xmath1 is supported on a simplicial refinement @xmath37 of @xmath1 satisfying @xmath335 . the first part of the proposition follows since the proof of theorem [ theorem_mori_nonsimp ] ( the non - simplicial case of our main theorem ) requires the existence of a simplicial refinement @xmath37 of @xmath1 with the following properties : * @xmath161 . * @xmath538 is described using primitive inequalities . if we assume conjecture [ casagrande ] , then the second bullet is automatically true , which means that corollary [ cor_simpl_ref_general ] gives the needed simplicial subdivision of @xmath1 . for the final assertion , observe that the proof of proposition [ prim_lemma ] applies without change since the first part of theorem [ thrm_simpl_ref_general ] gives the required simplicial refinement of @xmath1 without needing to assume quasi - projective . one way to think about conjecture [ casagrande ] is that once this conjecture is proved , the results of this paper would apply to any fan in @xmath50 whose support is convex of dimension @xmath5there would be no requirement that @xmath1 be quasi - projective . however , the proofs of the simplicial case given in theorem [ theorem_mori_primitive ] make essential use of extremal rays , which exist only in the quasi - projective case . the result of casagrande ( * ? ? ? 5.6 ) mentioned above is a good first step , but it is likely that some significantly new ideas will be needed to prove conjecture [ casagrande ] in general . this paper is based on the second author s phd thesis at the university of massachusetts @xcite , written under the direction of eduardo cattani and the first author . both authors are very grateful to eduardo cattani for many stimulating discussions . the first author would also like to thank anna scaramuzza for asking some interesting questions about primitive collections . we would also like to thank cinzia casagrande for bringing @xcite to our attention and for giving a counterexample to a conjecture in the original version of the paper . d. cox , _ recent developments in toric geometry _ , in _ algebraic geometry santa cruz 1995 _ ( j. kollr , r.lazarsfeld and d. morrison , eds . ) , proceedings of symposia in pure mathematics * 62 * , amer . soc . , providence , ri , 1997 , 390435

we review tools that have been developed in recent years to maximize our ability to discover and characterize new physics appearing in lhc events with missing transverse momentum .

1110.4502

the 7 tev run of the large hadron collider ( lhc ) is now well underway and a plethora of searches for new physics has already been carried out . this short review concerns searches for new physics in lhc events with missing transverse momentum . many of these searches are based on techniques that have been developed in the last few years , specifically for the advent of the lhc , and the main purpose of this review is to describe some of those techniques and the principles that underlie them . the review will also cover developments in the strategies that are planned to measure the properties of any new physics that we may be lucky enough to discover . this characterization of new physics is not wholly divorced from the discovery process itself , but the two processes are sufficiently different that they merit separate consideration . the experimental analyses with which this review concerns itself , like the machine and detectors themselves , are imponderably complicated . this review contains no discussion whatsoever of issues related to hardware , calibration , particle isolation and identification , triggering , jet reconstruction , pile - up , underlying event , and so on . instead , the starting point will be a set of high - level objects ( jets , leptons , missing transverse momentum , etc ) in an event , together with their associated measured momenta and uncertainties thereon . given these , we shall ask which observable or observables , considered as functions of the momenta ( visible or missing ) , optimize our ability to either make a discovery of new physics or to measure some property thereof . even with such a narrow focus , the discussion will , necessarily , be incomplete . the literature is , moreover , large and ever - growing , so that one can not even hope to catalogue it in a definitive fashion . rather , the hope is to provide , at least , a useful , if idiosyncratic , introduction to the subject and its methodology . before closing this introductory section , it is perhaps useful to spell out the reasons for concentrating on events with missing transverse momentum . one is that methods for dealing for lhc events without missing transverse momentum can often be straightforwardly copied from those used previously at hadron colliders or elsewhere . to take an example which we shall discuss in more detail below , in searching for new resonances it suffices to `` blindly '' compute invariant mass distributions of some combination of final state momenta . but in events with missing transverse momentum , we shall argue that the optimal observable strongly depends on the details of the signal dynamics . so , even though we have , by now , a great deal of experience in studying missing transverse momentum events ( for example , in the discoveries of the @xmath0-boson and top quark at cern and fermilab respectively ) , we still have to work hard to be sure that we have optimized our ability to discover physics at the lhc whose nature is not yet known to us . a second reason , of course , is that many interesting scenarios of new physics , both within the standard model ( sm ) and beyond it , do predict missing transverse momentum events at the lhc . indeed , we can expect to see missing transverse momentum whenever a neutrino is produced in the decay of a higgs boson . moreover , we hope to be able to produce new , invisible particles at the lhc , most notably dark matter . a third reason is purely pragmatic . data is produced in lhc collisions at such a rate that one must either throw most of it away at random with a `` pre - scaled '' trigger , or one must trigger on some characteristic . one suitable characteristic is missing transverse momentum , not least because its presence tells us that the event can not be some tedious qcd process , which overwhelmingly dominate the total cross - section . unfortunately , most of the time such events will be qcd , but badly measured , and indeed a large part of the remaining work will be to separate this background from a signal of events with genuine missing transverse momentum . the outline is as follows . in the next section , we discuss the issue of finding an optimal observable and why there is a necessity to do so in the first place . in section [ sec : kinematics ] , we consider one approach to finding such an optimal observable in the context of what is perhaps the most simple example , namely leptonic decays of the @xmath0-boson . we then generalize the same methodology , which essentially amounts to the assumption that the most important distinguishing feature of the signal ( compared to backgrounds ) is its kinematics , to more complicated situations , including decays of a higgs boson to @xmath0-bosons or @xmath1-mesons and pair production of supersymmetric partners of sm particles . the @xmath2 example is particularly appropriate for the purposes of our general discussion , since a number of more sophisticated approaches seem to be possible . even focussing only on kinematics , it is possible to fully reconstruct events at the lhc , up to a discrete ambiguity , meaning that one can , essentially , carry over the same bump hunting methods used for events without missing transverse momentum . but @xmath2 searches are also interesting because they give real - world examples of analyses which try to go beyond kinematics and try to use dynamical information to discriminate signal from background . we discuss analyses proposed or in use by both atlas and cms in section [ sec : dynamics ] . we then go on to consider how assumptions about the dynamics may help in susy searches , in a variety of ways . finally , we turn to the issue of measuring the properties of new physics , such as masses , spins , and couplings . there already exist reviews covering these subjects , so our discussion will be brief . we also discuss more generally the problem of how we can make inferences about dark matter on the basis of lhc measurements and how best to characterize new physics in its nascent stages . to begin with , one might wonder why there is any need to devote a review to discussing optimal observables for events with missing transverse momentum . indeed , the question of ` which observable is optimal ? ' was answered for us long ago by the statisticians . the answer is that the optimal observable is given by the likelihood function . in particle collisions , this amounts to the matrix element for a particular scattering process , including signal and background contributions , convoluted with the appropriate detector response and marginalized with respect to quantities that go unmeasured . unfortunately , this observable is nigh - on impossible to compute in practice : not only is our modelling of the matrix element ( for either signal or background ) and detector response insufficient , but also the computational effort required to carry out such an analysis at the lhc is inconceivable . even then , one might wonder why there is any need for phenomenologists to scratch their heads trying to find an optimal observable . at least when one is dealing with high - level objects after event selection , any observable may be written as a function of only a few basic observable momenta . so one might object that again the statisticians have already done the hard work in designing multivariate analyses , such as neural networks , boosted decision trees , etc , which will find the optimal observable for us , once we have trained them up sufficiently . there are two counter - objections to this objection . the first , as de rjula and galindo@xcite so charmingly put it , is that there are times when one would like to have the `` pleasure of understanding with use of one s own neural network '' in order to have a clear picture of what is going on . the second is that the sophisticated multivariate analyses on the market are perhaps not so sophisticated , in that they typically only form linear combinations of the initial observables . as such , they miss variables constructed , for example , as products or ratios of the initial observables . indeed , one does not have to look too hard to find examples in the recent literature where humans have succeeded in finding useful observables where multivariate methods have failed.@xcite so , _ nolens volens _ , there is still work for us to do . even though the true likelihood function is unattainable , it does suggest a useful way to approach the problem of finding an optimal observable . indeed , imagine that one is able to identify which aspect or aspects of the physics most affects our ability to make a discovery or a measurement in a particular channel . one could then try to find the observable that best takes account of this aspect . in the next section , we shall develop this idea in the context of several examples in which it is the differing kinematic properties of the signal and background which are assumed to be paramount . we have already mentioned how one might search for a resonance in events without missing transverse momentum . the tried and trusted way to do so is to plot the invariant mass of some combination of final state particles and to look for a bump . this is a useful thing to do , not only because lorentz - invariance guarantees that signal events will pile up at the resonance mass ( increasing the discovery potential ) , but also because it is rather hard to imagine effects coming from smoothly - varying backgrounds and detector response which could give rise to a bump . thus , even with almost complete ignorance about the nature or size of the backgrounds and detector response , one might confidently claim a discovery if a bump were seen . in this example , the aspect of the physics which is most important is the ( assumed ) presence of a narrow resonance peak in the invariant mass distribution of the signal and so the natural choice of observable is the invariant mass . the assumption is optimal not only for discovery , but also for measuring the mass of the resonance , since signal events pile up at the place where the measurement is performed . in a sense , this way of proceeding is akin to following the likelihood approach , albeit in a cavalier fashion . in effect , one is making the drastic approximation that `` the physics '' is encapsulated by the chosen aspect ( kinematics ) and nothing else and then making a crude guess as to the resulting likelihood variable . it is important to note that making drastic approximations of this kind can not invalidate the subsequent analysis , even if the approximations made are badly violated in reality . indeed , the only purpose of the procedure is to define ( or rather , motivate ) an observable . but some observables are better than others and so the price that one pays for a poor approximation is an observable that is far from being optimal . so , one can already can make a lot of progress just by using kinematic features of the signal , choosing an observable whose signal distribution is as `` peaked '' ( or piled - up ) as possible . of course , in order for this to be a useful course of action , one should take care that the background not be equally sculpted at the same time ! this idea , as we will see , generalizes in a very natural way to events with missing transverse momentum . when one ( or more ) of the final state particles is ( or are ) invisible , we can not hope to construct a resonance bump in an invariant mass , since the invariant mass is no longer observable . but we can try to construct an observable for which the signal is as peaked as possible . to do so , let us begin with the case where all final state particles are visible , for which the invariant mass observable is always equal to the resonance mass . to put it another way , the observable is bounded both above and below by the mass of the resonance . with one or more final state particles invisible , we shall not be able to find an observable which is bounded both above and below , but we shall be able to find observables which are bounded above or below . by finding the observable which gives the strongest upper or lower bound , we obtain an observable for which the signal is as peaked as possible , and which may be considered optimal in the sense of being discoverable in the presence of a smoothly ( and slowly ) varying background . what is more , if the location of that peak ( or other sharp feature ) is strongly correlated with some physical property of the signal ( such as the mass of the resonance ) , then it may be that the same observable is optimal for the purpose of that measurement . we now consider an example in which this observable is nothing other than the transverse mass . consider decays of a @xmath0-boson into a charged lepton and an invisible neutrino . one way to define the transverse mass , @xmath4,@xcite is as the observable that provides the greatest lower bound on the ( _ a priori _ unknown ) mass of a resonance which decays into some combination of visible and invisible final state particles . the classic example is the decay of a @xmath0-boson into a charged lepton and an invisible neutrino , for which the above prescription yields @xmath5 where @xmath6 is the transverse momentum of the charged lepton and @xmath7 the missing transverse momentum . in this way , @xmath4 is seen to be the optimal variable ( in the sense described above ) to discover the @xmath0-boson ( although it was not employed in that way in the original analysis@xcite ) . in fact , @xmath4 is even better than may appear on the basis of the above considerations . because @xmath4 is able to separate the signal from some of the backgrounds . indeed , there are plenty of ways to arrive at an event containing a charged lepton , missing transverse momentum and zero or more jets and each of these is a source of background . for example , every time one produces a @xmath8-meson , then ten per cent of the time these decay leptonically , resulting in precisely the same final state . however , for these events , @xmath4 can not exceed the mass of the @xmath8-meson , namely a few gev . thus , @xmath4 has the additional advantage of providing a separation between the signal events ( which may take values all the way up to @xmath9 ) and background events coming from leptonic @xmath8-decays , which lie below @xmath10 , if well - measured . if , instead of a charged lepton , the visible particle in the final state had been a jet , then another dominant source of background would come from events in which the energy , but not the direction , of the jet had been mismeasured , leading to apparent , fake , missing transverse momentum , aligned with the jet . but for such a configuration , @xmath4 is given , roughly , by the jet mass , which again is typically much smaller than @xmath9 . it is important to note that one does not have a theorem to the effect that there is a clean separation between the signal and all backgrounds . in typical final states there will be many background components and not all of them will enjoy this property . but this does not render the above observations worthless . the point is that , in any given analysis , one expects that certain background components will be well understood while others will be less so . one can also , therefore , try to proceed by finding an observable which is insensitive to the background with the largest uncertainty . one can immediately apply these ideas to searches for new physics at the lhc , namely to searches for a @xmath3-boson . now , the dominant background comes from the @xmath0-boson itself , but again one is guaranteed a clean separation of signal and background , at least for well - measured events , since for the background events @xmath11 . this strategy is exactly that which is employed in recent lhc searches.@xcite as an aside , the fact that @xmath4 gives the greatest lower bound on the parent particle mass means that it contains all of the information which follows from the kinematic constraints ( namely conservation of energy - momentum and the mass - shell constraints ) alone.@xcite the above approach generalizes straightforwardly to decays of a resonance into multiple visible particles , with the observable @xmath4 in ( [ mt ] ) being replaced by @xmath13 where @xmath6 is now the transverse momentum of the visible particle system and @xmath14 is its measured invariant mass . for decays with multiple invisible particles , things are a little more subtle . navely , one would define the transverse mass by @xmath15 but this is not an observable , since the invariant mass of the invisible system , @xmath16 , is not observable . however , since @xmath4 in ( [ mtww2 ] ) is a monotonically increasing function of @xmath16 , one immediately sees that the best lower bound on the mass of the decaying resonance is given by replacing @xmath16 by its minimum value , namely the sum of the invisible particle masses , if known , or , if not , by its minimum possible value , namely zero . an immediate application@xcite ( see also refs . ) is to decays of a heavy ( above 135 gev or so ) sm higgs boson into two charged leptons and two neutrini via ( possibly virtual ) @xmath0-bosons . the resulting observable , called @xmath17 in is just given by the right hand side of ( [ mtww ] ) and is central to state - of - the - art analyses carried out by atlas@xcite and cms@xcite . in more detail , atlas , for example , performs@xcite a cut - and - count analysis in the interval @xmath18 , for varying higgs mass @xmath19 ( after other cuts ) . in accordance with the arguments given above , this maximizes the signal contribution in the cut region compared to the dominant background ( which comes from continuum @xmath12 production . ) , as the figure of merit . this was based on the observation that , for a spinless resonance ( such as the sm higgs ) of mass @xmath20 decaying ( at threshold ) to @xmath12 , conservation of angular momentum forces the two charged leptons to be parallel . away from threshold , the two charged leptons remain well - correlated . this is a neat example of an analysis in which dynamics , rather than kinematics , has been used to motivate the choice of observable . ] in fact , for higgs decays in this channel , one can do even better . for @xmath19 above @xmath20 , both @xmath0-bosons will typically be produced on - shell , whilst even for @xmath19 below @xmath20 , one or other @xmath0-boson will typically be produced on - shell , at least for higgs masses large enough to give sizable cross - section times branching fraction in the @xmath12 channel . this observation can be used in the following way . one can define two new observables , @xmath21 and @xmath22 , as the ones that give the greatest lower bound on the mass of the ( higgs ) signal resonance , subject to the condition that either one or other ( for @xmath23)@xcite , or both ( for @xmath24)@xcite of the @xmath0-bosons be on shell . , @xmath23 , and @xmath24had been called @xmath25 , @xmath26 , and @xmath27 , respectively . ] in both cases , since one is applying additional constraints , one obtains a greater lower bound event - by - event , resulting in an even more sharply peaked signal distribution . unfortunately , one can not give an explicit , analytic form for these two variables , but they can be straighforwardly computed using an algorithmic computer code . perhaps unsurprisingly , each gives a significant improvement only in the region in which the assumptions that are made in the definition hold to a good approximation . the variable @xmath22 was first introduced in the context of higgs decays to @xmath29,@xcite which each , in turn , decay either leptonically , producing two neutrinos , or hadronically , producing only one neutrino . one apparent disadvantage of this variable for the @xmath30 channel is that the measured momenta must satisfy the condition @xmath31 in order that @xmath22 be well - defined . this is guaranteed for well - measured signal events , but in the presence of detector mismeasurements , events may easily fail to satisfy this condition . indeed , navely , a mismeasurement of @xmath32 by @xmath33 would lead to a similar mismeasurement in @xmath34 ; since @xmath33 typically far exceeds @xmath35 at the lhc , one would expect the condition would rarely be satisfied . in reality , the situation is not so bad,@xcite but still one finds that @xmath22 does not exist for about a third of events in simulations , leading to a significant reduction in statistics . ) for these events , so they need not be discarded . ] there are other approximations which might be made for @xmath29 decays , resulting in other observables . for now , we restrict our attention to those which are solely concerned with kinematics , deferring consideration of dynamics to the next section . firstly , given existing limits from lep on the mass of the higgs boson , it is likely that the produced @xmath1s will be significantly boosted in the lab frame , in which case one could assume that their decay products are collinear.@xcite if one makes this assumption , then in fact there are as many constraints as unknowns , such that the momenta ( and the invariant mass of the @xmath29 system ) can be reconstructed unambiguously . to wit , the two unknown proportionality constants between the visible and invisible momenta in each @xmath1 decay can both be determined from the two missing tranverse momentum constraints . so one ends up not with a lower bound on the invariant mass of the @xmath29 system , but its precise value , assuming that things are well - measured . in practice , this collinear approximation seems to be not such a good one : the presence of finite detector resolution and acceptance results in solutions with unphysical energies in a significant number of events ( as many as half in simulations carried out by the experimental collaborations ) . an alternative kinematic strategy for hadronic @xmath1 decays is to use the extra information which is available in such events from of the location of the secondary ( @xmath1 decay ) vertices.@xcite the best measured attributes of these are their impact parameters , namely the displacements of the secondary vertices measured in a direction perpendicular to the visible decay momenta . the invisible momentum in each @xmath1 decay must lie in the plane of the impact parameter and the visible momentum , leading to a quartic equation ( and a fourfold ambiguity ) in the reconstructed higgs mass . the problem arises of how to deal with the resulting discrete ambiguity ( which in other cases may be compounded by a combinatorial ambiguity as to which of multiple , indistinguishable final state particles should be assigned to which decay or part of a decay . a similar situation ( but for cascade decays of supersymmetric particles ) was studied in ref . . there it was empirically observed that the wrong solutions appeared to be correlated with the right solution . an explanation was given in ref . . in the specific case of @xmath1 decays considered above , for example , it was pointed out@xcite that , in the limit that the higgs mass becomes large compared to the mass of the @xmath1 , the quartic equation reduces to a linear equation , with the four solutions coalesced in a single solution . in the real world , we are slightly away from the limit , but the solutions remain close together . the fact that the wrong solutions are correlated with the right one means that one can simply retain all solutions ( indeed in ref . , it was pointed out that it is even useful to retain complex solutions , to increase the available statistics ) : the correlation means that the `` signal '' ( in the form of the correct solution ) will not be overwhelmed , but rather will be reinforced , by the `` background '' coming from the wrong solutions . these ideas can all be extended straightforwardly to events with new invisible particles , such as the lsp of a supersymmetric theory or another dark matter candidate , rather than common or garden neutrinos . if one is only interested in kinematics , the only relevant property of an invisible particle is its mass . as we shall see below , this poses an important complication when it comes to making mass or other measurements , since it introduces another unknown to be measured . but for discovery purposes , the only effect of a massive invisible is that the scale of the visible particles in the final state is set , roughly speaking , by the mass difference between the parent particle ( or particles ) produced in the hard process and the invisible daughter particles . this is significant for search strategies of the type discussed above , which aim for a kinematic separation between signal and background , since the smaller this mass difference becomes , the more the signal overlaps with the sm backgrounds . so even heavy ( but approximately degenerate ) new particles may be difficult to discover in this way . to make progress in such a scenario , one would need to use information about the dynamics as well . we postpone discussion of this to [ sec : dynamics ] . the second important complication of supersymmetric theories is that the parent particles are pair produced , with each decaying to one ( or more ) invisible daughter particles . thus , rather than assume , as we have done above , that the observed missing transverse momentum can be attributed to a single decay , we should only assume that it is partitioned in some way between the two decays . once we do so , we can again blindly follow our procedure of asking which observable gives the greatest event - by - event lower bound on the mass of the decaying parents . the resulting variable , whose definition remains algorithmic except in special cases,@xcite goes by the name of @xmath34.@xcite just as for @xmath4 , @xmath34 gives us a variable which is as `` peaked '' as possible for the signal , leading us to hope that a significant signal excess can be identified on top of backgrounds . even better , just like @xmath4 it has properties which serendipitously ensure a good separation between the signal and several of the dominant sm backgrounds , provided that the mass difference between the parents and invisible daughters is large.@xcite this makes our understanding of the backgrounds that appear in the signal region and the systematic uncertainties thereon more robust , enabling a discovery or exclusion to be made even in the absence of a sharp feature in the signal distribution . both general - purpose collaborations have now presented susy searches employing @xmath34.@xcite again , it follows that @xmath34 captures all of the information that is contained in the energy - momentum conservation and mass - shell constraints alone.@xcite this will be important in our discussion of mass measurement , in that if invisible particle masses can not be measured using @xmath34 , then they can not be measured at all using kinematics alone . happily , they can . once particles are pair produced , we must consider the ambiguity that arises in deciding which of the visible particles in the final state should be grouped together and associated to the decay of the parent . this problem is , of course , most acute when the decays are identical . again , one can take this ambiguity into account in designing the discovery observable , by asking which observable gives the greatest lower bound on the parent mass , after minimizing in addition with respect to the possible combinations . the resulting variable is called @xmath36.@xcite a different procedure has been suggested in ref . . finally , one can use a similar trick to deal with ambiguities that arise in deciding which jets are likely to have come from the final state and which from radiation in the initial state ( insofar as these are well - defined notions).@xcite this simple but far - reaching idea , of optimizing discovery potential by finding the observable that provides the strongest bound on the mass of a new particle has also been applied in the context of searches for charged higgs bosons in top quark decays@xcite and for pair - produced third - generation leptoquarks@xcite decaying to a @xmath37 final state.@xcite the kinematic ideas discussed above could , in principle , be applied to events with arbitrarily complicated topologies . we have already seen , however , that even in relatively simple cases , such as @xmath38 , it is not possible to find an explicit , analytic expression for the observable that encodes all of the information in kinematics . kim@xcite has noted that the notion of a bounding variable arises because of endpoint singularities that appear when one projects the full phase space of some event topology onto its observable subspace ( a simple example was given in ref . , in which the full phase space is a sphere and the observable phase space is a disk , singular on its edge ) . he has noted , furthermore , that these singularities show more general behaviour than simply endpoints ( such as cusps@xcite ) ; these more general features may also provide a way to identify a signal on top of a smoothly - varying background . once the singularities corresponding to a given signal topology have been found , one may try to construct an optimal observable in the region of the singularity , according to various criteria of optimality . de rjula and galindo@xcite have attacked this problem in the simplest case of @xmath0-boson decays , defining observables which are optimal in a statistical sense . all of the methods dicussed above use only the kinematic properties of the signal and backgrounds . we now discuss analyses which try , either separately or in addition , to exploit what is known , or may be assumed , about the dynamics . for @xmath2 searches ( either in the sm or mssm ) , both atlas@xcite and cms@xcite have employed more sophisticated approaches using _ ad hoc _ likelihood functions . neither of the approaches implemented so far uses information from the secondary vertices , such that the kinematics of the signal are underconstrained . the analyses are similar in that they both allow the unconstrained parameters , namely the momenta and invariant masses ( for hadronic decays ) of the invisible systems in each @xmath1 decay to vary , subject to the mass shell and missing transverse momentum constraints . the higgs mass is then reconstructed by maximizing a likelihood function on this space , whose nature differs in the two analyses . the atlas likelihood function@xcite contains two ingredients . the first is a probability density function for the @xmath1 decay angles , which is obtained by fitting a simulation of @xmath39 decays . the second ingredient is a gaussian smearing of the measured missing transverse momentum.@xcite cms are more reticent about the details of their likelihood function ; we are at least told that it `` [ takes ] into account the tau - decay phase space and the probability density in the tau transverse momentum , parametrized as a function of the tau - pair mass''.@xcite there have been several , rather different , efforts to use dynamical information to optimize discovery of theories with pair production , such as supersymmetry . the most basic of these continue the spirit of the kinematic discussion above , but either supplement or substitute it with dynamical assumptions . as an example , one might make the dynamical assumption that , because parton distribution functions fall off steeply with increasing @xmath40 , parents will tend to be pair produced at rest relative to one another in their cm frame , so as to minimize the resulting @xmath41 . if one makes this assumption , then the natural observable is defined in ref . , where it is denoted @xmath42 . since pairs of squarks are produced in a @xmath43-wave state at leading order in qcd in quark - antiquark collisions , it is clear that this assumption can not be valid in certain cases . but , as discussed above , this does not render the observable invalid ; rather , it only makes it possibly sub - optimal . , designed to deal with this issue . ] it certainly is true , however , that pairs of parents are produced back - to - back in their cm frame . if one made , furthermore , made the dynamical assumption that the cm frame of the parents was dominantly boosted only longitudinally with respect to the colliding beams , then the optimal observable might be one which was invariant under both longitudinal boosts of the cm frame of the parents and back - to - back boosts of the parents in their cm frame . indeed , one would expect the bulk of signal events to occur at the same value of such an observable ( in exactly the same way that lorentz invariance guarantees a peak in the invariant mass in decays without missing transverse momentum ) . the appropriate variable goes by the name of @xmath44.@xcite these approaches could be applied to arbitrary pair decay processes . for longer cascade decay chains , we have seen that it is kinematically possible to reconstruct events . however , one might wonder how reliable such a reconstruction is in the presence of detector mismeasurements . one way to deal with this is to construct a likelihood which takes the detector response into account.@xcite a completely different approach deals with the aforementioned difficulty of discovering susy using naive kinematic observables if the superpartner spectrum is roughly degenerate , or `` compressed '' . at least if the superpartners are heavy , one important feature of the signal is that it leads to a harder pattern of radiation in the initial state than what is typically obtained for sm backgrounds . thus , the optimal way to discover such scenarios may be to focus on events with more jets than expected from the signal final state and with large missing transverse momentum , arising when the parents recoil strongly against the initial state radiation.@xcite by incorporating a number of selection cuts , one can achieve a good coverage of the signal parameter space , without the need for sophisticated observables.@xcite one can also try to tag initial state radiation explicitly.@xcite yet another direction is to focus instead on the dynamics of the background rather than that of the signal . so far , efforts have been directed at the dominant background of mismeasured qcd events . here , the presence of the background is purely a detector effect of which one can hope to build up a good understanding . as a simple example , the fact that the largest mismeasurements come from jet energies ( rather than jet directions ) means that one can suppress much of this background by cutting on an observable with known properties under such mismeasurements . one such observable , @xmath45,@xcite works well for di - jet events , though the signal rejection is high . more challenging are events with multiple jets , since our understanding of the sm background prediction is very poor . atlas@xcite has performed a search in events with between six and eight jets in which the sm background is determined by extrapolation from measurements in events with fewer jets . the key observable is @xmath46 , where @xmath47 is the scalar sum of the transverse momenta of all jets within some fiducial region . this ratio measures the missing transverse momentum relative to the resolution due to stochastic variations in the measured jet energies that are assumed to follow the central limit theorem . it is then expected that the distribution of this observable will be roughly invariant under changes in the jet multiplicity , such that the background in the event sample with high jet multiplicity can be estimated from the measured distribution in the low jet multiplicity sample , which is assumed to be background dominated . a similar search was carried out by cdf,@xcite using a cut on this variable . cdf s search also employs a technique in which the total missing transverse momentum is compared with that observed using the charged particle spectrometer only , the rationale being that these should be aligned for signal events , but may be anti - aligned for background events arising from calorimeter jet energy mismeasurements . finally , one could attempt a more rigorous likelihood method using genuine matrix elements for signal process . though this has not been explored in the context of discovery , a study of the prospects for mass measurement has been carried out.@xcite we now turn to a discussion of methods that have been developed to measure the properties of new physics , once it has been discovered . as we remarked in the introduction , it is clear that the process of making measurements is not wholly divorced from the process of making a discovery , just as the process of excluding models currently underway at the lhc depends on where in the parameter space one sits . indeed , many of the observables and stategies proposed for making measurements coincide ( or nearly coincide ) with discovery strategies . this is hardly surprising , in that observables which do a good job of distinguishing signals from backgrounds will also facilitate making clean measurements . furthermore , the topics of mass and spin measurements have already been reviewed in the recent literature.@xcite our discussion will , therefore , be brief . the first measurement priority will be to establish the kinematic properties of the signal , namely the masses of any new particles . just like the discovery process , this is complicated by the fact that kinematic information is lost in events with invisible particles . it has long been known ( beginning with the pioneering study of endpoints in invariant mass distributions in ref . ) that some information about particle masses ( roughly speaking , mass differences ) can be obtained using kinematics alone . in events involving cascade decays , with on - shell intermediate states , it may be possible , as discussed above , to reconstruct events . this approach was pioneered in ref . and has been developed in refs . . the business of counting the various unknowns and constraints is straightforward and we shall not describe it here . as an example of how many constraints is required , for pair decays one needs two , on - shell intermediates in each chain in order to be able to reconstruct the masses . a crucial question is whether masses can be measured absolutely in decays with fewer on - shell intermediates , using kinematics alone . a definitive answer to this has been provided as follows : when a parent particle ( or a pair of identical parent particles ) decays into a system of visible daughter particles and a system of invisible daughter particles , there is enough information in kinematics to measure the mass of the parent and the sum of the masses of the invisible daughters , even when there are no on - shell intermediates.@xcite in particular , if a decay process involves only a single invisible dark matter particle , then the mass of that particle , can , in principle , be measured , no matter what decay produces that particle . what has not been answered definitively is the question of how best to measure those masses , or indeed whether it can be achieved in practice . of course , the answer to this question depends to a large extent to the nature of the signal and many different methods have been put forward . these have been painstakingly reviewed in ref . , to which the reader is referred for details . regrettably , along with this plethora of methods has come a _ smrgsbord _ of different observables with confusingly similar names . ref . counted over fifty different variables in the literature which go under the name of `` transverse mass '' , with or without further qualification . this state of affairs is probably unavoidable ; in any case , there would seem to be an onus upon authors and workers in the field to take care over the definition of observables . spin measurements are challenging even in the absence of missing transverse momentum , because spins ( being the generators of rotations ) manifest themselves in angular effects . in order for such effects to be present in a decay , for example , one first needs to create a preferred axis of some kind . so to have observable spin effects often requires a complicated event topology . for example , in a cascade decay , in order that non - trivial angular correlations exist , one requires at least that the intermediate particle in question be polarized , and if a fermion , that its decay be chiral.@xcite things are even more difficult given the presence of missing transverse momentum , because one needs to be able to reconstruct reference frames in order to be able to measure angles explicitly . one way to circumvent this problem is to focus on lorentz - invariant quantities , which take the same value in all frames . alternatively , one needs to reconstruct events , which can only be achieved for certain topologies , as described above . there are , nonetheless , many ideas on the market , most of which apply only to a specific decay topology . they are reviewed in refs . . besides measuring masses and spins , one would eventually like to measure all of the parameters appearing in the lagrangian that describes physics at the tev scale . these include the gauge and global charges of new particles ( with respect to whatever symmetry group is present at that scale ) , as well as the coupling strengths of the interactions that are permitted by that symmetry group . the method by which one might achieve this is , of course , greatly contingent upon the nature of the new physics and relatively little work has been appeared in the literature . to give just one example of what has been done , ref . explored in detail the procedure by which one might measure couplings and mixing angles , in addition to masses and spins , in cascade decays using invariant mass observables . we now describe a rather different measurement that has been proposed recently,@xcite namely to count invisible particles in events . if we are fortunate enough to discover new physics associated with missing transverse momentum at the lhc , there will surely be a great temptation to identify it with the dark matter that dominates the matter in the cosmos . the question arises , though , of how to make that identification concrete . the holy grail would , presumably , be to compute the expected relic density using fundamental parameters of the lagrangian measured at the lhc or in other terrestrial experiments and to compare with the observed cosmic abundance . unfortunately , it would appear that such a comparison is out of reach in the lhc era.@xcite for one thing , such a comparison would require one to measure a large number of parameters , each one of which would , from the previous discussion , require a herculean effort at the lhc . indeed , a computation of the relic abundance would require one not only to discover all the new particles appearing at the tev scale , but also to determine their masses , spins , and couplings to all other particles . for another , the observed relic abundance implies that it is the weak interaction that is responsible for setting the dark matter density , but weak interaction effects are inevitably overwhelmed by strong interaction effects at a hadron collider and so near impossible to disentangle . as an explicit example , cross sections for processes involving production of new particles followed by their decay are set in the narrow width approximation by the product of the production cross section and the branching fraction for the decay , but the production cross - section will usually be dominated by the strong interaction . ref . made the rather different suggestion of trying to gain information about a dark matter candidate by simply counting invisible particles in events with missing energy . indeed , multiple production would already constitute evidence for the presence of a symmetry , as required to stabilize dark matter on hubble timescales . similarly , if one could establish evidence for production of an odd number of invisible particles , one could rule out the simplest ( and most commonly employed in models ) symmetry , namely @xmath48 . but how can invisible particles be counted in lhc events ? the available energy in collisions is , of course , shared out in a random fashion between the particles produced and , as a result , the shapes of various observable distributions vary depending on how many invisible particles are present . as a simple example , if a parent particle is singly produced at the lhc and decays to visible particles and @xmath49-invisible particles whose masses may be neglected on collider scales , then the transverse mass ( [ mt ] ) has a power - law behaviour near its maximum , with the power given by @xmath50.@xcite this result holds independently of the number of visible particles , or indeed of whether there are intermediate particles on - shell . at least in this simple example , it should be rather easy to distinguish between production of a single , invisible particle ( for which the distribution diverges near the endpoint ) from production of multiple , invisible particles ( for which the distribution vanishes ) . a related question is how we might try to distinguish between different stabilizing symmetries for dark matter . consider , for example , the two simplest symmetry groups , namely @xmath48 and @xmath51 . counting invisible particles provides one way , in that an odd number of invisible particles can only be associated with @xmath51 and not @xmath48 . previous work@xcite attacked this question in a different way , focussing on the observation that models with @xmath51 symmetries might contain final states with the same visible particle content , but with different numbers of invisible particles , which would lead to distributions containing two , rather than one , components , which might be picked out in data . even once we have made a discovery of new physics and possibly measured a limited number of its properties , we shall be faced the question of how best to try to characterize that physics . the holy grail is , of course , to write down the lagrangian , in which the fundamental principles of symmetry , etc . , are as manifest as we know how to make them . unfortunately , it seems unlikely that we will succeed in guessing the correct form of the lagrangian soon after discovering new physics . indeed , it may not be possible at all to home in on the lagrangian in the lhc era , let alone to perform precise measurements of its parameters . but some characterization of the new physics would be invaluable in terms of suggesting where to look for corroborations of discoveries , to perform cross - checks , or indeed to further enhance our knowledge about the nature of the physics in a sort of bootstrap process . there seem to be two possible approaches . the first is make an educated guess ( perhaps on the basis of one s theoretical bias ) as to what the exact lagrangian is , modulo the values of a few parameters . one can then either exclude the model in an increasingly large region of parameter space ( in the absence of a signal excess ) or attempt to measure those parameters ( in the presence of one ) . this approach has been very popular , understandably so since it gives one a concrete hypothesis which ( one hopes ) will eventually be accepted or rejected . one downside is that one s guess must be a good one and , though it is a tale which is too long for us to tell here , our theoretical bias has not in the recent past proven to be a very good indicator for what new physics does lie beyond the standard model . a perhaps more pernicious downside is that focussing on a specific lagrangian can lead to the development of search strategies which , though optimized for that particular lagrangian , fail spectacularly when one considers other new physics , that is not dramatically different from the original lagrangian . a salutary example is provided by supersymmetric models with unification of gaugino masses at the gut scale . the rg evolution leads to a fixed , large ( 6:1 ) mass ratio between the gluino and the bino at the tev scale and , since the bino is often the lsp in such models , a fixed signal kinematics . searches which were developed with such models in mind have proved to be rather ineffective in probing models with compressed spectra.@xcite the second approach avoids this by via a more _ ad hoc _ strategy of making a coarse guess for new physics in the form of a simplified model@xcite which , it is hoped , captures the essence of the physics that is relevant for lhc collisions . this can hardly be described as a novel philosophy , in that physics has been done in much this way for millennia , but the details might be . typically , a simplified model might contain only a few new degrees of freedom , with masses allowed to float as well as the cross - sections for production and decay . the advantage of such an approach is that it enables a huge variety of models to be rapidly scanned and compared with lhc data ; the disadvantage is that such a model is in no way fundamental and it is hard to see how it could be used to make detailed predictions for new physics phenomena elsewhere , for example in the cosmos . presumably though , once one had narrowed down the possibilities to a small set of simplified models , one could return to the approach of guessing concrete lagrangians , for which this can be done . i am greatly indebted to a. j. barr and c. g. lester , who were instrumental in inculcating in me much of the material discussed . i also thank m. pierini for correspondence regarding @xmath52 . [ 2]#2 a. j. barr _ et . _ , _ guide to transverse projections and mass - constraining variables _ , http://arxiv.org/abs/1105.2977 [ 1105.2977 ] . a. abdesselam _ et . _ , _ boosted objects : a probe of beyond the standard model physics _ , _ eur . phys . j. _ * c71 * ( 2011 ) 1661 [ http://arxiv.org/abs/1012.5412[1012.5412 ] ] . a. rujula and a. galindo , _ measuring the w - boson mass at a hadron collider : a study of phase - space singularity methods _ , _ jhep _ * 1108 * ( 2011 ) 023 [ http://arxiv.org/abs/1106.0396[1106.0396 ] ] . collaboration , g. arnison _ et . _ , _ experimental observation of isolated large transverse energy electrons with associated missing energy at s**(1/2 ) = 540-gev _ , _ phys . * b122 * ( 1983 ) 103116 . collaboration , m. banner _ et . _ , _ observation of single isolated electrons of high transverse momentum in events with missing transverse energy at the cern anti - p p collider _ , _ phys . lett . _ * b122 * ( 1983 ) 476485 . collaboration , s. chatrchyan _ et . al . _ , _ search for a w boson decaying to a muon and a neutrino in pp collisions at sqrt(s ) = 7 tev _ , _ phys . _ * b701 * ( 2011 ) 160179 [ http://arxiv.org/abs/1103.0030[1103.0030 ] ] . collaboration , g. aad _ et . _ , _ search for a heavy gauge boson decaying to a charged lepton and a neutrino in 1 fb-1 of pp collisions at sqrt(s ) = 7 tev using the atlas detector _ , http://arxiv.org/abs/1108.1316[1108.1316 ] . a. j. barr , b. gripaios and c. g. lester , _ transverse masses and kinematic constraints : from the boundary to the crease _ , _ jhep _ * 11 * ( 2009 ) 096 [ http://arxiv.org/abs/0908.3779[0908.3779 ] ] . a. j. barr , b. gripaios and c. g. lester , _ measuring the higgs boson mass in dileptonic w - boson decays at hadron colliders _ , _ jhep _ * 07 * ( 2009 ) 072 [ http://arxiv.org/abs/0902.4864[0902.4864 ] ] . v. d. barger , t. han and j. ohnemus , _ heavy leptons at hadron supercolliders _ * d37 * ( 1988 ) 1174 . e. w. n. glover , j. ohnemus and s. s. d. willenbrock , _ higgs boson decay to one real and one virtual w boson _ , * d37 * ( 1988 ) 3193 . atlas - conf-2011 - 111 , aug , 2011 . a. j. barr , b. gripaios and c. g. lester , _ re - weighing the evidence for a higgs boson in dileptonic w - boson decays _ , http://arxiv.org/abs/1108.3468[1108.3468 ] . a. j. barr , b. gripaios and c. g. lester , _ finding higgs bosons heavier than @xmath53 in dileptonic w - boson decays _ , http://arxiv.org/abs/1110.2452[1110.2452 ] . a. j. barr , s. t. french , j. a. frost and c. g. lester , _ speedy higgs boson discovery in decays to tau lepton pairs : @xmath54 _ , http://arxiv.org/abs/1106.2322[1106.2322 ] . r. ellis , i. hinchliffe , m. soldate and j. van der bij , _ higgs decay to @xmath55 : a possible signature of intermediate mass higgs bosons at the ssc _ , _ nucl.phys . _ * b297 * ( 1988 ) 221 . b. gripaios , k. sakurai and b. webber , _ polynomials , riemann surfaces , and reconstructing missing- energy events _ , http://arxiv.org/abs/1103.3438[1103.3438 ] . cheng , d. engelhardt , j. f. gunion , z. han and b. mcelrath , _ accurate mass determinations in decay chains with missing energy _ , _ phys . rev . * 100 * ( 2008 ) 252001 [ http://arxiv.org/abs/0802.4290[0802.4290 ] ] . c. g. lester and d. j. summers , _ measuring masses of semi - invisibly decaying particles pair produced at hadron colliders _ * b463 * ( 1999 ) 99103 [ http://arxiv.org/abs/hep-ph/9906349 [ hep - ph/9906349 ] ] . w. s. cho , k. choi , y. g. kim and c. b. park , _ measuring superparticle masses at hadron collider using the transverse mass kink _ , _ jhep _ * 02 * ( 2008 ) 035 [ http://arxiv.org/abs/0711.4526[0711.4526 ] ] . w. s. cho , j. e. kim and j .- h . kim , _ amplification of endpoint structure for new particle mass measurement at the lhc _ , _ phys.rev . _ * d81 * ( 2010 ) 095010 [ http://arxiv.org/abs/0912.2354[0912.2354 ] ] . c. g. lester , _ the stransverse mass , @xmath34 , in special cases _ , http://arxiv.org/abs/1103.5682[1103.5682 ] . a. barr , c. lester and p. stephens , _ @xmath34 : the truth behind the glamour _ , _ j. phys . _ * g29 * ( 2003 ) 23432363 [ http://arxiv.org/abs/hep-ph/0304226[hep-ph/0304226 ] ] . a. j. barr and c. gwenlan , _ the race for supersymmetry : using @xmath34 for discovery _ , _ phys . * d80 * ( 2009 ) 074007 [ http://arxiv.org/abs/0907.2713[0907.2713 ] ] . collaboration , j. b. g. da costa _ et . _ , _ search for squarks and gluinos using final states with jets and missing transverse momentum with the atlas detector in sqrt(s ) = 7 tev proton - proton collisions _ , http://arxiv.org/abs/1102.5290[1102.5290 ] . collaboration , _ search for supersymmetry in all - hadronic events with @xmath34 _ , . cms - pas - sus-11 - 005 . m. serna , _ a short comparison between @xmath34 and @xmath44 _ , _ jhep _ * 06 * ( 2008 ) 004 [ http://arxiv.org/abs/0804.3344[0804.3344 ] ] . cheng and z. han , _ minimal kinematic constraints and @xmath34 _ , _ jhep _ * 12 * ( 2008 ) 063 [ http://arxiv.org/abs/0810.5178 [ 0810.5178 ] ] . c. lester and a. barr , _ mtgen : mass scale measurements in pair - production at colliders _ , _ jhep _ * 12 * ( 2007 ) 102 [ http://arxiv.org/abs/0708.1028[0708.1028 ] ] . k. choi , d. guadagnoli and c. b. park , _ reducing combinatoric uncertainties : a new technique based on @xmath34 variables _ , http://arxiv.org/abs/1109.2201[1109.2201 ] . j. alwall , k. hiramatsu , m. m. nojiri and y. shimizu , _ novel reconstruction technique for new physics processes with initial state radiation _ , _ phys . * 103 * ( 2009 ) 151802 [ http://arxiv.org/abs/0905.1201[0905.1201 ] ] . e. gross and o. vitells , _ transverse mass observables for charged higgs boson searches at hadron colliders _ , _ phys . * d81 * ( 2010 ) 055010 [ http://arxiv.org/abs/0907.5367[0907.5367 ] ] . b. gripaios , _ composite leptoquarks at the lhc _ , _ jhep _ * 02 * ( 2010 ) 045 [ http://arxiv.org/abs/0910.1789[0910.1789 ] ] . b. gripaios , a. papaefstathiou , k. sakurai and b. webber , _ searching for third - generation composite leptoquarks at the lhc _ , _ jhep _ * 01 * ( 2011 ) 156 [ http://arxiv.org/abs/1010.3962[1010.3962 ] ] . kim , _ algebraic singularity method for mass measurement with missing energy _ * 104 * ( 2010 ) 081601 [ http://arxiv.org/abs/0910.1149[0910.1149 ] ] . b. m. gripaios , in g. brooijmans _ et . _ , _ new physics at the lhc . a les houches report : physics at tev colliders 2009 - new physics working group _ , http://arxiv.org/abs/1005.1229[1005.1229 ] . t. han , i .- kim and j. song , _ kinematic cusps : determining the missing particle mass at colliders _ , _ phys . lett . _ * b693 * ( 2010 ) 575579 [ http://arxiv.org/abs/0906.5009[0906.5009 ] ] . k. agashe , d. kim , m. toharia and d. g. e. walker , _ distinguishing dark matter stabilization symmetries using multiple kinematic edges and cusps _ , _ phys . * d82 * ( 2010 ) 015007 [ http://arxiv.org/abs/1003.0899[1003.0899 ] ] . collaboration , _ search for neutral mssm higgs bosons decaying to tau+tau- pairs in proton - proton collisions at sqrt(s ) = 7 tev with the atlas detector _ , atlas - conf-2011 - 132 . collaboration , s. chatrchyan _ et . _ , _ search for neutral mssm higgs bosons decaying to tau pairs in @xmath56 collisions at sqrt(s)=7 tev _ * 106 * ( 2011 ) 231801 [ http://arxiv.org/abs/1104.1619[1104.1619 ] ] . a. elagin , p. murat , a. pranko and a. safonov , _ a new mass reconstruction technique for resonances decaying to di - tau _ , _ nucl . _ * a654 * ( 2011 ) 481489 [ http://arxiv.org/abs/1012.4686 [ 1012.4686 ] ] . collaboration , _ reconstruction and calibration of missing transverse energy and performance in z and w events in atlas proton - proton collisions at @xmath57 tev _ , atlas - conf-2011 - 080 . c. rogan , _ kinematical variables towards new dynamics at the lhc _ , http://arxiv.org/abs/1006.2727[1006.2727 ] . d. r. tovey , _ on measuring the masses of pair - produced semi - invisibly decaying particles at hadron colliders _ , _ jhep _ * 04 * ( 2008 ) 034 [ http://arxiv.org/abs/0802.2879[0802.2879 ] ] . c. g. lester , in b. allanach _ et . al . _ , _ les houches physics at tev colliders 2003 beyond the standard model working group : summary report _ , http://arxiv.org/abs/hep-ph/0402295[hep-ph/0402295 ] . b. webber , _ mass determination in sequential particle decay chains _ , _ jhep _ * 09 * ( 2009 ) 124 [ http://arxiv.org/abs/0907.5307 [ 0907.5307 ] ] . m. m. nojiri , k. sakurai and b. r. webber , _ reconstructing particle masses from pairs of decay chains _ , _ jhep _ * 06 * ( 2010 ) 069 [ http://arxiv.org/abs/1005.2532[1005.2532 ] ] . j. alwall , m .- le , m. lisanti and j. g. wacker , _ searching for directly decaying gluinos at the tevatron _ , _ phys . * b666 * ( 2008 ) 3437 [ http://arxiv.org/abs/0803.0019[0803.0019 ] ] . j. alwall , m .- le , m. lisanti and j. g. wacker , _ model - independent jets plus missing energy searches _ , _ phys . _ * d79 * ( 2009 ) 015005 [ http://arxiv.org/abs/0809.3264[0809.3264 ] ] . d. s. m. alves , e. izaguirre and j. g. wacker , _ where the sidewalk ends : jets and missing energy search strategies for the 7 tev lhc _ , http://arxiv.org/abs/1102.5338[1102.5338 ] . d. krohn , l. randall and l .- t . wang , _ on the feasibility and utility of isr tagging _ , http://arxiv.org/abs/1101.0810[1101.0810 ] . l. randall and d. tucker - smith , _ dijet searches for supersymmetry at the lhc _ , _ phys . lett . _ * 101 * ( 2008 ) 221803 [ http://arxiv.org/abs/0806.1049[0806.1049 ] ] . collaboration , g. aad _ et . _ , _ search for new phenomena in final states with large jet multiplicities and missing transverse momentum using @xmath57 tev pp collisions with the atlas detector _ , [ http://arxiv.org/abs/1110.2299[1110.2299 ] ] . j. alwall , a. freitas and o. mattelaer , _ the matrix element method and qcd radiation _ , _ phys . _ * d83 * ( 2011 ) 074010 [ http://arxiv.org/abs/1010.2263[1010.2263 ] ] . wang and i. yavin , _ a review of spin determination at the lhc _ , _ int _ * a23 * ( 2008 ) 46474668 [ http://arxiv.org/abs/0802.2726[0802.2726 ] ] . a. de roeck _ et . _ , _ from the lhc to future colliders _ , _ eur . phys . j. _ * c66 * ( 2010 ) 525583 [ http://arxiv.org/abs/0909.3240[0909.3240 ] ] . a. j. barr and c. g. lester , _ a review of the mass measurement techniques proposed for the large hadron collider _ , _ j.phys.g _ * g37 * ( 2010 ) 123001 [ http://arxiv.org/abs/1004.2732[1004.2732 ] ] . f. e. paige , _ determining susy particle masses at lhc _ , http://arxiv.org/abs/hep-ph/9609373[hep-ph/9609373 ] . m. m. nojiri , g. polesello and d. r. tovey , _ proposal for a new reconstruction technique for susy processes at the lhc _ , http://arxiv.org/abs/hep-ph/0312317[hep-ph/0312317 ] . k. kawagoe , m. m. nojiri and g. polesello , _ a new susy mass reconstruction method at the cern lhc _ , _ phys . * d71 * ( 2005 ) 035008 [ http://arxiv.org/abs/hep-ph/0410160[hep-ph/0410160 ] ] . cheng , j. f. gunion , z. han and b. mcelrath , _ accurate mass determinations in decay chains with missing energy : ii _ , _ phys . _ * d80 * ( 2009 ) 035020 [ http://arxiv.org/abs/0905.1344 [ 0905.1344 ] ] . w. s. cho , k. choi , y. g. kim and c. b. park , _ gluino stransverse mass _ , _ phys . * 100 * ( 2008 ) 171801 [ http://arxiv.org/abs/0709.0288[0709.0288 ] ] . b. gripaios , _ transverse observables and mass determination at hadron colliders _ , _ jhep _ * 02 * ( 2008 ) 053 [ http://arxiv.org/abs/0709.2740[0709.2740 ] ] . a. j. barr , b. gripaios and c. g. lester , _ weighing wimps with kinks at colliders : invisible particle mass measurements from endpoints _ , _ jhep _ * 02 * ( 2008 ) 014 [ http://arxiv.org/abs/0711.4008[0711.4008 ] ] . wang and i. yavin , _ spin measurements in cascade decays at the lhc _ , _ jhep _ * 04 * ( 2007 ) 032 [ http://arxiv.org/abs/hep-ph/0605296[hep-ph/0605296 ] ] . m. burns , k. kong , k. t. matchev and m. park , _ a general method for model - independent measurements of particle spins , couplings and mixing angles in cascade decays with missing energy at hadron colliders _ , _ jhep _ * 10 * ( 2008 ) 081 [ http://arxiv.org/abs/0808.2472[0808.2472 ] ] . g. f. giudice , b. gripaios and r. mahbubani , _ counting dark matter particles in lhc events _ , http://arxiv.org/abs/1108.1800 [ 1108.1800 ] . e. a. baltz , m. battaglia , m. e. peskin and t. wizansky , _ determination of dark matter properties at high - energy colliders _ , _ phys.rev . _ * d74 * ( 2006 ) 103521 [ http://arxiv.org/abs/hep-ph/0602187 [ hep - ph/0602187 ] ] . d. g. e. walker , _ dark matter stabilization symmetries and long - lived particles at the large hadron collider _ , http://arxiv.org/abs/0907.3142[0907.3142 ] . k. agashe , d. kim , d. g. e. walker and l. zhu , _ using @xmath34 to distinguish dark matter stabilization symmetries _ , http://arxiv.org/abs/1012.4460[1012.4460 ] . n. arkani - hamed _ et . _ , _ marmoset : the path from lhc data to the new standard model via on - shell effective theories _ , http://arxiv.org/abs/hep-ph/0703088[hep-ph/0703088 ] . d. alves _ et . _ , _ simplified models for lhc new physics searches _ , http://arxiv.org/abs/1105.2838[1105.2838 ]

we study axion - dilaton cosmologies derived from the low - energy string effective action . we present the classical homogeneous friedmann - robertson - walker solutions and derive the semi - classical perturbation spectra in the dilaton , axion and moduli fields in the pre - big bang scenario . by constructing the unique s - duality invariant field perturbations for the axion and dilaton fields we derive s - duality invariant solutions , valid when the axion field is time - dependent as well as in a dilaton - vacuum cosmology . whereas the dilaton and moduli fields have steep blue perturbation spectra ( with spectral index ) we find that the axion spectrum depends upon the expansion rate of the internal dimensions ( ) which allows scale - invariant ( ) spectra . we note that for the metric is non - singular in the conformal frame in which the axion is minimally coupled . sussex - th-97 - 001 + pu rcg-97/1 + hep - th/9701082 + _ phys rev d , in press _ + e. j. copeland , richard easther and david wands +centre for theoretical physics , university of sussex , brighton , bn1 9qh , u.k . +newman laboratory of nuclear studies , cornell university , ithaca ny14853 , u.s.a . +department of mathematics , university of waikato , private bag , hamilton , new zealand +department of physics , waseda university , 3 - 4 - 1 okubo , shinjuku - ku , tokyo 169 , japan +school of mathematical studies , university of portsmouth , portsmouth po1 2eg , u.k .

hep-th9701082

the dramatic progress that has been claimed in understanding black holes in the context of string theory @xcite in the past year focuses attention upon the implications that strings might have for cosmology . the early universe provides a natural arena in which to seek observational evidence for string theory as a fundamental theory . general relativity describes all currently observed gravitational physics with remarkable accuracy @xcite , but its description of the very early universe is expected to break down near the planck scale , and possibly at lower energies as well . most investigations of the cosmological consequences of string theory have focussed on the role of the dilaton field , @xmath10 , which provides a varying effective gravitational constant . in the absence of other matter , the low energy action @xcite leads to an effective theory of gravity of the form proposed by brans and dicke @xcite . although the predicted value of the brans - dicke parameter @xmath11 is incompatible with present day post - newtonian tests @xcite , it may be that loop corrections yield an experimentally acceptable general relativistic limit @xcite , or that the dilaton s present day value may be fixed by it acquiring a mass ( for a detailed list of references see @xcite ) . the pre - big bang scenario @xcite is a specific example of a cosmological model based on string theory which differs radically from that predicted by general relativity because of the presence of the dilaton field . the weakly coupled , expanding , `` pre - big bang '' solutions have many similarities with conventional inflation models @xcite , most notably the approach to flatness and homogeneity on large scales by stretching the quantum vacuum state on small scales up to large ( super - horizon ) scales . solutions derived from the low - energy effective action run into a `` big bang '' singularity @xcite but the hope is that in the full string theory , including higher - order terms , the pre - big bang solution may be smoothly connected to a solution describing an expanding , `` post - big bang '' universe @xcite and , ultimately , general relativistic evolution . our intention in this paper is to draw attention to the crucial role that the antisymmetric tensor field , @xmath12}$ ] , may have in cosmological scenarios based on the low - energy limit of the superstring action . an antisymmetric tensor inevitably appears , along with the graviton and dilaton , in the neveu - schwarz bosonic sector of the low - energy string effective action , @xmath13 where @xmath14 , and @xmath15 is newton s constant in @xmath16-dimensions . previous studies @xcite-@xcite have considered the classical evolution of the antisymmetric tensor field in simple cosmologies . the role of the additional tensor fields derived from the ramond - ramond sector of type iia and type iib string theory has recently been considered in @xcite . in this paper we calculate for the first time the spectrum of perturbations about the classical background field that may be produced due to vacuum fluctuations in the antisymmetric tensor field . we will consider spacetimes which contain a four dimensional homogeneous and isotropic external metric . in four spacetime dimensions the antisymmetric tensor field has only one degree of freedom which may be represented by a pseudo - scalar axion field , @xmath17 . for a homogeneous and isotropic metric we must have a homogeneous axion field , @xmath18 this has been referred to as the solitonic anstaz due to the close connection with solitonic p - brane solutions @xcite . an alternative ansatz , analogous to elementary p - brane solutions @xcite , taking the tensor potential @xmath19 to be time - dependent leads to an inhomogeneous axion field and hence will be incompatible with a @xmath20 friedmann - robertson - walker ( frw ) cosmology @xcite . in section 2 we review the classical evolution of frw cosmologies both with and without an evolving axion , stressing the important role of the axion when the scale factor in the string frame becomes small . incorporating the evolution of the axion leads to a cosmology that is very different from the classical pre - big bang scenario where the axion field is fixed . we explore the differences between the cosmological evolution as seen in different conformally related metrics in section 3 , drawing attention to the evolution in the axion frame ( where the axion field is minimally coupled ) . specifically , we find that the singular evolution of the metric in the string frame can be non - singular in the axion frame . these axion - dilaton solutions of the low - energy action can be related by a duality transformation which reduces to a scale factor duality in the absence of the axion field , as described in section 4 . in section 5 we set out our formalism for describing inhomogeneous linear perturbations about the homogeneous and isotropic 4-d background solutions . even when the background axion field is set to zero , there will inevitably be quantum fluctuations in the field . in section 6 we calculate the spectrum of semi - classical axion perturbations as well as dilaton and moduli perturbation spectra produced in the pre - big bang scenario . in section 7 we extend this calculation to more general axion - dilaton cosmologies by constructing s - duality invariant combinations of the field perturbations that enable us to derive s - duality invariant solutions . moreover we demonstrate that the late - time dilaton and axion spectra turn out to be independent of the preceding evolution along different but s - duality related classical solutions . importantly , the tilt of the axion spectrum can be significantly different from the steep `` blue '' spectra of dilatons and gravitons predicted by the pre - big bang scenario . here we consider cosmological solutions derived from the low - energy string action where we take the full @xmath16-dimensional spacetime to have a metric of the form @xmath21 @xmath22 , @xmath23 run from @xmath24 to @xmath25 , and @xmath26 , @xmath27 run from @xmath24 to @xmath28 . we will allow for the variation of the @xmath29 compactified dimensions by including a single modulus field , @xmath30 , proportional to the volume of this internal space , but neglect any curvature or anisotropy so that @xmath31 . the effective dilaton in the four dimensional external spacetime is then @xmath32 and the antisymmetric tensor field in four dimensions can be written in terms of the pseudo - scalar axion field @xmath17 as @xmath33 where @xmath34 is the covariant antisymmetric 4-form such that @xmath35 . the low - energy string effective action , given in eq . ( [ dilaction ] ) , then becomes @xmath36 where @xmath37 determines the effective value of the planck mass when @xmath38 , and @xmath39 is the ricci scalar of the four dimensional external spacetime . we assume the external four dimensional spacetime is described by a flat frw metric with the line element @xmath40 where @xmath41 is the scale factor . in addition , frw solutions with non - zero spatial curvature can also be found @xcite . to be compatible with a homogeneous and isotropic metric , all the fields must be homogeneous and the action then reduces to ( up to a total derivative ) @xmath42 we refer to models with a constant axion field ( @xmath43 ) as _ dilaton - vacuum _ solutions . these are the well - known monotonic power - law solutions . ] @xmath44 where the integration constants @xmath45 and @xmath46 determine the rate of change of the effective dilaton and internal volume respectively . note that there is a constraint equation ) in the next section . ] which requires @xmath47 the dilaton - vacuum solutions are shown in figs . 1 to 3 . if stable compactification has occurred and the volume of the internal spaces is fixed ( @xmath48 , or @xmath20 ) we have @xmath49 . these dilaton - vacuum solutions can be expressed in terms of the proper time , @xmath50 , giving @xmath51 all these solutions have semi - infinite proper lifetimes . those starting from a singularity at @xmath52 for @xmath53 , are denoted the @xmath54 branch in ref . @xcite , while those which approach a singularity at @xmath52 for @xmath55 are referred to as the @xmath56 branch . our choice of origin for the time coordinate is arbitrary . a more fundamental definition of the @xmath57 branches may be given by considering the evolution of the shifted dilaton @xcite @xmath58 its time derivative , @xmath59 is always positive on the @xmath56 branch ( or @xmath60 ) due to the constraint on the value of @xmath61 [ eq . ( [ r ] ) ] , and always negative on the @xmath54 branch ( or @xmath62 ) . these @xmath57 branches do _ not _ refer to the choice of sign for @xmath61 in eq . ( [ r ] ) . on either the @xmath56 or @xmath54 branches of the dilaton - vacuum cosmologies we still have two distinct solutions corresponding to the choice of the exponent @xmath63 or @xmath64 , which determines whether @xmath65 goes to negative or positive infinity , respectively , as @xmath66 . the generalisation of eqs.([dilphi][dilbeta ] ) to the more general axion - dilaton cosmology where @xmath67 is particularly simple @xcite : @xmath68 where the exponents are related via @xmath69 . the time - dependent axion solutions are plotted in figs . 1 to 4 . the presence of the axion places a lower bound on the value of the dilaton , @xmath70 . in doing so it interpolates between two dilaton - vacuum solutions with an asymptotically constant axion field . when @xmath71 the solutions approach the @xmath72 dilaton - vacuum solution and as @xmath66 the solution approaches the @xmath73 dilaton - vacuum solution . we shall see that the asymptotic approach to dilaton - vacuum solutions at early and late times leads to a particularly simple form for the semi - classical perturbation spectra , independent of the intermediate evolution . the dynamical effect of the axion field is negligible except near @xmath74 , when it leads to a bounce in the dilaton , @xmath75 . et al _ @xcite have recently drawn attention to the connection between these cosmological solutions and solitonic @xmath76-brane solutions . if @xmath77 then this also leads to a bounce of the scale factor , @xmath78 . however we still have the two disconnected branches , as defined by eq . ( [ defbranch ] ) , corresponding to an increasing shifted dilaton , approaching a singularity on the @xmath56 branch , or a decreasing shifted dilaton , on the @xmath54 branch . thus far we have written all the solutions in terms of the string frame . if stringy matter is minimally coupled in this frame then stringy test particles will follow geodesics with respect to this metric . however , in order to understand the physical evolution in these models it is revealing to look at conformally related metrics , @xmath79 . if the conformal factor @xmath80 is itself homogeneous then the transformed metric remains a frw metric but with scale factor @xmath81 . this transformation of the scale factor can lead to some ambiguities in the interpretation of the cosmological solutions @xcite . for instance inflation is often defined as accelerated expansion ( @xmath82 ) . however such a definition is dependent upon the choice of conformal frame in which one chooses to evaluate the acceleration . note that the proper time also changes under a conformal transformation , @xmath83 . one must not assume that a finite proper time interval in one frame necessarily coincides with a finite time in another frame and , in particular , we shall see that what looks like a singular evolution in one frame may appear non - singular in another frame . by choosing a conformal factor @xmath84 we can work in a frame @xmath85 where the dilaton is minimally coupled to the external metric . thus the gravitational part of the action in eq . ( [ 4daction ] ) reduces to the einstein - hilbert action @xcite @xmath86 and hence this is known as the einstein frame - dimensional metric is minimally coupled to the dilaton as the einstein frame . this corresponds to a kaluza - klein gravity theory , which will not in general coincide with einstein gravity in the four - dimensional external metric . ] . both the dilaton and the moduli fields have standard kinetic terms in this frame and thus moduli particles and dilatons ( in a constant axion field ) would therefore follow geodesics in the four dimensional external spacetime of the einstein frame . the familiar field equations of general relativity @xmath87 apply in this frame . this may help one s cosmological intuition , which is rooted in four dimensional general relativity , but also assists mathematically by decoupling the equations for the evolution of the metric from the value of @xmath65 . the stress - energy tensor @xmath88 for homogeneous fields reduces to that for a perfect fluid with a stiff equation of state @xcite , i.e. , pressure equal to density , @xmath89 the evolution equations for homogeneous fields in an frw metric are then @xmath90 plus the constraint @xmath91 where @xmath92 . from eqs . ( [ hprime ] ) and ( [ h ] ) we see that in the general axion - dilaton case where we allow the dilaton , axion and/or moduli fields to evolve , the expansion in the einstein frame always obeys @xmath93 leading to the simple solution for the scale factor @xmath94 the scalar field equations of motion , eqs . ( [ phieom][betaeom ] ) , can then be integrated to give the solutions presented in eqs . ( [ axiphi][axisigma ] ) . even in spatially curved frw models the equations of motion remain integrable @xcite despite the apparently non - trivial couplings between the fields , because we can make a conformal transformation to the einstein frame where all the fields are minimally coupled to the metric and , so long as they are all homogeneous , their combined dynamical effect is no different from a single massless field @xcite . because there is no interaction potential for the fields , the strong energy condition ( @xmath95 ) is always satisfied [ see eq . ( [ rho ] ) ] and the general relativistic singularity theorems must hold . thus we know there is no way to construct a non - singular evolution in the einstein frame with these massless fields . in the string frame the usual general relativistic results do not hold . even without an interaction potential we can obtain an accelerated expansion in eq . ( [ dilat ] ) for the @xmath56 dilaton - vacuum branch ( @xmath96 ) with @xmath97 ( or @xmath77 in the axion - dilaton solution ) . however unlike conventional power - law inflation , @xmath98 with @xmath99 , we have `` pole - inflation '' with @xmath100 , and we approach a curvature singularity with @xmath101 and @xmath102 as @xmath103 . in the einstein frame we see that @xmath66 on the @xmath56 branch always corresponds to a collapsing universe with @xmath104 . however this still fulfils one definition of inflation , namely , that the comoving hubble length ( @xmath105 inevitably becomes larger than the hubble scale in that frame as @xmath66 . this allows one to produce perturbations in the dilaton , moduli and graviton fields on scales much larger than the present hubble scale from quantum fluctuations in flat spacetime at earlier times , as we shall discuss in more detail later . in both the string frame and the einstein frame we either reach a curvature singularity in a finite proper time in the future for @xmath60 or emerge from a curvature singularity at a finite time in the past for @xmath62 the only exception to this is the axion - dilaton solutions in the string frame when @xmath106 ( @xmath107 ) in eqs . ( [ axiphi][axisigma ] ) which bounce at exactly @xmath108 @xcite . however even in this case the dilaton and moduli become infinite at a finite proper time . both the dilaton and moduli fields are minimally coupled in the einstein frame ( i.e. they have standard kinetic terms ) . however the axion s kinetic term retains a non - minimal coupling to the dilaton . this can be removed by a conformal transformation to another conformally related metric , the `` axion frame '' , given by @xmath109 and hence @xmath110 the axion field is a minimally coupled massless scalar field in this frame and thus axionic test particles would follow geodesics with respect to this metric . although conformally related to the string and einstein frames , the metric the axions `` see '' may behave very differently from the metrics in the string or einstein frames . in terms of conformal time , the axionic scale factor for the dilaton - vacuum solutions given by eqs . ( [ dilphi ] ) and ( [ dila ] ) when @xmath43 , evolves as @xmath111 we see that the proper time in the axion frame is given by @xmath112 so it takes an infinite proper time to reach @xmath108 for @xmath113 ( and thus @xmath114 ) and the scalar curvature for the axion metric , @xmath115 , vanishes as @xmath66 . however , these same dilaton - vacuum solutions then reach @xmath71 in a finite proper time where @xmath116 diverges . because the conformal factor diverges as @xmath66 it stretches out the curvature singularity in the string metric into a non - singular evolution in the axion frame . but as @xmath117 as @xmath71 the non - singular evolution in the string frame gets compressed into a curvature singularity in the axion frame . similar behaviour has previously been noted in the case of black holes in the low - energy limit of string theory @xcite . astronauts made of axionic matter falling into an axion - dilaton black hole in @xmath20 would take an infinite proper time ( measured by their axionic clocks ) to fall into what appears , in the einstein frame , to be the singularity . in terms of the proper time @xmath118 in the axion frame we have @xmath119 for @xmath113 we have conventional power - law inflation ( not pole - inflation ) with @xmath120 with @xmath121>1 $ ] . we shall see that this has important consequences for the tilt of the power - spectrum of semi - classical perturbations in the axion field produced on large scales . these dilaton - vacuum solutions still have a curvature singularity as @xmath122 so the solutions still have only a semi - infinite lifetime in the axion frame is an exception as they correspond to de sitter expansion with @xmath123 . ] , but for @xmath113 this now coincides with @xmath71 , so the identification of the @xmath56 and @xmath54 branches as solutions approaching or leaving a singularity is interchanged for the @xmath64 solution in the axion frame when @xmath113 . however this implies that the axion - dilaton solutions with a time - dependent axion field will be non - singular in the axion frame if @xmath124 . remember that the axion - dilaton solutions with @xmath67 match @xmath73 dilaton - vacuum solutions at @xmath66 onto @xmath72 solutions as @xmath125 . thus for @xmath124 , the axion - dilaton solutions are non - singular in the axion frame as @xmath66 ( because @xmath126 ) _ and _ non - singular as @xmath127 ( because @xmath128 ) . the general evolution for the axion - dilaton system , eqs . ( [ axiphi][axisigma ] ) is given in the axion frame by eq . ( [ bara ] ) , so @xmath129 from this we can extract the proper time , @xmath118 , @xmath130\ ] ] which , as shown above , is semi - infinite ( @xmath131 ) for @xmath132 but unbounded ( @xmath133 ) for @xmath134 . equations ( [ generalaxionsf ] ) and ( [ genaxiontime ] ) give us a parametric solution for the axion frame scale factor in terms of the proper time in that frame . representative examples showing the behaviour of @xmath135 for different values of @xmath45 are given in fig . note that the scale factor , @xmath136 , has a non - zero minimum value ( i.e. , a bounce ) whenever @xmath137 . when @xmath124 , @xmath136 becomes infinite as @xmath138 , passes through a non - zero minimum value and then expands indefinitely as @xmath139 . in particular , if stable compactification has occured so that the moduli field is fixed ( @xmath48 ) , or if @xmath20 ( so that @xmath140 ) , then @xmath141 and the axion - dilaton solution is _ always _ non - singular in the axion frame . when @xmath142 , @xmath136 does have a bounce but is singular , since it becomes infinitely large in a finite proper time , as @xmath122 . finally , when @xmath143 , the solutions are monotonic and there is a singularity when @xmath136 vanishes at @xmath144 . the constant axion solutions given in eqs . ( [ dilphit])([dilbetat ] ) are related by the scale factor duality transformation @xcite @xmath145 which corresponds to a change in the parameters @xmath146 in eqs . ( [ dilphit][dilbetat ] ) . this is a particular case of a more general @xmath147 duality @xcite where the axion field remains constant . when we can neglect the evolution of the moduli fields ( @xmath48 and hence @xmath49 ) this coincides with @xmath148 . note that this scale factor duality does _ not _ take one from the @xmath56 to @xmath54 branch or vice versa . this would require a time reversal . the pre - big bang scenario postulates a non - singular universe by linking the semi - infinite lifetime expanding @xmath56 branch , @xmath149 , starting at @xmath150 to the semi - infinite expanding @xmath54 branch , @xmath151 , travelling off to @xmath152 via a scale factor duality transformation , plus time reversal near , the singularity at @xmath153 . the presence of a time - dependent axion field @xmath154 ( the solitonic ansatz ) breaks the @xmath147-invariance which requires that it is the antisymmetric potential @xmath155 which is homogeneous ( the elementary ansatz ) . the elementary ansatz is only compatible for a restricted class of metrics in anisotropic 4-d spacetimes @xcite , or if the axion is constant . solutions with a time - dependent axion field do respect the invariance of the low energy string action under the @xmath156 transformation : @xmath157 where @xmath158 , @xmath159 , @xmath160 and @xmath161 are real constants subject to @xmath162 , and @xmath163 is the complex dilaton field @xmath164 this leads to @xmath165 in the underlying string theory this represents the modular invariance of the complex dilaton @xcite , but working only with the classical fields in the low energy action , it represents a transformation between the dilaton and axion fields which leaves invariant @xmath166 in terms of @xmath65 and @xmath17 , or indeed @xmath163 , it is not immediately apparent that @xmath167 should remain invariant under the transformation given in eq . ( [ sduality ] ) . it is rather more transparent if we define the matrix @xmath168 which obeys @xmath169 , where @xmath170 and thus is a member of @xmath156 . the particular @xmath156 transformation given in eq . ( [ sduality ] ) is given by @xmath171 where @xmath172 is also a member of @xmath156 . then we can write @xmath173tr@xmath174 and , noting that @xmath175 , it is straightforward to verify that @xmath167 is invariant . we will also find this notation particularly convenient later to construct explicitly s - duality invariant dilaton and axion field perturbations . the lagrange density of the axion and dilaton fields in the einstein frame @xmath176 is s - duality invariant and , because the dilaton and axion are minimally coupled to the other fields in the einstein frame , the evolution of the moduli field , @xmath159 , and scale factor , @xmath177 , are unaffected by the s - duality transformation . however the scale factor in the original string frame must transform and will not remain invariant under a non - trivial transformation . if we choose @xmath178 , when @xmath179constant , the solutions given in eqs . ( [ dilphi])([dilbeta ] ) are mapped by the transformation in eq . ( [ sduality ] ) to @xmath180 which leaves @xmath17 constant . in particular , for @xmath181 we have @xmath182 and hence this is a transformation between strong and weak coupling . the form of the solutions given in eqs . ( [ dilphi])([dilbeta ] ) are unchanged , but the parameters @xmath183 comparing with eq . ( [ rsfdual ] ) , we find that in the particular case when @xmath184 , and hence @xmath49 , this coincides with the scale factor duality given in eq . ( [ sfdual ] ) . the more general s - duality transformations of @xmath185 and @xmath17 given in eqs . ( [ saxiphi ] ) and ( [ saxisig ] ) can be shown to relate the dilaton - vacuum cosmologies , given in eqs . ( [ dilphi][dilbeta ] ) , to the more general axion - dilaton cosmologies with a time dependent axion field , given in eqs . ( [ axiphi][axisigma ] ) , with a fixed value of @xmath186 . thus the s - duality transformation allows one to generate the general axion - dilaton solutions with a given value of @xmath45 starting from only with the dilaton - vacuum solution with @xmath187 . thus far we have considered only homogeneous classical solutions to the equations of motion . in the next section we will consider inhomogeneous perturbations that may be generated due to vacuum fluctuations . in order to follow their evolution we will set up in this section the formalism required to describe linear perturbations about the homogeneous background metric . we shall consider perturbations of the four - dimensional metric in the spatially flat gauge ( or in more general frw models , the uniform spatial curvature gauge @xcite ) , using the einstein frame , so that to first - order the perturbed line element can be written as @xmath188 dx^i dx^j \right\ } \ , , \ ] ] where @xmath189 and @xmath190 are the scalar metric perturbations ( in the notation of ref . @xcite ) and @xmath191 represents a transverse and traceless tensor perturbation . linear perturbations about the homogeneous background fields can be decomposed as a sum of fourier modes with comoving wavenumber @xmath192 ( and in the case of the tensor perturbations two independent polarisations ) which evolve independently of other wavenumbers . the advantage of splitting the metric perturbations into scalar and tensor parts is that the scalar and tensor modes evolve independently to first order with only the scalar perturbations being coupled to scalar field fluctuations @xcite . in the spatially flat gauge we have the added simplification that the evolution equations for linear perturbations about homogeneous scalar fields are decoupled from the metric perturbations , although they are still related by a constraint equation . the field equations for the linearised scalar perturbations are @xmath193 plus the constraints @xmath194 note that the scalar metric perturbations are not invariant under a conformal transformation . even the spatially flat nature of the line element in eq . ( [ dds ] ) is not preserved under a conformal transformation back to the string frame due to the first - order perturbation in the conformal factor @xmath195 . however the tensor perturbation remains invariant under both conformal transformations and gauge transformations @xmath196 . the evolution equation for the scalar metric perturbations , eq . ( [ aeom ] ) is independent of the evolution of the different scalar fields and is dependent only on the evolution of the einstein frame scale factor @xmath197 given by eq . ( [ esf ] ) . this in turn is determined solely by the stiff fluid equation of state for the homogeneous fields in the einstein frame , regardless of the time dependence of the axion field . equation ( [ aeom ] ) can be integrated to give the general solution @xmath198 \ , , \ ] ] where @xmath199 and @xmath200 are hankel functions of the first and second kind . using the recurrence relation between bessel functions , we obtain from eqs . ( [ aofb ] ) and ( [ gena ] ) , @xmath201 \ , .\ ] ] our scalar metric perturbations can be written in terms of the gauge invariant metric potentials @xcite @xmath202 note that the gauge transformation @xmath203 brings the metric of eq . ( [ dds ] ) into the more commonly used longitudinal gauge @xcite where @xmath204 dx^i dx^j \right\ } \ , .\end{aligned}\ ] ] the curvature perturbation on uniform energy density hypersurfaces ( as @xmath205 ) is commonly denoted by @xmath206 @xcite and is given by @xmath207 and hence with @xmath208 given by eq . ( [ esf ] ) for the scale factor in the einstein frame , we have @xmath209 in any dilaton - vacuum or axion - dilaton cosmology . @xmath206 is a particularly useful quantity to calculate as it becomes constant on scales much larger than the hubble scale ( @xmath210 ) for purely adiabatic perturbations , even through changes in the equation of state . in single - field inflation models this allows one to compute the density perturbation at late times , during the matter or radiation dominated eras , by equating @xmath206 at `` re - entry '' ( @xmath211 ) with that at horizon crossing during inflation . thus previous studies have calculated the spectrum of @xmath189 , and hence @xmath206 , in order to predict the density perturbations induced in the pre - big bang scenario @xcite . however , the situation is not really so straightforward in the pre - big bang scenario as in single - field inflation , because in the full low - energy string effective action there will be many fields present which can lead to non - adiabatic perturbations . we must be aware of the fact that density perturbations at late times may not be simply related to @xmath206 alone , but may also be dependent upon fluctuations in other fields . one such field is the axion field , and we shall see that it may have a markedly different spectrum from @xmath206 . the scalar field perturbations themselves transform under the gauge transformation @xmath196 giving @xmath212 . thus the scalar field perturbations in the longitudinal gauge ( @xmath213 ) are related to those in the spatially flat gauge ( @xmath214 ) under the gauge transformation in eq . ( [ gt ] ) as @xmath215 fortunately , the gravitational wave perturbations @xmath191 are both gauge and conformally invariant . they decouple from the scalar perturbations in the einstein frame to give a simple evolution equation for each fourier mode @xmath216 the growing mode in the long wavelength ( @xmath217 ) limit is @xmath218 . ( we have not considered gravitational waves propagating in the @xmath29 internal dimensions . see ref . the spectrum depends solely on the dynamics of the scale factor in the einstein frame given in eq . ( [ esf ] ) , which as we have seen is the same regardless of the time - dependence of the moduli or axion fields . it leads to a spectrum of primordial gravitational waves steeply growing on short scales , with a spectral index @xmath219 @xcite . this is in contrast to conventional inflation models which require @xmath220 @xcite . the graviton spectrum appears to be a robust and distinctive prediction of any pre - big bang type evolution based upon the low - order string effective action . this has been discussed extensively elsewhere @xcite , so we now turn to discuss in more detail the spectra corresponding to scalar perturbations . while the solutions for the homogeneous dilaton , axion and scale factor in the different frames may lead to interesting behaviour in the early universe , the success of the standard big bang model suggests that the evolution should closely approach the conventional general relativistic evolution at least by the time of nucleosynthesis . if we are to see any trace of the earlier evolution it will be in the primordial spectrum of inhomogeneities present on large - scales that we observe today . such large - scale structure can only be generated by some unconventional physics , such as inflation , topological defects or a pre - big bang epoch . during a period of accelerated expansion the comoving hubble length @xmath221 decreases and vacuum fluctuations which are assumed to start in the flat - spacetime vacuum state may be stretched up to exponentially large scales . the precise form of the spectrum depends on the expansion of the homogeneous background and the couplings between the fields . we have seen that the comoving hubble length does indeed decrease in the einstein frame during the contracting phase when @xmath60 . because the dilaton , moduli fields and graviton are minimally coupled to this metric , this ensures that small - scale vacuum fluctuations will eventually be stretched beyond the comoving hubble scale during this epoch . the production of scalar and tensor metric perturbations in the pre - big bang scenario has been studied by various authors ( see for example @xcite ) . as we remarked earlier , the axion field is taken to be a constant in these solutions . however , while a constant axion field may be a consistent particular solution when describing the background classical field , one can not necessarily neglect quantum fluctuations in this field . in this section we will consider for the first time the production of axions during a pre - big bang type evolution ( where the background axion field is constant ) and then go on to discuss the perturbation spectrum in the more general case with @xmath67 . we will also analyse the behaviour of these cosmological vacuum states to first - order under s - duality transformations . first of all , let us consider the perturbation spectra produced when the background axion field remains constant , @xmath43 . the evolution of the homogeneous background fields is given in eqs . ( [ dilphi][dilbeta ] ) and the dilaton and moduli fields both evolve as minimally coupled massless fields in the einstein frame . in particular , the dilaton perturbations are decoupled from the axion perturbations and the equations of motion in the spatially flat gauge , eq . ( [ dphieom][dbetaeom ] ) , become @xmath222 plus we have the constraint eq . ( [ aofb ] ) @xmath223 from eq . ( [ pbbacon ] ) we see that , to first - order , the metric perturbation , @xmath189 , is determined solely by the dilaton and moduli field perturbations . the canonically normalised field perturbations are @xcite @xmath224 which , from eqs . ( [ pbbdphieom ] ) and ( [ pbbdbetaeom ] ) , obey the wave equations @xmath225 after inserting the simple solution for the einstein frame scale factor given in eqs . ( [ esf ] ) we find that these equations give the general solutions @xmath226 \ , , \\ w & = & |k\eta|^{1/2 } \left [ w_+ h_0^{(1)}(|k\eta| ) + w_- h_0^{(2)}(|k\eta| ) \right ] \ , .\end{aligned}\ ] ] on the @xmath56 branch , i.e. , when @xmath60 , we can normalise modes at early times , @xmath227 , where all the modes are far inside the hubble scale , @xmath228 , and can be assumed to be in flat - spacetime vacuum . just as in conventional inflation , this produces perturbations on scales far outside the horizon , @xmath229 , at late times , @xmath66 . conversely , the solution for the @xmath54 branch with @xmath62 is dependent upon the initial state of modes far outside the horizon , @xmath229 , at early times where @xmath66 . the role of a period of inflation , or of the pre - big bang @xmath56 branch , is precisely to set up this initial state which otherwise appears as a mysterious initial condition in the conventional ( non - inflationary ) big bang model . allowing only positive frequency modes in the flat - spacetime vacuum state at early times for the pre - big bang @xmath56 branch requires @xcite that , as @xmath230 , @xmath231 and similarly for @xmath232 , giving @xmath233 the power spectrum for perturbations is commonly denoted by @xmath234 and thus for modes far outside the horizon ( @xmath205 ) we have @xmath235 ^ 2 \,,\\ \label{pbbbeta } { \cal p}_{\delta\beta } & \to & { 2\over n\pi^3 } \kappa^2\widetilde{h}^2 ( -k\eta)^3[\ln(-k\eta)]^2 \,,\end{aligned}\ ] ] where @xmath236 is the hubble rate in the einstein frame , and recall @xmath29 is the number of compact dimensions . the amplitude of the perturbations grows towards small scales , but only becomes large for modes outside the horizon ( @xmath237 ) when @xmath238 , i.e. , the planck scale in the einstein frame . the spectral tilt of the perturbation spectra is given by @xmath239 which from eqs . ( [ pbbphi ] ) and ( [ pbbbeta ] ) gives @xmath240 ( where we neglect the logarithmic dependence ) . we need also to compute the amplitude of the scalar metric perturbations , to check the validity of our linear perturbation analysis . normalising the amplitude of the spectrum for the metric perturbation @xmath189 in eq . ( [ gena ] ) from the constraint eq . ( [ aconstraint ] ) , using eqs . ( [ dila][dilbeta ] ) for the background fields and eqs.([pbbphi ] ) and ( [ pbbbeta ] ) for their perturbations , we have @xmath241 ^ 2 \,.\ ] ] [ remember that we are adding independent random variables . the @xmath25 comes from @xmath242 . ] note that this spectrum of scalar metric perturbations in entirely independent of the integration constants that parameterise the solutions given in eqs . ( [ dila][dilbeta ] ) . the scalar spectrum , just like the spectrum of tensor perturbations is a robust prediction of any pre - big bang scenario where the universe collapses in the einstein frame , and becomes dominated by homogeneous scalar fields . just like the field perturbations , the scalar metric perturbations have a steep blue spectrum , @xmath243 , which becomes large on superhorizon scales @xmath237 only near the planck scale , @xmath238 . note that bardeen s gauge invariant perturbations @xmath244 and @xmath245 , defined in eqs . ( [ aphipsi ] ) and ( [ bphipsi ] ) , actually become large much earlier @xcite , but the fact that the perturbations remain small in our choice of gauge implies that our linear calculation is in fact valid up until the planck epoch @xcite . unfortunately this leaves us with such a steeply tilted spectrum of metric perturbations that there would be effectively no primordial metric perturbations on large ( super - galactic ) scales in our present universe if the post - big bang era began close to the planck scale . the metric fluctuations are of order unity on the planck scale ( @xmath246 cm ) when @xmath247k in the standard post - big bang model . this corresponds to a comoving scale of about @xmath248 cm today ( when @xmath249k ) , about @xmath250 times the scale of perturbations observed on the microwave background sky . thus the microwave background temperature anisotropies should be of order @xmath251 rather than the observed @xmath252 . however , it turns out that the presence of the axion field could provide an alternative spectrum of perturbations more suitable as a source of large - scale structure . while the dilaton and moduli fields evolve as massless minimally coupled fields in the einstein frame , the axion evolves as a massless minimally coupled field in the axion frame and the canonically normalised field perturbation is @xmath253 in this section we are considering the axion spectrum in the pre - big bang scenario where the background axion field is constant . as a result density perturbations are only second - order in the axion perturbation and so we can neglect the back - reaction from the metric to linear order . the field perturbation @xmath254 is gauge invariant when @xmath43 [ see eq . ( [ gaugephi ] ) ] and in any gauge , the axion perturbation obeys the decoupled wave equation given in eq . ( [ pbbdsigmaeom ] ) which can be re - written in terms of @xmath255 as @xmath256 as we have just mentioned , whereas the dilaton and moduli evolve as massless minimally coupled fields in the einstein frame , the axion is minimally coupled in the axion frame , whose evolution given in eq . ( [ axionsf ] ) is significantly different . in fact , substituting eq . ( [ axionsf ] ) in eq . ( [ vpp ] ) we have @xmath257 \ , , \ ] ] where we have used @xmath258 . once again , we can only normalise this using the flat spacetime vacuum state at early times as @xmath259 on the @xmath56 branch , as in eq . ( [ shortwave ] ) , which gives @xmath260 and hence we have @xmath261 at late times , as @xmath262 , we find and @xmath263 the dilaton remains constant and the axion frame and einstein frame coincide , upto a constant factor . thus the axion spectrum behaves like that for the dilaton and moduli fields and the late time evolution in this case is that logarithmic with respect to @xmath264 , as given in eqs . ( [ pbbphi ] ) and ( [ pbbbeta ] ) . ] @xmath265 where the numerical coefficient @xmath266 approaches unity for @xmath267 . the expression for the axion power spectrum can be written in terms of the field perturbation when each mode crosses outside the horizon @xmath268 ^ 2 \left ( { \bar{h}_c \over 2\pi } \right)^2 \ , , \ ] ] where @xmath269 is the hubble rate when @xmath270 . this is the power spectrum for a massless scalar field during power - law inflation which approaches the famous result @xmath271 as @xmath272 , and the expansion in the axion frame becomes exponential arises due to our dimensionless definition of @xmath17 . ] . more importantly , the spectral index @xmath273 depends crucially upon the value of @xmath186 . the spectrum becomes the classic scale - invariant harrison - zeldovich spectrum as @xmath274 . the lowest possible value of the spectral tilt @xmath275 is @xmath276 which is obtained when stable compactification has occurred and the moduli field @xmath159 is fixed . the more rapidly the internal dimensions evolve , the steeper the resulting axion spectrum until for @xmath277 and @xmath263 we have @xmath278 like the dilaton and moduli spectra . note that the condition for a negatively tilted spectrum coincides exactly with the requirement for conventional power - law inflation , rather than pole inflation , in the axion frame . of course , when the background axion field is constant these perturbations , unlike the dilaton or moduli perturbations , do not affect the scalar metric perturbations ( i.e. , these are isocurvature perturbations ) . however , if the axion field does affect the energy density at late times ( for instance , by the axion field acquiring a mass ) then the spectrum of density perturbations need not have a steeply tilted blue spectrum like the dilaton perturbations , but rather could have a nearly scale - invariant spectrum as required for large - scale structure formation @xcite . when we allow the background homogeneous axion field to be time - dependent we must allow for the interaction between the dilaton and axion field and the metric to first - order . in fact we have seen that in the spatially flat gauge the evolution equations for both the scalar and tensor metric perturbations [ eqs . ( [ aeom ] ) and ( [ heom ] ) ] are independent of the evolution of the different scalar fields and are determined solely by the evolution of the einstein scale factor given in eq . ( [ esf ] ) . because the moduli field perturbations remain decoupled from both the axion and dilaton , their evolution equation , eq . ( [ pbbdbetaeom ] ) , is also unaffected . thus the spectral tilts of the scalar and tensor metric perturbations and the moduli spectrum , eq . ( [ pbbbeta ] ) , remain the same as in the pre - big bang scenario . we can understand this in terms of the s - duality transformations that relate the general axion - dilaton solutions to the dilaton - vacuum solutions . these transformations leave the einstein frame metric and moduli field invariant and thus not only the homogeneous fields , but also their perturbations , are identical in s - duality related cosmologies . however the dilaton and axion fields and their perturbations will in general be affected by s - duality transformations . the dilaton and axion perturbation field equations ( [ dphieom ] ) and ( [ dsigmaeom ] ) become coupled to first order when @xmath67 , and the chances of obtaining analytic solutions might appear to be remote . however , we can exploit the s - duality symmetry which relates the general axion - dilaton cosmologies to the much simpler dilaton - vacuum cosmologies in order to find linear combinations of the axion and dilaton perturbations which remain straightforward to integrate even in the more general case . we define two new s - duality invariant variables @xmath279 in terms of @xmath280 and @xmath281 the perturbation equations decouple and the field equations ( [ dphieom ] ) and ( [ dsigmaeom ] ) then become @xmath282 x & = & 0 \\ \label{yeom } y '' + 2\widetilde{h } y ' + k^2 y & = & 0 \ , .\end{aligned}\ ] ] it is far from obvious on first inspection that these variables should be invariant under the s - duality transformation given in eq . ( [ sduality ] ) . however written in terms of the matrix @xmath283 defined in eq . ( [ n ] ) we have @xmath284 and we can see that these variables are the unique s - duality invariant linear combinations of the axion and dilaton perturbations . they reduce to the ( decoupled ) axion and dilaton perturbations in the pure dilaton - vacuum background , as @xmath285 , where we have @xmath286 note that @xmath280 is not only s - duality invariant , but also gauge invariant . that is , it does nt matter which gauge we choose to calculate @xmath254 and @xmath287 , the combination which defines @xmath280 remains unchanged . it is proportional to the axion perturbation on uniform - dilaton hypersurfaces , @xmath288 . by symmetry , it is also proportional to the dilaton perturbation on constant axion hypersurfaces , though this perturbation diverges in the limit that the background axion field becomes constant . having found s - duality invariant variables , one can verify that the evolution equations for these variables , eqs . ( [ xeom ] ) and ( [ yeom ] ) are themselves invariant under s - duality . remembering that the general axion - dilaton cosmological solutions can always be related to the dilaton - vacuum solutions by an s - duality transform , we see that the evolution equations for @xmath280 and @xmath281 in an arbitrary axion - dilaton cosmology are exactly the same as those for the axion and dilaton perturbations in the dilaton - vacuum case . just as in the constant axion case , we can define canonically normalised variables @xmath289 which coincide with the definitions given in eqs . ( [ defu ] ) and ( [ defv ] ) in the dilaton - vacuum case . in general , @xmath290 obeys the s - duality invariant equation of motion given in eq . ( [ upp ] ) and whose general solution is given by eq . ( [ usol ] ) . the equation of motion for @xmath255 given in eq . ( [ vpp ] ) , however , is not invariant under an s - duality transformation . instead the s - duality invariant version of the equation of motion is @xmath291 which coincides with eq . ( [ vpp ] ) when @xmath43 . the general solution for @xmath255 is thus still given by eq . ( [ vsol ] ) . we can still normalise cosmological vacuum perturbations at early times on the @xmath56 branch as @xmath227 because we have seen that in this limit the general axion - dilaton solution given in eqs . ( [ axiphi][axisigma ] ) approach the constant axion solutions with @xmath72 . thus the constants @xmath292 and @xmath293 are given by eqs . ( [ uplus ] ) and ( [ vplus ] ) . by picking s - duality invariant field perturbations we have been able to calculate the general axion - dilaton cosmological perturbation spectra using the pure dilaton - vacuum cosmological vacuum states . we have @xmath294 ^ 2 \,.\ ] ] and the generalised axion perturbation spectrum is given by @xmath295 to recover the actual ( though gauge and s - duality dependent ) axion and dilaton perturbations we can invert eqs . ( [ defx ] ) and ( [ defy ] ) to give @xmath296 however , at late times on the @xmath56 branch , as @xmath66 the general axion - dilaton solutions approach dilaton - vacuum solutions with @xmath73 , and hence @xmath297 and @xmath298 . note that the change of sign from @xmath72 to @xmath73 between the early and late time dilaton - vacuum solutions leads to a phase shift @xmath299 with respect to the late - time behaviour of the pure dilaton - vacuum solutions . but the final power spectrum for the dilaton and axion perturbations as @xmath66 in the general axion - dilaton cosmologies is identical to that given in eqs . ( [ pbbphi ] ) and ( [ pbbsigma ] ) for the s - duality related dilaton - vacuum case . the tilt and amplitude of the spectra are determined solely by the parameter @xmath186 and are insensitive to the specific time dependence of the axion field in different , but s - duality related , solutions . the constraint equation for @xmath189 , eq . ( [ aconstraint ] ) , includes only @xmath281 and @xmath300 ; @xmath301 from eqs . ( [ py ] ) and ( [ pbbbeta ] ) we see that the spectrum of scalar metric perturbations is unaffected by the time dependence of the axion field and is the same as that obtained in the constant axion case , given in eq . ( [ aspectrum ] ) . the low - energy limit of string theory , or m - theory , contains many different degrees of freedom . in this paper we have considered a very simple model containing only a @xmath20 spatially flat frw metric with a dilaton , a single massless modulus field ( representing the volume of @xmath29 internal dimensions ) and a pseudo - scalar axion field derived from the neveu - schwarz antisymmetric tensor potential . the axion - dilaton solutions can be generated from the dilaton - vacuum solutions by an s - duality transformation . they generalise the power - law dilaton - vacuum solutions in a particularly simple way , interpolating between two asymptotically dilaton - vacuum regimes , which are themselves related by an s - duality transformation . although the general axion - dilaton solutions do not alter the singular nature of the cosmological solutions in the string or einstein frame , we draw attention to the fact that the evolution in the conformally related axion frame ( in which the axion is minimally coupled ) can become non - singular when the axion field is allowed to be time - dependent . the worldlines of axionic observers can have an infinite proper lifetime in this frame . there is no graceful exit @xcite from the pre - big bang @xmath56 branch to the post - big bang @xmath54 branch , but the @xmath56 or @xmath54 branches themselves can have an infinite proper lifetime . the shrinking comoving hubble length during a pre - big bang era generates a spectrum of perturbations about the homogeneous background fields from quantum fluctuations . we have calculated the spectrum of large scale perturbations produced in the axion field . the axion spectral index can lie anywhere in the range @xmath302 to @xmath303 , which includes the possibility of the nearly scale - invariant ( @xmath304 ) spectrum required for structure formation . this is in contrast to the dilaton and moduli perturbations which have a steep blue spectrum with an index of @xmath0 , making them incapable of seeding large scale structure in our present universe . the actual value of the axion spectral index depends on the rate of expansion of the internal dimensions . if stable compactification has already occured , leading to an effective @xmath303-dimensional spacetime , the spectral index is @xmath305 . in the simplest case where the background axion field is constant , the axion perturbations are isocurvature perturbations during the pre - big bang epoch . whether these axion perturbations are able to seed large - scale structure in the post - big bang universe depends crucially on the coupling between the axion and the matter which dominates the universe today . nonetheless , it is intriguing that , in principle , the axion could give rise to a nearly scale - invariant spectrum and that the tilt of that spectrum is dependent on the compactification of the internal dimensions . we have seen that s - duality is a powerful tool for calculating not only the classical background solutions in general axion - dilaton cosmologies but also the semi - classical perturbation spectra . by constructing explicitly s - duality invariant field perturbations we are able to calculate the perturbation spectra in the more general axion - dilaton cosmologies as well as the dilaton - vacuum case . it is not surprising that by taking s - duality invariant field perturbations we can derive s - duality invariant solutions . more remarkably , however , the late time dilaton and axion spectra turn out to be independent of the preceding evolution along different , but s - duality related , classical solutions . this results from the fact that s - duality related axion - dilaton solutions all approach the same dilaton - vacuum solution at late times . by contrast , other symmetries present in the low energy string action , such as the symmetry which mixes the moduli field with the dilaton and axion @xcite , do not relate solutions with the same late time behaviour and so will not leave the perturbation spectra invariant . despite the well - known problems associated with achieving a graceful exit from the pre - big bang era , it is worth noting a couple of advantages that the pre - big bang predictions have over conventional ( potential - dominated ) inflation models . firstly , the perturbations originate as vacuum fluctuations at early times , where their amplitude is normalised in a low - curvature , weakly coupled regime in the infinite past , and not at arbitrarily small scales during the planck epoch when the correct vacuum state may be uncertain . secondly , one can give analytic expressions for the asymptotic perturbations on large scales without having to invoke any slow - roll type approximations as must usually be done in conventional inflation models . this is possible not only in the presence of the dilaton alone , but also when one incorporates the moduli and axion fields . we are grateful to b. de carlos , m. giovannini , r. gregory , and j. lidsey for useful conversations . dw and re thank sussex , and re thanks portsmouth , cornell and waikato universities for their hospitality while this work was in progress . re was supported by the japan society for the promotion of science . ejc is grateful to the physics department at dartmouth college for their hospitality while some of this work was carried out . 99 a. strominger and c. vafa , phys . b * 379 * , 99 ( 1996 ) ; j. m. maldacena , _ black holes in string theory _ , hep - th/9607235 ( 1996 ) . c. m. will , _ theory and experiment in gravitational physics _ , cambridge university press ( 1993 ) . e. s. fradkin and a. a. tseytlin , phys . b * 158 * , 316 ( 1985 ) ; c. g. callan , d. friedan , e. j. martinec , and m. j. perry , nuc . phys . * b262 * , 593 ( 1985 ) ; c. lovelace , nuc . phys . * b273 * , 413 ( 1986 ) . c. brans and r. h. dicke , phys . rev . * 124 * , 925 ( 1961 ) . j. a. casas , j. garca - bellido , and m. quirs , nuc . phys . * b361 * , 713 ( 1991 ) ; j. garca - bellido and m. quirs , nuc . phys . * b385 * , 558 ( 1992 ) . t. damour and a. m. polyakov , nucl . phys . * b423 * , 532 ( 1994 ) . m. b. green , j. h. schwarz and e. witten , _ superstring theory : vol . 1 _ ( cambridge university press , 1987 ) m. gasperini and g. veneziano , astropart . phys . * 1 * , 317 ( 1993 ) . m. gasperini and g. veneziano , mod . a8 * , 3701 ( 1993 ) . r. brustein and g. veneziano , phys . b * 329 * , 429 ( 1994 ) . n. kaloper , r. madden and k. a. olive , nucl . phys . * b452 * , 677 ( 1995 ) ; phys . lett . b * 371 * , 34 ( 1996 ) . r. easther , k. maeda and d. wands , phys . d * 53 * , 4247 ( 1996 ) . r. easther and k. maeda , phys . d * 54 * , 7572 ( 1996 ) . m. gasperini , m. maggiore and g. veneziano , to appear nucl . b , hep - th/9611039 ( 1996 ) . a. r. liddle , r. g. moorhouse , and a. b. henriques , nuc . b311 * , 719 ( 1989 ) . a. a. tseytlin , int . d * 1 * , 223 ( 1992 ) . d. s. goldwirth and m. j. perry , phys . d * 49 * , 5019 ( 1994 ) . e. j. copeland , a. lahiri , and d. wands , phys . d * 50 * , 4868 ( 1994 ) . e. j. copeland , a. lahiri , and d. wands , phys . d * 51 * , 1569 ( 1995 ) . n. a. batakis , phys . b * 353 * , 39 ( 1995 ) ; n. a. batakis and a. a. kehagias , nucl . phys . * b449 * , 248 ( 1995 ) . j. d. barrow and k. kunze , phys . d * 55 * , 623 ( 1997 ) ; j. d. barrow and m. p. dabrowski , phys . d * 55 * , 630 ( 1997 ) . a. lukas , b. a. ovrut and d. waldram , phys . b * 393 * , 65 ( 1997 ) ; _ string and m - theory cosmological solutions with ramond forms _ , hep - th/9610238 ( 1996 ) . j. e. lidsey , _ on the cosmology and symmetry of dilaton - axion gravity _ , gr - qc/9609063 ( 1996 ) . n. kaloper , phys . d * 55 * , 3394 ( 1997 ) ; h. l , s. mukherji , c. n. pope and k .- w . xu , _ cosmological solutions in string theory _ , hep - th/9610107 ; r. poppe and s. schwager , phys . lett . b * 393 * , 51 ( 1997 ) . g. veneziano , phys . b * 265 * , 287 ( 1991 ) . d. wands , class . quantum grav . * 11 * , 269 ( 1994 ) . r. tabensky and a. h. taub , comm . * 29 * , 61 ( 1973 ) . j. p. mimoso and d. wands , phys . d * 51 * , 477 ( 1995 ) ; _ ibid _ * 52 * , 5612 ( 1995 ) . k. behrndt and s. frste , nuc . phys . * b430 * , 441 ( 1994 ) . a. shapere , s. trivedi and f. wilczek , mod . lett . a * 6 * , 2677 ( 1991 ) ; a. sen , _ ibid_. * 8 * , 2023 ( 1993 ) . k. meissner and g. veneziano , phys . b * 267 * , 33 ( 1991 ) ; a. sen , phys . b * 271 * , 295 ( 1991 ) . r. brustein , m. gasperini , m. giovannini , v. f. mukhanov and g. veneziano , phys . d * 51 * , 6744 ( 1995 ) . j. hwang , _ structure formation in string cosmology _ , hep - th/9608041 ( 1996 ) ; _ density spectrums from kinetic inflations _ , gr - qc/9612065 ( 1996 ) . j. m. bardeen , phys . d * 22 * , 1882 ( 1980 ) . v. f. mukhanov , h. a. feldman and r. h. brandenberger , phys . rep . * 215 * , 203 ( 1992 ) . m. gasperini and m. giovannini , class . quantum grav . * 14 * , 735 ( 1997 ) ; m. giovannini , phys . d * 55 * , 595 ( 1997 ) . a. r. liddle and d. h. lyth , phys . rep . * 231 * , 1 ( 1993 ) . r. brustein , m. gasperini , m. giovannini and g. veneziano , phys . b*361 * , 45 ( 1995 ) ; m. galluccio , m. litterio and f. occhionero , _ graviton spectra in string cosmology _ , gr - qc/9608007 ( 1996 ) . v. f. mukhanov , sov . jetp * 68 * , 1297 ( 1988 ) . n. d. birrel and p. c. w. davies . , _ quantum fields in curved space _ , cambridge university press ( 1982 ) . * figure 1 : * the string frame scale factor for the axion - dilaton solution ( solid line ) given in eq . ( [ axia ] ) . it approaches the @xmath73 dilaton - vacuum solution given in eq . ( [ dila ] ) ( dashed line ) as @xmath306 and the @xmath72 dilaton - vacuum solution ( dot - dashed line ) as @xmath307 . the @xmath56 and @xmath54 branches are also labelled . the specific parameters for the axion - dilaton solution are @xmath308 , @xmath309 and @xmath310 . * figure 2 : * the dilaton , @xmath65 , with @xmath311 and the same notation and parameter values as in fig . 1 . the axion - dilaton solution [ eq . ( [ axiphi ] ) ] approaches the @xmath64 dilaton - vacuum solution [ eq . ( [ dilphi ] ) ] at small @xmath312 and the @xmath63 solution at large @xmath312 . * figure 3 : * the modulus , @xmath159 , with @xmath313 and the same parameter values as in fig . 1 and @xmath314 . the evolution is the same , for both the dilaton - vacuum in eq . ( [ dilbeta ] ) and axion - dilaton solutions in eq . ( [ axibeta ] ) . * figure 6 : * the axion frame scale factor , @xmath136 , plotted against proper time in the axion frame for axion - dilaton cosmologies ( solid lines ) with three different values of @xmath45 , while the other constants are the same as in fig . 1 . the asymptotic dilaton - vacuum solutions with @xmath73 ( dashed lines ) and @xmath72 ( dot - dashed lines ) are also shown . when @xmath316 the semi - infinite interval in @xmath317 is mapped onto an unbounded interval for the proper time , so the @xmath56 and @xmath54 branch solutions are displayed separately in the two top figures .

we derive new constraints on the mass of the milky way s dark matter halo , based on a set of halo stars from sdss as kinematic tracers . our sample comprises rigorously selected blue horizontal - branch ( bhb ) halo stars at kpc , and with distances from the galactic center up to kpc , with photometry and spectra drawn from sdss dr-6 . with distances accurate to , this sample enables construction of the full line - of - sight velocity distribution at different galactocentric radii . to interpret these distributions , we compare them to matched mock observations drawn from two different cosmological galaxy formation simulations designed to resemble the milky way , which we presume to have an appropriate orbital distribution of halo stars . specifically , we select simulated halo stars in the same volume as the observations , and derive the distributions of their line - of - sight velocities for different radii , normalized by the simulation s local circular velocity . we then determine which value of brings the observed distribution into agreement with the corresponding distributions from the simulations . these values are then adopted as observational estimates for , after a small jeans equation correction is made to account for slight data / simulation differences in the radial density distribution . this procedure results in an estimate of the milky way s circular velocity curve to kpc , which is found to be slightly falling from the adopted value of at the sun s location , and implies mm . the radial dependence of , derived in statistically independent bins , is found to be consistent with the expectations from an nfw dark matter halo with the established stellar mass components at its center . if we assume an nfw halo profile of characteristic concentration holds , we can use the observations to estimate the virial mass of the milky way s dark matter halo , mm , which is lower than many previous estimates . we have checked that the particulars of the cosmological simulations are unlikely to introduce systematics larger than the statistical uncertainties . this estimate implies that nearly 40% of the baryons within the virial radius of the milky way s dark matter halo reside in the stellar components of our galaxy . a value for m of onlym also ( re-)opens the question of whether all of the milky way s satellite galaxies are on bound orbits .

0801.1232

the visible parts of galaxies are , in the current paradigm for galaxy formation , concentrations of baryons at the center of much larger dark matter halos , which have assembled through hierarchical merging and gas cooling . understanding the properties of these dark matter host halos , their virial masses , concentrations , and radial mass profiles , _ vis - a - vis _ the luminous properties of the main galaxy at their center , is crucial for modeling the dynamics of the central galaxy , for connecting observations of galaxies to large - scale cosmological dark matter simulations , and for understanding what fraction of baryons in the halo ended up as stars in the central galaxy . in turn , the extended stellar distributions of galaxies , in particular the stellar halos of nearby galaxies , offer some of the best probes to test generic predictions about the nature of dark matter mass profiles ( navarro et al . 1996 ) . the milky way and its surrounding halo are of particular interest , as our internal position permits the placement of unique constraints on the galaxy s stellar mass content , on its dark matter profile at large radii , and on the 3-d shape of its dark matter halo . yet , our location within the galaxy also complicates some measurements , such as the extended rotation curve of gas in its disk . as a result , the dark mass profile for the milky way between @xmath13 kpc and @xmath14 kpc and the halo s virial mass have not been previously constrained to better than a factor of 2 to 3 . in practice , it has proven useful to quantify the halo mass profile by either a circular velocity curve , @xmath15 , or by the escape velocity curve , @xmath16 . in previous work , the most common tools used to estimate the milky way halo mass are the escape velocity and the velocity dispersion profile of the tracer populations ( i.e. , halo stars , or the milky way s satellite galaxies and globular clusters ) . the escape velocity provides constraints on the gravitational potential at the relevant position ( little & tremaine 1987 ; zaritsky et al . 1989 ; kulessa & lynden - bell 1992 ; kochanek 1996 ) . recent work has shown that the total mass of the halo is around @xmath17 . wilkinson & evans ( 1999 ) used the velocities of 27 satellite galaxies and globular clusters to find a halo mass of @xmath18 by adopting a truncated , flat rotation curve halo model . sakamoto , chiba & beers ( 2003 ) used a sample including 11 satellite galaxies , 137 globular clusters , and 413 solar neighborhood field horizontal - branch stars , along with a flat rotation curve model , to obtain a total halo mass of @xmath19 or @xmath20 , depending on the inclusion ( or not ) of leo i. recently , smith et al . ( 2007 ) estimated a halo mass of @xmath21 , based on a sample of high - velocity stars from the rave survey and two published databases , and an adiabatically contracted nfw halo model . battaglia et al . ( 2005 , 2006 ) used a derived velocity dispersion profile to determine a total mass of @xmath22 , with some dependence on the model adopted for the halo profile ( see also dehnen , mclaughlin & sachania 2006 ) . in order to improve the precision of the mass estimate of the halo , the fundamental first step is to begin with high - quality data ( tracers with accurate distances and radial velocities ) , augmented with an efficient method of analysis capable of extracting the maximum amount of information . for example , wilkinson & evans ( 1999 ) comment that a data set of @xmath23 radial velocities of bhb stars reduces the uncertainty in the mass estimate of the halo to @xmath24% . this goal has proven elusive , even with data samples twice this size ( e.g. sakamoto et al . 2003 ) , due to the use of a relatively nearby sample of tracers . however , we also expect a large and distant data set to constrain the gravitational potential at different positions , which is ultimately a more reliable probe of the dark matter halo . the planned or already - underway kinematically unbiased surveys , such as the european space agency s astrometric mission gaia ( turon et al . 2005 ) , the space interferometry mission ( sim ; unwin et al . 2000 ) , the radial velocity experiment ( rave ; steinmetz et al . 2006 ) , the sloan digital sky survey ( sdss ; york et al . 2000 ) , and the sloan extension for galactic understanding and exploration ( segue ; newberg et al . 2003 , beers et al . 2004 ) , will finally make it feasible to obtain the required large sample of tracers with well - measured parameters . blue horizontal - branch ( bhb ) stars are excellent tracers of galactic halo dynamics , because they are luminous and have a nearly constant absolute magnitude within a restricted color range ( see , e.g. , sirko et al . 2004 for bhb stars in sdss ) . sdss and segue specifically targeted bhb stars for spectroscopy ; we employ these stars ( and other serendipitously discovered bhb stars with available spectra , e.g. , mis - identified qsos ) to derive precision constraints on the mass of the milky way s dark matter halo . in 2 we describe the assembly of a particularly conservative ( i.e. , low contamination ) selection of bhb stars from the spectra of candidate bhbs in sdss dr-6 . two different cosmological simulations , one of a galaxy resembling the milky way with a @xmath25 halo , and one of a milky - way - like galaxy simulation with a @xmath26 halo , are presented in 3 . this section also describes the estimation of the circular velocity curve as a function of galactocentric radius , @xmath15 , obtained by comparing the observed bhb radial velocities to the kinematics of the simulations . in 4 we present the resulting estimates of the milky way s halo circular velocity curve , and discuss implications for the virial mass of the milky way s halo . our conclusions are summarized in 5 . the sdss is an imaging and spectroscopic survey covering more than a quarter of the sky ( york et al . although its main focus was ( and is ) extragalactic science , there are still large numbers of stars not only imaged , but also targeted spectroscopically . in 2005 the project entered a new phase ( sdss - ii ) , in which segue , a sub - survey of sdss - ii , addresses fundamental questions about the formation and evolution of our own galaxy ( http://www.sdss.org ) . in addition to further ( low - latitude ) imaging , segue in particular provides for a more systematic and extensive acquisition of spectroscopy for stars in the milky way , from which stellar parameters and radial velocities can be derived from the application of a well - calibrated and well - tested set of procedures , the segue stellar parameter pipeline ( sspp ; see lee et al . 2008a , b ; allende prieto et al . 2008 ) . the sloan digital sky survey uses a ccd camera ( gunn et al . 1998 ) on a dedicated 2.5 m telescope ( gunn et al . 2006 ) at apache point observatory , new mexico , to obtain images in five broad optical bands ( @xmath27 ; fukugita et al . the survey data - processing software measures the properties of each detected object in the imaging data in all five bands , and determines and applies both astrometric and photometric calibrations ( lupton et al . 2001 ; pier et al . 2003 ; ivezi et al . photometric calibration is provided by simultaneous observations with a 20-inch telescope at the same site ( hogg et al . 2001 ; smith et al . 2002 ; stoughton et al . 2002 ; tucker et al . this work is based on stellar spectra that are part of the sdss dr-6 ( adelman - mccarthy et al . 2008 ) . the radial velocities are taken from the sspp , which ( primarily ) uses matches of the observed spectra to a set of @xmath28 elodie template spectra , corrected to the heliocentric standard - of - rest ( hsr ) frame . note that the sspp automatically corrects for a systematic offset in sdss stellar radial velocities of 7.3 km s@xmath29 ( see lee et al . 2008a , b for derivation of this offset ) . after this correction is applied , the sdss radial velocities exhibit negligible systematic errors , and have an accuracy between 5 and 20 km @xmath30 , depending on the s / n of the spectrum and the stellar spectral type . we aim to select as `` pure '' a set of true bhb stars as possible , where the contamination , e.g. , from halo blue - straggler stars , is minimized , even if this selection procedure results in a smaller sample . we have not made any effort to construct a complete sample of bhb stars . therefore , we have not simply followed the dr-6 classification procedures , but have employed a very stringent approach combining previously established color cuts with a set of balmer - line profile selection criteria . our technique is similar to that of sirko et al . ( 2004 ) , but we use slightly different criteria in the adopted color cuts . throughout this paper all magnitudes are corrected for galactic extinction , and the colors are corrected for reddening , both based on the procedures of schlegel et al . ( 1998 ) . we start our selection of bhb stars by adopting the color cuts for identification of bhb candidates used in yanny et al . ( 2000 ) : @xmath31 @xmath32 these color cuts are shown in figure [ f : f1 ] ; the rectangle is the region of bhb candidates . this color cut produces @xmath33 bhb photometric candidates with existing spectra , but with a considerable contamination by both blue straggler ( bs ) stars and warm main - sequence stars ( ms ) . the subsequent spectroscopic analysis for bhb candidates is aimed at eliminating , or at least greatly reducing , contamination from such stars . in the ranges of effective temperature considered herein ( roughly 7000 k to 10,000 k ) , bhb stars have lower surface gravities than bs stars and higher temperatures than ( old , halo population ) ms stars . the balmer - line profiles of warm stars are sensitive to both gravity and temperature ; their analysis provides a powerful method to select bhb stars with confidence . we analyze the line profiles after normalizing the continuum for all stars , as illustrated in figure [ f : f2 ] . then we combine two independent methods to identify non - bhb stars , as described below . the @xmath34 method ( pier 1983 ; sommer - larsen , christensen 1986 ; arnold & gilmore 1992 ; flynn , sommer - larsen , & christensen 1994 ; kinman , suntzeff , & kraft 1994 ; wilhelm , beers , & gray 1999 ) discriminates bhb stars from bs stars by determining the value of @xmath35 , the width of the balmer line at 20% below the local continuum , and distinguishes bhb stars from ms stars by measuring the value of @xmath36 , the flux relative to the continuum at the line core ( beers et al . 1992 ; sirko et al . figure [ f : f3 ] shows how @xmath37 can distinguish a bhb star from a bs star . a plot of @xmath37 versus @xmath36 of the @xmath38 line for stars brighter than @xmath39 , and passing the initial color cuts , is shown in figure [ f : f4 ] . the concentration of stars centered at @xmath40 ) represents the bhb stars ; the stars with larger @xmath37 are bs stars , and the remaining stars are ms stars . figure [ f : f4 ] indicates that the sample contamination resulting from the color - selection only is rather severe , on the order of 50% . the criteria for confirmation of a bhb star , based on the h@xmath41 line analysis , are : @xmath42 the _ scale width vs. shape _ method ( clewley et al . 2002 ) is based on a srsic profile ( srsic 1968 ) fit to the balmer lines : @xmath43},\ ] ] where @xmath44 is the normalized flux density and @xmath45 is assumed to be the nominal wavelength of the balmer line . to account for small radial velocity corrections and the imperfect normalization of spectra , we fit the normalized extracted spectrum to the srsic profile with five free parameters : @xmath46 , @xmath47 , @xmath48 and @xmath49 . @xmath50}\ ] ] the set of stars that passed the initial color cuts exhibits a bimodal distribution in the @xmath51 versus @xmath52 plane ( figure [ f : f5 ] ) , where @xmath53 refers to the h@xmath54 line . this gap allows one to quite cleanly separate bhb stars from bs stars , according to : @xmath55 @xmath56 as the color coding of the points in figure [ f : f5 ] shows , most bhb stars selected by the @xmath57 & @xmath58 method already lie in the appropriate region of the _ scale width vs. shape _ method applied to @xmath59 . the combination of these two stringent criteria indeed appears to eliminate most stars that are not _ bona - fide _ bhb stars . the combination of figure [ f : f4 ] and figure [ f : f5 ] suggests that the contamination is well below 10% . note that the two spectroscopic criteria are determined based on the spectra of bright stars ( @xmath60 ) . because of the high quality spectra available for such stars , the criteria can be identified easily by eye from figure [ f : f4 ] and figure [ f : f5 ] . for fainter stars ( @xmath61 ) we adopt the same spectroscopic criteria as for the bright stars . there are a total of @xmath62 stars that survive the color cuts , and _ both _ of the balmer - line profile cuts described above . this sample forms the basis of our remaining analysis . our basic approach is to identify bhb stars from their spectra ( 2.1 ) , then estimate their absolute magnitude from photometry alone . bhb stars have similar , but not identical , absolute magnitudes , as they are affected slightly by temperature and metallicity ( e.g. , wilhelm , beers , & gray 1999 ; sirko et al . figure [ f : f6 ] shows five theoretical absolute magnitudes @xmath63 in the @xmath64 plane , taken from sirko et al . ( 2004 ) . for each bhb star in our sample we define the most probable absolute magnitude associated with its @xmath64 colors by simply finding the absolute magnitude of the point on the theoretical track that is closest to the observed star in this color - color space . ) , and use it instead of picking the `` closest '' model . however , the difference between the two methods is less than the theoretical uncertainties in the derived luminosities . ] the absolute magnitude error of a given star derived from this method is on the order of 0.2 mag ( sirko et al . 2004 ) , corresponding to a distance accuracy of 10% ; errors in the measured photometry are much smaller than this error . the distance from the sun , and from the galactic center can then be determined from : @xmath65 @xmath66 where @xmath67 is the extinction - corrected magnitude in the @xmath67 band , @xmath68 is the absolute magnitude in the @xmath67 band , @xmath69 is the distance to the sun , @xmath70 is the distance from the galactic center , @xmath71 and @xmath72 are the galactic latitude and longitude respectively ; we take @xmath73 , the distance of the sun from the galactic center , to be @xmath74 kpc . for ease of the subsequent analysis , we convert the heliocentric radial velocities to the galactic standard of rest ( gsr ) frame by adopting a value of @xmath75 km s@xmath29 for the local standard of rest ( v@xmath76 ) and a solar motion of @xmath77 km s@xmath29 in ( u , v , w ) , which are defined in a right - handed galactic system with u pointing towards the galactic center , v in the direction of rotation , and w towards the north galactic pole ( dehnen & binney 1998 ) . hereafter , @xmath78 stands for the radial velocity in the gsr frame ( i.e. , the radial velocity component along the star - sun direction , corrected for galactic rotation ) . if @xmath79 is the heliocentric radial velocity , then : @xmath80 our sample of @xmath62 bhb stars may contain some thick - disk stars , with @xmath81 , so we impose an additional geometric constraint @xmath82 , which reduces the sample to @xmath0 ( presumed ) _ halo _ bhb stars within 60 kpc , and with radial velocity error less than 30 km @xmath30 . the spatial distribution of our @xmath0 halo bhb stars is shown in figure [ f : f7 ] and figure [ f : f8 ] , with all stars located at least @xmath83 kpc from the galactic disk and at @xmath84 kpc from the galactic center . figure [ f : f9 ] ( upper panel ) shows the derived [ fe / h ] estimates for the bhb sample reported by the sspp ( see lee et al . 2008a , b for details ) . most of our bhb stars are metal - poor ( [ fe / h ] @xmath852 ) , as expected for a sample of halo stars . this is further testament to the quality of the sdss spectra and the rigor of the sample selection . the observed distribution of line - of - sight velocities is well - fitted by a gaussian distribution with @xmath86 , as shown in the lower panel of figure [ f : f9 ] . the upper panel of figure [ f : f10 ] shows the distribution of @xmath78 vs. @xmath87 for the _ halo _ bhb stars . it reveals a nearly equal number of stars with positive and negative @xmath78 at a given radius : the bhb population exhibits very little net rotation , so we will subsequently analyze only @xmath88 . the lower panel of figure [ f : f10 ] shows the binned velocity dispersion , @xmath89 , of the sample as a function of radius . it is well described by @xmath90 where r is the galactocentric radius . we will use eqn 6 subsequently for jeans equation modeling . a _ maximum likelihood _ fit to estimate @xmath91 and @xmath92 yields @xmath93 , which is consist with eqn 6 within 1-@xmath94 . this indicates that the choice of the bins has small effect on the inferred @xmath95 . we also checked that these values of @xmath96 are insensitive to small changes in the assumed @xmath97 . the radial number density profile of halo stars in the milky way in the range @xmath98 kpc can be approximated by @xmath99 ( e.g. , bell et al 2007 ) . accounting for the @xmath100 volume effect , inspection of figure [ f : f10 ] and figure [ f : f12 ] reveals that the radial distribution of our bhb sample falls off much more rapidly than this , in particular at large distances . this is mostly attributable to the sdss spectroscopic target selection . the chances of a candidate bhb star to be targeted for spectroscopy in the course of sdss and segue is not a simple function of its apparent magnitude , but depends on many ( often operational ) factors , which have also evolved over the course of the survey . the net result is that only under very favorable circumstances are distant ( hence faint ) bhbs targeted , and have spectra obtained that are of sufficient s / n to pass our quality criteria . as a result , the radial distribution of our sample falls off much more steeply than the parent population of halo bhb stars . we account for these effects in the subsequent analysis , as described below . overall , our present sample of distant halo stars with available kinematics is nearly an order of magnitude larger than that of battaglia et al . ( 2005 ) , which has 240 stars . because our tracers are bhb stars , their distances are also known more accurately . however , the battaglia et al . sample does extend to larger radii , up to @xmath101 kpc . we now describe our approach to convert the @xmath102 measurements ( as a function of galactocentric radius , r ) for the @xmath103 bhb stars into estimates of @xmath5 of the milky way halo , and ultimately to estimate @xmath104 for the milky way s halo . with the full line - of - sight velocity data set , sampling radii of @xmath105 kpc , we can obtain both the velocity dispersion , @xmath106 , and identify a set of exceptionally high - velocity stars at various radii that might be suitable to estimate @xmath16 . some previous work ( e.g. , battaglia et al 2005 ) considered only velocity dispersions , while others focused only on high - velocity stars in the solar neighborhood ( sakamoto et al . 2003 ; smith et al . we consider the full velocity distribution ; as we shall see later , it is close to a gaussian and hence most of the pertinent information is contained in @xmath107 . to link the observables , @xmath78 and @xmath87 , to @xmath15 , we must not only account for the particular survey volume , but also need to make at least an implicit assumption about the nature of the halo - star distribution function , in particular its ( an-)isotropy . in contrast to , e.g. , battaglia et al . ( 2005 ) , we restrict ourselves not only to jeans equation modeling , but choose to account for these issues also by investigation of a comparison with cosmologically motivated galaxy simulations , to make `` mock observations '' within these simulations , and then match test results to the observations . one can not expect the halo stars in the simulations to have exactly the sample density profile as the actual stars in the halo of the milky way ; we account for such differences as described below . based on prior estimates of the halo mass , we chose two sph simulations corresponding to the formation of milky way - like haloes , from which `` pseudo - observations '' are constructed for comparison with the data . both halos were picked from low - resolution cosmological dark matter simulations and were re - simulated at higher resolution , including the effects of gas dynamics , and star formation and stellar feedback . simulation i was run using the smoothed particle hydrodynamics code grapesph ( steinmetz 1996 ) , and included a moderately efficient stellar feedback model in which the supernova energy was added to the thermal energy of the gas , as described in abadi et al . this simulated halo has a present day virial mass of about @xmath108 and a virial radius of @xmath109 kpc . simulation ii was run using the smoothed particle hydrodynamics code gadget-2 ( springel 2005 ) , assuming the standard star - formation prescription of springel et al . ( 2003 ) , but without stellar feedback . this simulated halo has a present - day virial radius of @xmath110 kpc , and a virial mass of @xmath111 ( these values differ from naab et al . 2007 , as we use a different virial contrast in this paper ) . for further details of the simulations we refer the interested reader to abadi et al . ( 2003 , simulation i ) and to naab et al . ( 2007 , simulation ii ) , and references therein . for each particle in these simulations we have the @xmath112-d positions and @xmath112-d velocities , the circular velocity , @xmath113 , and the escape velocity , @xmath114 , to the virial radius of the simulation at each given position . because sdss / segue only observed @xmath115 @xmath116 steradians , not the entire sky ( see figure [ f : f11 ] ) , in order to compare the observations to the simulations we must select simulated star particles in the same region as the sdss footprint . as a first step , we specified `` galactic coordinates '' in the simulation by defining the `` galactic plane '' by the net angular momentum of all ( stellar ) particles within 10 kpc of the center . in this coordinate system one has ( by definition ) @xmath117 kpc ( @xmath118 kpc ) . we can then calculate the galactocentric radial velocity as seen from the `` sun '' for each simulation particle as : @xmath119 since the simulated `` observer '' is at rest in the galactocentric coordinate system , no assumptions about the lsr need be made in the simulations . to create `` pseudo - observations '' , we first remove the satellites in the simulation and then assume 10 positions for the sun that all have @xmath120 and @xmath121 , but different azimuthal angles , @xmath122 . the corresponding @xmath123 and @xmath124 in this coordinate system can then be used to select the star particles in the same region of the simulation as our observed bhb sample . the effective flux limit of the bhb sample , @xmath125 , implies a maximum distance from the sun of about @xmath126 kpc , a distance which we impose as a selection limit in the simulations . to select `` halo '' stars , we also impose that the stars have to be at least @xmath83 kpc above or below the disk plane . finally , we average these 10 samples of simulated halo stars . figure [ f : f11 ] shows the distribution of these selected simulated stars and the observed halo bhb stars in @xmath123 and @xmath124 . this procedure results in a sample of simulated halo stars ( each position produces an `` observational data set '' ) with galactocentric radial velocities , galactocentric radii , escape velocities , and circular velocities , whose distribution is shown in figure [ f : f12 ] . this figure makes it clear that , even in a large sample , the galactocentric radial velocity rarely approaches the escape velocity ( one obvious reason being that we measure only the projected component of the space velocity ) . we analyze the implications of the observed bhb kinematics for the milky way halo s mass distribution in two steps . we first estimate @xmath5 from the data / simulation comparison in a set of statistically independent radial bins , effectively constructing a circular velocity curve extending to 60 kpc . we then fit the circular velocity curve with nfw halo ( and bulge+disk ) models , resulting in estimates of m@xmath11 . it should be noted that the milky way halo s presumed virial radius extends about a factor of four beyond the most distant bhb stars in our observed sample of tracers . for the data / model comparison we construct the distributions , @xmath127 , for the simulations and the data respectively . for the simulations , we use the procedure described in the above section to obtain the @xmath78 and @xmath128 at each particle position . we compare those @xmath129 distributions to analogous ones constructed from the data for a sequence of trial values @xmath15 . as the best observational estimate of @xmath5 , we then take that value for which the probability that @xmath130 and @xmath131 were drawn from the same distribution is maximal . as we have no _ a - priori _ functional form for these distributions , we define this best match as that which maximizes the probability in a two - sided k - s test . to define confidence limits on the @xmath5 estimate , we repeat this procedure with bootstrapped versions of @xmath130 , and take as @xmath132 the variance of the resulting @xmath128 distribution . we divide the galactocentric radius ( @xmath87 ) into 10 bins and apply this method , which results in 10 ( statistically ) independent radial estimates of @xmath5 . specifically , the bins we adopted were @xmath133 kpc , @xmath134 kpc , @xmath135 kpc , @xmath136 kpc , @xmath137 kpc , @xmath138 kpc , @xmath139 kpc , @xmath140 kpc , @xmath141 kpc , and @xmath142 kpc ( all distances are from the galactic center , not from the sun ) . the best matched distributions of @xmath143 are illustrated in figure [ f : f13 ] and figure [ f : f14 ] . in these figures , the red line shows @xmath143 of the simulations , while the black line shows that of the data . overall , the velocity distributions agree well with one another . to estimate the statistical uncertainties on these values , we use bootstrap resampling on our bhb sample ( typically 100 times ) , repeat the above procedure , and indicate the resulting 68% ( 1-@xmath94 ) confidence region as error bars . in a given gravitational potential , more centrally concentrated kinematic tracer populations will exhibit smaller velocity dispersions than more extended ones . in the simple case of a spherical potential , with tracers of @xmath144 and an isotropic velocity dispersion that varies only slowly with radius , the jeans equation , for a given @xmath5 , yields the relative velocity dispersions of the two populations , @xmath145 , where the indices refer to two different hypothetical tracer populations . in the radial range of 10 - 60 kpc , the density profile of halo stars in simulation ii is approximately @xmath146 , while that of simulation i is @xmath147 . this , however , should not be compared to the radial distribution of the stars for which we actually have velocities , but to the radial profile of halo stars from which they were drawn ( see 2.3 ) . this requires the reasonable assumption that the measured velocities are uncorrelated with the spectral targeting even a more complete sampling of bhb stars at a given radius would have yielded the same velocity distribution . as an estimate for the actual density profile for the milky way s stellar halo at 10 - 60 kpc , we take the estimates ( for main sequence turn - off stars ) of @xmath148 ( bell et al . 2007 ) , based on sdss . using the above correction based on the jeans equation , we must subsequently revise the derived velocity scales for the milky way halo upward by @xmath149 for simulation ii and downward by @xmath150 for simulation i. matching @xmath151 , and applying the above correction , we obtain the estimates summarized in figure [ f : f15 ] and in table 3 . the filled circles in this figure reflect the @xmath15 estimates based on simulation i , while the filled squares stand for the @xmath5 estimates based on simulation ii . as mentioned before , the error bars are from bootstrapping . we have explored whether the derived estimates of @xmath5 depend on our adopted value for the distance of the sun from the galactic center , @xmath152 , or on the adopted local rotation velocity , @xmath97 . if we vary @xmath152 adopted in the observation and simulations ( see eqn 4 and eqn 7 ) , from @xmath153 to @xmath154 , @xmath5 changes by only about 1% . our results for the derived @xmath5 are also insensitive to the choice of the @xmath97 adopted in the observation ( see eqn 5 ) . taking @xmath97 as @xmath155 or @xmath156 changes the estimated @xmath5 by only 3% . for reference , we show how these estimates of @xmath5 compare to those derived from the jeans equation and the fit to @xmath157 shown in figure [ f : f10 ] . from the jeans equation , @xmath5 can be estimated from the velocity dispersion , @xmath158 ( binney & tremaine 1987 ) as follows , @xmath159 with@xmath160 where @xmath161 and @xmath162 are the radial and tangential velocity dispersions , respectively , in spherical coordinates , and @xmath163 is the stellar density . the distribution of the halo stars in the simulations are anisotropic , with @xmath164 , and the simulations exhibit @xmath165 for this particular survey volume . taking from figure [ f : f10 ] the fit to @xmath166 of the bhb stars ( eqn 6 ) , and assuming @xmath167 for bhb stars , we can derive two circular velocity curves , for @xmath164 ( anisotropic ) , and @xmath168 ( isotropic ) , respectively , by adopting @xmath169 ( see figure [ f : f15 ] ) . they both agree well with the simulation - based estimates , and are not used any more in the following analysis . figure [ f : f16 ] and figure [ f : f17 ] present the main result of our analysis , @xmath5 , an estimate of the circular velocity curve from @xmath98 kpc . this represents the first time that the circular curve for the galaxy has been estimated to such large distances at this accuracy . note that at small radii this estimate , though derived from halo stars , agrees well ( within @xmath13% ) with established determinations at the solar radius ( @xmath170 km @xmath171 ) . beyond the solar radius , the circular velocity curve appears to be gently falling to 175 @xmath172 at @xmath2 kpc . note also that the circular velocity curve is a conceptually more robust estimate than @xmath114 , which depends more sensitively than v@xmath173 on @xmath174 at radii beyond the measurements . using the functional form for @xmath5 expected for an nfw halo and the stellar component ( see below ) as a means to interpolate the individual circular velocity curve estimates , one obtains @xmath175 km @xmath30 , or m@xmath176m@xmath9 . this is the largest radius for which the data directly constrain @xmath177 or @xmath178 . yet this radius is only one - fourth of the expected virial radius of the milky way s halo . we therefore proceed with a separate step , to use these @xmath5 estimates to constrain parameterized models for the overall dark matter halo . we assume that the galactic potential is represented by three components , a spherical hernquist ( 1990 ) bulge , an exponential disk for the stellar components , and describe the halo by an nfw profile ( navarro et al . the total potential can then be simply expressed as @xmath179 with an assumed potential , presumed to be spherically symmetric , for the disk and bulge of @xmath180 @xmath181 where @xmath182 , @xmath183 kpc , @xmath184 , and @xmath185 kpc ( similar to smith et al . the radial potential for a spherical nfw density profile can be expressed as @xmath186 where @xmath47 is a concentration parameter , defined as the ratio of the virial radius to the scale radius . for standard @xmath187cdm cosmogonies we do not attempt to constrain halo flattening . the parameter @xmath188 is a characteristic density given by @xmath189 where @xmath190 is the critical density of the universe , @xmath191 is the contribution of matter to the critical density , and @xmath192 is the critical overdensity at virialization . the virial mass can then be determined from the virial radius , using @xmath193 for our analysis we adopt @xmath194 , @xmath195 , and h@xmath196 = 65@xmath197 @xmath198 . given recent discussions ( and doubts raised ) of whether the baryons modify the dark matter profile , as expected from `` adiabatic contraction '' ( dutton et al . 2007 ) , we consider both an unaltered and an adiabatically contracted nfw profile in the fit of @xmath199 . by fitting the observed @xmath5 with @xmath200 from @xmath201 shown as eqn 10 we can constrain the halo mass of the milky way . in this fit , we simply adopt an unaltered nfw profile and a present - day relation between the mean value of @xmath47 and @xmath104 , @xmath202 this relation is accurate over the range @xmath203 , and is based on the model of macci et al . ( 2007 ) with @xmath204 , @xmath205 , @xmath206 , and @xmath207 . therefore , the @xmath104 is derived as a 1-parameter fit ( fit only @xmath104 , presuming @xmath208 ) . the results are summarized in figure [ f : f16 ] . specifically , for the circular velocity estimates resulting from simulation i , we find @xmath209 with @xmath210 kpc and @xmath211 . for the circular velocity estimates based on simulation ii we find @xmath212 with @xmath213 kpc and @xmath214 . note that the error bars of @xmath104 , @xmath215 , and c are determined from 1-@xmath94 confidence intervals in a chi - square test . for the two cases above we have adopted an unaltered nfw profile and an average relation between @xmath47 and @xmath104 . if we fit an adiabatically contracted nfw profile ( using the prescription of blumenthal et al . 1986 and mo et al . 1998 ) and the same disk and bulge as in eqn 11 and eqn 12 , taking the concentration parameter ( @xmath47 ) and virial mass ( m@xmath11 ) of the nfw profile as independent parameters ( i.e. , we do not require that they follow the relationship in eqn 16 ) , the @xmath104 can be derived as a 2-parameter fit ( @xmath104 and @xmath47 ) , as shown in figure [ f : f17 ] . for the circular velocity estimates resulting from simulation i @xmath216 with @xmath217 kpc and @xmath218 . the virial mass calculated by the circular velocity estimates based on simulation ii is @xmath219 with @xmath220 kpc and @xmath221 . the error bars on @xmath104 and @xmath215 are determined for 1-@xmath94 confidence intervals in a chi - square test , fixing the best - fit value of c , while the error bars on c are determined for 1-@xmath94 confidence intervals in a chi - square test , fixing the best - fit value of @xmath104 . note that although contracted and uncontracted halo fits differ quite strongly in their ( initial ) concentration , reassuringly the m@xmath222 estimates remain relatively unaffected . the lower concentrations for the contracted halo fits are still reasonally consistent with the concentration scatter expected from cosmological simulations ( e.g. , nfw 96 ) . the results here show that the milky way s circular velocity curve must be gently falling to distances of 60 kpc from its value of @xmath223 km / s at the solar radius ; the null hypothesis @xmath224 has a constant value of @xmath225 is rejected by our fits with very high statistical significance ( at a level of 0.01 ) . a direct comparison with earlier work , at the data or @xmath166 level , is not straightforward to carry out , as each sample has distinct selection effects , such as differing radial distributions . our estimate of @xmath104 , taken to be the average of the four estimates derived in this section , @xmath226m@xmath9 , falls well within the ( considerably larger ) range of values estimated by battaglia et al . ( 2005 , 2006 ) : @xmath227 . this value is also reasonably consistent with the recent estimate by li & white ( 2008 ) , based on local group dynamics . in general , however , the new mass estimate presented here lies at the lower limit of most previous estimates . the method to estimate @xmath5 can also be used to derive the escape velocity curve @xmath228 . we have carried out an analogous procedure to that described in 3.2 , using @xmath229 , and found lower limits on m@xmath230 , which do not appear to be as stringent as the @xmath5 estimates . we have constrained the mass distribution of the milky way s dark matter halo , by analyzing the kinematics of nearly @xmath231 bhb stars drawn from sdss dr-6 , which reach to galactocentric distances of @xmath232 kpc . to obtain a `` clean sample '' of bhb stars , we have re - analyzed all candidate bhb spectra , following the prescription by sirko et al . 2004 , which should result in a contamination fraction ( mostly by bs stars ) of well below 10% . the metallicity distribution , centered on [ fe / h ] @xmath233 dex , confirms that most sample members must be halo stars . the distances to the bhb stars are known to @xmath234 . to account for the complex survey geometry and for plausible orbital distributions of our sample of bhb stars , we have compared the observed radial velocities to analogous quantities drawn from the `` star particles '' in galaxy formation simulations that resemble the milky way . in particular , we have placed a virtual `` observer '' at 8.0 kpc from the simulation centers , looking in the actual sdss directions and sampling radial velocities for stars that are more than 4 kpc above and below the disk plane . we then explored to what mass scale , or @xmath5 , the simulations need to be scaled to in order to match the observed line - of - sight velocity distribution in a set of radial bins . in this analysis , we adjusted this scaling using the jeans equation , to account for the slight difference in the radial profile of observed halo stars ( @xmath235 ) and simulated halo stars ( @xmath236 ) over this radial range . this procedure results in direct estimates of @xmath5 from @xmath13 kpc to @xmath2 kpc , the best such estimates to date over this range . the circular velocity estimate varies slightly with radius , dropping from @xmath237 at 10 kpc to @xmath238 in the most distant two bins . applying this procedure to two independent cosmological simulations ( simulation i and simulation ii , respectively ) results in consistent estimates of @xmath5 . the mass enclosed within 60 kpc , constrained quite directly by the data , is found to be @xmath239m@xmath9 . as a result of the much more extensive data set provided by sdss , the uncertainties on this estimate are substantially lower than those obtained by previous comparable work . a simple , jeans- equation based modeling approach , assuming ( an-)isotropies of either @xmath240 ( as found for the halo stars in the cosmological simulations ) yields results that are consistent with these values . although each of the @xmath5 points were estimated independently , the implied overall profile is consistent with both the mass profile in the simulations and with a parameterized mass model that combines a fixed disk and bulge model with an nfw dark matter halo , whose concentration c corresponds to the expected mean value for its virial mass . we have also explored halo fits with the concentration c as a free parameter , as well as halo profiles that have been modified by adiabatic contraction . we find that our data can not discriminate well whether adiabatic contraction occurs or not an uncontracted halo of higher than average concentration and a contracted halo of ( initially ) low concentration fit comparably well . the resulting virial masses , @xmath241m@xmath9 , are consistent for both fitting approaches . we have also checked that these results are quite robust with respect to distance errors , modest sample contamination ( @xmath242 ) , and the choice of a different , independent galaxy formation simulation . the estimate of @xmath104 , which does imply an extrapolation from r@xmath243 kpc to the virial radius of @xmath244 kpc , is consistent with a recent estimate from a much smaller sample of halo stars ( battaglia et al . 2006 ) , but it is lower than previous estimates that also rely on the kinematics of satellite galaxies ( e.g. , kochanek et al . however , recent results on the lmc ( kallivayalil et al . 2006 ; besla et al . 2007 ) indicate that not even the magellanic clouds may have been bound to the milky way for long , posing a potential conceptual problems for the use of satellite dynamics . it should be interesting to explore how our new constraint modifies the local group timing argument and its implication for m31 s halo mass ( li & white 2008 ) . the estimate of @xmath245m@xmath9 , together with an estimated total cold baryonic mass of @xmath246m@xmath9 and a global baryon mass fraction of @xmath247 , implies that nearly 40% of all baryons within the virial radius have cooled to form the milky way s stars and ( cold ) gas , consistent with recent estimates for galaxies of that mass scale , based on statistical arguments ( van den bosh et al . 2007 ; gnedin et al . 2007 ) . we note that parts of our analysis have been performed under the assumption that the stellar halo of the galaxy is considered as a single relaxed population , or one that matches the simulations . our data on the overall dynamics are consistent with that hypothesis . recent evidence from carollo et al . ( 2007 ) and miceli et al . ( 2007 ) suggest that the halo may well be more complex , and comprise two distinct populations of inner- and outer - halo stars , with ( slightly ) different net rotational velocities and spatial profiles . we defer all analysis of kinematic and spatial sub - structure in this particular sample to a future paper . the sdss and segue surveys have shown in this context that they can provide unparalleled sets of kinematic tracers for the milky way , enabling a direct `` circular velocity curve '' estimate of the milky way extending to 60 kpc . once the full set of spectroscopy from segue is available , a much larger set of stars for such an analysis should be available . the proposed extension of sdss , known as sdss - iii ( and using a more sensitive , 1000 fiber spectrograph ) , will provide the opportunity to obtain higher quality spectra for fainter , more distant bhb stars , thus extending the reach of our analysis to over 100 kpc . this work was made possible by the support of the chinese academy of sciences and the max - planck - institute for astronomy , and is supported by the national natural science foundation of china under grants nos . 10433010 and 10521001 . tcb and ysl acknowledge partial support from the us national science foundation under grants ast 06 - 0715 and ast 07 - 07776 , as well as from grant phy 02 - 15783 ; physics frontier center / joint institute for nuclear astrophysics ( jina ) . funding for the sloan digital sky survey ( 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 , and the max planck society , and the higher education funding council for england . 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 american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , the university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , the 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 . although subtle , one can notice that the bs stars s balmer lines are wider at @xmath248 below the local continuum than the bhb star . these effects arise due to the higher gravity of the bs star . ] and @xmath37 , as determined from the @xmath38 line , for stars brighter than @xmath249 and passing the color cuts shown in figure [ f : f1 ] . the trail of stars with @xmath250 are too cool to be bhb stars , while the concentration of stars with @xmath251 is due to blue stragglers with higher surface gravity . the stars that lie well outside the main locus are the result of poor parameter determinations due to missing spectroscopic data at the location of the @xmath38 line . the region enclosed by the box is used as the bhb selection criterion for the @xmath38 , @xmath57 & @xmath58 method . ] , as a function of apparent magnitude , for the entire sample of halo bhb stars . ( lower panel ) the distribution of line - of - sight velocities , corrected to the gsr , for the entire sample of bhb stars . a gaussian of width @xmath254 km @xmath30 centered on the local standard of rest is shown for reference . , title="fig:",height=377 ] , as a function of apparent magnitude , for the entire sample of halo bhb stars . ( lower panel ) the distribution of line - of - sight velocities , corrected to the gsr , for the entire sample of bhb stars . a gaussian of width @xmath254 km @xmath30 centered on the local standard of rest is shown for reference . , title="fig:",height=377 ] as a function of galactocentric distance , @xmath87 , for the entire sample of halo bhb stars . ( lower panel ) the velocity dispersion , @xmath255 , as a function of galactocentric distance . a best fit exponentially falling relationship is plotted.,title="fig:",height=377 ] as a function of galactocentric distance , @xmath87 , for the entire sample of halo bhb stars . ( lower panel ) the velocity dispersion , @xmath255 , as a function of galactocentric distance . a best fit exponentially falling relationship is plotted.,title="fig:",height=377 ] between the halo star particles in the simulation i and the observations , shown here for all bins . the dashed line is @xmath256 , and the solid line is @xmath257 , after finding the best matching velocity scaling , @xmath128 , listed in table 3 ; the best match in this context is obtained from a k - s test . ] between the halo star particles in the simulation ii and the observations , shown here for all bins . the dashed line is @xmath256 , and the solid line is @xmath257 , after finding the best matching velocity scaling , @xmath128 , listed in table 3 ; the best match in this context is obtained from a k - s test . ] , for two different simulated galaxies . the filled circles are the v@xmath258 estimates for the observed halo bhb stars based on simulation i and the filled squares are the v@xmath258 estimates based on simulation ii . the two lines are the circular velocity curve estimates derived from the velocity dispersion profile ( figure [ f : f10 ] ) and the jeans equation with @xmath164 and @xmath168 . ] estimates , while the large symbols in the two plots are the v@xmath259 estimates . contributions of the adopted model components ( i.e. disk , bulge , and halo ) and the circular velocity curves based on the jeans equation are plotted in different linestyles . estimates of virial mass , m@xmath260 , virial radius , r@xmath260 and concentration parameter , @xmath47 , are labeled on the plots.,title="fig:",height=377 ] estimates , while the large symbols in the two plots are the v@xmath259 estimates . contributions of the adopted model components ( i.e. disk , bulge , and halo ) and the circular velocity curves based on the jeans equation are plotted in different linestyles . estimates of virial mass , m@xmath260 , virial radius , r@xmath260 and concentration parameter , @xmath47 , are labeled on the plots.,title="fig:",height=377 ] , but here the circular velocity curves were derived under the assumption of a contracted nfw profile . the solid line is the best - fit circular velocity curve to the v@xmath259 estimates , while the large symbols in the two plots are the v@xmath259 estimates . contributions of the adopted model components ( i.e. disk , bulge , and halo ) and the circular velocity curves based on the jeans equation are plotted in different linestyles . estimates of virial mass , m@xmath260 , virial radius , r@xmath260 and concentration parameter , @xmath47 are labeled on the plots.,title="fig:",height=377 ] , but here the circular velocity curves were derived under the assumption of a contracted nfw profile . the solid line is the best - fit circular velocity curve to the v@xmath259 estimates , while the large symbols in the two plots are the v@xmath259 estimates . contributions of the adopted model components ( i.e. disk , bulge , and halo ) and the circular velocity curves based on the jeans equation are plotted in different linestyles . estimates of virial mass , m@xmath260 , virial radius , r@xmath260 and concentration parameter , @xmath47 are labeled on the plots.,title="fig:",height=377 ] 51602 - 0266 - 225 & sdssj094218.23 - 002519.7 & 145.575943 & -0.422125 & 236.196579 & 36.896130 & 15.46 & -0.10 & 1.20 & 0.02 & 0.02 & 0.01 & -95.7 & 2.1 & -245.5 & 0.24 & 26.93 & 0.84 & 9.97 & bhb + 51602 - 0266 - 397 & sdssj094138.17 + 001821.5 & 145.409058 & 0.305967 & 235.328827 & 37.175690 & 18.48 & -0.11 & 1.22 & 0.04 & 0.02 & 0.02 & -36.0 & 12.8 & -183.8 & 0.20 & 27.78 & 1.12 & 10.57 & bhb + 51602 - 0266 - 634&sdssj094840.23 + 002818.0&147.167633&0.471673&236.434402 & 38.711887&18.01&-0.02&1.02&0.03&0.02&0.01&9.4&10.5 & -136.8&0.29 & 25.80 & 0.92&10.01&bhb + 51609 - 0292 - 102&sdssj125223.54 - 003708.2&193.098083 & -0.618937 & 303.444397&62.251862 & 14.29 & -0.24 & 1.27 & 0.02 & 0.02&0.02 & 227.3 & 1.6 & 148.7 & 0.23 & 27.00 & 1.03 & 10.16 & bhb + 51609 - 0292 - 155&sdssj125050.87 - 000806.1&192.711960 & -0.135032 & 302.609894&62.736347&17.84 & -0.00 & 1.06&0.02 & 0.02&0.02 & 190.8 & 7.2&112.8 & 0.24 & 27.46 & 0.85 & 10.27 & bhb + 51609 - 0292 - 232 & sdssj124759.81 - 000456.2&191.999207 & -0.082266 & 301.051117 & 62.776917&18.16 & -0.04&1.05 & 0.03&0.02 & 0.02 & -121.4 & 9.3 & -200.9 & 0.28 & 30.32 & 0.86 & 10.31 & bs + 51609 - 0292 - 269 & sdssj124721.11 - 002931.5 & 191.837952 & -0.492089 & 300.729431 & 62.362186 & 19.17 & -0.09&1.12 & 0.05 & 0.03 & 0.02 & 232.0 & 19.5 & 151.0 & 0.30 & 32.62 & 0.93 & 10.10 & ms + 51609 - 0292 - 329&sdssj124641.66 + 003751.2&191.673569 & 0.630884 & 300.275543&63.478165&17.15&-0.10&1.23 & 0.02 & 0.01&0.02 & -49.0 & 3.9 & -127.2 & 0.23 & 25.03 & 0.81&8.25&bhb + 51609 - 0292 - 351&sdssj124449.35 + 002157.4&191.205612 & 0.365958 & 299.263092&63.190571 & 17.64 & -0.26 & 1.13 & 0.02 & 0.01 & 0.01 & 4.2 & 5.7&-75.8 & 0.24 & 24.54 & 1.15 & 9.51 & bhb + 51609 - 0292 - 367 & sdssj124805.12 + 010113.5 & 192.021347&1.020428 & 301.028198 & 63.879784&16.38 & -0.17 & 1.22 & 0.03 & 0.02 & 0.02 & 84.8 & 2.8 & 8.6 & 0.24 & 34.34 & 0.87 & 11.53 & bs + 51602 - 0266 - 125 & sdssj094317.57 - 011021.2 & 145.823196 & -1.172550 & 237.138931 & 36.662491 & 16.48 & 1.17 & -0.22 & 0.55 & 26.60 & 0.21 & 1.02 & 10.07 & 15.3 & 20.2 & 14.7 & 10.3 & 9.2 & 178.0 & 3.5 & 26.2 & 8315 & 3.12 & -2.03 & bhb + 51602 - 0266 - 397 & sdssj094138.17 + 001821.5 & 145.409058 & 0.305967 & 235.328827 & 37.175690 & 18.48 & 1.22 & -0.11 & 0.55 & 27.78 & 0.20 & 1.12 & 10.57 & 38.5 & 42.8 & 25.5 & 25.3 & 23.3 & -36.0 & 12.8 & -183.8 & 8306 & 3.81 & -1.97 & bhb + 51602 - 0266 - 634 & sdssj094840.23 + 002818.0 & 147.167633 & 0.471673 & 236.434402 & 38.711887 & 18.01 & 1.02 & -0.02 & 0.60 & 25.80 & 0.29 & 0.92 & 10.01 & 30.3 & 34.6 & 21.1 & 19.7 & 19.0 & 9.4 & 10.5 & -136.8 & 7850 & 4.20 & -1.27 & bhb + 51609 - 0292 - 102 & sdssj125223.54 - 003708.2 & 193.098083 & -0.618937 & 303.444397 & 62.251862 & 14.29 & 1.27 & -0.24 & 0.55 & 27.00 & 0.23 & 1.03 & 10.16 & 5.6 & 8.5 & 6.6 & 2.2 & 5.0 & 227.3 & 1.6 & 148.7 & 8668 & 2.61 & -1.96 & bhb + 51609 - 0292 - 329 & sdssj124641.66 + 003751.2 & 191.673569 & 0.630884 & 300.275543 & 63.478165 & 17.15 & 1.23 & -0.10 & 0.55 & 25.03 & 0.23 & 0.81 & 8.25 & 20.9 & 20.6 & 3.3 & 8.1 & 18.7 & -49.0 & 3.9 & -127.2 & 7596 & 2.95 & -2.00 & bhb + 51609 - 0292 - 351 & sdssj124449.35 + 002157.4 & 191.205612 & 0.365958 & 299.263092 & 63.190571 & 17.64 & 1.13 & -0.26 & 0.60 & 24.54 & 0.24 & 1.15 & 9.51 & 25.6 & 25.1 & 2.4 & 10.1 & 22.8 & 4.2 & 5.7 & -75.8 & 9247 & 3.11 & -1.43 & bhb + 51609 - 0292 - 582 & sdssj125254.01 + 002903.2 & 193.225037 & 0.484220 & 303.747009 & 63.353657 & 16.22 & 1.23 & -0.13 & 0.55 & 26.21 & 0.25 & 0.85 & 9.09 & 13.6 & 14.0 & 4.6 & 5.1 & 12.2 & -42.9 & 3.2 & -118.0 & 7953 & 3.14 & -1.66 & bhb + 51609 - 0304 - 219 & sdssj141723.88 - 002220.8 & 214.349503 & -0.372432 & 343.356750 & 55.602268 & 16.64 & 1.05 & -0.31 & 0.70 & 23.05 & 0.20 & 1.19 & 9.68 & 15.4 & 13.0 & -0.3 & 2.5 & 12.7 & -25.0 & 3.8 & -50.1 & -9999 & -9.99 & -9.99 & bhb + 51613 - 0305 - 243 & sdssj142408.24 - 002930.7 & 216.034348 & -0.491848 & 345.608093 & 54.511181 & 15.71 & 1.17 & -0.02 & 0.60 & 25.01 & 0.25 & 0.87 & 8.94 & 10.5 & 8.9 & 2.1 & 1.5 & 8.6 & -78.4 & 2.4 & -99.4 & 7754 & 3.68 & -2.26 & bhb + 51613 - 0305 - 488 & sdssj142826.28 + 002915.4 & 217.109497 & 0.487611 & 348.109955 & 54.617874 & 17.54 & 1.17 & -0.26 & 0.60 & 24.71 & 0.22 & 1.12 & 9.55 & 24.4 & 21.0 & -5.8 & 2.9 & 19.9 & 74.7 & 4.2 & 59.4 & 9338 & 2.87 & -2.95 & bhb + 51614 - 0281 - 438 & sdssj112513.99 + 004207.5 & 171.308289 & 0.702082 & 261.294983 & 56.438610 & 18.51 & 1.13 & -0.23 & 0.60 & 24.60 & 0.22 & 1.16 & 9.44 & 38.2 & 39.7 & 11.2 & 20.9 & 31.8 & 252.4 & 9.6 & 134.5 & 8609 & 3.13 & -1.56 & bhb + lcc [ t : tbl3 ] 7.5 & @xmath261 & @xmath262 + 12.5 & @xmath263 & @xmath264 + 17.5 & @xmath265 & @xmath266 + 22.5 & @xmath267 & @xmath268 + 27.5 & @xmath269 & @xmath270 + 32.5 & @xmath271 & @xmath272 + 37.5 & @xmath273 & @xmath274 + 42.5 & @xmath275 & @xmath276 + 47.5 & @xmath277 & @xmath278 + 55.0 & @xmath279 & @xmath280 +

consider the problem of joint parameter estimation and prediction in a markov random field : i.e. , the model parameters are estimated on the basis of an initial set of data , and then the fitted model is used to perform prediction ( e.g. , smoothing , denoising , interpolation ) on a new noisy observation . working under the restriction of limited computation , we analyze a joint method in which the _ same convex variational relaxation _ is used to construct an m - estimator for fitting parameters , and to perform approximate marginalization for the prediction step . the key result of this paper is that in the computation - limited setting , using an inconsistent parameter estimator ( i.e. , an estimator that returns the `` wrong '' model even in the infinite data limit ) can be provably beneficial , since the resulting errors can partially compensate for errors made by using an approximate prediction technique . en route to this result , we analyze the asymptotic properties of m - estimators based on convex variational relaxations , and establish a lipschitz stability property that holds for a broad class of variational methods . we show that joint estimation / prediction based on the reweighted sum - product algorithm substantially outperforms a commonly used heuristic based on ordinary sum - product . * inconsistent parameter estimation in markov random fields : benefits in the computation - limited setting * + department of statistics , uc berkeley + technical report 690

cs0602092

graphical models such as markov random fields ( mrfs ) are widely used in many application domains , including spatial statistics , statistical signal processing , and communication theory . a fundamental limitation to their practical use is the infeasibility of computing various statistical quantities ( e.g. , marginals , data likelihoods etc . ) ; such quantities are of interest both bayesian and frequentist settings . sampling - based methods , especially those of the markov chain monte carlo ( mcmc ) variety @xcite , represent one approach to obtaining stochastic approximations to marginals and likelihoods . a disadvantage of sampling methods is their relatively high computational cost . for instance , in applications with severe limits on delay and computational overhead ( e.g. , error - control coding , real - time tracking , video compression ) , mcmc methods are likely to be overly slow . it is thus of considerable interest for various application domains to consider less computationally intensive methods for generating approximations to marginals , log likelihoods , and other relevant statistical quantities . variational methods are one class of techniques that can be used to generate deterministic approximations in markov random fields ( mrfs ) . at the foundation of these methods is the fact that for a broad class of mrfs , the computation of the log likelihood and marginal probabilities can be reformulated as a convex optimization problem ( see @xcite for an overview ) . although this optimization problem is intractable to solve exactly for general mrfs , it suggests a principled route to obtaining approximations namely , by relaxing the original optimization problem , and taking the optimal solutions to the relaxed problem as approximations to the exact values . in many cases , optimization of the relaxed problem can be carried out by `` message - passing '' algorithms , in which neighboring nodes in the markov random field convey statistical information ( e.g. , likelihoods ) by passing functions or vectors ( referred to as messages ) . estimating the parameters of a markov random field from data poses another significant challenge . a direct approach for instance , via ( regularized ) maximum likelihood estimation entails evaluating the cumulant generating ( or log partition ) function , which is computationally intractable for general markov random fields . one viable option is the pseudolikelihood method @xcite , which can be shown to produce consistent parameter estimates under suitable assumptions , though with an associated loss of statistical efficiency . other researchers have studied algorithms for ml estimation based on stochastic approximation @xcite , which again are consistent under appropriate assumptions , but can be slow to converge . as illustrated in figure [ figestpred ] , the problem domain of interest in this paper is that of joint estimation and prediction in a markov random field . more precisely , given samples @xmath0 from some unknown underlying model @xmath1 , the first step is to form an estimate of the model parameters . now suppose that we are given a noisy observation of a new sample path @xmath2 , and that we wish to form a ( near)-optimal estimate of @xmath3 using the fitted model , and the noisy observation ( denoted @xmath4 ) . examples of such prediction problems include signal denoising , image interpolation , and decoding of error - control codes . disregarding any issues of computational cost and speed , one could proceed via route a in figure [ figestpred]that is , one could envisage first using a standard technique ( e.g. , regularized maximum likelihood ) for parameter estimation , and then carrying out the prediction step ( which might , for instance , involve computing certain marginal probabilities ) by monte carlo methods . this paper , in contrast , is concerned with the _ computation - limited _ setting , in which both sampling or brute force methods are overly intensive . with this motivation , a number of researchers have studied the use of approximate message - passing techniques , both for problems of prediction @xcite as well as for parameter estimation @xcite . however , despite their wide - spread use , the theoretical understanding of such message - passing techniques remains limited , especially for parameter estimation . consequently , it is of considerable interest to characterize and quantify the loss in performance incurred by using computationally tractable methods versus exact methods ( i.e. , route b versus a in figure [ figestpred ] ) . more specifically , our analysis applies to variational methods that are based _ convex relaxations_. this class includes a broad range of extant methods among them the tree - reweighted sum - product algorithm @xcite , reweighted forms of generalized belief propagation @xcite , and semidefinite relaxations @xcite . moreover , it is straightforward to modify other message - passing methods ( e.g. , expectation propagation @xcite ) so as to `` convexify '' them . at a high level , the key idea of this paper is the following : given that approximate methods can lead to errors at both the estimation and prediction phases , it is natural to speculate that these sources of error might be arranged to partially cancel one another . the theoretical analysis of this paper confirms this intuition : we show that with respect to end - to - end performance , it is in fact beneficial , even in the infinite data limit , to learn the `` wrong '' the model by using _ inconsistent _ methods for parameter estimation . en route to this result , we analyze the asymptotic properties of m - estimators based on convex variational relaxations , and establish a lipschitz stability property that holds for a broad class of variational methods . we show that joint estimation / prediction based on the reweighted sum - product algorithm substantially outperforms a commonly used heuristic based on ordinary sum - product . the remainder of this paper is organized as follows . section [ secbackground ] provides background on markov random fields and associated variational representations , as well as the problem statement . in section [ secconvsurr ] , we introduce the notion of a convex surrogate to the cumulant generating function , and then illustrate this notion via the tree - reweighted bethe approximation @xcite . in section [ secjoint ] , we describe how any convex surrogate defines a particular joint scheme for parameter estimation and prediction . section [ secanal ] provides results on the asymptotic behavior of the estimation step , as well as the stability of the prediction step . section [ secmixgauss ] is devoted to the derivation of performance bounds for joint estimation and prediction methods , with particular emphasis on the mixture - of - gaussians observation model . in section [ secexperiments ] , we provide experimental results on the performance of a joint estimation / prediction method based on the tree - reweighted bethe surrogate , and compare it to a heuristic method based on the ordinary belief propagation algorithm . we conclude in section [ secdiscussion ] with a summary and discussion of directions for future work . consider an undirected graph @xmath5 , consisting of a set of vertices @xmath6 and an edge set @xmath7 . we associate to each vertex @xmath8 a multinomial random variable @xmath9 taking values in the set @xmath10 . we use the lower case letter @xmath11 to denote particular realizations of the random variable @xmath9 in the set @xmath12 . this paper makes use of the following exponential representation of a pairwise markov random field over the multinomial random vector @xmath13 . we begin by defining , for each @xmath14 , the @xmath15-valued indicator function @xmath16 & \defn & \begin{cases } 1 & \mbox{if $ x_s = j$ } \\ 0 & \mbox{otherwise } \end{cases}\end{aligned}\ ] ] these indicator functions can be used to define a potential function @xmath17 via @xmath18\end{aligned}\ ] ] where @xmath19 is the vector of exponential parameters associated with the potential . our exclusion of the index @xmath20 is deliberate , so as to ensure that the collection of indicator functions @xmath21 , \ ; j = 1 , \ldots , { \ensuremath{m}}- 1\}$ ] remain affinely independent . in a similar fashion , we define for any pair @xmath22 the pairwise potential function @xmath23 \ ; \ind_k[x_t],\end{aligned}\ ] ] where we use @xmath24 to denote the associated collection of exponential parameters , and @xmath25 \ , \ind_k[x_s ] , \ ; j , k=1,2 , \ldots , { \ensuremath{m}}- 1 \}$ ] for the associated set of sufficient statistics . overall , the probability mass function of the multinomial markov random field in exponential form can be written as @xmath26 here the function @xmath27\end{aligned}\ ] ] is the logarithm of the normalizing constant associated with @xmath28 . the collection of distributions thus defined can be viewed as a regular and minimal exponential family @xcite . in particular , the exponential parameter @xmath29 and the vector of sufficient statistics @xmath30 are formed by concatenating the exponential parameters ( respectively indicator functions ) associated with each vertex and edge viz . [ eqndefnover ] @xmath31 this notation allows us to write equation more compactly as @xmath32 . a quick calculation shows that @xmath33 , where @xmath34 is the dimension of this exponential family . the following properties of @xmath35 are well - known : [ lembasic ] ( a ) the function @xmath35 is convex , and strictly so when the sufficient statistics are affinely independent . + ( b ) it is an infinitely differentiable function , with derivatives corresponding to cumulants . in particular , for any indices @xmath36 , we have @xmath37 , \qquad \pdifftwo{\partc}{\eparam_\sumind}{\eparam_\beta } = \covv{\clipot_\sumind(x)}{\clipot_\beta(x)}{\eparam},\ ] ] where @xmath38 and @xmath39 denote the expectation and covariance respectively . we use @xmath40 to denote the vector of _ mean parameters _ defined element - wise by @xmath41 $ ] for any @xmath42 . a convenient property of the sufficient statistics @xmath30 defined in equations and is that these mean parameters correspond to marginal probabilities . for instance , when @xmath43 or @xmath44 , we have respectively [ eqnover ] @xmath45 = { \ensuremath{p(x_s = j \ , ; \ , \eparam ) } } , \quad \mbox{and } \\ \label{eqnoverb } \meanpar_{st;jk } & = & \exs_\eparam \big \ { \ind_j[x_s]\ , \ind_k[x_t ] \big \ } = { \ensuremath{p(x_s = j , x_t = k \ , ; \ , \eparam)}}.\end{aligned}\ ] ] this section is devoted to a systematic procedure for constructing convex functions that represent approximations to the cumulant generating function . we begin with a quick development of an exact variational principle , one which is intractable to solve in general cases . ( more details on this exact variational principle can be found in the papers @xcite . ) nonetheless , this exact variational principle is useful , in that various natural relaxations of the optimization problem can be used to define convex surrogates to the cumulant generating function . after a high - level description of such constructions in general , we then illustrate it more concretely with the particular case of the `` convexified '' bethe entropy @xcite . since @xmath35 is a convex and continuous function ( see lemma [ lembasic ] ) , the theory of convex duality @xcite guarantees that it has a variational representation , given in terms of its conjugate dual function @xmath46 , of the following form @xmath47 in order to make effective use of this variational representation , it remains determine the form of the dual function . a useful fact is that the exponential family arises naturally as the solution of an entropy maximization problem . in particular , consider the set of linear constraints @xmath48 \ , \defn \ ; \sum_{x \in \statespn } p(x ) \clipot_\sumind(x ) & = & \meanpar_\sumind \quad \mbox{for $ \sumind = 1 , \ldots , \df$},\end{aligned}\ ] ] where @xmath40 is a set of target mean parameters . letting @xmath49 denote the set of all probability distributions with support on @xmath50 , consider the _ constrained entropy maximization problem _ : maximize the entropy @xmath51 subject to the constraints . a first question is when there any distributions @xmath52 that satisfy the constraints . accordingly , we define the set @xmath53 \quad \mbox{for some $ p \in \mathcal{p}$ } \big \},\end{aligned}\ ] ] corresponding to the set of @xmath54 for which the constraint set is non - empty . for any the optimal value of the constrained maximization problem is @xmath55 ( by definition , since the problem is infeasible ) . otherwise , it can be shown that the optimum is attained at a unique distribution in the exponential family , which we denote by @xmath56 . overall , these facts allow us to specify the conjugate dual function as follows : @xmath57 see the technical report @xcite for more details of this dual calculation . with this form of the dual function , we are guaranteed that the cumulant generating function @xmath35 has the following variational representation : @xmath58 however , in general , solving the variational problem is intractable . this intractability should not be a surprise , since the cumulant generating function is intractable to compute for a general graphical model . the difficulty arises from two sources . first , the _ constraint set _ @xmath59 is extremely difficult to characterize exactly for a general graph with cycles . for the case of a multinomial markov random field , it can be seen ( using the minkowski - weyl theorem ) that @xmath59 is a polytope , meaning that it can be characterized by a finite number of linear constraints . the question , of course , is how rapidly this number of constraints grows with the number of nodes @xmath60 in the graph . unless certain fundamental conjectures in computational complexity turn out to be false , this growth must be non - polynomial ( see deza and laurent @xcite for an in - depth discussion of the binary case ) . tree - structured graphs are a notable exception , for which the junction tree theory @xcite guarantees that the growth is only linear in @xmath60 . second , the _ dual function _ @xmath61 lacks a closed - form representation for a general graph . note in particular that the representation is not explicit , since it requires solving a constrained entropy maximization problem in order to compute the value @xmath62 . again , important exceptions to this rule are tree - structured graphs . here a special case of the junction tree theory guarantees that any markov random field on a tree @xmath63 can be factorized in terms of its marginals as follows @xmath64 consequently , in this case , the negative entropy ( and hence the dual function ) can be computed explicitly as @xmath65 where @xmath66 and @xmath67 are the singleton entropy and mutual information , respectively , associated with the node @xmath68 and edge @xmath69 . for a general graph with cycles , in contrast , the dual function lacks such an explicit form , and is not easy to compute . given these challenges , it is natural to consider approximations to @xmath61 and @xmath59 . as we discuss in the following section , the resulting relaxed optimization problem defines a convex surrogate to the cumulant generating function . we now describe a general procedure for constructing convex surrogates to the cumulant generating function , consisting of two main ingredients . given the intractability of characterizing the marginal polytope @xmath59 , it is natural to consider a relaxation . more specifically , let @xmath70 be a convex and compact set that acts as an outer bound to @xmath59 . we use @xmath71 to denote elements of @xmath70 , and refer to them as _ pseudomarginals _ since they represent relaxed versions of local marginals . the second ingredient is to designed to sidestep the intractability of the dual function : in particular , let @xmath72 be a strictly convex and twice continuously differentiable approximation to @xmath61 . we require that the domain of @xmath72 ( i.e. , @xmath73 ) be contained within the relaxed constraint set @xmath70 . by combining these two approximations , we obtain a convex surrogate @xmath74 to the cumulant generating function , specified via the solution of the following relaxed optimization problem @xmath75 note the parallel between this definition and the variational representation of @xmath35 in equation . the function @xmath74 so defined has several desirable properties , as summarized in the following proposition : [ propbasicprop ] any convex surrogate @xmath74 defined via has the following properties : 1 . for each @xmath76 , the optimum defining @xmath74 is attained at a unique point @xmath77 . the function @xmath74 is convex on @xmath78 . it is differentiable on @xmath78 , and more specifically : @xmath79 \(i ) by construction , the constraint set @xmath70 is compact and convex , and the function @xmath72 is strictly convex , so that the optimum is attained at a unique point @xmath77 . + ( ii ) observe that @xmath74 is defined by the maximum of a collection of functions linear in @xmath29 , which ensures that it is convex @xcite . + ( iii ) finally , the function @xmath80 satisfies the hypotheses of danskin s theorem @xcite , from which we conclude that @xmath74 is differentiable with @xmath81 as claimed . given the interpretation of @xmath77 as a pseudomarginal , this last property of @xmath74 is analogous to the well - known cumulant generating property of @xmath35namely , @xmath82as specified in lemma [ lembasic ] . the following example provides a more concrete illustration of this constructive procedure , using a tree - based approximation to the marginal polytope , and a convexifed bethe entropy approximation @xcite . as with the ordinary bethe approximation @xcite , the cost function and constraint set underlying this approximation are exact for any tree - structured markov random field . [ [ relaxed - polytope ] ] relaxed polytope : + + + + + + + + + + + + + + + + + we begin by describing a relaxed version @xmath70 of the marginal polytope @xmath83 . let @xmath84 and @xmath85 represent a collection of singleton and pairwise pseudomarginals , respectively , associated with vertices and edges of a graph @xmath86 . these quantities , as locally valid marginal distributions , must satisfy the following set of local consistency conditions : @xmath87 by construction , we are guaranteed the inclusion @xmath88 . moreover , a special case of the junction tree theory @xcite guarantees that equality holds when the underlying graph is a tree ( in particular , any @xmath89 can be realized as the marginals of the tree - structured distribution of the form ) . however , the inclusion is strict for any graph with cycles ; see appendix [ appstrict ] for further discussion of this issue . [ [ entropy - approximation ] ] entropy approximation : + + + + + + + + + + + + + + + + + + + + + + we now define an entropy approximation @xmath90 that is finite for any pseudomarginal @xmath71 in the relaxed set @xmath91 . we begin by considering a collection @xmath92 of spanning trees associated with the original graph . given @xmath89 , there is for each spanning tree @xmath93a unique tree - structured distribution that has marginals @xmath84 and @xmath85 on the vertex set @xmath94 and edge set @xmath95 of the tree . using equations and , the entropy of this tree - structured distribution can be computed explicitly . convexified bethe entropy _ approximation is based on taking a convex combination of these tree entropies , where each tree is weighted by a probability @xmath96 $ ] . doing so and expanding the sum yields @xmath97 where @xmath98 $ ] are the _ edge appearance probabilities _ defined by the distribution @xmath99 over the tree collection . by construction , the function @xmath90 is differentiable ; moreover , it can be shown @xcite that it is strictly convex for any vector @xmath100 of strictly positive edge appearance probabilities . [ [ bethe - surrogate - and - reweighted - sum - product ] ] bethe surrogate and reweighted sum - product : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + we use these two ingredients the relaxation @xmath91 of the marginal polytope , and the convexified bethe entropy approximation to define the following convex surrogate @xmath101 since the conditions of proposition [ propbasicprop ] are satisfied , we are guaranteed that @xmath102 is convex and differentiable on @xmath78 , and moreover that @xmath103 , where ( for each @xmath76 ) the quantity @xmath77 denotes the unique optimum of problem . perhaps most importantly , the optimizing pseudomarginals @xmath77 can be computed efficiently using a _ tree - reweighted variant _ of the sum - product message - passing algorithm @xcite . this method operates by passing `` messages '' , which in the multinomial case are simply @xmath104-vectors of non - negative numbers , along edges of the graph . we use @xmath105 to represent the message passed from node @xmath106 to node @xmath107 . in the tree - reweighted variant , these messages are updated according to the following recursion @xmath108^{\rho_{ut } } } { \big[{\ensuremath{m}}_{st}(x_t)\big]^{1-\rho_{st}}}.\end{aligned}\ ] ] upon convergence of the updates , the fixed point messages @xmath109 yield the unique global optimum of the optimization problem via the following equations [ eqntrwmarg ] @xmath110^{\rho_{us } } , \quad \mbox{and } \\ \taupar_{st}(x_s , x_t ; \eparam ) & \propto & \exp \big\{\eparam_s(x_s ) + \eparam_t(x_t ) + \frac{\eparam_{st}(x_s , x_t)}{\rho_{st } } \big \ } \frac{\prod \limits_{u \in \neigh(s ) } \big[{\ensuremath{m}}_{us}(x_s)\big]^{\rho_{us } } \ ; \prod \limits_{v \in \neigh(s ) } \big[{\ensuremath{m}}_{vs}(x_s)\big]^{\rho_{vs } } } { { \ensuremath{m}}_{st}(x_t ) \ ; { \ensuremath{m}}_{ts}(x_s ) } \end{aligned}\ ] ] further details on these updates and their properties can be found in the paper @xcite . we now turn to consideration of how convex surrogates , as constructed by the procedure described in the previous section , are useful for both approximate parameter estimation as well as prediction . suppose that we are given i.i.d . samples @xmath111 from an mrf of the form , where the underlying true parameter @xmath112 is unknown . one standard way in which to estimate @xmath112 is via maximum likelihood ( possibly with an additional regularization term ) ; in this particular exponential family setting , it is straightforward to show that the ( normalized ) log likelihood takes the form @xmath113 where function @xmath114 is a regularization term with an associated ( possibly data - dependent ) weight @xmath115 . the quantities @xmath116 are the empirical moments defined by the data . for the indicator - based exponential representation , these empirical moments correspond to a set of singleton and pairwise marginal distributions , denoted @xmath117 and @xmath118 respectively . it is intractable to maximize the regularized likelihood directly , due to the presence of the cumulant generating function @xmath35 . thus , a natural thought is to use the convex surrogate @xmath74 to define an alternative estimator obtained by maximizing the regularized _ surrogate likelihood _ : @xmath119 by design , the surrogate @xmath74 and hence the surrogate likelihood @xmath120 , as well as their derivatives , can be computed in a straightforward manner ( typically by some sort of message - passing algorithm ) . it is thus straightforward to compute the parameter @xmath121 achieving the maximum of the regularized surrogate likelihood ( for instance , gradient descent would a simple though naive method ) . for the tree - reweighted bethe surrogate , we have shown in previous work @xcite that in the absence of regularization , the optimal parameter estimates @xmath122 have a very simple closed - form solution , specified in terms of the weights @xmath123 and the empirical marginals @xmath124 . ( we make use of this closed form in our experimental comparison in section [ secexperiments ] ; see equation . ) if a regularizing term is added , these estimates no longer have a closed - form solution , but the optimization problem can still be solved efficiently using the tree - reweighted sum - product algorithm @xcite . using such an estimator , we now consider a joint approach to estimation and prediction . recalling the basic set - up , we are given an initial set of i.i.d . samples @xmath125 from @xmath126 , where the true model parameter @xmath112 is unknown . these samples are used to form an estimate of the markov random field . we are then given a noisy observation @xmath127 of a new sample @xmath128 , and the goal is to use this observation in conjunction with the fitted model to form a near - optimal estimate of @xmath129 . the key point is that the same convex surrogate @xmath74 is used both in forming the surrogate likelihood for approximate parameter estimation , and in the variational method for performing prediction . for a given fitted model parameter @xmath76 , the central object in performing prediction is the posterior distribution @xmath130 . in the exponential family setting , for a fixed noisy observation @xmath127 , this posterior can always be written as a new exponential family member , described by parameter @xmath131 . ( here the term @xmath132 serves to incorporate the effect of the noisy observation . ) with this set - up , the procedure consists of the following steps : + examples of the prediction task in the final step include smoothing ( e.g. , denoising of a noisy image ) and interpolation ( e.g. , in the presence of missing data ) . we provide a concrete illustration of such a prediction problem in section [ secmixgauss ] using a mixture - of - gaussians observation model . the most important property of this joint scheme is that the _ convex surrogate _ @xmath74 underlies both the parameter estimation phase ( used to form the surrrogate likelihood ) , and the prediction phase ( used in the variational method for computing approximate marginals ) . it is this matching property that will be shown to be beneficial in terms of overall performance . in this section , we turn to the analysis of the surrogate - based method for estimation and prediction . we begin by exploring the asymptotic behavior of the parameter estimator . we then prove a lipschitz stability result applicable to any variational method that is based on a strongly concave entropy approximation . this stability result plays a central role in our subsequent development of bounds on the performance loss in section [ secmixgauss ] . we begin by considering the asymptotic behavior of the parameter estimator @xmath121 defined by the surrogate likelihood . since this parameter estimator is a particular type of @xmath133-estimator , its asymptotic behavior can be investigated using standard methods , as summarized in the following : [ propvan ] let @xmath74 be a strictly convex surrogate to the cumulant generating function , defined via equation with a strictly concave entropy approximation @xmath134 . consider the sequence of parameter estimates @xmath135 given by @xmath136 where @xmath114 is a non - negative and convex regularizer , and the regularization parameter satisfies @xmath137 . then for a general graph with cycles , the following results hold : 1 . we have @xmath138 , where @xmath139 is ( in general ) distinct from the true parameter @xmath112 . the estimator is asymptotically normal : @xmath140 & { \stackrel{d}{\longrightarrow } } & n \biggr(0 , \ , \big(\nabla^2 { \ensuremath{b}}({\ensuremath{\estim{\eparam}}})\big)^{-1 } \nabla^2 \partc(\eparams ) \big(\nabla^2 { \ensuremath{b}}({\ensuremath{\estim{\eparam}}})\big)^{-1 } \biggr)\end{aligned}\ ] ] by construction , the convex surrogate @xmath74 and the ( negative ) entropy approximation @xmath72 are a fenchel - legendre conjugate dual pair . from proposition [ propbasicprop ] , the surrogate @xmath74 is differentiable . moreover , the strict convexity of @xmath74 and @xmath72 ensure that the gradient mapping @xmath141 is one - to - one and onto the relative interior of the constraint set @xmath70 ( see section 26 of rockafellar @xcite ) . moreover , the inverse mapping @xmath142 exists , and is given by the dual gradient @xmath143 . let @xmath144 be the moment parameters associated with the true distribution @xmath112 ( i.e. , in the limit of infinite data , the asymptotic value of the parameter estimate is defined by @xmath145 note that @xmath144 belongs to the relative interior of @xmath59 , and hence to the relative interior of @xmath70 . therefore , equation has a unique solution @xmath146 . by strict convexity , the regularized surrogate likelihood has a unique global maximum . let us consider the optimality conditions defining this unique maximum @xmath121 ; they are given by @xmath147 , where @xmath148 denotes an arbitrary element of the subdifferential of the convex function @xmath114 at the point @xmath121 . we can now write @xmath149 - \lambda^n \partial { \ensuremath{r}}({\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}).\end{aligned}\ ] ] taking inner products with the difference @xmath150 yields @xmath151^t \ ; \left[{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}\right ] \ ; \leq \ ; \left[{\ensuremath{\estim{\meanpar}}}^{\ensuremath{n}}- \meanpar^ * \right]^t \left[{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}\right ] + \lambda^n \partial { \ensuremath{r}}({\ensuremath{\estim{\eparam}}}^{\ensuremath{n}})^t \left[{\ensuremath{\estim{\eparam}}}- { \ensuremath{\estim{\eparam}}}^{\ensuremath{n}}\right],\ ] ] where inequality ( a ) follows from the convexity of @xmath74 . from the convexity and non - negativity of @xmath114 , we have @xmath152 \ ; \leq \ ; \lambda^{\ensuremath{n}}\left [ { \ensuremath{r}}({\ensuremath{\estim{\eparam } } } ) - { \ensuremath{r}}({\ensuremath{\estim{\eparam}}}^{\ensuremath{n}})\right ] \ ; \leq \ ; \lambda^{\ensuremath{n}}{\ensuremath{r}}({\ensuremath{\estim{\eparam}}}).\ ] ] applying this inequality and cauchy - schwartz to equation yields @xmath153^t \ ; \left[\frac{{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}}{\|{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}\| } \right ] \ ; \leq \ ; \|{\ensuremath{\estim{\meanpar}}}^{\ensuremath{n}}- \meanpar^ * \| + \lambda^n { \ensuremath{r}}({\ensuremath{\estim{\eparam}}})\ ] ] since @xmath154 by assumption and @xmath155 by the weak law of large numbers , the quantity @xmath156^t \ ; \left[\frac{{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}}{\|{\ensuremath{\estim{\eparam}}}^{\ensuremath{n}}- { \ensuremath{\estim{\eparam}}}\| } \right]$ ] converges in probability to zero . by the strict convexity of @xmath74 , this fact implies that @xmath121 converges in probability to @xmath139 , thereby completing the proof of part ( a ) . to establish part ( b ) , we observe that @xmath157 { \stackrel{d}{\longrightarrow}}n(0 , \nabla^2 { \ensuremath{a}}(\eparams))$ ] by the central limit theorem . using this fact and applying the delta method to equation yields that @xmath158 { \stackrel{d}{\longrightarrow}}n \left(0 , \nabla^2 \partc(\eparams ) \right),\ ] ] where we have used the fact that @xmath159 . the strict convexity of @xmath74 guarantees that @xmath160 is invertible , so that claim ( b ) follows . a key property of the estimator is its _ inconsistency_i.e . , the estimated model differs from the true model @xmath112 even in the limit of large data . despite this inconsistency , we will see that the approximate parameter estimates @xmath121 are nonetheless useful for performing prediction . a desirable property of any algorithm particularly one applied to statistical data is that it exhibit an appropriate form of stability with respect to its inputs . not all message - passing algorithms have such stability properties . for instance , the standard sum - product message - passing algorithm , although stable for relatively weakly coupled mrfs @xcite , can be highly unstable in other regimes due to the appearance of multiple local optima in the non - convex bethe problem . however , previous experimental work has shown that methods based on convex relaxations , including the reweighted sum - product ( or belief propagation ) algorithm @xcite , reweighted generalized bp @xcite , and log - determinant relaxations @xcite appear to be very stable . for instance , figure [ figunstable ] provides a simple illustration of the instability of the ordinary sum - product algorithm , contrasted with the stability of the tree - reweighted updates . wiegerinck @xcite provides similar results for reweighted forms of the generalized belief propagation . here we provide theoretical support for these empirical observations : in particular , we prove that , in sharp contrast to non - convex methods , any variational method based on a strongly convex entropy approximation is globally stable . this stability property plays a fundamental role in providing a performance guarantee on joint estimation / prediction methods . we begin by noting that for a multinomial markov random field , the computation of the exact marginal probabilities is a globally lipschitz operation : [ lemexact ] for any discrete markov random field , there is a constant @xmath161 such that @xmath162 this lemma , which is proved in appendix [ appexact ] , guarantees that small changes in the problem parameters that is , `` perturbations '' @xmath163lead to correspondingly small changes in the computed marginals . our goal is to establish analogous lipschitz properties for variational methods . in particular , it turns out that any variational method based on a suitably concave entropy approximation satisfies such a stability condition . more precisely , a function @xmath164 is _ strongly convex _ if there exists a constant @xmath165 such that @xmath166 for all @xmath167 . for a twice continuously differentiable function , this condition is equivalent to having the eigenspectrum of the hessian @xmath168 be uniformly bounded below by @xmath169 . with this definition , we have : [ proplipstable ] consider any strictly convex surrogate @xmath74 based on a strongly concave entropy approximation @xmath134 . then there exists a constant @xmath170 such that @xmath171 from proposition [ propbasicprop ] , we have @xmath172 , so that the statement is equivalent to the assertion that the gradient @xmath141 is a lipschitz function . applying the mean value theorem to @xmath141 , we can write @xmath173 where @xmath174 $ ] . consequently , in order to establish the lipschitz condition , it suffices to show that the spectral norm of @xmath175 is uniformly bounded above over all @xmath176 . since @xmath74 and @xmath72 are a strictly convex legendre pair , we have @xmath177^{-1}$ ] . by the strong convexity of @xmath72 , we are guaranteed that the spectral norm of @xmath178 is uniformly bounded away from zero , which yields the claim . many existing entropy approximations can be shown to be strongly concave . in appendix [ appconv ] , we provide a detailed proof of this fact for the convexified bethe entropy . [ lemconv ] for any set @xmath179 of strictly positive edge appearance probabilities , the convexified bethe entropy is strongly concave . we note that the same argument can be used to establish strong concavity for the reweighted kikuchi approximations studied by wiegerinck @xcite . moreover , it can be shown that the gaussian - based log - determinant relaxation considered in wainwright and jordan @xcite is also strongly concave . for all of these variational methods , then , proposition [ proplipstable ] guarantees that the pseudomarginal computation is globally lipschitz stable , thereby providing theoretical confirmation of previous experimental results @xcite . in this section , we develop theoretical bounds on the performance loss of our approximate approach to joint estimation and prediction , relative to the unattainable bayes optimum . so as not to unnecessarily complicate the result , we focus on the performance loss in the infinite data limit corrections , hold for the finite data setting . ] ( i.e. , for which the number of samples @xmath180 ) . in the infinite data setting , the bayes optimum is unattainable for two reasons : 1 . it is based on knowledge of the exact parameter @xmath112 , which is not easy to obtain . 2 . it assumes ( in the prediction phase ) that computing exact marginal probabilities @xmath54 of the markov random field is feasible . of these two difficulties , it is the latter assumption regarding the computation of marginal probabilities that is the most serious . as discussed earlier , there do exist computationally tractable estimators of @xmath112 that are consistent though not statistically efficient under appropriate conditions ; one example is the pseudolikelihood method @xcite mentioned previously . on the other hand , mcmc methods may be used to generate stochastic approximations to marginal probabilities , but may require greater than polynomial complexity . recall from proposition [ propvan ] that the parameter estimator based on the surrogate likelihood @xmath120 is _ inconsistent _ , in the sense that the parameter vector @xmath139 returned in the limit of infinite data is generally not equal to the true parameter @xmath112 . our analysis in this section will demonstrate that this inconsistency is beneficial . although the ideas and techniques described here are more generally applicable , we focus here on a special observation model so as to obtain a concrete result . [ [ observation - model ] ] observation model : + + + + + + + + + + + + + + + + + + in particular , we assume that the multinomial random vector @xmath181 defined by the markov random field is a label vector for the components in a finite mixture of gaussians . for each node @xmath68 , we specify a new random variable @xmath182 by the conditional distribution @xmath183 so that @xmath182 is a mixture of @xmath104 gaussians . such gaussian mixture models are widely used in spatial statistics as well as statistical signal and image processing @xcite . now suppose that we observe a noise - corrupted version of @xmath184 namely , a vector @xmath4 of observations with components of the form @xmath185 where @xmath186 is additive gaussian noise , and the parameter @xmath187 $ ] specifies the signal - to - noise ratio ( snr ) of the observation model . note that @xmath188 corresponds to pure noise , whereas @xmath189 corresponds to completely uncorrupted observations . [ [ optimal - prediction ] ] optimal prediction : + + + + + + + + + + + + + + + + + + + our goal is to compute an optimal estimate @xmath190 of @xmath129 as a function of the observation @xmath191 , using the mean - squared error as the risk function . the essential object in this computation is the posterior distribution @xmath192 , where the conditional distribution @xmath193 is defined by the observation model . as shown in the sequel , the posterior distribution ( with @xmath127 fixed ) can be expressed as an exponential family member of the form @xmath194 ( see equation ) . disregarding computational cost , it is straightforward to show that the optimal bayes least squares estimator ( blse ) takes the form @xmath195,\end{aligned}\ ] ] where @xmath196 denotes the marginal probability associated with the posterior distribution , and @xmath197 is the usual blse weighting for a gaussian with variance @xmath198 . [ [ approximate - prediction ] ] approximate prediction : + + + + + + + + + + + + + + + + + + + + + + + since the marginal distributions @xmath196 are intractable to compute exactly , it is natural to consider an approximate predictor , based on a set @xmath71 of pseudomarginals computed from a variational relaxation . more explicitly , we run the variational algorithm on the parameter vector @xmath199 that is obtained by combining the new observation @xmath127 with the fitted model @xmath139 , and use the outputted pseudomarginals @xmath200 as weights in the approximate predictor @xmath201,\end{aligned}\ ] ] where the weights @xmath202 are defined in equation . we now turn to a comparison of the bayes least - squares estimator ( blse ) defined in equation to the surrogate - based predictor . since ( by definition ) the blse is optimal for the mean - squared error ( mse ) , using the surrogate - based predictor will necessarily lead to a larger mse . our goal is to prove an upper bound on the maximal possible increase in this mse , where the bound is specified in terms of the underlying model @xmath112 and the snr parameter @xmath203 . more specifically , for a given problem , we define the mean - squared errors @xmath204 of the bayes - optimal and surrogate - based predictors , respectively . we seek upper bounds on the increase @xmath205 of the approximate predictor relative to bayes optimum . before providing a technical statement and proof , we begin with some intuition underlying the bounds , and the role of lipschitz stability . first , consider the low snr regime ( @xmath206 ) in which the observation @xmath4 is heavily corrupted by noise . in the limit @xmath207 , the new observations are pure noise , so that the prediction of @xmath3 should be based simply on the estimated model namely , the true model @xmath126 in the bayes optimal case , and the `` incorrect '' model @xmath208 for the method based on surrogate likelihood . the key point here is the following : by properties of the mle and surrogate - based estimator , the following equalities hold : @xmath209 here equality ( a ) follows from lemma [ lembasic ] , whereas equality ( b ) follows from the moment - matching property of the mle in exponential families . equalities ( c ) and ( d ) hold from the proposition [ propbasicprop ] and the pseudomoment - matching property of the surrogate - based parameter estimator ( see proof of proposition [ propvan ] ) . as a key consequence , it follows that the combination of surrogate - based estimation and prediction is _ functionally indistinguishable _ from the bayes - optimal behavior in the limit of @xmath188 . more specifically , in the limiting case , the errors systematically introduced by the inconsistent learning procedure are cancelled out exactly by the approximate variational method for computing marginal distributions . of course , exactness for @xmath188 is of limited interest ; however , when combined with the lipschitz stability ensured by proposition [ proplipstable ] , it allows us to gain good control of the low snr regime . at the other extreme of high snr ( @xmath210 ) , the observations are nearly perfect , and hence dominate the behavior of the optimal estimator . more precisely , for @xmath203 close to @xmath211 , we have @xmath212 for all @xmath213 , so that @xmath214 . consequently , in the high snr regime , accuracy of the marginal computation has little effect on the accuracy of the predictor . although bounds of this nature can be developed in more generality , for simplicity in notation we focus here on the case of @xmath215 mixture components . we begin by introducing the factors that play a role in our bound on the performance loss @xmath216 . first , the lipschitz stability enters in the form of the quantity : @xmath217 where @xmath218 denotes the maximal singular value . following the argument in the proof of proposition [ proplipstable ] , it can be seen that @xmath219 is finite . second , in order to apply the lipschitz stability result , it is convenient to express the effect of introducing a new observation vector @xmath127 , drawn from the additive noise observation model , as a perturbation of the exponential parameterization . in particular , for any parameter @xmath33 and observation @xmath127 from the model , the conditional distribution @xmath220 can be expressed as @xmath221 , where the exponential parameter @xmath222 has components @xmath223 see appendix [ appgamform ] for a derivation of these relations . third , it is convenient to have short notation for the gaussian estimators of each mixture component : @xmath224 with this notation , we have the following [ thmbound ] the mse increase is upper bounded by @xmath225 before proving the bound , we begin by considering its behavior in a few special cases . [ [ effect - of - snr ] ] effect of snr : + + + + + + + + + + + + + + first , consider the low snr limit in which @xmath226 . in this limit , it can be seen that @xmath227 , so that the the overall bound @xmath228 tends to zero . similarly , in the high snr limit as @xmath229 , we see that @xmath230 for @xmath231 , which drives the differences @xmath232 , and in turn the overall bound @xmath228 to zero . thus , the surrogate - based method is optimal in both the low and high snr regimes ; its behavior in the intermediate regime is governed by the balance between these two terms . [ [ effect - of - equal - variances ] ] effect of equal variances : + + + + + + + + + + + + + + + + + + + + + + + + + + now consider the special case of equal variances @xmath233 , in which case @xmath234 . thus , the difference @xmath235 simplifies to @xmath236 , so that the bound reduces to @xmath237 as shown by the simpler expression , for @xmath238 , the mse increase is very small , since such a two - component mixture is close to a pure gaussian . [ [ effect - of - mean - difference ] ] effect of mean difference : + + + + + + + + + + + + + + + + + + + + + + + + + + finally consider the case of equal means @xmath239 in the two gaussian mixture components . in this case , we have @xmath240 \ ; [ y_s - { \ensuremath{\nu}}]$ ] , so that the bound reduces to @xmath241 ^ 2 \ ; \exs \left \ { \min \left ( 1 , \ ; { \ensuremath{l}}(\eparams ; { \ensuremath{\estim{\eparam } } } ) \frac{\|{\ensuremath{\gamma}}(\ysca ; { \ensuremath{\alpha}})\|_2}{\sqrt{{\ensuremath{n } } } } \right ) \sqrt{\frac{\sum_s ( \ysca_s - { \ensuremath{\nu}})^4}{{\ensuremath{n } } } } \right \}.\end{aligned}\ ] ] here the mse increase depends on the snr @xmath203 and the difference @xmath242 \ ; \left[{\ensuremath{\alpha}}^2 \sigma^2_1 + ( 1-{\ensuremath{\alpha}}^2)\right]}.\end{aligned}\ ] ] observe , in particular , that the mse increases tends to zero as the difference @xmath243 decreases . by the pythagorean relation that characterizes the bayes least squares estimator @xmath244 , we have @xmath245 using the definitions of @xmath246 and @xmath247 , some algebraic manipulation yields @xmath248 ^ 2 & = & \left[\tau_s({\ensuremath{\estim{\eparam}}}+{\ensuremath{\gamma } } ) - \mu_s(\eparams + { \ensuremath{\gamma } } ) \right]^2 \ ; \left [ { \ensuremath{{\ensuremath{g}}_1}}(y_s ) - { \ensuremath{{\ensuremath{g}}_0}}(y_s ) \right]^2 \\ & \leq & \left |\tau_s({\ensuremath{\estim{\eparam}}}+ { \ensuremath{\gamma } } ) - \mu_s(\eparams+ { \ensuremath{\gamma } } ) \right | \left [ { \ensuremath{{\ensuremath{g}}_1}}(y_s ) - { \ensuremath{{\ensuremath{g}}_0}}(y_s ) \right]^2,\end{aligned}\ ] ] where the second inequality uses the fact that @xmath249 since @xmath250 and @xmath251 are marginal probabilities . next we write @xmath252 ^ 2 \nonumber \\ \label{eqninter } & \leq & \frac{1}{\sqrt{{\ensuremath{n } } } } \|\tau({\ensuremath{\estim{\eparam}}}+ { \ensuremath{\gamma } } ) - \mu(\eparams + { \ensuremath{\gamma } } ) \|_2 \ ; \sqrt { \frac{\sum_{s=1}^{\ensuremath{n}}\left where the last line uses the cauchy - schwarz inequality . it remains to bound the 2-norm @xmath253 . an initial naive bound follows from the fact @xmath254 $ ] implies that @xmath255 , whence @xmath256 an alternative bound , which will be better for small perturbations @xmath257 , can be obtained as follows . using the relation @xmath258 guaranteed by the definition of the ml estimator and surrogate estimator , we have @xmath259 + \left [ \mu(\eparams ) - \mu(\eparams + { \ensuremath{\gamma}})\right ] \right \|_2 \\ & = & \left \| \left [ \nabla^2 { \ensuremath{b}}({\ensuremath{\estim{\eparam}}}+ s { \ensuremath{\gamma } } ) - \nabla^2 \partc(\eparams + t { \ensuremath{\gamma } } ) \right ] { \ensuremath{\gamma}}\right \|_2,\end{aligned}\ ] ] for some @xmath260 $ ] , where we have used the mean value theorem . thus , using the definition of @xmath261 , we have @xmath262 combining the bounds and and applying them to equation , we obtain @xmath263 taking expectations of both sides yields the result . in order to test our joint estimation / prediction procedure , we have applied it to coupled gaussian mixture models on different graphs , coupling strengths , observation snrs , and mixture distributions . here we describe both experimental results to quantify the performance loss of the tree - reweighted sum - product algorithm @xcite , and compare it to both a baseline independence model , as well as a closely related heuristic method that uses the ordinary sum - product ( or belief propagation ) algorithm . in section [ secjointprocedure ] , we described a generic procedure for joint estimation and prediction . here we begin by describing the special case of this procedure when the underlying variational method is the tree - reweighted sum - product algorithm @xcite . any instantiation of the tree - reweighted sum - product algorithm is specified by a collection of edge weights @xmath123 , one for each edge @xmath264 of the graph . the vector of edge weights must belong to the spanning tree polytope ; see wainwright et al . @xcite for further background on these weights and the reweighted algorithm . given a fixed set of edge weights @xmath99 , the joint procedure based on the tree - reweighted sum - product algorithm consists of the following steps : 1 . given an initial set of i.i.d . data @xmath0 , we first compute the empirical marginal distributions @xmath265 , \qquad \estim{\meanpar}_{st}(j , k ) \defn \frac{1}{{\ensuremath{n } } } \sum_{i=1}^{\ensuremath{n}}\ind[x^i_s = j ] \ ; \ind[x^i_t = k],\ ] ] and use them to compute the approximate parameter estimate @xmath266 as shown in our previous work @xcite , the estimates are the global maxima of the surrogate likelihood based on the convexified bethe approximation without any regularization term ( i.e. , @xmath267 ) . 2 . given the new noisy observation @xmath4 of the form , we incorporate it by by forming the new exponential parameter @xmath268 where equation defines @xmath269 for the gaussian mixture model under consideration . we then compute approximate marginals @xmath270 by running the trw sum - product algorithm with edge appearance weights @xmath123 , using the message updates , on the graphical model distribution with exponential parameter @xmath199 . we use the approximate marginals ( see equation ) to construct the prediction @xmath271 in equation . we evaluated the tree - reweighted sum - product based on its increase in mean - squared error ( mse ) over the bayes optimal predictor . moreover , we compared the performance of the tree - reweighted approach to the following alternatives : 1 . as a baseline , we used the _ independence model _ in which the mixture distributions at each node are all assumed to be independent . in this case , ml estimates of the parameters are given by @xmath272 , with all of the coupling terms @xmath273 equal to zero . the prediction step reduces to computing the bayes least squares estimate at each node independently , based only on the local data @xmath274 . the _ standard sum - product or belief propagation _ ( bp ) approach is closely related to the tree - reweighted sum - product method , but based on the edge weights @xmath275 for all edges . in particular , we first form the approximate parameter estimate @xmath139 using equation with @xmath275 . as shown in our previous work @xcite , this approximate parameter estimate uniquely defines the markov random field for which the empirical marginals @xmath276 and @xmath277 are fixed points of the ordinary belief propagation algorithm . we note that a parameter estimator of this type has been used previously by other researchers @xcite . in the prediction step , we then use the ordinary belief propagation algorithm ( i.e. , again with @xmath275 ) to compute approximate marginals of the graphical model with parameter @xmath199 . finally , based on these approximate bp marginals , we compute the approximate predictor using equation . although our methods are more generally applicable , here we show representative results for @xmath278 components , and two different types of gaussian mixtures . 1 . mixture ensemble a is bimodal , with components @xmath279 and @xmath280 . mixture ensemble b was constructed with mean and variance components @xmath281 and @xmath282 ; these choices serve to mimic heavy - tailed behavior . in both cases , each mixture component is equally weighted ; see figure [ figmixensemble ] for histograms of the resulting mixture ensembles . [ cols="^,^,^ " , ] panel ( a ) provides the comparison ensembles of type a , with fixed variances @xmath283 and mean vectors @xmath284 ranging from @xmath285 to @xmath286 . note how the bounds capture the qualitative behavior for low snr , for which the difficulty of the problem increases as the mean separation is increased . in contrast , in the high snr regime , the bounds are extremely conservative , and fail to predict that the sharp drop - off in error as the snr parameter @xmath203 approaches one . this drop - off is particularly pronounced for the ensemble with largest mean separation ( marked with @xmath287 ) . panel ( b ) provides a similar comparison for ensembles of type b , with fixed mean vectors @xmath288 , and variances @xmath289 ranging from @xmath290 to @xmath291 . in this case , although the bounds are still very conservative in quantitative terms , they reasonably capture the qualitative behavior of the error over the full range of snr . key challenges in the application of markov random fields include the estimation ( learning ) of model parameters , and performing prediction using noisy samples ( e.g. , smoothing , interpolation , denoising ) . both of these problems present substantial computational challenges for general markov random fields . in this paper , we have described and analyzed methods for joint estimation and prediction that are based on convex variational methods . our central result is that using inconsistent parameter estimators can be beneficial in the computation - limited setting . indeed , our results provide rigorous confirmation of the fact that using parameter estimates that are `` systematically incorrect '' is helpful in offsetting the error introduced by using an approximate method during the prediction step . in concrete terms , we demonstrated that a joint prediction / estimation method using the tree - reweighted sum - product algorithm yields good performance across a wide range of experimental conditions . although our work has focused on a particular scenario , we suspect that similar ideas and techniques will be useful in related applications of approximate methods for learning and prediction . this work was supported by an alfred p. sloan foundation fellowship , an okawa foundation research fellowship , an intel corporation equipment grant , and nsf grant dms-0528488 . as an illustration on the single cycle on @xmath292 vertices , the pseudomarginal vector with elements @xmath293 belongs to @xmath91 for all choices @xmath294 $ ] , but fails to belong to @xmath59 , for instance , when @xmath295 . using lemma [ lembasic ] and the mean value theorem , we write @xmath296 for some @xmath297 . hence , it suffices to show that the eigenspectrum of the hessian is uniformly bounded above by @xmath298 . the functions @xmath30 are all 0 - 1 valued indicator functions , so that the diagonal elements of @xmath299 are bounded above in particular , @xmath300 for any index @xmath301 . consequently , we have @xmath302 as required . consider a spanning tree @xmath93 of @xmath86 with edge set @xmath95 . given a vector @xmath303 , we associate with @xmath93 a subvector @xmath304 formed by those components of @xmath71 associated with vertices @xmath94 and edges @xmath95 . note that by construction @xmath305 . the mapping @xmath306 can be represented by a projection matrix @xmath307 with the block structure @xmath308 in this definition , we are assuming for convenience that @xmath71 is ordered such that the @xmath309 components corresponding to the tree @xmath93 are placed first . with this notation , we have @xmath310 . by our construction of the function @xmath102 , there exists a probability distribution such that @xmath311 , where @xmath312 denotes the negative entropy of the tree - structured distribution defined by the vector of marginals @xmath304 . hence , the hessian of @xmath102 has the decomposition @xmath313 to check dimensions of the various quantities , note that @xmath314 is a @xmath315 matrix , and recall that each matrix @xmath316 . now by lemma [ lemexact ] , the eigenvalues of the @xmath317 are uniformly bounded above ; hence , the eigenvalues of @xmath318 are uniformly bounded away from zero . hence , for each tree @xmath93 , there exists a constant @xmath319 such that for all @xmath320 @xmath321 substituting this relation into our decomposition and expanding the sum over @xmath93 yields @xmath322 \sum_{s \in \vertex } \|z\{s\}\|^2 + \sum_{(s , t ) \in \edge } \big [ \sum_{\treegr \in \tract } { \ensuremath{\rho}}(\treegr ) { \ensuremath{c}}_\treegr \ind[(s , t ) \in \edge(\treegr ) ] \ ; \big ] \ ; \|z\{(s , t ) \ } \|^2 . \qquad\end{aligned}\ ] ] defining @xmath323 , we have the lower bounds @xmath324 & \geq & { \ensuremath{c}}^ * \sum_{\treegr \in \tract } { \ensuremath{\rho}}(\treegr ) \ ; = \ ; { \ensuremath{c}}^ * \ ; > \ ; 0\\ \sum_{\treegr \in \tract } { \ensuremath{\rho}}(\treegr ) { \ensuremath{c}}_\treegr \ind[(s , t ) \in \edge(\treegr ) ] & \geq & { \ensuremath{c}}^ * { \treegr \in \tract } { \ensuremath{\rho}}(\treegr ) \ind[(s , t ) \in \edge(\treegr ) ] \ ; = \ ; { \ensuremath{c}}^ * { \ensuremath{\rho}}_{st } \ ; \geq \ ; { \ensuremath{c}}^ * { \ensuremath{\rho}}^ * > 0,\end{aligned}\ ] ] where @xmath325 . applying these bounds to equation yields the final inequality @xmath326 with @xmath327 , which establishes that the eigenvalues of @xmath328 are bounded away from zero consider the observation model @xmath329 , where @xmath330 and @xmath184 is a mixture of two gaussians @xmath331 and @xmath332 . conditioned on the value of the mixing indicator @xmath333 , the distribution of @xmath274 is gaussian with mean @xmath334 and variance @xmath335 . let us focus on one component @xmath336 in the factorized conditional distribution @xmath337 . for @xmath338 , it has the form @xmath339 } } \exp \big \ { - \frac{1}{2 \big[{\ensuremath{\alpha}}^2 \sigma^2_j + ( 1 - { \ensuremath{\alpha}}^2 ) \big ] } ( y_s - { \ensuremath{\alpha}}{\ensuremath{\nu}}_{j } ) ^2 \big \}.\end{aligned}\ ] ] we wish to represent the influence of this term on @xmath11 in the form @xmath340 for some exponential parameter @xmath341 . we see that @xmath341 should have the form @xmath342 } { \big[{\ensuremath{\alpha}}^2 \sigma^2_1 + ( 1 - { \ensuremath{\alpha}}^2 ) \big ] } + \frac{(y_s - { \ensuremath{\alpha}}{\ensuremath{\nu}}_0)^2 } { 2 \big[{\ensuremath{\alpha}}^2 \sigma^2_0 + ( 1 - { \ensuremath{\alpha}}^2 ) \big ] } - \frac{(y_s - { \ensuremath{\alpha}}{\ensuremath{\nu}}_1)^2}{2 \big[{\ensuremath{\alpha}}^2 \sigma^2_1 + ( 1 - { \ensuremath{\alpha}}^2 ) \big ] } \end{aligned}\ ] ] m. j. wainwright , t. s. jaakkola , and a. s. willsky . tree - reweighted belief propagation algorithms and approximate ml estimation by pseudomoment matching . in _ workshop on artificial intelligence and statistics _ , january 2003 .

suppose we make a series of measurements on a chosen quantum system . the outcomes of the measurements form a sequence of random events which occur in a particular order . the system , together with a meter or meters , can be seen as following the paths of a stochastic network connecting all possible outcomes . the paths are shaped from the virtual paths of the system , and the corresponding probabilities are determined by the measuring devices employed . if the measurements are highly accurate , the virtual paths become `` real '' , and the mean values of a quantity ( a functional ) is directly related to the frequencies with which the paths are travelled . if the measurements are highly inaccurate , the mean ( weak ) values are expressed in terms of the relative probabilities amplitudes . for pre- and post - selected systems they are bound to take arbitrary values , depending on the chosen transition . this is a direct consequence of the uncertainty principle , which forbids one to distinguish between interfering alternatives , while leaving the interference between them intact .

1606.01845

a sequence of measurements made on a quantum system leads to number of random outcomes , observed by an experimentalist . in this sense , the composite system+meter(s ) may be seen as following a path across a stochastic network , connecting all possible measurement results . the same can be said about a purely classical system , and the quantum nature of the experiment reveals itself in the manner , in which the transition probabilities are calculated . while a classical theory allows for non - invasive observations , a quantum meter plays an active role in `` shaping '' the network . in the simplest case of von neumann ( vn ) measurements @xcite , the transition probabilities , as well as the observed events themselves , are determined by the virtual paths available to the studied system , the nature of the measured quantity , and the accuracy of the meter(s ) . perhaps the most studied network is produced by an accurate vn measurement of a projector on a state @xmath0 ( preparation , or pre - selection ) , followed by a not necessarily accurate vn measurement of an operator @xmath1 , followed by an accurate measurement of a projector on a final state @xmath2 ( post - selection ) @xcite . post - selection may succeed or fail , and a sub - ensemble is selected by collecting the statistics only in the case it is successful . if @xmath1 has @xmath3 distinct eigenvalues , there a @xmath3 virtual paths connecting the two states , for which quantum mechanics provides probability amplitudes , but not probabilities , if the system is considered in isolation . if the measurement of @xmath1 is accurate ( ideal , strong ) , the meter destroys the interference between the paths , which can be now equipped with probabilities ( squared of the moduli of the corresponding amplitudes ) , and become `` real '' . each time the experiment is repeated , the pointer points at one of the eigenvalues of the operator @xmath1 , whose mean value is then calculated using the relative frequencies of these occurrences . the probability to end up in @xmath2 is not what it would be , had the meter not perturbed the system s evolution . more controversial is the opposite limit , where the intermediate measurement is made inaccurate ( weak ) , in order to avoid the said perturbation . the weakness of the measurement inevitably causes the meter s readings to be spread , covering the whole real axis in the limit when the perturbation is sent to zero @xcite . this gives an operational meaning to the uncertainty principle , which states that the value of @xmath1 in a superposition of its eigenstates must be _ indeterminate_. indeed , trying to measure it without destroying the superposition would yield any value at all . much of the controversy resides , however , in the description of the mean of a weak meter s publication of @xcite , it gave rise a number of intriguing concepts . these include `` negative kinetic energy '' @xcite , `` negative number of particles '' @xcite , `` having one particle in several places simultaneously '' @xcite , `` photons disembodied from its polarisation '' @xcite , `` electrons with disembodied charge and mass '' @xcite , and `` an atom with the internal energy disembodied form the mass '' @xcite , all supported by the `` evidence '' of weak measurements . recently we have shown @xcite that the above statements amount to over - interpretations , easily dismissed once the weak values are identified with the transition amplitudes on the paths connecting the initial and final states , and are not seen as the values of the measured quantities . in this paper we will continue to look at the problem of consecutive quantum measurements , and the classical statistical ensembles produced whenever a meter , or meters , interact with the observed quantum system . we will pay special attention to the importance of the quantum uncertainty principle , and emphasise the role quantum interference plays in the loss of information about the system s past . throughout the paper , we will use the simplest case of a two - level system as an example , and refer the reader to refs . @xcite -@xcite for a more general analysis . the rest of the paper is organised as follows . in sect . ii we compare two of the best known formulations of the uncertainty principle . in sect . iii we briefly describe the work of a von neumann meter . in section iv we introduce a simple classical toy model , to be compared with its quantum counterparts later . section v returns to the simplest quantum network produced by a vn measurement of a quantity , with pre- and post - selection . in section vi we use our approach to describe a measurement of the difference of two physical quantities , in such a way that the values themselves remain indeterminate . section vii sets the rules for combining virtual paths . accurate ( strong ) , and inaccurate ( weak ) limits of the measurements are analysed in sections viii and ix , respectively . in sect . x the uncertainty principle is used to explain the properties of the weak values . common misconceptions related with the issue are discussed in sect . section xii contains our conclusions . there are several formulations of the uncertainty principle ( up ) covering various aspects of the problem ( for a recent review see @xcite ) . perhaps best known is the one relating the respective standard deviations , @xmath4 and @xmath5 , of two variables , represented by non - commuting operators @xmath1 and @xmath6 , measured in the same state . the so - called robertson s uncertainty relation @xcite reads @xmath7\ra|}{2 } , \end{aligned}\ ] ] where @xmath8 , @xmath9\equiv \hat{a}\hat{b}-\hat{b}\hat{a}$ ] is the commutator , and the angular brackets indicate averaging in the pure state @xmath10 the system is supposed to be in , @xmath11 . equation ( [ q1 ] ) is often interpreted as an indication of the fact that a measurement of @xmath12 must disturb a measurement of @xmath13 , and vice versa . a somewhat different approach to the up can be found in @xcite , where one reads : _ any determination of the alternative taken by a process capable of following more that one alternative destroys the interference between alternatives_. this principle is complemented by the rule for assigning probabilities to alternative scenarios @xcite : _ when an event can occur in several alternative ways , the probability amplitude for the event is the sum of the probability amplitudes for each way considered separately . if an experiment is performed which is capable of determining when one or another alternative is actually taken , the probability of the event is the sum of the probabilities of each alternative_. from the above statements , to which we will refer as the `` feynman s uncertainty principle '' , one may conclude that interfering alternatives can not be told apart and must , therefore , form a single indivisible pathway @xcite . another corollary to the principle is that interference must be destroyed by a physical agent , e.g. , a meter coupled to the observed system . thus , construction of probabilities by means of squaring the moduli of the corresponding amplitudes , pre - supposes the existence of a suitable meter , as well as the fact that the meter has already been deployed @xcite . it is the feynman s up to which we will appeal in what follows . despite the obvious difference between the formulations , the feynman s principle is not all that different from the up expressed by eq.([q1 ] ) . consider , for example , a system with a zero hamiltonian , @xmath14 , prepared at @xmath15 in a state @xmath0 . an accurate measurement of @xmath6 at @xmath16 yields an eigenvalue @xmath17 with the probability @xmath18 , where @xmath19 is the corresponding eigenstate . suppose we also want to learn something about the value of @xmath1 at @xmath20 . the system must be , in some sense , in one of the eigenstates of @xmath1 . to see in what sense , precisely , we insert the unity @xmath21 into the amplitude @xmath22 . the result is @xmath23 . there are now different ways ( paths , pathways , routes ) to reach the final state by `` passing through a state @xmath24 . without a meter , we can not determine which of them is actually taken . thus , despite the abundance of virtual paths , there is only a single ' ' real `` one , ' ' travelled " with the probability @xmath25 . if , on the other hand , we employ a meter , capable of telling us which of the eigenvalues @xmath26 has occurred at @xmath20 , the virtual paths will become real , with the @xmath27-th path travelled with the probability @xmath28 . we also find that the measurement of @xmath1 has disturbed the result of measuring @xmath6 , since @xmath29 quantifying the disturbance would lead to estimates similar to ( [ q1 ] ) . we will not follow this matter any further , and in the next section discuss possible realisation of quantum measurements . in the following we will often refer to von neumann ( vn ) measurements @xcite , whose definition we will revisit in this section . a vn meter , designed to measure a variable @xmath1 of a system governed by a hamiltonian @xmath30 , consists of a pointer ( e.g. , a massive particle on a line ) whose position is @xmath31 , and which is coupled to the observed system via ( @xmath32 ) @xmath33 if the value of @xmath1 is measured at some @xmath34 , the function @xmath35 is chosen to be @xmath36 , where @xmath37 is the dirac delta . we will assume that @xmath1 has discrete eigenvalues @xmath38 , and @xmath39 are the corresponding eigenstates , @xmath40 . prior to the interaction , at @xmath15 , the system is described by a state @xmath41 , and the pointer is in the state @xmath42 , such that @xmath43 is a real function ( e.g. , a gaussian ) , peaked around @xmath44 , and rapidly tending to zero as @xmath45 @xcite . immediately after the interaction has taken place , the pointer becomes entangled with the system , and their wave function is given by @xmath46 where @xmath47\psi\ra$ ] we assume further that the final position of the pointer can be determined precisely , leaving out the fundamental question of how exactly this is done , and whether a collapse of the wave function has occurred @xcite . in one way or another , after the interaction has taken place , we may `` look '' at the pointer , and find a meter s reading @xmath31 . perhaps the best known practical realisation of a vn measurement is the stern - gerlach apparatus ( see , for example , @xcite and @xcite ) . other vn meters can be constructed by using the spin - orbit interaction @xcite- @xcite , or bose - einstein condensates trapped in double - well potentials @xcite . if nothing else is done to the system , the probability ( density ) to find a reading @xmath48 is ( @xmath49 $ ] stands for the trace of the operator inside the brackets ) @xmath50= \sum_i |c_i(t_0)|^2g^2(\xi_0-a_i),\end{aligned}\ ] ] where @xmath51|\psi(t+0)\ra$ ] . alternatively , at some @xmath52 we may try to find ( post - select ) the system in some state @xmath2 , and keep the meter s readings only if the post - selection is successful . with this additional condition the probability to find a reading @xmath48 becomes @xmath53=\n |\sum_i \la \phi |\exp[-i\hat{h}(t - t_0)]|i\ra g(\xi_0-a_i)\la i|\exp(-i\hat{h}t_0)|\psi\ra|^2,\end{aligned}\ ] ] thus , apart from the preparation ( pre - selection ) of the system in a state @xmath0 , two other events have occurred : the pointer was found pointing at @xmath48 , and the system was later found in a final state @xmath2 . quantum mechanics provides the probabilities of such occurrences by means of expressions similar to eqs.([n3 ] ) and ( [ n4 ] ) . we can imagine , for example , performing more measurements between @xmath15 and @xmath16 , thus creating a network of scenarios , each of which would occur with certain frequency if the experiment is repeated many times . before discussing such `` quantum '' networks , we briefly turn to their classical counterparts . we start with a simple toy model which , we hope , will be helpful for the discussion of the following sections . suppose one has at his / her disposal a kit containing purely classical elements , little balls , tubes and connectors , which can be joined together . a connector @xmath54 has two entrances , and @xmath55 is the probability to enter via the inlet @xmath56 , and subsequently leave via the outlet @xmath57 . if one of the outlets is blocked , the probability to exit via the remaining one is unity for both entrances . connecting the elements as shown in fig . 1 , one can `` engineer '' a network of paths , which a ball inserted at the top will travel at random , until ending up in one of the final states @xmath58 and @xmath59 . ascribing variables ( numbers ) @xmath60 , @xmath61 , @xmath62 , @xmath63 to each of the connectors , one can also associate numbers , ( functionals ) with each path . for example , in fig.1a there are eight paths @xmath64 , @xmath65 . four of them end in the final state @xmath58 , @xmath66 : @xmath67 , travelled with the probability @xmath68=w_{b_1}(1|1)w_{a_1}(1|1)w_{in}(1|1)$ ] , @xmath69 : @xmath70 , travelled with the probability @xmath71=w_{b_1}(1|2)w_{a_2}(2|1)w_{in}(2|1)$ ] , @xmath72 : @xmath73 , travelled with the probability @xmath74=w_{b_2}(2|2)w_{a_1}(2|1)w_{in}(1|1)$ ] , @xmath75 : @xmath76 , travelled with the probability @xmath77=w_{b_2}(2|1)w_{a_2}(1|1)w_{in}(2|1)$ ] , and another four , @xmath78 , @xmath79 , end in @xmath59 . it is always possible to consider sub - networks containing only few paths , regardless of what is happening at later times , or elsewhere in the diagram . for example , we may limit our attention only to the cases where the ball goes to @xmath62 . as a result of this `` post - selection '' there are only two paths : @xmath66 , through @xmath80 and @xmath69 , through @xmath81 , travelled with probabilities @xmath68=w_{a_1}(1|1)w_{in}(1|1)$ ] and @xmath71=w_{a_1}(2|2)w_{in}(2|1)$ ] , respectively . there is a single random variable , a functional @xmath82 $ ] taking the values @xmath38 , @xmath83 , and whose mean value , evaluated over many runs by counting the times the ball passes through @xmath60 and @xmath61 , is @xmath84f[1]+p[2]f[2]\}/\ { p[1]+p[2]\}$ ] . alternatively , we may focus only on those cases where the ball ends up in @xmath58 , and ask for the difference between the values of @xmath85 and @xmath86 , @xmath87 , which is recorded for each run . let @xmath88 and @xmath89 . now the said difference is a functional @xmath90 $ ] defined on the four paths introduced above , and taking the values @xmath91=f_2[4]=0 $ ] , @xmath92=-2 $ ] and @xmath93=2 $ ] . since two of the values of @xmath94 are the same , we can reduce the number of paths by one , by combining @xmath66 and @xmath75 into a single pathway , @xmath95 , travelled with a probability @xmath96=p[1]+p[4]$ ] , and on which @xmath97 $ ] is understood to be zero . the resulting equivalent system is shown in fig.1b , and for @xmath98 , evaluated by writing down the value of @xmath94 for each run of the ball , adding up the results , and dividing by the number of runs , we find @xmath99f[1]+p[3]f[2]}{p[1 + 4]+p[2]+p[3]}.\end{aligned}\ ] ] we have invoked the simple toy model in fig.1 in order to emphasise its resemblances to and differences from the quantum case outlined at the end of previous section . as in the classical case , a set of vn measurements creates sequences of events occurring with certain probabilities , which can be seen as classical pathways `` travelled '' with certain frequencies , if the the experiment is repeated many times . the quantum origin of the experiment becomes evident in the way the probabilities of the events are calculated , as well as in the nature of these events . a meter recombines and divides the system s virtual paths , endowed with probability amplitudes only , thereby producing `` real '' ones , to which probabilities can be assigned . in particular , two interfering paths are combined into a single one in a much stronger sense , than in the classical case . for example , in fig.1b the paths @xmath66 and @xmath75 are combined into @xmath95 , since we chose to be interested only in the difference @xmath85 of @xmath38 , and not in their actual values . we will show that in the quantum case , these actual values may not even exist . finally , in the classical case , it is understood that two different setups built from the elements of the same kit can be completely different , and need not share any statistical properties . in the quantum case this is less clear , and we will return to the problem of `` contextuality '' later . as the first example , we consider a vn measurement of an operator @xmath100 , for a two - level system , pre- and post - selected in the states @xmath0 and @xmath2 , at @xmath15 and @xmath16 , respectively . for simplicity we put the system s hamiltonian to zero , @xmath14 , so that nothing happens between @xmath15 and @xmath34 , as well as between @xmath34 and @xmath16 . our purpose is to show that this creates a classical statistical ensemble not too dissimilar from the one built from the classical elements in the previous section . to see how this is done , we consider the two _ virtual _ paths connecting @xmath0 and @xmath2 , @xmath64 , @xmath83 . these are @xmath66 : @xmath101 , with an amplitude @xmath102=\la \phi |1\ra\la1|\psi\ra$ ] , and @xmath69 : @xmath103 , with an amplitude @xmath104=\la \phi |2\ra\la 2|\psi\ra$ ] . we also have @xmath105=\la \phi|\psi\ra$ ] . the value of @xmath1 at @xmath106 is a functional @xmath107 $ ] defined on the two paths and , obviously , taking the values @xmath60 and @xmath61 @xmath108=a_1,\q \text{and } \q f[2]=a_2.\end{aligned}\ ] ] without a meter , we may define an _ amplitude distribution _ @xmath109 for the value @xmath110 taken by @xmath111 , @xmath112\delta(f - a_1)+a[2]\delta(f - a_2 ) , \q \int \phi(f)df=\sum_{i=1}^2a[i]=\la \phi|\psi\ra \end{aligned}\ ] ] but can not say what is the _ probability _ for having , e.g. , the value @xmath60 . with a meter ( [ q1 ] ) coupled to the system , the final pointer s state @xmath113 is easily found to be given by @xmath114g(\xi - a_1)+a[2]g(\xi - a_2),\q\q\end{aligned}\ ] ] and the meter reads @xmath48 with a ( unnormalised ) probability @xmath115g(\xi_0-a_1)+a[2]g(\xi_0-a_2)|^2 . \end{aligned}\ ] ] equation ( [ y5 ] ) may be interpreted in the following way . by turning the meter on , we have created a continuum of _ real _ paths , all labelled by @xmath31 . these connect the three observed events , @xmath116 @xmath117 @xmath118 @xmath117 @xmath119 , assuming the meter is read before the post - selection is performed . the range of possible scenarios is determined by the width of the initial pointer s state @xmath120 in eq.([n1a ] ) , which also determines the accuracy of the measurement . for example , if @xmath120 is chosen to be @xmath121 the possible values of @xmath31 would lie inside the the interval @xmath122 $ ] , assuming @xmath123 . the path labelled @xmath31 is travelled with a probability ( density ) @xmath124 . far from passively observing the motion of the system , a meter actively participates in shaping the observed statistical ensemble . accordingly , the probabilities in eq.([y5 ] ) contain both the amplitudes @xmath125 $ ] , which characterise the unobserved system , and the meter s own @xmath120 . in the general case we may learn little about the the quantity of interest , @xmath1 . instead of extracting its value , we can only conclude that if the meter reading is @xmath48 , the system s state after interaction with the meter was @xcite @xmath126 for which the precise value of @xmath1 remains indeterminate . it is , however , possible to say that the measurement `` drives '' the system through a state @xmath127 , with a probability @xmath128 . to see that this is , indeed , the case , we could make an accurate measurement of the projector onto @xmath129 just after @xmath130 . repeating the procedure many times will show that whenever the reading of the first meter lies in a narrow interval around @xmath48 , the additional projection will also succeed . thus , while every attempt to measure @xmath1 leads to a particular stochastic network , not every measurement allows us to determine the value of @xmath1 . we will return to the question of the accuracy shortly after considering yet another example . suppose that , as in previous section , we have a two - level system pre- and post - selected in the states @xmath0 and @xmath2 at @xmath15 and @xmath16 , respectively . now we want to learn something about the difference of the quantities represented by operators @xmath131 and @xmath132 , which may or may not commute , at @xmath133 and @xmath134 , such that @xmath135 . [ classically , we could consider , for example , the @xmath54-and @xmath136-components of the momentum , and ask for @xmath137 . quantally , there is more than one way to approach this task . \(i ) we could employ two vn meters , set to measure @xmath1 and @xmath6 separately , and evaluate the difference of their readings . \(ii ) we could also construct a hermitian operator formally representing the difference of interest , @xmath138 \hat{b}\exp[i\hat{h}(t_2-t_1)]-\hat{a}$ ] , and use a single vn meter to measure it at @xmath133 . as discussed in @xcite , a classical analog of this procedure is a predictive measurement , which uses the fact that knowing the system s coordinates and momenta at @xmath133 allows one to restore the system s subsequent trajectory and , with it , all future values of all quantities . \(iii ) a procedure unique to quantum mechanics allows one to evaluate the difference of interest , leaving the actual values of @xmath1 and @xmath6 indeterminate . in general , all three methods give different results @xcite , and here we are only interested in the option ( iii ) . with this in mind , we couple a vn pointer ( position @xmath31 ) , to @xmath1 and @xmath6 at @xmath133 and @xmath134 , respectively , so that the interaction with the meter is given by @xmath139,\end{aligned}\ ] ] and the pointer s initial state , @xmath120 has , as before , a gaussian shape , centred at @xmath44 . the measurement works as follows : after the first coupling the pointer is shifted by one of the eigenvalues of @xmath1 , and after the second one by an eigenvalue of @xmath6 , albeit in the opposite sense . the resulting shift is , thus , the difference of the two values . this is not exactly the scheme used by von neumann @xcite , and we will call this meter von neuman - like @xcite . without a meter there are four virtual paths , @xmath64 , @xmath140 , which connect @xmath0 and @xmath2 , and pass through different intermediate states . for example , the path going through @xmath141 , and then through @xmath142 has a probability amplitude @xmath143 |2_b\ra\la2_b|\exp[-i\hat{h}(t_2-t_1)]|1_a\ra\la1_a|\exp(-i\hat{h}t_1)]|\psi\ra$ ] . putting , as before , @xmath14 we , therefore , have the paths @xmath66 : @xmath144 , with an amplitude @xmath102=\la \phi |1_b\ra\la1_b| 1_a\ra\la1_a|\psi\ra$ ] , @xmath69 : @xmath145 , with an amplitude @xmath104=\la \phi|1_b\ra\la1_b| 2_a\ra\la2_a|\psi\ra$ ] , @xmath72 : @xmath146 , with an amplitude @xmath147=\la \phi|2_b\ra\la2_b| 1_a\ra\la1_a|\psi\ra$ ] , @xmath75 : @xmath148 , with an amplitude @xmath149=\la \phi|2_b\ra\la2_b| 2_a\ra\la2_a|\psi\ra$ ] , and also the condition @xmath150=\la \phi|\psi\ra$ ] . the four paths are easily identified in the diagram in fig.3a . the difference between @xmath1 and @xmath6 is a functional @xmath90 $ ] defined on the four paths , where it may take four different values @xmath151=b_1-a_1,\q f_2[2]=b_1-a_2,\q f_2[3]=b_2-a_1,\q \text{and } \q f_2[4]=b_2-a_2.\end{aligned}\ ] ] let @xmath1 and @xmath6 represent components of the spin ( without a factor of 1/2 ) along two different axes , so that @xmath88 , and @xmath89 . now @xmath94 takes three different values , @xmath152 , @xmath153 and @xmath154 . with no meter employed , we can not say what is the chance of having , e.g. , @xmath155 , but , as before , may write down an amplitude distribution @xmath109 for the value @xmath110 of the functional @xmath94 , @xmath156+a[4]\}\delta(f)+a[2]\delta(f+2)+a[3]\delta(f-2),\end{aligned}\ ] ] normalised by the condition @xmath157=\la \phi|\psi\ra$ ] . as in the previous section , this so far purely cosmetic re - arrangement of the virtual paths is useful for describing the effects of a meter , once it is coupled to the system as prescribed by eq . ( [ x1 ] ) . after a successful post - selection the pointer is in a pure state @xmath113 , and it is a simple matter to check that @xmath158+a[4]\}g(\xi)+a[2]g(\xi+2)+a[3]g(\xi-2).\q\q\end{aligned}\ ] ] we can now look at eq.([x4 ] ) from the point of view adopted in the previous section . as in sect . iv , by external manipulation ( connecting the system to a meter ) we have created a classical statistical ensemble , where the system can reach the final state vie a continuum of real paths , labelled by the the variable @xmath31 . again , the meter plays an active role , and the ( unnormalised ) probability to travel the @xmath48-th path is determined by the shape of @xmath159 , which is the property of the pointer , as well as the amplitudes of the virtual paths , which represent the system `` on its own '' , @xmath160f[1]+p[3]f[2]}{p[1 + 4]+p[2]+p[3]}. p(\xi_0)=| \{a[1]+a[4]\}g(\xi)+a[2]g(\xi+2)+a[3]g(\xi-2)|^2.\end{aligned}\ ] ] with this , the mean reading of the meter over many trials is readily seen to be given by @xmath161 as in the previous example , the value of @xmath94 remains indeterminate , if there is more than one term in r.h.s . of eq.([x5 ] ) . to establish a good correlation between the pointer readings @xmath31 and the values of @xmath94 we must , therefore , try to avoid the overlap between different @xmath159 s in ( [ x5 ] ) . we will do so after discussing the general rules for combining virtual paths in the next section . the feynman s uncertainty principle , cited in sect . ii , implies that virtual paths can be combined and recombined into new sets of paths , with interference being responsible for the loss of information about certain aspects of quantum motion . next we briefly discuss the rules for creating such combinations . we start with the well known case of changing the basis functions in which a state of the system can be expanded at a given time . let a two - state system be in the state @xmath162 , where @xmath163 , and @xmath164 is the amplitude with which @xmath165 enters in @xmath10 . consider a different basis , @xmath166 , @xmath167 , which is related to the first basis by a unitary transformation , @xmath168 . now the amplitude @xmath85 , with which @xmath166 enters in @xmath10 , is @xmath169 , @xmath83 . these equations can be read as follows : if a new basis state is a weighted sum of the old ones , the corresponding amplitude is the weighted sum of the corresponding amplitudes . the same rules can be applied to virtual paths , and we can describe the action of the meter in sect.v in a similar manner . if the pointer reading is @xmath48 , the meter combines virtual paths @xmath66 and @xmath69 with amplitudes @xmath102 $ ] and @xmath104 $ ] , into a single real path @xmath170 , * h the amplitudes @xmath102 $ ] and @xmath104 $ ] , are combined into a new path , @xmath171 , with complex valued weights @xmath172 and @xmath173 , then the amplitude for the new pathway is @xmath174=\alpha a[1 ] + \beta a[2].\end{aligned}\ ] ] if a functional @xmath175 $ ] takes different definite values on @xmath66 and @xmath69 , @xmath176\ne f[2]$ ] , its value on @xmath171 is indeterminate , ( we will elaborate on this statement in sect.x ) . that the width of @xmath120 , @xmath177 , determines the resolution , or the accuracy , of a measurement is readily seen if @xmath120 is chosen to have the form of a `` rectangular window ( [ y5b ] ) . inserting ( [ y5b ] ) in eq . ( [ x4 ] ) yields @xmath178 and we can be sure that if the reading if @xmath48 , the value of the measured functional , @xmath110 , is no smaller than @xmath179 , and no greater than @xmath180 . we are , however , ' ' banned " from looking inside the interval @xmath181 $ ] , where all information about @xmath110 is lost to the interference which the meter has failed to destroy . next we look at the question of accuracy in more detail . suppose we can change the width @xmath177 of the initial pointer state , e.g. , by choosing @xmath182 where the factor @xmath183 ensures that the state is properly normalised , @xmath184 . since , by assumption , the pointer s final position is evaluated accurately , it is the uncertainty of its initial setting that determines precision of the measurement . for a highly accurate measurements we should choose @xmath177 as small as possible , thus requiring @xmath185 if so , in the example of sect . v , eq.([y5 ] ) reduces to @xmath186|^2 \delta(\xi - a_1)+|a[2]|^2 \delta(\xi - a_2)\equiv p[1 ] \delta(\xi - a_1)+p[2 ] \delta(\xi - a_2 ) . \end{aligned}\ ] ] and of all real paths survive only two . equivalently , we can say that the action of the meter has destroyed all interference between the virtual paths @xmath66 and @xmath69 , which can now be equipped with the probabilities @xmath68 $ ] and @xmath71 $ ] , respectively . the paths are `` real '' in the sense that each time the experiment is repeated , the pointer will indicate which of the two paths has been travelled , with @xmath68/(p[1]+p[2])$ ] and @xmath71/(p[1]+p[2]$ ] giving the relative frequencies of each occurrence . note that the appearance of @xmath187|^2 $ ] , later interpreted as probabilities , is not postulated , but is a consequence of the change in the physical state of the system , caused by its interaction with the meter @xcite . recalling that whenever post - selection in the state @xmath2 fails , the system ends up in its orthogonal compliment , @xmath188 , @xmath189 , we can add two new paths , @xmath72 and @xmath75 , connecting @xmath0 with @xmath188 , and travelled with probabilities @xmath190=|\la \phi_\perp |1\ra\la1|\psi\ra|^2 , \end{aligned}\ ] ] and @xmath191=|\la \phi_\perp |2\ra\la2|\psi\ra|^2 . \end{aligned}\ ] ] respectively . we now have an exact equivalent of the ( upper part of the ) model in fig.1 . the only difference is that this time it is built from quantum `` elements '' , rather than from pieces of a classical kit . the rest is a simple exercise in classical statistics . in both cases we can calculate the averages of the functional @xmath111 . for example , conditioned by successful post - selection in @xmath2 , the mean value of @xmath111 is given by @xmath192p[1]+f_1[2]p[2])/(p[1]+p[2])$ ] . we can also choose the operator @xmath1 to be the projector on the state @xmath193 , so that its eigenvalues are @xmath194 and @xmath153 , and we have @xmath195=1 $ ] and @xmath196=0 $ ] . now the mean value of @xmath111 yields precisely the probability to travel the first path , @xmath197}{p[1]+p[2]}.\end{aligned}\ ] ] classically , we would arrive at the same result by writing down @xmath194 whenever the ball passes through @xmath60 , @xmath153 if it passes through @xmath61 , add up what is written , and divide by the number of the balls we used . the case case of measuring the difference of operators @xmath198 and @xmath199 , discussed in sect.vi , can be analysed in a similar manner . there an accurate meter creates three real paths connecting @xmath0 and @xmath2 . the three pathways are @xmath95 , travelled with a probability @xmath96=|a[1]+a[4]|^2 $ ] , @xmath69 , travelled with a probability @xmath71=|a[2]|^2 $ ] , and @xmath72 , travelled with a probability @xmath74=|a[3]|^2 $ ] . adding three more paths connecting @xmath0 with @xmath188 , we obtain a classical stochastic system shown in fig.3 , similar to that built from the tubes and connectors in fig.1 . the mean value of the functional @xmath94 defined in eq.([x2 ] ) , conditioned by arriving in @xmath2 , is now twice the difference of the relative frequncies with which the paths @xmath69 and @xmath72 are travelled , @xmath200p[1 + 4]+f_2[2]p[2]+f_2[3]p[3]}{p[1 + 4]+p[2]+p[3]}= \frac{2(p[2]-p[3])}{p[1 + 4]+p[2]+p[3]}.\end{aligned}\ ] ] to conclude , we return to the remark made at the end of sect . iv . joining the connectors and tubes in different ways , one may construct completely different statistical networks , which have nothing in common , except their constituent parts . in a similar way , by applying different measurement schemes , and measuring different operators to different accuracies , one may engineer equally different stochastic ensembles , having nothing in common except their parent quantum system . ignoring this fact may lead to various counterfactual paradoxes , some of which disappear once the above remark is taken into account ( for more detail see @xcite and the refs . therein ) . next we look at what happens in the opposite limit , @xmath201 there are several reasons for doing it . firstly , as in the case of the accurate limit considered in the previous section , the result is universal in the sense that it no longer depends on the particular choice of @xmath120 , and appears to represent some `` intrinsic '' property of the studied system . secondly , observation of the system s history changes , in general , the probability to end up in the state @xmath2 which , when an accurate meter is employed , is different from that without a meter , @xmath202\ne|\la \phi |\exp(-i\hat{h}t)|\psi\ra|^2.\end{aligned}\ ] ] this is a well known example of `` one measurement perturbing the other '' , and it is tempting to try to get rid of the perturbation , and to see what will then occur . the perturbation disappears if , and and only if , all @xmath159 s in eq.([x4 ] ) have approximately the same value , @xmath203 , so that @xmath124 in eq.([x5 ] ) is @xmath204 . this can only happen in the limit ( [ v1 ] ) , when @xmath120 is made very broad . one immediate consequence of taking the limit ( [ v1 ] ) is that the range of possible meter s readings has widened to occupy the whole real axis , @xmath205 this gives an operational meaning to the word `` indeterminate '' we have often used above . if one wishes to keep interference between paths intact , he / she _ must _ choose the meter so inaccurate , that its reading will be arbitrary , leaving the researcher completely ignorant of the value of the functional in question . since it appears that not much can be learnt from an individual inaccurate measurement and , we follow @xcite and move on to look at the _ average _ meter reading , @xmath206 , obtained as the number of trials @xmath3 tends to infinity . nowt @xmath206 must be expressed in the terms of the path amplitudes @xmath125 $ ] . the authors of @xcite have shown that in the case of sect.v the correct expression is @xmath207a[i]}{\sum_i a[i]}= \text{re}\{a_1\alpha[1]+a_2\alpha[2]\}.\end{aligned}\ ] ] where @xmath208=a[i]/\sum_i a[i]\end{aligned}\ ] ] is the _ relative amplitude _ for the path number @xmath57 . in the special case of @xmath1 being the projector onto the state @xmath193 , @xmath209 , we have @xmath210,\end{aligned}\ ] ] in @xcite the result was extended to any real functional for which , in the absence of a meter , the amplitude for taking a value @xmath110 is @xmath109 , @xmath211 similarly , the average reading of a very inaccurate meter measuring the difference of @xmath198 and @xmath199 in the example of sect.vi , is given by @xmath212a[1 + 4]+f[2]a[2]+f[3]a[3]}{a[1 + 4]+a[2]+a[3]}= 2\text{re}(\al[2]-\al[3])\end{aligned}\ ] ] in general , an average reading of a very inaccurate meter is expressed in terms of the relative amplitudes for the virtual paths connecting the initial and final states . equations ( [ v3a ] ) and ( [ v6 ] ) bear uncanny resemblance to eqs.([w4 ] ) and ( [ w5 ] ) , in which the probabilities @xmath213 $ ] have been replaced by the corresponding probability amplitudes @xmath125 $ ] , and the real part was chosen to make the result real valued . this may be taken as a hint that the quantity under the @xmath214 sign , the _ weak value _ of a functional @xmath215 , @xmath216a[i]}{\sum_ia[i]}\end{aligned}\ ] ] has the universal importance of its accurate ( `` strong '' ) counterpart in eq.([w5 ] ) , @xmath217p[i]}{\sum_ip[i ] } , \end{aligned}\ ] ] in eqs.([w4 ] ) and ( [ w5 ] ) . in the next section we will show that this can not be the case , and for a good reason . as discussed in sect.viii , an accurately measured mean value of the projector onto an intermediate state gives the relative frequency with which the path created by the meter is travelled , and so serves to distinguish it from other real paths present . what would the result of such measurement mean in the most inaccurate limit , where the path in question exists only as a virtual one ? the uncertainty principle states that two virtual paths can not be distinguished by physical means inside the single real pathway they are combined into . there appears to be a logical rather than a mathematical difficulty . to address it , consider a purely classical conundrum . suppose there are two drops of water on a glass surface , and we put a grain of send in the drop number one . the drops move closer to each other , and coalesce . in which of the two drops is the sand grain now ? if asked , one would certainly reject the question as meaningless , given the new circumstances . but if an answer is required under pain of death , the individual might reply `` inside the water drop number @xmath218 , for all i care '' . the answer to a meaningless question may be _ anything at all_. the above passage , illustrated in fig.3 , is not entirely out of place in the present discussion . with two real paths created by an accurate meter , in each trial the pointer would point at zero or one , indicating the realisation of one of the two real possibilities . if the meter is highly inaccurate , and the interference between two virtual paths is not destroyed , the pointer is equally likely to point at any number . if asked `` was the first or the second virtual path travelled in this transition ? '' , the meter , unable to politely decline to answer a meaningless question , points at , say , @xmath218 . importantly , this behaviour is a necessity , rather than a paradox . it is the uncertainty principle that makes the question meaningless , and any other kind of answer would lead to a conflict with the principle . fully aware of the randomness of an individual inaccurate result , the authors of @xcite focused on the meter s average reading which , given the same initial an final states , is a uniquely defined quantity , which should come up the same in every long series of trials . this is given by the real part ( or the imaginary part , if a slightly different meter is used @xcite,@xcite ) of what looks like an average ( [ w5 ] ) , but obtained with a complex valued distribution , constructed from the path amplitudes @xmath125 $ ] . we wish examine its meaning , bearing in mind the restrictions imposed by the up . firstly , it should be remembered that eq.([v5 ] ) _ expresses _ the average @xmath206 _ defined _ in a standard way by eq.([x6 ] ) , in terms of terms of the systems amplitudes @xmath219 $ ] . secondly , as discussed in sect.viii , mean pointer reading in an accurate measurement of a projector yields the relative frequency with which the real path in question is travelled . thus with @xmath220 trials performed , @xmath221 gives the number of trials in which the system has chosen the path . the result makes sense , since @xmath213\ge 0 $ ] , and @xmath222 . the same can not be said about the amplitudes @xmath125 $ ] . choosing @xmath102=1 $ ] and @xmath104=-1.01 $ ] , and inserting them into eq.([v7 ] ) , we obtain @xmath223 . an attempt to interpret @xmath221 as the number of cases where the first virtual paths is travelled in a series of @xmath224 trials gives a bizarre value of @xmath225 , another arbitrary answer to a meaningless question . we need not limit ourselves to measuring projectors . every functional taking the values @xmath175 $ ] can be written as a sum @xmath226=\sum_jf[j ] f^j[path]$ ] , where @xmath227 $ ] takes the values @xmath228=\delta_{ij}$ ] . in an accurate measurement , the mean value of @xmath227 $ ] gives the relative frequency with which the @xmath56-th path is travelled . in an inaccurate measurement , its value will remain indeterminate , and with it also the value of @xmath226 $ ] . suppose a group of researchers measure the @xmath229-components of the spin of an electron , @xmath230 , at @xmath20 , in a setup where the spin is pre- and post - selected at @xmath15 and @xmath16 . moreover , all researchers have different initial and final states @xmath0 and @xmath2 . if the measurement is accurate , all researchers will be obtaining the results @xmath231 and @xmath194 , albeit in different proportions . they can all agree that the quantity they measure may take only these two values , and use this knowledge to develop theories of spin-1/2 particles , if needed . if the measurement is weak , they can only agree that the spin they measure is able to take any real value at all . suppose further that the equipment is such that each researcher only has access to the average values , and not to the distributions . if the measurements are accurate , for all choices of @xmath0 and @xmath2 @xmath206 would lie between @xmath231 and @xmath194 . this gives the knowledge of a useful fact that @xmath232 represents a quantity whose value can not be larger than one , or less than minus one , although its discreet nature must be learned from some other source . if the measurement is weak , it can be shown @xcite , that there is always a choice of @xmath0 and @xmath2 to give @xmath206 any real value , positive , negative or zero . thus , after trying all possible @xmath0 and @xmath2 , the experimentators will remain in the state of maximal ignorance regarding the properties of the measured quantity . the mathematical trick which allows the up to conceal the information about individual virtual paths is based on replacing the path probabilities with path amplitudes when calculating a weak value . it is not so much the fact that @xmath125 $ ] are in general complex valued , but rather the absence of restrictions on the signs of their real and imaginary parts , that can cause the weak value of a projector to be @xmath233 @xcite . the mean and variance of a non - negative distribution give the estimates for the centre and the width of its region of support . the mean and variance of a distribution which can change sign do not have this property , allowing @xmath206 lie anywhere on the real axis , or obtain a zero variance even when the distribution is not a dirac delta @xcite . conclusions of the previous section can be summarised as follows @xcite . a highly inaccurate `` weak '' measurement ( wm ) on a pre- and post - selected system is a perturbative scheme , in which an additional system ( a vn meter ) is weakly coupled to the observed one ( e.g. , a spin ) . the coupling is such that the mean value of one of the meter s variables ( pointer position @xmath31 ) coincides with the real ( or imaginary , if necessary @xcite , @xcite ) part of a weighted sum of the relative amplitudes on the virtual paths connecting the system s initial and final states , @xmath234\al[i ] , \q \al[i]\equiv a[i]/ \sum_{i}a[i].\end{aligned}\ ] ] in eq.([r1 ] ) the paths @xmath64 , @xmath235 and the weighs @xmath175 $ ] are determined by the nature of the measured quantities . if is is possible ( it may not be ) to measure directly a functional @xmath236 , whose values are zero on all paths except the path number @xmath56 , @xmath228=\delta_{ij}$ ] the mean reading of the weak pointer will yield precisely the real ( imaginary ) part of the amplitude @xmath237 . in short , weak measurements measure probability amplitudes . a certain amount of confusion , which led to a plethora of recent publications on this still fashionable subject ( for a recent review see @xcite and refs . therein ) , has its roots in ref . @xcite where the weak values , defined in @xcite as _ " ... complex numbers that one can assign to the powers of a quantum observable operator @xmath1 using two states , an initial state @xmath39 called _ the preparation _ or _ preselection , _ and a final state @xmath238 called the _ postselection `` , were first introduced . the authors of @xcite considered a weak measurement of the @xmath229-component of a spin-1/2 , similar to that described in sect.v , and concluded that _ `` ... the usual measurement procedure for preselected and postselected ensembles of quantum systems gives unusual results . '' _ the title question of the ref.@xcite ' ' how the result of measurement of a component of the spin of a spin-1/2 particle can turn out to be 100 ? `` has , however , a simple answer . this would happen if the meter measures the difference of relative amplitudes @xmath239-\al[2]$ ] , for an improbable transition , for which @xmath240+a[2]|^2=|\la \phi|\psi\ra|^2 $ ] is small . surprise at the ' ' unusual results `` is also misplaced : the authors of @xcite might as well obtain any other value , large or small , negative , positive or zero . indeed , the opposite would be surprising , as finding only the values between @xmath231 and @xmath194 would contradict the uncertainty principle , as discussed in the previous section . with weak values described only vaguely as a ' ' new concept `` @xcite , some authors proposed to use them as a tool for resolving ' ' counterfactual paradoxes `` . one simple example is the so - called ' ' three - box case `` @xcite , @xcite , @xcite , where the final state can be reached via three virtual pathways with amplitudes @xmath241=c , \q a[2]=-c , \q \text{and } \q a[3]=c,\end{aligned}\ ] ] where @xmath242 is a complex number . an accurate measurement of a functional @xmath243 , @xmath244=1 , \q f^1[2]=f^1[3]=0 , \end{aligned}\ ] ] always finds the system taking the first path . on the other hand , an accurate measurement of a functional @xmath245 , @xmath246=1 , \q f^3[1]=f^3[2]=0 , \end{aligned}\ ] ] always finds the system taking the third path , and the authors of @xcite alert the reader to _ `` ... the peculiarity of having one particle in several places simultaneously even in the stronger sense than in the double slit experiment ... ''_. admitting that the two results were obtained in different physical circumstances , the authors of @xcite rely on the wm _ `` ... to test ... the assertions otherwise regarded a counterfactual''_. their reasoning is as follows . since wm perturb the system only slightly , it is possible to conduct several of them at the same time @xcite ( or , we add , even perform full tomography of the studied transition @xcite ) . simultaneous wm of @xmath111 and @xmath247 by two meters yield @xmath248 and @xmath249 , and seemingly confirm the particle s presence in two places at the same time . a closer look suggests a far simpler explanation of this ' ' peculiarity `` . different strong measurements ' ' engineer `` two completely different classical ensembles ( cf . the passage at the end of sect.viii ) . this becomes clear if one adds all remaining final states , and the real pathways leading to them @xcite . the fact that in two different setups the same final state is always reached via paths of a different kind is not surprising at all . ( we would not give it a second thought if the setups were built from the tubes and connectors of sect.iv . ) so what is tested by the simultaneous wm of @xmath111 and @xmath247 ? we have seen that their results would be @xmath250 % \frac{a[1]}{a[1]+a[2]+a[3 ] } = 1 , \q \text { and}\q \overline{\xi}_2=re\al[3 ] % \frac{a[3]}{a[1]+a[2]+a[3 ] } = 1.\end{aligned}\ ] ] since the relative amplitudes always add up to one , @xmath239+\al[2]+\al[3]=1 $ ] , we also learn that @xmath251=-1 $ ] . far from asserting that the particle is in ' ' two places `` , the weak values give us the values of the real parts of the relative amplitudes . this information is not yet sufficient even for predicting the results of the two accurate measurements , since for this we also need the imaginary parts of @xmath252 $ ] . suppose that by making a wm of different kind we established that @xmath253=0 $ ] , @xmath254 . now we know that all @xmath255 s are real , and @xmath239=-\al[2]=\al[3]$ ] , and the standard rules of quantum mechanics tell us what would happen in two _ mutually exclusive _ situations , where the paths @xmath69 and @xmath72 are combined into a real pathway @xmath256 , or @xmath66 and @xmath69 are combined into @xmath257 , as happens in the two accurate measurements mentioned above . the last shred of mystery , if there was one , disappears when we recall that the perturbation theory typically gives information about both the moduli and the phases of amplitudes and wave function , and the wm scheme is the rule rather than the exception @xcite . finally a much criticised @xcite , @xcite paper entitled ' ' how the result of a single coin toss can turn out to be 100 heads `` @xcite , the authors argue that _ `` weak values are not inherently quantum but rather a purely statistical feature of pre- and postselection with disturbance''_. without going into the details of the author s reasoning , or into the arguments of their critics , one sees that the claim of @xcite must be wrong . to obtain a mean value which lies outside the region of support of a distribution , the distribution must change sign @xcite . such distributions naturally appear in quantum mechanics , whose basic building blocks are complex valued probability amplitudes @xcite . they may not , however , appear in a classical theory where all physically meaningful averages are taken with non - negative probabilities . the parallel between the titles of @xcite , and the ref.@xcite , which followed it twenty six years on , is not coincidental . the fallacy of weak classical values in @xcite stems naturally from the suggestion @xcite that the combinations of the transitions amplitudes , which arise when a measurement is extremely inaccurate , amount to a ' ' new concept " of a quantum variable , capable of shedding light on a new physical reality . in summary , we have shown that the application of a sequence of quantum measurements creates a stochastic network . a network is classical in the sense that the probabilities for all observable events have , of course , all the usual properties of probabilities in a classical theory . a network is quantum in its origin , and the action of a meter , or meters , can be seen as a way of `` engineering '' real scenarios , endowed with probabilities , from the virtual paths of the system , for which only probability amplitudes are available . a network is conveniently described in terms of functionals , which take numerical values on the virtual paths , and whose values the measurements are set to determine . in the case of von neumann and von neumann - like meters considered here , the real pathways are determined by : \(i ) the set of virtual paths available for the system with if no meter is present . ( ii ) the values of the measured functional . two paths sharing the same value can not be distinguished by the meter(s ) . \(iii ) the form and width of the initial state of the pointer , @xmath120 , which determine the accuracy of the measurement . highly accurate measurements turn all virtual paths , that can be distinguished , into real ones , travelled with the probabilities proportional to the squared moduli of the corresponding probability amplitudes . if a functional whose value is @xmath194 on one of the virtual paths , and @xmath153 on all others , is measured accurately , its mean value yields the relative frequency with which the path is travelled . if the same functional is measured in a highly inaccurate manner , its mean ( weak ) value coincides with the real ( or imaginary ) part of the relative amplitude , corresponding to the selected path . since there no _ apriori _ restrictions on the values of the relative amplitudes , it is always possible to find a transition in which the result of a weak measurement would be any desired number , large , small , positive or negative . this is a necessary consequence of the uncertainty principle , and the `` unusual '' weak values lying outside the spectrum of the operator @xmath1 in the example given in sect . v , are by no means unusual . many of exotic interpretations of the weak measurements results stem from the failure to identify the weak values with the path amplitudes , or their linear combinations , and are easily dismissed once the problem is reformulated in the language of standard quantum mechanics . support of the project grupos consolidados upv / ehu del gobierno vasco ( it-472 - 10 ) and the mineco grant no . fis2015 - 67161-p , as well as useful discussions with prof . e. akhmatskaya are gratefully acknowledged . 10 j. v. neumann , _ mathematical foundations of quantum mechanics _ ( princeton university press , princeton , 1955 ) , p. 183 . y. aharonov , d. z. albert , and l. vaidman , phys . lett . * 60 * , 1351 ( 1988 ) . y. aharonov , s. popescu , d. rorhlich and l. vaidman , phys . a * 48 * , 4084 ( 1993 ) . y. aharonov , a. botero , s . popescu , b. reznik and j. tollaksen , phys . a * 301 * , 130 ( 2002 ) . y. aharonov and l. vaidman , in time in quantum mechanics , edited by g. muga , r. sala mayato , and i. egusquiza ( springer , berlin / heidelberg , 2008 ) , 2nd ed . , vol . 1 . y. aharonov , s. popescu , d. rohrlich , and p. skrzypczyk , new j. phys . , * 15 * , 113015 ( 2013 ) . d. sokolovski , phys . a , * 380 * , 1593 ( 2016 ) . d. sokolovski and r. sala mayato , phys . a * 71 * , 042101 ( 2005 ) ; * 73 * , 052115 ( 2006 ) . * 76 * , 039903(e ) ( 2006 ) . d. sokolovski , phys . a , * 84 * , 062117 ( 2011 ) . d. sokolovski , phys . a , * 76 * , 042125 ( 2007 ) . d. sokolovski and e. akhmatskaya , , phys . a , * 84 * , 0422104 ( 2011 ) . d. sokolovski and e. akhmatskaya , ann . * 339 * 307 , ( 2013 ) . d. sokolovski , phys . d , * 87 * , 076001 ( 2013 ) . d. sen , current science , * 107 * , 203 ( 2014 ) . robertson , phys . a , * 34 * , 163 ( 1929 ) . feynman and a.r . hibbs , _ quantum mechanics and path integrals _ ( mcgraw - hill , new york 1965 ) , chap r. p. feynman , r. leighton , and m. sands , _ the feynman lectures on physics iii _ ( dover publications , inc . , new york , 1989 ) , ch.1 . the choice of @xmath120 as a real function ensures that both the mean momentum and the mean position of the pointer are zero . in the classical limit , a von neumann meter does not perturb the motion of the monitored system , provided its momentum is set exactly zero . quantally , this condition can be met only approximately . the spread of the momenta around the zero mean accounts , therefore , for the inevitable perturbation the meter would incur . making @xmath159 broad in the coordinate space would minimise this perturbation , at the expense of the measurement s accuracy . d. bohm , _ quantum theory _ ( addison - wesley , reading , ma , 1965 ) . d. sokolovski and e.ya . sherman , phys . a , * 84 * , 030101 ( 2011 ) . d. sokolovski and e.ya . sherman , phys . a , * 89 * , 043614 ( 2014 ) . sherman and d. sokolovski , new j. phys . , * 16 * , 015013 ( 2014 ) . d. sokolovski , phys . lett , * 102 * , 230405 ( 2009 ) . d. sokolovski , i. puerto gimenez , and r. sala mayato , phys . lett . a , * 372 * , 6578 ( 2008 ) . equation ( [ y6 ] ) follows upon multiplication of the pure state @xmath258 in eq.([n2 ] ) by a projector @xmath259 , and tracing out the pointer s variable . d. sokolovski , proc . a , * 460 * , 1505 ( 2004 ) . the so - called hardy s paradox is analysed in terms of feynman paths in d. sokolovski , i. puerto gimenez , and r. sala mayato , phys . a , * 372 * , 3784 ( 2008 ) . d. sokolovski , phys . lett . a , * 379 * , 1097 ( 2015 ) . j. dressel , m. malik , f. m. miatto , a. n. jordan , and r. w. boyd , rev . phys , _ 86 _ , 307 ( 2014 ) . y. aharonov , l. vaidman , j. phys . a * 24 * , 2315 ( 1991 ) . hoffmann , m. inuma and y. shikano , arxiv 1410.7126v2 ( 2014 ) ; l. vaidman , arxiv 1409.5386v1 ( 2014 ) ; y. aharonov and d. rohrlich , arxiv 1410.0381v1 ( 2014 ) ; j. dressel , arxiv 1410.0943v1 ( 2014 ) ; d. sokolovski , arxiv 1410.0570v1 ( 2014 ) . c. ferrie and j. combes , phys . * 113 * , 120404 ( 2014 ) .

we investigate the intergenerational shape dynamics of single _ caulobacter crescentus _ cells using a novel combination of imaging techniques and theoretical modeling . we determine the dynamics of cell pole - to - pole lengths , cross - sectional widths , and medial curvatures from high accuracy measurements of cell contours . moreover , these shape parameters are determined for over 250 cells across approximately 10000 total generations , which affords high statistical precision . our data and model show that constriction is initiated early in the cell cycle and that its dynamics are controlled by the time scale of exponential longitudinal growth . based on our extensive and detailed growth and contour data , we develop a minimal mechanical model that quantitatively accounts for the cell shape dynamics and suggests that the asymmetric location of the division plane reflects the distinct mechanical properties of the stalked and swarmer poles . furthermore , we find that the asymmetry in the division plane location is inherited from the previous generation . we interpret these results in terms of the current molecular understanding of shape , growth , and division of _ c. crescentus_. cell shape both reflects and regulates biological function . the importance of cell shape is exemplified by bacteria , which rely on specific localization of structural proteins for spatiotemporal organization . bacteria take forms resembling spheres , spirals , rods , and crescents . these shapes are defined by cell walls consisting of networks of glycan strands cross - linked by peptide chains to form a thin peptidoglycan meshwork . super - resolution imaging is now revealing the internal positions of associated proteins . these include cytoskeletal proteins such as mreb , a homolog of actin , intermediate filament - like bundles of cres ( crescentin ) , and ftsz , a homolog of tubulin . however , due to the inherently stochastic nature of molecular processes , understanding how these proteins act collectively to exert mechanical stresses and modulate the effects of turgor pressure and other environmental factors requires complementary methods such as high - throughput , quantitative optical imaging . multigenerational imaging data for bacterial cells can now be obtained from microfluidic devices of various designs . still , a common limitation of most devices is that the environmental conditions change throughout the course of the experiment , particularly as geometric growth of the population results in crowding of the experimental imaging spaces . we previously addressed this issue by engineering a _ c. crescentus _ strain in which cell adhesion is switched on and off by a small molecule ( and inducible promoter ) , allowing measurements to be made in a simple microfluidic device . this technology allows imaging generations of growth of an identical set of 250500 single cells distributed over fields of view . thus cell density is low and remains constant . these studies afforded sufficient statistical precision to show that single _ c. crescentus _ cells grow exponentially in size and divide upon reaching a critical multiple (.8 ) of their initial sizes . satisfaction of a series of scaling laws predicted by a simple stochastic model for exponential growth indicates that these dynamics can be characterized by a single time scale . in this paper , we use more advanced image analysis methods to extract cell shape contours from these data . the resulting geometric parameters , together with mathematical models , provide insights into growth and division in _ c. crescentus _ and the plausible role of cell wall mechanics and dynamics in these processes . specifically , we identify natural variables for tracking cell dynamics , and develop a minimal mechanical model that shows how longitudinal growth can arise from an isotropic pressure . we then examine the dynamics of cell constriction and unexpectedly find that it is governed by the same time constant as exponential growth . this important finding can be understood in terms of an intuitive geometric model that relates the constriction dynamics to the kinetics of the growth of septal cell wall . we further suggest that the site of constriction can arise from differences in materials properties of the poles and show that it is established in the previous generation i.e . , the location of the site of division can be predicted before formation of the divisome . we relate our results to the known dynamics of contributing molecular factors and existing models for bacterial growth and division .

1503.07147

* the length is sufficient to characterize the exponential growth of each cell*. various techniques have been put forth to analyze cell morphology gathered from single cell images @xcite . recent work on image analysis of single cells has attempted to optimize two problems : separation of distinct ( but potentially overlapping ) cells and accurate determination of the edge of each cell @xcite . because crowding is not an issue in our setup , we could focus solely on constructing an algorithm to delineate each cell contour accurately and precisely . as shown in fig . 1a , b and described in the methods section , we first segment each cell using pixel - based edge detection similar to @xcite , then perform spline interpolation to determine the cell contour at sub - pixel resolution . the sequence of such images for each single cell constitutes a trajectory in time @xmath3 that serves as the basis for quantitative analysis . division events are then detected in an automated fashion using custom python code , and used to divide time trajectories for each cell into individual generations . all data shown here were obtained by observing 260 single _ c. crescentus _ cells perfused in complex medium ( peptone - yeast extract ; pye ) at 31@xmath4c over the course of 2 days ( corresponding to 9672 separate generations ) . under these conditions , the mean population growth rate and division time remain constant , so we treat the trajectories of individual generations as members of a single ensemble . in other words , we segment each cell trajectory by generation and take the resulting initial frame ( i.e. , immediately following division ) as @xmath5 minutes . in order to average over the ensemble , we then bin quantitative information according to time since division , @xmath3 , normalized by the respective division time @xmath6 . the normalized time , @xmath7 , serves as a cell - cycle phase variable . for our quantitative analysis , we focus on a set of three intuitive and independent parameters that characterize cell shape at each stage of growth : length @xmath8 , width @xmath9 , and radius of curvature @xmath10 ( fig . 1c ) . they are calculated directly from each splined contour as follows ( see also supplementary fig . 1 ) : * we define the length , @xmath11 , as the pole - to - pole distance along the contour of the cell medial axis at the normalized time @xmath12 ( fig . 2a ) . * we assign a single radius of curvature , @xmath13 , to each cell based upon the best - fit circle to the medial axis ( fig . although stalked ( @xmath14 ) and swarmer ( @xmath15 ) portions may be described by different radii of curvature toward the end of the cell cycle , the average radius obtained by averaging the contributions of each portion yields the same value , i.e. , @xmath16 ( see supplementary fig . 2c ) . * we define the width , @xmath17 as the length of the perpendicular segment spanning from one side of the cell contour to the other at each position @xmath18 along the medial axis , which runs from @xmath19 at the stalked pole to @xmath20 at the swarmer pole . furthermore , we spatially averaged the width over positions along the medial axis , @xmath21 , to obtain a characteristic width at each time point ( fig . 2c ) . the mean division time is @xmath22 min , where @xmath23 indicates a population average . we find that @xmath24 increases exponentially with time constant @xmath25 min , essentially the same time constant that we previously observed for the cross - sectional area @xcite , while @xmath26 and @xmath27 remain approximately constant for @xmath28 and each shows a dip for @xmath29 when cell constriction becomes prominent . the sharp rise in @xmath27 seen for @xmath30 results from independent alignment of the stalked and swarmer portions with the microfluidic flow as they become able to move independently ( i.e. , fluctuate easily about the plane of constriction ) . these observations confirm the assumptions in @xcite that the length is sufficient to describe the growth of cell size . moreover , we can track the dynamics of the spanning angle , @xmath31 , using the relation @xmath32 . * mechanical model for cell shape and growth*. there are many details of cell growth and shape that require interpretation . for example , it is not obvious _ a priori _ that growth should be almost exclusively longitudinal . therefore , we have developed a minimal mechanical model that can explain these observations . we parametrize the geometry of the cell wall by a collection of shape variables @xmath33 , where @xmath34 , @xmath35 , and @xmath36 are the parameters introduced above ( fig . as the cell grows in overall size , we postulate that the rate of growth in the shape parameter @xmath37 is proportional to the net decrease in cell wall energy , @xmath38 , per unit change in @xmath37 @xcite . assuming linear response , the configurational rate of strain , @xmath39 , is proportional to the corresponding driving force @xmath40 , in analogy with the constitutive law of newtonian flow @xcite : @xmath41 where the constant @xmath42 describes the rate of irreversible flow corresponding to the variable @xmath37 . according to equation , exponential growth occurs if @xmath43 is constant , whereas @xmath37 reaches a steady - state value if @xmath44 along with the condition @xmath45 . it thus remains to specify the form of @xmath46 . for a _ c. crescentus _ cell of total volume @xmath47 and surface area @xmath48 , our model for the total energy in the cell wall is given by @xmath49 where @xmath50 is a constant pressure driving cell wall expansion ; @xmath51 is the tension on the surface of the cell wall ; @xmath52 is the energy required to maintain the cell width ; @xmath53 represents the mechanical energy required to maintain the crescent cell shape ; @xmath54 is the energy driving cell wall constriction . traditionally @xmath50 was taken to be the turgor pressure @xcite ; while the importance of the turgor pressure has recently been questioned @xcite , an effective pressure must still arise from the synthesis and insertion of peptidoglycan strands that constitute the cell wall . we note that a purely elastic description of cell wall mechanics would lead to a curvature - dependent surface tension @xcite . however , if growth is similar to plastic deformation , the tension is uniform @xcite . the effective tension in our model depends on the local surface curvatures through the energy terms @xmath52 and @xmath53 , that describe harmonic wells around preferred values of surface curvatures . the mechanical energy for maintaining width is given by @xmath55 where the constant @xmath56 is the preferred radius of curvature , @xmath57 is the bending rigidity and @xmath58 is a differential area element @xcite . contributions to @xmath57 can come from the peptidoglycan cell wall as well as membrane - associated cytoskeletal proteins like mreb , mrec , rodz , etc . , which are known to control cell width @xcite . in addition to maintaining a constant average width , _ c. crescentus _ cells exhibit a characteristic crescent shape , which relies on expression of the intermediate filament - like protein crescentin @xcite . although the mechanism by which crescentin acts is not known , various models have been proposed , including modulation of elongation rates across the cell wall @xcite and bundling with a preferred curvature @xcite . we assume the latter and write the energy for maintaining the crescent shape as @xmath59 where @xmath60 is the arc - length parameter along the crescentin bundle attached to the cell wall , @xmath61 is the local curvature , @xmath62 is the preferred radius of curvature , @xmath63 is the contour length , and @xmath64 is the linear bending rigidity . equation accounts for the compressive stresses generated by the crescentin bundle on one side of the cell wall , leading to a reduced rate of cell growth , according to equation . as a result , the cell wall grows differentially and maintains a non - zero curvature of the centerline . in the absence of crescentin ( @xmath65 ) , our model predicts an exponential decay in the cell curvature that leads to a straight morphology , consistent with previous observations @xcite . finally , one must also account for the energy driving cell wall constriction . constriction proceeds via insertion of new peptidoglycan material at the constriction site . this process leads to the formation of daughter pole caps @xcite . we take constriction to be governed by an energy of the form @xmath66 , where @xmath67 is the surface area of the septal cell wall , and @xmath68 is the energy per unit area released during peptidoglycan insertion . * there exists an optimal cell geometry for a given mechanical energy*. to apply the model introduced above ( equations and ) to interpreting the data in fig . 2 , we assume a minimal cell geometry given by a toroidal segment with uniform radius of curvature @xmath13 , uniform cross - sectional width @xmath21 and the spanning angle @xmath69 . to this end , we estimate as many mechanical parameters as we can from the literature and then determine the rest by fitting our experimentally measured values . turgor pressure in gram - negative bacteria has been measured to be in the range @xmath70 mpa @xcite . we use a value for the effective internal pressure close to the higher end of the measured values for turgor pressure , @xmath71 mpa , in order to account for peptidoglycan insertion . we estimate the surface tension as @xmath72 nn@xmath73 m ( see supplementary model section ) and multiply it by the cell surface area @xmath74 to obtain the cell wall surface energy . first , we neglect cell constriction ( setting @xmath75 ) and assume that the crescentin structure spans the length of the cell wall ( excluding the endcaps ) @xcite , with a contour length @xmath76 . the mechanical properties of mreb and crescentin are likely similar to those of f - actin and intermediate filaments , respectively @xcite . however , due to a lack of direct measurements , we obtain the mechanical parameters @xmath57 and @xmath64 by fitting the model to the experimental data . as desired , we find that the total energy @xmath46 has a stable absolute minimum at particular values of the cross - section diameter @xmath77 and the centerline radius @xmath10 , given by solution of @xmath78 ( see supplementary fig . 4 ) . the measured values are @xmath79 m and @xmath80 m ( @xmath28 ) , and , as indicated by the red solid curves in fig . 2b , c , the model reproduces them with @xmath81 nn@xmath82 m and @xmath83 nn@xmath82m@xmath84 . while the fitted value for @xmath64 is numerically close to the estimate based on the known mechanical properties of intermediate filaments ( @xmath85 nn@xmath82m@xmath84 ) , the value for @xmath57 is much higher than the bending rigidity of mreb bundles ( see supplementary information ) . this indicates that @xmath57 is only determined in part by mreb and can have contributions from the cell wall . given stable values for @xmath77 and @xmath10 , growth is completely described by the dynamics of the angle variable @xmath69 . consequently , we write the total energy in the scaling form @xmath86 , with @xmath87 the energy density along the longitudinal direction . the condition for growth then becomes @xmath88 , such that the energy is minimized for increasing values of @xmath69 . from our experimental data , the angle spanned by the cell centerline increases by an amount @xmath89 during the entire cycle . using our parameter estimates and fitting the data in fig . 2 , we obtain a numerical value for the energy density @xmath90 nn@xmath82 m . we relate the angle dynamics to the length by @xmath91 where @xmath92 ( @xmath88 ) is the rate of longitudinal growth , which can be interpreted as resulting from remodeling of peptidoglycan subunits with a mean current @xmath93 , across the cell surface area @xmath94 . from an exponential fit to the data for cell length ( fig . 2a ) , we obtain @xmath95 ( nn @xmath82 m min)@xmath96 , which gives us an estimate of the friction coefficient , @xmath97 nn @xmath82m@xmath96 min , associated with longitudinal growth ; e.g. , mreb motion that is known to correlate strongly with the insertion of peptidoglycan strands @xcite . our results are consistent with previous observations of _ c. crescentus _ cells with arrested division but continued growth @xcite . * constriction begins early and proceeds with the same time constant as exponential growth*. having characterized the dynamics of growth , we now turn to constriction at the division plane . as mentioned above , we obtain the experimental width at each point along each cell s medial axis . the typical width profile is non - uniform along its length , exhibiting a pronounced invagination near the cell center ( with width @xmath98 ; fig . this invagination , which ultimately becomes the division plane , is readily identifiable early in the cell cycle , even before noticeable constriction occurs . we discuss the kinetics of constriction in this section , and focus on its location later in the manuscript . as shown in fig . 3a ( black points ) , @xmath99 progressively decreases towards zero until pinching off at @xmath100 . due to the limited spatial resolution of our imaging ( phase contrast microscopy ) , the pinch - off process occurring for @xmath30 could not be captured , but @xmath99 at earlier times ( i.e. , @xmath101 ) is precisely determined as a function of @xmath12 . to model the dynamics of constriction , we assume as in ref . @xcite ( fig . 3a , inset ) : ( i ) the shape of the zone of constriction is given by two intersecting and partially formed hemispheres with radii @xmath102 ; and ( ii ) constriction proceeds by completing the missing parts of the hemisphere such that the newly formed cell wall surface maintains the curvature of the pre - formed spherical segments . as a result , a simple geometric formula is obtained that relates the width of the constriction zone , @xmath98 , to the surface area @xmath103 of the newly formed cell wall , @xmath104 where @xmath105 is the maximum surface area achieved by the caps as the constriction process is completed , i.e. , when @xmath106 . we assume that the addition of new cell wall near the division plane initiates with a rate , @xmath107 , and thereafter grows exponentially with a rate , @xmath108 , according to , @xmath109 subject to the initial condition @xmath110 . the first term on the right - hand side of equation follows from equation , using @xmath103 as the shape variable , after incorporating the constriction energy @xmath111 . the rate of septal peptidoglycan synthesis , @xmath108 , is thus directly proportional to the energy per unit area released during constriction , @xmath68 . the solution , @xmath112 , can then be substituted into equation to derive the time - dependence of @xmath98 , whose dynamics is controlled by two time scales : @xmath113 and @xmath114 . fitting equation with the data for @xmath99 , we obtain @xmath115 min and @xmath116 min . the fitted values for the time constants controlling constriction dynamics ( @xmath113 and @xmath114 ) are remarkably similar to that of exponential cell elongation ( @xmath117 min ) . this shows that septal growth proceeds at a rate comparable to longitudinal growth . therefore , one of the main conclusions that we draw is that cell wall constriction ( fig . 3a ) is controlled by the same time constant as exponential longitudinal growth ( fig . 2a ) . having determined the dynamics of @xmath98 , we compute the average width across the entire cell @xmath21 using the simplified shape of the constriction zone as shown in fig . 3a ( inset ) . the resultant prediction ( blue solid curve in fig . 2c ) is in excellent agreement with the experimental data and captures the dip in @xmath26 seen for @xmath118 . constriction also leads to a drop in the average radius of curvature of the centerline , as shown by the experimental data in fig . in the supplementary material we derive a relation between the centerline radius of curvature @xmath13 and the minimum width @xmath98 , given by @xmath119 , predicting that cell curvature increases at the same rate as @xmath98 drops . using this relation , we are able to quantitatively capture the dip in @xmath27 seen for @xmath118 ( solid blue curve in fig . 2b ) without invoking any additional fitting parameters . * origin of the asymmetric location of the primary invagination*. we now consider the position of the division plane and its interplay with cell shape . as shown in fig . 3b , the distance of the width minimum from the stalked pole ( @xmath120 ) increases through the cell cycle at the same rate as the full length of the growing cell ( @xmath11 ) , such that their ratio remains constant with time - averaged mean @xmath121 . the presence of the primary invagination early in the cell cycle is reiterated in fig . 3c , which shows the width profile constructed by ensemble - averaging over each cell at the timepoint immediately following division . in addition to the width minimum @xmath98 , there are two characteristic maxima near either pole , @xmath122 and @xmath123 , respectively ( fig . 3c , inset ) . as evident in fig . 3c , the stalked pole diameter @xmath124 is on average larger than its swarmer counterpart @xmath125 ( also see supplementary fig . we show that the asymmetric location of the invagination ( and the asymmetric width profile ) can originate from the distinct mechanical properties inherent to the pole caps in _ c. crescentus_. the shapes of the cell poles can be explained by laplace s law that relates the pressure difference , @xmath50 , across the cell wall to the surface tensions in the stalked or the swarmer pole , @xmath126 . the radii of curvature of the poles then follow from laplace s law @xmath127 where the superscript ( @xmath128 ) denotes the stalked or the swarmer pole . thus a larger radius of curvature in the poles has to be compensated by a higher surface tension to maintain a constant pressure difference @xmath50 . assuming that the poles form hemispheres , we have @xmath129 . our data indicate that the early time ratio for @xmath130 ( @xmath131 ) shows a strong positive correlation with the ratio @xmath132 , with an average value @xmath133 ( see supplementary fig 3a ) . laplace s law then requires that the stalked pole be mechanically stiffer than the swarmer pole ; @xmath134 . this observation suggests that the asymmetry in the lengths of the stalked and swarmer parts of the cell depends upon different mechanical properties of the respective poles . to quantitatively support this claim , we investigate an effective contour model for the cell shape . to this end , we assume that the fluctuations in cell shape relax more rapidly than the time scale of growth . this separation of timescales allows us to derive the equation governing the cell contour by minimizing the total mechanical energy ( equation ) . from the solution we compute the resultant width profile for the entire cell ( see supplementary model section ) . as shown in fig . 3d , the model with asymmetric surface tensions of the poles causes the primary invagination to occur away from the cell mid - plane . the spatial location of the invagination relative to the cell length depends linearly on the ratio @xmath135 . symmetry is restored for @xmath136 , as shown in fig . 3d ( blue dashed curve ) . we note that a gradient in @xmath51 along the cell body would imply differences in longitudinal growth rates between the stalked and the swarmer portions of the cell ( eq . ) . our data exclude this possibility since both @xmath120 and @xmath137 grow at the same rate @xmath138 , as evidenced by the constancy of their ratio ( fig . 3b and supplementary fig . because _ c. crescentus _ does not exhibit polar growth , the _ polar stiffness model _ is consistent with the observed uniformity in longitudinal growth rate . in addition , the non - uniformity in cell width comes from the differences in mechanical response in the cell wall due to preferential attachment of crescentin along the concave sidewall . for a cres mutant cell ( where @xmath65 ) , our model predicts a uniform width profile before the onset of constriction . * cell shape evolution during wall constriction*. the experimental width profiles show that the growing and constricting cells typically develop a second minimum in width ( fig . these secondary invaginations are observed in both the stalked and swarmer portions of single cells in the predivisional stage ( @xmath139 ) , although they are more common in the stalked portions ( fig . we show here that these secondary minima become the primary minima in each of the daughter cells . to study the dynamics of the development of the secondary minimum we introduce a new quantity , @xmath140 , defined as the distance from the stalked pole to the secondary minimum in the stalked part ( see fig . 5c , inset ) we find that the ratio @xmath141 has a mean value of @xmath142 0.55 at later points in the cell cycle ( fig . 5b ) , equal to the constant ratio maintained by the distance from the stalked pole to the primary minimum , @xmath143 . in fact the kymograph of width profiles ( shown over 2 generations for a representative single cell ) in fig . 5c demonstrates that the predivisional secondary invaginations are inherited as primary invaginations after division . this mechanism provides continuity and inheritance of the invaginations across generations and is an intrinsic element of the mechanism for cell division in _ c. crescentus_. to quantitatively explain the experimental width profiles during constriction , we use our mechanical model to determine the instantaneous cell shape by minimizing the total energy ( equation ) at the specified time points ( see supplementary model section ) . to take constriction into account , we impose the constraint that @xmath144 , where @xmath145 is determined by equations and . in addition , we assume non - uniform materials properties in the cell wall by taking the tension in the cell poles ( @xmath146 ) and the septal region to be higher than the rest of the cell . as constriction proceeds and @xmath145 decreases , we compute the shape of the cell contours ( fig . 4c ) and the corresponding width profiles ( fig . the computed width profiles faithfully reproduce the secondary invaginations , which become more pronounced as the daughter pole caps become prominent . an example of the experimental width profiles is shown in fig . 4b at evenly - spaced intervals in time for a single generation , and the corresponding model width profiles are shown in fig . we note that the experimental cell contours in the predivisional stage ( @xmath30 ) bend away from the initial midline axis and develop an alternate growth direction ( fig . 4a , blue contour ) . these bend deformations are induced by the microfluidic flow about the pinch - off plane ; the cells become increasingly `` floppy '' as the constriction proceeds . the consistent propagation of a specific shape through the processes of growth and division relies upon an intricate interplay between the controlled spatiotemporal expression and localization of proteins , and cytoskeletal structural elements . the high statistical precision of our measurements allows us to gain new insights into cell morphology . from precise determination of cell contours over time , we observe that a typical cell width profile is non - uniform at all times with a pronounced primary invagination appearing during the earliest stages of the cell cycle . during cell constriction , the decrease in the minimum width is governed by the same time constant as exponential axial growth ( fig . furthermore , the location of the primary invagination divides the cell contour into its stalked and swarmer compartments , such that the ratio of the length of the stalked part @xmath147 to the total pole - to - pole length @xmath11 remains constant during the cycle with a mean value @xmath148 ( fig . these observations and our mechanical model lead to two important conclusions : first , _ the dynamics of cell wall constriction and septal growth occur concomitantly _ , and second , _ the asymmetric location of the primary invagination can be explained by the differences in mechanical properties in the stalked and swarmer poles_. a corollary of the first conclusion is that the size ratio threshold at division occurs naturally without requiring a complex timing mechanism @xcite . in addition to the primary septal invagination , the cell contours exhibit a pronounced secondary invagination during the predivisional stages ( fig . remarkably , the secondary invaginations develop at a precise location relative to the total length of the stalked compartments , @xmath149 ( fig . the data thus allow a third conclusion : _ these secondary invaginations are inherited as primary invaginations in each of the daughter cells , directing the formation of the division plane in the next generation_. thus , through consistent and controlled nucleation of invaginations across generations , _ c. crescentus _ cells maintain a constant ratio of the sizes of stalked and swarmer daughter cells . our experimental observations and the parameters in the cell shape model can be related to the current molecular understanding for gram - negative bacteria , in particular _ c. crescentus_. before the onset of noticeable constriction , cell shape is dictated by the mechanical properties of the peptidoglycan cell wall in addition to various shape - controlling proteins such as mreb , mrec , rodz and cres . single molecule tracking studies have revealed that mreb forms short filamentous bundles anchored to the inner surface of the cell wall and moves circumferentially at a rate much faster than the rate of cell growth @xcite . _ in vitro _ experiments show that mreb filaments can induce indentation of lipid membranes , suggesting that they may have a preferred radius of curvature @xcite . thus on time scales comparable to cell growth , @xmath52 is determined in part by the energy cost of adhering mreb bundles to the cell wall ( see supplementary model section ) . bacterial cell division is driven by a large complex of proteins , commonly known as divisomes that assemble into the z - ring structure near the longitudinal mid - plane of the cell @xcite . the z - ring contains ftsz protofilaments that are assembled in a patchy band - like structure @xcite . ftsz protofilaments are anchored to the cell membrane via ftsa and zipa , and play a crucial role in driving cell wall constriction @xcite . during constriction , the divisome proteins also control peptidoglycan synthesis and direct the formation of new cell wall via the activity of penicillin - binding proteins ( pbps ) @xcite . thus the divisome plays a two - fold role by concomitantly guiding cell wall constriction and growth of the septal peptidoglycan layer . according to our model the constriction of the cell wall is driven by the synthesis of septal cell wall at a rate @xmath108 ( @xmath150 ) , which can be directly related to the activity of pbps triggered by the divisome assembly . furthermore , in our model it is sufficient that the divisome guide the curvature of cell wall growth in the septal region ( see fig . 3a , inset ) . while the mechanism behind the precise asymmetric location of the division plane in _ c. crescentus _ cells is not well understood , it is likely that the atpase mipz helps division site placement by exhibiting an asymmetric concentration gradient during the predivisional stage @xcite . mipz activity inhibits ftsz assembly ; as a result of polar localization of mipz , z - ring assembly is promoted near the mid - cell @xcite . our cell shape model suggests that the early time asymmetric location of the primary invagination , which develops into the division plane , is controlled by the differences in surface tensions maintained in the poles . the presence of this invagination at @xmath151 , as inherited from the secondary invaginations in the previous generation , aids in z - ring assembly at the site of the invagination . the curvature - sensing capability of the z - ring may be enabled by the minimization of the ftsz polymer conformational energy that is determined by the difference between cell surface curvature and ftsz spontaneous curvature @xcite . a higher tension in the stalked pole can be induced by asymmetric localization of polar proteins , such as popz , early in the cell cycle . experiments have shown that popz localizes to the stalked pole during the initial phase of the cell cycle and increasingly accumulates at the swarmer pole as the cell cycle proceeds @xcite . consistent with this observation , our data show that the correlation between the pole sizes ( determined by the ratio of surface tension to pressure ) and the stalked and swarmer compartment lengths tend to disappear later in the cycle ( supplementary fig . 3 ) , as cell constriction proceeds . a recent experimental study also demonstrates that molecular perturbation of clp proteases can destroy the asymmetry of cell division in _ c. crescentus _ @xcite , suggesting the interplay of subcellular protease activity with the physical properties of the cell wall . earlier theoretical models have predicted that a small amount of pinch - off force from the z - ring ( @xmath152 pn ) is sufficient to accomplish division by establishing a direction along which new peptidoglycan strands can be inserted @xcite . in contrast , our data combined with the mathematical model allows the interpretation that _ the early time asymmetric invagination in the cell wall can set the direction for the insertion of new peptidoglycan strands_. constriction results from exponential growth of surface area in the septum ( at the same rate as longitudinal extension ) . the instantaneous cell shape is determined by minimizing the energy functional at given values of the cell size parameters . finally , from our estimate of the cell wall energy density @xmath87 ( @xmath153 nn@xmath82 m ) , we predict that a net amount @xmath154 nn@xmath82 m of mechanical energy is used by the peptidoglycan network for cell wall growth . for a _ c. crescentus _ cell of surface area @xmath155m@xmath84 , layered with glycan strands of length @xmath15 nm and cross - linked by peptide chains with maximally stretched length @xmath14 nm @xcite , there are roughly @xmath156 peptidoglycan subunits . thus on average , each peptidoglycan subunit can consume mechanical energy of @xmath12.4@xmath157 nn@xmath82 m , or @xmath10.6 @xmath158 at a temperature @xmath159c . cell wall remodeling and insertion of new peptidoglycan material can likely create defects in the peptidoglycan network @xcite . one thus expects cellular materials properties to change over time , as a result of these molecular scale fluctuations . although we neglect such variations in our mean field model , it nonetheless quantitatively captures the average trends in cell shape features . in future work we plan to more closely connect the energy terms of the continuum model with molecular details . * acquisition of experimental data*. data were acquired as in ref . @xcite . briefly , the inducibly - sticky _ caulobacter crescentus _ strain fc1428 was introduced into a microfluidic device and cells were incubated for one hour in the presence of the vanillate inducer . the device was placed inside a homemade acrylic microscope enclosure ( @xmath160 ) equilibrated to 31@xmath4c ( temperature controller : csc32j , omega and heater fan : hgl419 , omega ) . at the start of the experiment , complex medium ( peptone - yeast extract ; pye ) was infused through the channel at a constant flow rate of 7 @xmath82l / min ( phd2000 , harvard apparatus ) , which flushed out non - adherent cells . a microscope ( nikon ti eclipse with the `` perfect focus '' system ) and robotic xy stage ( prior scientific proscan iii ) under computerized control ( labview 8.6 , national instrument ) were used to acquire phase - contrast images at a magnification of 250x ( emccd : andor ixon+ du888 1k @xmath161 1k pixels , objective : nikon plan fluor 100x oil objective plus 2.5x expander , lamp : nikon c - hfgi ) and a frame rate of 1 frame / min for 15 unique fields of view over 48 hours . in the present study we use a dataset consisting of 260 cells , corresponding to 9672 generations ( division events ) . * analysis of single cell shape*. the acquired phase - contrast images were analyzed using a novel routine we developed ( written in python ) . each image was processed with a pixel - based edge detection algorithm that applied a local smoothing filter , followed by a bottom - hat operation . the boundary of each cell was identified by thresholding the filtered image . a smoothing b - spline was interpolated through the boundary pixels to construct each cell contour . each identified cell was then tracked over time to build a full time series . we chose to include only cells that divided for more than 10 generations in the analysis . a minimal amount of filtering was applied to each growth curve to remove spurious points ( e.g. , resulting from cells coming together and touching , or cells twisting out of plane ) . the timing of every division was verified by visual inspection of the corresponding phase contrast images , so that the error in this quantity is approximately set by the image acquisition rate of 1 frame / min . 10 url # 1`#1`urlprefix[2]#2 [ 2][]#2 & . _ _ * * , ( ) . & . _ _ * * , ( ) . & _ _ * * , ( ) . _ _ ( , , ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . & . _ _ * * , ( ) . , & . _ _ * * , ( ) . _ _ * * , ( ) . , , & . _ _ * * , ( ) . & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . _ _ ( , , ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . . _ _ * * , ( ) . & . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . & _ _ * * , ( ) . & _ _ * * , ( ) . , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . , , & . _ _ * * , ( ) . _ _ * * , ( ) . , , , & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , , & . _ _ * * , ( ) . _ et al . _ . _ _ ( ) . , , & . _ _ * * , ( ) . , & . _ _ * * , ( ) . & . _ _ * * , ( ) . , , , & _ _ * * , ( ) . & . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . _ et al . _ . _ _ * * , ( ) . , , & . _ _ * * , ( ) . & . _ _ * * , ( ) . we gratefully acknowledge funding from nsf physics of living systems ( nsf phy-1305542 ) , nsf materials research science and engineering center ( mrsec ) at the university of chicago ( nsf dmr-1420709 ) , the w. m. keck foundation and the graduate program in biophysical sciences at the university of chicago ( t32 eb009412/eb / nibib nih hhs / united states ) . n.f.s . also thanks the office of naval research ( onr ) for a national security science and engineering faculty fellowship ( nsseff ) . c.s.w . , s.i.b . , a.r.d . , and n.f.s . designed the experiments ; s.b . , a.r.d . , and n.f.s . designed the model ; c.s.w . and s.i.b . performed the experiments and observed the phenomena reported ; c.s.w . designed and implemented custom software to automate cell shape image analysis ; c.s.w . and s.b . analyzed the data ; s.c . contributed reagents and materials ; c.s.w . , a.r.d . , and n.f.s . wrote the manuscript ; all authors discussed the results and commented on the manuscript . the authors declare no competing financial interests . * supplemental material * biblabel[1][s#1 ] a number of quantities can be immediately calculated from each splined cell contour ( supplementary fig . 1a ) , including the cross - sectional area . the cell medial axis was determined by calculating the voronoi diagram of the cell contour [ s@xcite ] , pruning the branches ( supplementary fig . 1b ) , and extending this skeleton to the edges of the cell contour in a manner that preserves average curvature of the medial axis ( supplementary fig . the intersections between the cell medial axis and contour represent the stalked and swarmer poles , respectively . the cell length was calculated by evaluating the distance along the medial axis between either pole . the cell widths were determined by creating ribs perpendicular to the medial axis along its length and determining the distances between their intersections with opposite sides of the cell contour ( supplementary fig . to calculate time - averaged quantities , we normalized trajectories for each generation by the respective division time @xmath6 ( thus converting each variable to a function of @xmath7 ) . we then split these data into 73 bins , as @xmath162 min under these conditions , and ensemble - averaged each of these bins over every generation . we can define five specific values of the width according to the local minima and maxima in the width profile ( @xmath163 , @xmath164 , @xmath165 , @xmath166 , @xmath167 ) , not all of which may be present in any given cell . where we identify a primary minimum in the width profile , @xmath163 , we can also determine two local maxima , @xmath164 and @xmath165 , corresponding respectively to the stalked and swarmer portions ( supplementary fig . these values are approximately constant throughout the cell cycle , with @xmath168 . however , at later times ( @xmath139 ) the value of @xmath123 increases until @xmath169 . in some cases , additional secondary local minima are observed , @xmath170 and @xmath171 , corresponding respectively to the stalked and swarmer portions ( supplementary fig . although we note the value of these quantities for early times here ( where they are approximately equal to their respective local maxima ) , these minima can only be determined with certainty at later times ( @xmath139 ) . there , we observe the presence of a secondary minimum in the stalked portions of most cells . p5.5em p6.5em p3em l + & mean @xmath172 s.d . & & @xmath12 + + @xmath173 & 0.74 @xmath172 0.02 & @xmath82 m & @xmath174 $ ] + @xmath175 & 0.85 @xmath172 0.02 & @xmath82 m & @xmath174 $ ] + @xmath176 & 0.84 @xmath172 0.03 & @xmath82 m & @xmath174 $ ] + @xmath177 & 0.83 @xmath172 0.02 & @xmath82 m & @xmath174 $ ] + @xmath178 & 0.82 @xmath172 0.02 & @xmath82 m & @xmath174 $ ] + @xmath179 & 4.44 @xmath172 2.12 & @xmath82 m & @xmath174 $ ] + @xmath180 & 4.34 @xmath172 2.50 & @xmath82 m & @xmath174 $ ] + @xmath181 & 3.94 @xmath172 2.42 & @xmath82 m & @xmath174 $ ] + @xmath182 & 54.3 @xmath172 5.2 & % & @xmath183 $ ] + @xmath184 & 45.7 @xmath172 5.2 & % & @xmath183 $ ] + + we split each cell into stalked and swarmer portions according to the location of the primary minimum in the width profile . the radius of curvature ( @xmath10 ) was calculated as the radius of the best - fit circle to the cell medial axis . we tested two methods of determining the radius of curvature : ( i ) fitting the whole cell to a circle or ( ii ) fitting the stalked and swarmer portions separately . at early times in the cell cycle , ( ii ) gives poorer results because of fewer data points . at later times in the cell cycle , ( i ) gives poorer results because the stalked and swarmer portions can indeed have differing radii of curvature , which we attribute to the alignment of the swarmer portion of the cell with the direction of fluid flow after the division plane has narrowed enough that it becomes mechanically decoupled from the stalked portion , i.e. , like a flexible hinge . however , in the mean the results of ( ii ) are equal to the value calculated by ( i ) for earlier times . therefore , we use only data from method ( i ) but exclude later time points ( supplementary fig . because of the variation in the values of the calculated radius of curvature was large , such that we can not define a reasonable arithmetic mean without arbitrarily filtering the dataset , we first averaged the corresponding unsigned curvature ( equivalent to @xmath185 ) and then converted to radius of curvature ( i.e. , what we report as @xmath186 is actually the harmonic mean , calculated as @xmath187 ) . the length was also split according to the locations of the local minima , into @xmath188 ( distance along cell medial axis from stalked pole to @xmath163 ) , @xmath189 ( distance along cell medial axis from swarmer pole to @xmath163 ) , @xmath190 ( distance along cell medial axis from stalked pole to @xmath166 ) , and @xmath191 ( distance along cell medial axis from swarmer pole to @xmath167 ) . the value of @xmath188 and @xmath190 are compared in supplementary fig . note that the length of the stalked portion @xmath188 grows exponentially , with the same time constant as the length @xmath8 ( as does @xmath189 , although it is not shown here for clarity ) , which is a necessary condition for the addition of peptidoglycan material along the entire length of the cell when the location of @xmath163 is set at early times . at later times ( @xmath139 ) , the length from stalked pole to secondary minimum @xmath190 also starts to increase . we relate the asymmetry in the length of stalked and swarmer portions at early times to asymmetries in the stalked and swarmer poles . the model predicts a linear relationship between the ratio of lengths @xmath192 and the ratio of the tensions at either pole @xmath193 . we can not directly measure the latter quantity , but from laplace s law it is equal to the inverse ratio of the mean curvatures at either pole , which we approximate as the inverse of half the maximum pole width , assuming that the poles are hemispheres with diameter @xmath164 and @xmath165 , respectively . supplementary fig . 3 shows scatter plots comparing @xmath194 to @xmath192 for three different time intervals . the red best - fit line is shown only for supplementary fig . 3a , for which @xmath195 ( linear fits to all other plots produced values of @xmath196 ) . this line runs from the point @xmath197 , corresponding to the dashed blue curve in fig . 3c of the main text ( which is the case of a symmetric cell ) to @xmath198 , corresponding to the red solid curve in fig . 3d of the main text ( which is the case of the average asymmetric _ c. crescentus _ cell ) . note that at times @xmath199 , any correlation disappears . in order to quantify the error in our width profiles , we imaged a single field of view of 24 _ c. crescentus _ cells perfused in complex medium at 31@xmath4c at a frame rate of 5 frames per second ( 300 times faster than the frame rate used to acquire all other data ) , and calculated the splined contours for each cell . we focused in particular on a single `` dead '' ( non - growing and non - dividing ) cell , and found the root - mean - square deviation ( rmsd ) of nearest points along the cell contour between subsequent frames to be 12 nm . in addition , we found the rmsd of equivalent points along the width profile between subsequent frames to be 28 nm , or a 3.2% pinch depth at the average value of @xmath164 = 0.85 @xmath82 m . * cell wall mechanics*. the total energy @xmath46 for the bacterial cell wall is given as the sum of contributions from an active internal pressure @xmath50 driving cell volume ( @xmath47 ) expansion , mechanical energy @xmath200 in the cell wall , and the mechanical energy of interactions with cytoskeletal proteins @xmath201 : @xmath202 the bacterial cell wall consists of a network of glycan strands cross - linked by peptide chains know as the peptidoglycan network . growth occurs via the insertion of new peptidoglycan strands into the existing network along with the breaking of existing bonds due to turgor pressure induced stretching . we assume that elastic equilibrium is reached rapidly as compared to the rate of synthesis of new material [ s@xcite ] . as a result of cell wall remodeling and irreversible elongation , growth can be understood as resulting from plastic deformations [ s@xcite ] . to understand the origin of the cell wall tension , @xmath51 , in the model introduced in the main text , we consider the cell wall as a thin elastic shell that deforms plastically when stretched beyond a maximum strain @xmath203 , the yield strain . a thin shell has two modes of elastic deformations , bending and stretching [ s@xcite ] , such that @xmath204 . in the limit of small thickness of the shell @xmath205 as compared to its radii of curvature , one can neglect the bending energy ( that scales as @xmath206 ) whereas the stretching energy @xmath207 is given by , @xmath208 where @xmath209 is the mechanical stress tensor and @xmath210 is the strain tensor . as yield strain is reached at the onset of growth , we have @xmath211 ( assuming isotropic stretching ) , where @xmath212 is the kronecker delta . furthermore , assuming a hookean constitutive relation for the stress tensor [ s@xcite ] @xmath213\;,\ ] ] where @xmath214 is the young s modulus and @xmath215 is the poisson ratio , we have @xmath216 , where , @xmath217 for a gram - negative bacterial cell wall of thickness @xmath218 nm , elastic modulus @xmath219 mpa [ s@xcite ] and average yield strain @xmath220 , the wall tension is estimated to be @xmath72 nn@xmath73 m . while the actual value for the turgor pressure counteracting this tension can be contested , our choice for the numerical value for internal pressure @xmath50 can be justified using a simple mechanical argument . radial force - balance dictates that in order to maintain an average cross - sectional radius @xmath221 @xmath82 m , a cell wall with surface tension @xmath222 50 nm/@xmath82 m has to balance an internal pressure of magnitude @xmath223 0.25 mpa , which is numerically very close to our choice for @xmath71 mpa . * cytoskeletal bundle mechanics*. next , we model the mechanical energy in the cell wall due to interactions with cytoskeletal proteins . our minimalist approach considers two crucial protein bundles that are directly responsible for maintaining the shape of _ c. crescentus mreb protein bundles contribute an effective energy @xmath52 , which favors a rod - like shape , and crescentin filament bundles contribute an energy @xmath53 , which favors a crescent - like shape . we thus have @xmath224 . mreb subunits form patchy filamentous bundles adherent to the cell wall and oriented perpendicular to the long axis of the cell [ s@xcite ] . the elastic energy stored in an adherent mreb subunit is given by @xmath225 where @xmath226 labels the subunit , @xmath227 is the circumferential radius of curvature of the cell wall where the subunit is attached , @xmath228 is the length of the subunit , @xmath56 is the intrinsic radius of curvature and @xmath229 is the bending rigidity of the associated mreb bundle . the total energy imparted by a collection of @xmath230 attached mreb subunits is given by @xmath231 . next , employing a continuum mean field assumption , we replace individual subunit lengths by their average length @xmath232 nm [ s@xcite ] and assume a uniform number density @xmath233 of mreb subunits in the cell surface to obtain @xmath234 where , @xmath235 is the effective bending modulus due to mreb induced traction forces . the bending rigidity of an mreb bundle is given by @xmath236 , where @xmath237 is the flexural rigidity of mreb filaments ( assumed to be similar to f - actin ) , @xmath238 is the number of mreb protofilaments per bundle , and @xmath239 is an exponent in the range 12 depending on the strength of crosslinking or bundling agents [ s@xcite ] . mreb filaments in a bundle appear to have strong lateral interaction with negligible filament sliding , so we assume @xmath240 . the diameter of an mreb protofilament is @xmath241 nm [ s@xcite ] and the average width of an mreb bundle has been determined from super - resolution imaging to be in the range 60 - 90 nm [ s@xcite ] , giving the estimate @xmath242 15 - 22 . estimating the surface number density of mreb subunits as @xmath243 @xmath82m@xmath244 and using @xmath245 nn@xmath82m@xmath84 [ s@xcite ] , we obtain @xmath57 in the range 5.612.1 nn@xmath82 m . crescentin proteins form a cohesive bundled structure anchored to the sidewall of _ c. crescentus _ cells . the energetic contribution due to crescentin is given by @xmath246 where @xmath64 is the bending rigidity , @xmath247 is an arc - length parameter , @xmath248 is the longitudinal curvature of the cell wall and @xmath62 is the preferred radius of curvature of crescentin bundles . the bending rigidity of crescentin can be expressed as @xmath249 , where @xmath250 is the young s modulus and @xmath251 is the area moment of inertia of the bundle ( with width @xmath252 ) given by @xmath253 . since crescentin is an intermediate filament homologue , we assume that @xmath250 is similar to the young s modulus of intermediate filament bundles given by @xmath254 mpa [ s@xcite ] . assuming @xmath255 @xmath82 m , we estimate the bending rigidity of a crescentin bundle to be @xmath256 nn@xmath82m@xmath84 . * mean field model for cell shape and size dynamics*. as described in the main text , the total mechanical energy in the cell wall of _ c. crescentus _ cells can be given as a sum of contributions from internal pressure , wall surface tension , mechanical energy of interactions with bundles of cytoskeletal proteins such as crescentin and mreb , and the constriction energy @xmath54 during cell division . in the mean field description , we neglect the contributions from @xmath54 and disregard any spatial variations in cell geometry . the mean field description is a good approximation of the cell shape dynamics for @xmath257 , where the average width and the midline radius of curvature remains constant ( fig . 2 in the main text ) . we approximate the shape of _ c. crescentus _ cells as the segment of a torus with radius of curvature @xmath258 , cross - sectional width @xmath259 and spanning angle @xmath260 . we also neglect the pole caps that are mechanically rigid and do not remodel during wall growth . the total energy is then given by @xmath261=-p ( \pi { \bar{w}}^2 r\theta)/4 + \gamma ( \pi r { \bar{w}}\theta ) + e_\text{width } + e_\text{cres}\;,\ ] ] where @xmath262 ^ 2 \;,\ ] ] @xmath263 ^ 2\;.\ ] ] from the above expressions , we see that the total energy has the scaling form @xmath264=\theta u[{\bar{w}},r]$ ] . according to eq . ( 1 ) in the main text , the dynamics of the radius of curvature @xmath10 , the spanning angle @xmath31 and the width @xmath9 are given by @xmath265 where @xmath266 , @xmath267 and @xmath268 are the rate constants . the steady - state values for the radius of curvature of the centerline and the cross - sectional width is given by the solutions to @xmath269 and @xmath270 . as shown in supplementary fig . 4a the energy density @xmath87 admits a stable absolute minimum in width , controlled by the stiffness parameter @xmath57 such that for lower values of @xmath57 ( red curve in supplementary fig . 4a ) , @xmath87 does not have a minimum and the width grows in time rendering the rod - like shape unstable . similarly , the parameter @xmath64 controls cell curvature , such that the energy density has a stable absolute minimum in @xmath10 beyond a critical stiffness @xmath64 ( supplementary fig . 4b ) . at lower values of @xmath64 the cell does not have a stable radius of curvature and the energy is minimized by increasing @xmath10 . this leads to straightening of the cell s medial axis . in the parameter range where the cell maintains a stable value for @xmath77 and @xmath10 , the energy density @xmath87 becomes a constant during cell elongation . from the growth law introduced in the main text , the spanning angle @xmath31 evolves in time according to the equation @xmath271 such that the condition for exponential growth becomes @xmath88 . * constriction dynamics*. as described in the main text , we assume that the shape of the constriction zone is given by two intersecting hemishperical segments with diameter @xmath272 . the total surface area of the septum is given by @xmath273 , where @xmath274 is the total length of the spherical segments ( see supplementary fig . 5a ) . using elementary geometry , we obtain the following relation between @xmath274 and the minimum width , @xmath275 : @xmath276 if septal growth occurs by addition of new peptidoglycan strands while maintaining the curvature of the spherical segments , then @xmath274 is the shape variable controlling the growth of septal surface area @xmath67 as well as the constriction dynamics of @xmath275 . to describe the dynamics of @xmath274 , we assume that initial phase of elongation occurs with a velocity @xmath277 , and thereafter growth follows an exponential law with rate @xmath108 , @xmath278 multiplying both sides of the equation by @xmath279 , we derive the growth dynamics of the surface area s(t ) , @xmath280 where @xmath281 . having determined the dynamics of @xmath282 , one can evaluate the time - dependence of the average width @xmath77 defined as , @xmath283 , where @xmath60 is the coordinate along the centerline . using the simplified geometry of the constricting cell , given by two toroidal segments ( peripheral regions ) connected by two intersecting hemispheres ( septal region ) , one gets , @xmath284\;,\ ] ] where @xmath285 , is the width in the septal region and @xmath77 is the average width of the stalked and swarmer components as determined by eq . . as shown in fig . 2c in the main text ( blue solid line ) , eq . is in excellent agreement with the experimental data and captures the dip in @xmath259 seen for @xmath118 . fig . 2b in the main text also shows that the radius of curvature of the centerline ( @xmath258 ) drops for @xmath118 when cell constriction is prominent . therefore we examine the role of constriction dynamics on the time - dependence of the centerline radius of curvature . the centerline spans a distance @xmath274 and an angle @xmath286 in the septal region , as shown in supplementary fig . consider a small segment @xmath287 along the centerline with a local radius of curvature @xmath10 and spanning angle @xmath288 . we then have the relation @xmath289 . this leads to the following identity , @xmath290 now the geometry of the septal region ( supplementary fig . 5a ) directly relates the rate of increase in the spanning angle @xmath291 to the rate of drop in the tangent angle ( @xmath292 ) at the constriction site , @xmath293 . furthermore , using the relations , @xmath294 and @xmath295 , we get @xmath296 . the last equality follows after variations of eq . . as a result we have the following kinetic relation , @xmath297 combining the identities in eq . and we get , @xmath298 , showing that the radius of curvature drops at the same rate as @xmath275 shrinks . * contour model for cell shape*. to quantitatively capture the experimentally observed spatial variations in cell shape we study an effective two - dimensional contour model for the cell shape . this approach facilitates closer comparison with the two - dimensional splined contours obtained from our experimental data . the contour description of the rod - like cell shape negelects circumferential variations in cell geometry . the model incorporates non - uniform materials properties and mechanical constraints across the cell wall , with the poles and the septal region being mechanically stiffer than the rest of the cell wall . by exploiting the rod - like geometry , one can use the centerline curve to divide the contour into two parts , the _ upper _ and the _ lower _ cell wall . as shown in supplementary fig . we further subdivide the cell contour into the stalked and swarmer poles ( @xmath299 ) , the upper ( @xmath300 ) and lower cell wall ( @xmath301 ) , and the septal region ( sep ) . we parametrize the instantaneous shape of the cell contour using the two - dimensional centerline curve @xmath302 , the length of the centerline @xmath303 and the width @xmath304 , where @xmath60 is the absolute distance along the centerline from the stalked pole such that @xmath305 $ ] . if @xmath306 denotes the outward unit normal vector on the centerline , the curves defining the upper and lower parts of the cell contour , @xmath307 , are given by the relation , @xmath308 , where @xmath309 represent the perpendicular distances of the top and bottom curves from the centerline . the total cell width is then given by @xmath310 . it is convenient to switch to polar coordinates , where the shape of the cell contour is given by the re - parametrized curve @xmath311 , where @xmath312 is the angular coordinate spanning the centerline , which can be approximated as the arc of a circle with radius @xmath258 . since the ratio @xmath313 is small at all times , one can approximate the local curvature as @xmath314 where prime denotes derivative with respect to @xmath315 and the subscripts @xmath172 represent the upper and the lower part of the cell contour respectively . furthermore , in the linear regime , the differential arc length can be approximated as @xmath316 . the dynamics of the shape parameters @xmath303 , @xmath260 , and @xmath258 are determined from the kinetic law in eq . ( 1 ) of the main text . the instantaneous width profile @xmath304 results from minimizing the total energy functional , which leads to the following shape equation , @xmath317 where , @xmath318 is the tension on the cell contour and @xmath319 and @xmath320 are the linear force densities on the contour due to maintenance of width and the crescent shape , respectively . in the contour model we simplify the energetic contribution due to maintenance of width ( acting on the sections @xmath300 and @xmath301 ) as @xmath321 where the elastic constant @xmath322 depends on the bending rigidity @xmath57 introduced in the main text as @xmath323 . this linear approximation holds if @xmath324 . from our data and mean field model fits we get @xmath325 . the resultant force density is given by @xmath326 . the bending energy induced by crescentin protein bundles anchored onto the cell wall ( regions @xmath300 ) is given by @xmath327 ^ 2\;,\ ] ] where @xmath328 is the spontaneous curvature of the crescentin bundle , and s is the arc - length parameter along the upper part of the cell contour ( @xmath300 ) which is related to @xmath60 as @xmath329 . to obtain the force density we consider an infinitesimal deformation , @xmath330 , of the upper contour as @xmath331 , where @xmath332 is the outward unit normal . accordingly the curvature and the differential arc - length changes , @xmath333 and @xmath334 , where @xmath335 and @xmath336 [ s@xcite ] . the resultant force density is obtained after variations of the energy functional , @xmath337 , where @xmath338 $ ] . this leads to the following non - linear force contribution : @xmath339 using eq . , @xmath340 can be linearized and expressed using the angular coordinate @xmath315 as @xmath341\;.\ ] ] * compartmentalizing the cell contour*. ( 1 ) pole caps : the cell poles are assumed to be mechanically inert in the sense that they do not interact with the active cytoskeletal proteins such that @xmath342 . the mechanical forces acting on the cell poles come from turgor pressure @xmath50 , and the tension @xmath343 . the shape of the cell poles are then described by the two - dimensional laplace s law , @xmath344 where the superscripts @xmath345 and @xmath346 denote respectively the stalked and swarmer poles . \(2 ) upper cell contour ( @xmath300 ) : the region @xmath300 in the upper cell contour obeys the force - balance equation : @xmath347 at the endpoints of the segments @xmath300 we impose the boundary condition that the curvature @xmath348 must equal the longitudinal curvature of the poles . \(3 ) bottom cell contour ( @xmath301 ) : in the absence of crescentin , the region @xmath301 in the bottom cell contour obeys the force - balance equation : @xmath349 \(4 ) septal region : we incorporate the effect of constriction in the cell shape equation by imposing the constraint ( boundary condition ) that @xmath350 , where @xmath275 evolves according to the kinetics described in eq . ( 6 ) and ( 7 ) of the main text . furthermore , the curvatures at the end points of the septal segments must conform to the curvature of the newly formed poles . subject to these boundary conditions , the width profile in the upper septal region ( @xmath351 ) is given by , @xmath352 with a surface tension ( @xmath353 ) of the newly formed poles chosen to be much higher than the peripheral region , @xmath354 . the contour below the centerline obeys the equation , @xmath355 the governing cell shape equation given in eq . , is then solved numerically in each part of the cell contour with matching boundary conditions in @xmath356 and its derivatives . 10 url # 1`#1`urlprefix[2]#2 [ 2][]#2

active galactic nuclei ( agn ) are extremely powerful cosmic objects , driven by accretion of hot gas upon super - massive black holes . the zoo of agn classes are divided into two major groups , with type-1 agn displaying broad balmer emission lines and type-2 narrow ones . for a long time it was believed that a type-2 agn is a type-1 agn viewed through a dusty kiloparsec - size torus , but an emerging body of observations suggests more than just the viewing angle matters . here we report significant differences in supernova counts and classes in the first study to date of supernovae near type-1 and type-2 agn host galaxies , using data from the intermediate palomar transient factory , the sloan digital sky survey data release 7 and galaxy zoo . we detect many more supernovae in type-2 agn hosts ( size of effect 5.1 ) compared to type-1 hosts , which shows that the two classes of agn are located inside host galaxies with different properties . in addition , type-1 and type-2 agn that are dominated by star formation according to wise colours and are matched in 22 m absolute magnitude differ by a factor of ten in[oiii] luminosity , suggesting that when residing in similar type of host galaxies type-1 agn are much more luminous . our results demonstrate two more factors that play an important role in completing the current picture : the age of stellar populations and the agn luminosity . this has immediate consequences for understanding the many agn classes and galaxy evolution .

1701.08647

the agn unification theory @xcite has been subject to many successes and , also , some controversies . in its simplest form only the viewing angle towards the torus matters . this basic picture has long been challenged by statistical tests @xcite . many statistical tests are often overlooked due to anisotropic sample selection or small sample sizes @xcite . recently , it was shown that even with isotropic selection criteria and large sample sizes of thousands of pairs , the galaxy neighbours to type-1 and type-2 agn differ significantly @xcite within a few hundred kiloparsec , including a difference in the number of close neighbour galaxies @xcite . this claim is not uncontroversial . seemingly , gordon et al . ( 2016 ) do not find any differences in the galaxy neighbours . however , their study suffers from poor statistics ( using only tens of pairs ) and does not mention some significant colour differences they find in neighbours , see @xmath6-values from the kolmogoroff - smirnoff tests in table 4 of @xcite . recent works @xcite support that factors beyond the viewing angle must play a role @xcite . some believe the torus is a nuclear stellar nursery driven by inflows and outflows @xcite or the outer parts of a disk - wind coming from the accretion disk @xcite , driving an evolutionary sequence from type-1 @xmath7 type 1.2/1.5 @xmath7 1.8/1.9 @xmath7 type-2 @xcite . presence of spectral features at 10 and 18 @xmath8 indicate that the agn has different clumpiness of the torus depending on if it is a type-1 or type-2 agn and if isolated or in merger @xcite . the absence of detected broad - line regions in low - luminosity agn @xcite , and lack of tori at the high - luminosity end , further complicate the picture . these findings open up for some important questions : what are the true physical differences between the type-1 and type-2 agn populations ? differences could lie in the physics of the central engines , in the structure of the tori @xcite , but also in the host galaxies themselves , e.g. @xcite . the star - formation histories of the host galaxies can be readily compared with the help of supernovae ( sne ) , requiring no assumptions about the composition of the galaxy spectra . the luminosity - weighted age of the stellar population is reflected in the occurrence of different sn types : progenitors of core - collapse ( c - c ) sne are massive stars ( @xmath9 ) with short life - times ( @xmath10 yrs ) and indicate recent or ongoing star formation . thermonuclear sne , whose progenitors are white dwarfs that take on average @xmath11 years to form , are indicators of earlier epochs of star formation . early - type galaxies which are dominated by old stellar populations are not known to host core - collapse sne , whereas thermonuclear sne are found in all types of galaxies @xcite . recent works indicate a larger fraction of core - collapse to thermonuclear sne in non - active , star - forming spirals than in spirals hosting agn . this fraction is connected to the earlier morphological type of the agn hosts @xcite and the stage of eventual merger in close pairs of galaxies @xcite . in agn unification theory the sn counts and types are expected to be the same for both classes of agn . in this study using sne from the intermediate palomar transient factory ( iptf ) , we test whether type-1 and type-2 agn host galaxies have the same or different occurence rates of sne . different sn rates would indicate these agn types reside in galaxies with different star - formation histories . the agn and galaxy samples are taken from the sloan digital sky survey ( sdss ) @xcite data release 7 @xcite . the sne are taken from the iptf catalogue . this survey is well suited for this work due to its coherent , untargeted mode of detecting transient sources with one and the same telescope . in section [ sec : methods ] we discuss the sample selection and methods . in section [ sec : results ] the main results , in section [ sec : biases ] different potential biases that can influence the results . in section [ sec : statistical ] we present the statistical analysis . finally , we present the conclusions in section [ sec : conclusions ] . the iptf is an un - targeted wide - field sky survey using the 1.2 m samuel oschin telescope ( p48 ) at palomar observatory to detect and follow up transient astronomical sources . the scientific scope of the iptf spans from small solar system objects to extragalactic phenomena . the iptf project has been running since 2013 , and it had a precursor , the palomar transient factory ( ptf ) , which was active during 2009 - 2012 . the technical background of ptf is presented in @xcite and the scientific motivation in @xcite . the iptf is presented in @xcite . for brevity , we will refer to our sn catalogue ( including sne found during the ptf period ) simply as the iptf catalogue . in this study , we use the iptf sn catalogue from the time window between 2009 march 2 until 2014 june 17 . this catalogue contains 2190 extragalactic sne with known right ascension ( @xmath12 ) , declination ( @xmath13 ) , spectroscopic redshift ( @xmath14 ) and spectroscopic classification . this sn catalogue contains 1494 thermonuclear sne ( i.e. type ia ) and 632 core - collapse sne . in the core - collapse category , we include sne types ib , ic , ib / c , ibn , ii , iip , iil , iin and iib . the remaining 64 sne are either of unclear spectroscopic type or superluminous sne . none of the 11 superluminous sne with @xmath15 in our catalogue was found in a galaxy with a spectrum in sdss dr7 . the superluminous sne can therefore , for our purposes , be included in the remainder category . the sne in this sample are located at declinations @xmath16 , mostly at galactic latitudes @xmath17 . the mean redshift of the sample is @xmath18 , with 95 % of the sne having @xmath15 . for a motivation of curtailing the sn catalogue at 2014 june 17 , [ sec : snbiasiptf ] . an advantage of the iptf catalogue is its cohesive nature . all the sne have been discovered during an untargeted search with the same telescope , and spectroscopic classifications have been made in a timely fashion . the compatibility with sdss in sky coverage makes the iptf sne sample suitable for our investigation . our sdss samples covers @xmath19 and galactic latitudes @xmath17 , comparable to the distribution of the sn locations . another sn catalogue that comes to mind is the sdss sn catalogue @xcite . the 902 confirmed sne in the sdss sn survey represents a sn sample of smaller scope than the iptf sample . not all sne are spectroscopically classified , but the survey is deep and homogenous . unfortunately , the sne are all located in the sdss southern equatorial stripe 82 , which is outside the area of the sdss dr7 sample from the central region used in this study . this renders the sdss sne unsuitable for our purposes . the asiago sn catalogue @xcite is a historically comprehensive compilation of extragalactic sne . the catalogue encompasses 6530 sne as of 2016 march 27 , in both hemispheres . the asiago catalogue is extensive but has uneven quality some of the sne in it lack spectroscopic classification , some lack spectroscopic redshift . the sn catalogue compiled by @xcite contains data for 5526 extragalactic sne , whereas the sternberg sn catalogue , presented by @xcite , contains less than 3000 extragalactic sne . the circumstance that they are compiled from a wide range of sources make them less suitable compared to the cohesive iptf catalogue . the same holds for the open supernova catalogue @xcite , with @xmath20 sne ( of which 12 % have spectra in the catalogue ) as of 2016 december , collected from different public sources . as discussed by @xcite , earlier sn searches have prioritised sn detection over completeness with respect to sn types or host galaxy types . this bias should affect such compilations as the asiago catalogue and other catalogues listing sne found before the start of untargeted sn searches . the samples of host galaxies were obtained through the sdss data release 7 . we select objects classified as either quasars or galaxies , within redshift @xmath21 , unless flagged for brightness ( flags&0x2=0 ) , saturation ( flags&0x40000=0 ) , or blending ( flags&0x8=0 ) ( * ? ? ? * their table 9 ) . the emission lines are obtained from the specline table in dr7 . we require that the objects have h@xmath12 in emission and select type-1 agn , type-2 agn and star - forming galaxies using optical emission line diagnostics . our type-1 agn are objects with @xmath1(h@xmath12 ) @xmath2210 ( or fwhm(h@xmath12 ) @xmath22 1000 km / s ) . the type-2 agn have narrow lines @xmath1(h@xmath12 ) @xmath23 10 fulfilling the kauffmann criterion @xcite : @xmath24/{\rm h}\beta ) > 0.61/(\log([\ion{n}{ii}]/{\rm h}\alpha))-0.05)+1.3 \label{eq , bpt}\ ] ] [ bptequation ] the star - forming galaxies are defined as all the other narrow - line objects . in this way , we classify the objects into type-1s , type-2s and star - forming galaxies , referred to as `` largest samples '' , using optical emission line diagnostics . for all objects we search for morphological classifications ( spiral , elliptical , uncertain ) from the project galaxy zoo 1 @xcite , emission line measurements , redshifts and celestial coordinates , leaving `` parent samples '' of 11632 type-1 agn ( 1864 spiral ) , 77708 type-2 agn ( 36720 spiral ) and 137489 star - forming galaxies ( 49072 spiral ) . for the vast majority of these objects we can also find wide - field infrared survey explorer ( wise ) magnitudes . as additional samples later used only for direct comparison of host galaxy properties , we also create some refined samples. starting from the parent samples , we select only face - on spiral hosts with high s / n in the emission lines from galaxy zoo data release 2 @xcite , minimizing dust extinction due to host galaxy inclination . we also require @xmath25 3 in h@xmath12 , minimum sdss gaussian line heights @xmath26(h@xmath12 ) @xmath22 10 * 10@xmath27erg / s/@xmath28/ and h(h@xmath29 ) @xmath22 5 * 10@xmath27erg / s/@xmath28/ in order to avoid effects of stellar absorption affecting weak lines in our classification . out of these , we select only those having wise colours . the refined samples will be used for comparing star formation with wise colours in type-1 and type-2 agn or the [ ] 5007 in host galaxies matched by amount of dust . the numbers of coherently collected sne are scarce ( 2190 in our sample as on 2014 june 17 ) . thus , we create pairwise matched subsamples of type-1 agn , type-2 agn and star - forming objects to compare objects as similar to each other as possible in redshift distribution and selected properties e.g. the luminosity @xmath4[]5007 . we compare ( i ) type-1 agn to type-2 agn , and ( ii ) type-2 agn to star - forming galaxies . the aim with the latter test is to probe whether the observed type-2 agn properties can be explained by star formation alone . for two samples of intrinsically similar objects the probability of detecting faint sne is the same if they have similar redshift distributions . therefore , for each galaxy in the parent sample , we select a galaxy from the second parent sample having the closest value in redshift and a specific property of interest . after the matching is done , we first throw away all matched pairs that differ more than 20% in the property of interest . four types of specific properties and matchings are explored : 1 . redshift only . this allows to remove biases in galaxy zoo morphology classifications as well as the malmquist bias . @xmath4[]5007 from the narrow - line region ( nlr ) . in the simplest agn unification , the nlr is believed to be isotropically distributed outside the torus and to be equally strong for agn of the same activity level irrespective of the viewing angle . selecting on @xmath4[]5007 meaning one selects all type-1 and type-2 agn above a selected certain line flux in a sample should give the same host galaxy properties under the conditions of isotropy @xcite . matching on @xmath4[]5007 is similar to selecting on @xmath4[]5007 if the two @xmath4[]5007 distributions are the same ( as predicted by the simplest unification ) but can be problematic if the distributions differ at the high - luminosity end . moreover , we expect the same line width @xmath1[]5007 in matched samples type-1 and type-2 agn samples . wise @xmath30 ( 22@xmath8 ) absolute magnitude . we use this match as our stellar - mass proxy assuming the dust emission traces the stellar mass . the 22@xmath8 magnitude is a good measure of heated - dust emission in the host galaxy , especially in galaxies where the torus contribution to the total 22@xmath8 is negligible in comparison , meaning galaxies having @xmath2 @xcite . a less favourable option is to use dust reddening @xmath31(h@xmath12/h@xmath29 ) , but if dust reddening in the blr and nlr differs @xcite , the matching will be biased . we therefore neither correct @xmath4[]5007 for dust reddening . this matching is only done for the galaxies in the parent samples that have wise magnitudes . 4 . exponential fit scale radius ( r - band ) . as the star - formation history depends on the available gas mass , matching by an apparent measure of the galaxy volume should minimize differences in star - formation histories under the assumption of a mass - size relation . we do a two - sample kolmogoroff - smirnoff test for each matched property to ensure the sample distributions are statistically similar . for the @xmath30-matched samples we had to do a finer match by throwing away matched pairs differing more than 5% in @xmath30 . the pairwise matched subsamples for examining sn counts are created by matching in three different parameters ( redshift @xmath32 , luminosity @xmath4[]5007 and @xmath33 ) as described earlier . the sizes of the pairwise matched subsamples can be found in table [ supernovaall ] . an example of the redshift and property distributions in the matched type-1 and type-2 samples can be seen in figure [ redshiftdistributions ] and [ otherdistributions ] . the similarity of all the matched samples is ensured through two - sample kolmogoroff - smirnoff where the null hypothesis ( that two matched samples are similar ) holds for the nominal value @xmath12=0.05 . the procedure is repeated for pairwise matched type-2 agn and star - forming galaxies . finally , the refined samples of face - on hosts with high signal - to - noise are also matched with the same method . in addition , the refined samples are matched by balmer decrement @xmath31(h@xmath12)/@xmath31(h@xmath29 ) . a sn is considered to be matched with a galaxy if the following two conditions are both satisfied : * the projected distance on the plane of the sky between the sn and the galaxy is computed ( plane approximation ) . the search radius @xmath34 is set by converting the desired physical search radius into a angular radius at a galaxy distance assumed to be @xmath35 . a hubble constant @xmath36 = 72 @xmath37 @xmath38 @xmath39 is used throughout this work . if the sn is found to lie inside the given search radius , the compliance with a redshift condition is also checked . * our redshift condition is @xmath40 and accounts for redshifts due to peculiar motions . this is more strict than the redshift matching condition of @xmath41 used in the study by @xcite . in their study , however , most ( 97 % ) of their sample of 620 cross - matches fulfills the @xmath40 condition . if both these conditions are fulfilled , the sn is considered to be associated with the galaxy . the sne inside a galaxy are collected by the @xmath42 10 kpc , and those inside a galaxy or in a close companion are found with the @xmath42 100 kpc criterion . if this is wrong and there is no association between the agn and the sn , then the distribution of different sne with different @xmath43 = @xmath44 ought to be a uniform distribution . this can be checked by plotting a histogram with the @xmath43 between the agn and the sne , see fig [ supernovagalaxydelta ] . it is clear that the distribution is bottom - heavy and that most of the sne are associated with the host galaxies . it is important to point out , that the sn redshift determination method brings in uncertainty regarding the redshift uncertainties . most of the time the redshift is measured from the host galaxy of the sn and is rather accurate . but sometimes , the sn redshift is measured directly from the sn itself and influenced by doppler broadening and expansion velocities , being slightly larger @xcite . the degree of association between sne and galaxies can also be checked by visually inspecting the sdss dr7 images of the galaxies , with the sn positions overplotted . we note the subjectivity involved , e.g. in cases where a sn occurs in an interacting pair of galaxies . for the largest sample of type-2 agns ( 77708 galaxies ) , and the 59 sne matched to them , about 10 % of the sne visually appears to be located in a companion galaxy of a sample galaxy . for the largest sample of starforming galaxies ( 137489 galaxies ) , and the 152 sne matched to them , a comparable fraction ( about 9 % ) appears to be located in a companion galaxy . these results from visual inspection of sn positions complements the conclusions drawn from fig . [ supernovagalaxydelta ] . we begin our analysis by counting sne within two different projected distances from the center of the host galaxies : 10 kpc ( within the typical radius of a spiral galaxy ) , 100 kpc ( within the galaxy or a possible close companion ) . we start with the largest samples , see table [ largest ] . the first thing to notice is the lack of sne around type-1 agn at projected separations d @xmath23 100 kpc . only one sn is found . the number of sne around type-2 agn are clearly systematically higher for the largest samples in table [ largest ] ( size of effect @xmath0 5.1@xmath1 ) . we verify that this is no bias due to potential difficulties in detecting sn in the immediate vicinity of bright or transient agn by excluding all sne within 3 arcseconds from the host centres , see section 4.3.2 . this still yields 1 sn near 11632 type-1 agn ( detection fraction @xmath0 8.6 @xmath45 10@xmath46 ) and 46 sn near 77708 type-2 agn ( detection fraction @xmath0 5.9 @xmath45 10@xmath47 ) a significant difference ( size of effect @xmath484.1 @xmath1 , see section [ sec : unc ] ) . visual inspection show that the majority of the sne ( @xmath22 90 % ) come directly from the agn hosts ( only a small fraction from the companion galaxies ) . for the type-1 and type-2 samples matched in redshift or @xmath30 the differences are significant . but for the @xmath4[]5007-matched samples we only find one sn in each sample , yielding no difference at all . does this mean the simplest unification is valid and there is no difference in galaxy properties and sn counts in samples selected and matched in @xmath4[]5007 ? if so , also the host morphologies for the matched samples should be the same . but the fraction of type-1 and type-2 agn in spiral hosts is 20 % vs 44 % , in disagreement with the simplest unification . this suggests that the lack of significant difference in sn counts stems from poor statistics . as an alternative test , we select on @xmath4[]5007 . this should be suitable as in the simplest unification no difference is expected in @xmath4[]5007 on the higher luminosity - end ( while matching removes potential differences at the high - luminosity end ) . we do this twice for the largest samples using both @xmath31[]5007 @xmath49 @xmath45 10@xmath50 erg / s/@xmath28 and @xmath31[]5007 @xmath51 @xmath45 10@xmath50 erg / s/@xmath28 as flux limits . the difference in @xmath4[]5007-selected type-1 and type-2 agn samples is noteable ( @xmath31[]5007 @xmath52 case : size of effect @xmath482.8 @xmath1 , @xmath31[]5007 @xmath51 case : 1.8 @xmath1 ) following the same trend of table [ largest ] . this reflects purely changes in the sample sizes , unless the difference in matched vs selected samples originates in a break - down of unification at the higher luminosities . as early - type objects are well - known @xcite for having few sne setting off , it would be ideal to use only spiral hosts . we do not find a single sn around spiral type-1 agn . but using a statistical hypothesis test @xcite ( see section [ sec : pairwise ] ) only a borderline - significant difference in sn counts for type-1 and type-2 agn at @xmath42 100 kpc ( p - value @xmath0 0.06 ) is found , showing the need of larger sn samples . for the pairwise matched samples the counts are too small to show any significance individually . the @xmath4[]5007-matched samples we commented upon earlier , while for the redshift - matched samples a significant difference is found for @xmath42 10 kpc . the lack of significant difference for @xmath42 100 kpc therefore stems from the poor statistics . the samples matched in redshift and apparent galaxy size yield small , insignificant differences between type-1 and type-2 agn . as the size of the galaxy depends on the star - formation history , no difference in counts could mean that we have either matched successfully in the star - formation history , or more realistically , that the sample sizes are too small . however , the collected results in table [ supernovaall ] are visually presented in figure [ supernovacountsplot ] where the left column reinforces our earlier conclusions : type-2 agn hosts have higher sn rates than type-1 agn hosts , much higher than from the expectations of the simplest unification theory ( represented by the grey line ) . also the insignificant differences fall into the same area of the plot . the larger count of sne could mean either a difference in stellar age or stellar mass ( or both ) . but samples unmatched in stellar mass lead to differences in clustering on mpc scale @xcite while we find no differences in sn counts on large scale , see sections [ sec : largescale1 ] and [ sec : largescale2 ] . moreover , the matching in @xmath30 our proxy for stellar mass still yields significant different sn counts . this suggests that the discrepancy in sn counts is due to differences in stellar age between type-1 and type-2 agn . the conclusion is supported by the much higher @xmath53 colour indices in type-2s indicating stronger star formation @xcite in samples consisting of face - on , spiral - host type-1 and type-2 agn that are dominated by dust emission from stars with @xmath2 @xcite and pair - wise matched in @xmath4[]5007 , see section [ sec : altsfr ] . less striking differences are found between type-2 agn and star - forming galaxies . star - forming spirals show higher core - collapse sn counts at d @xmath23 10 kpc ( @xmath0 3.3@xmath1 ) and even in the redshift - matched samples a difference is present . the right column in figure [ supernovacountsplot ] shows a relationship between the two classes , although offset in y - axis following the shape of the grey line . one may argue that the @xmath4[]5007 may be anisotropic or even have different physical origins in type-1 and type-2 agn . if so , one expects to find differences in the gaussian line width of the [ ] 5007 emission in type-1s and type-2s and/or differences in sn counts . as type-1 agn have a contribution of a non - stellar , power - law continuum component to their observed luminosity , this could influence the detections of type-1 and type-2 agn . however , the emission lines are independent of the continuum emission and in the simplest unification @xmath4[]5007 is expected to be an isotropic indicator of agn luminosity . type-1 and type-2 samples selected on , or matched in , @xmath4[]5007 should be the same in all other properties : they should have similar host galaxy types and similar nlr kinematics . however , in @xcite @xmath4[]5007-matched hosts showed different colours of their galaxy neighbours . using @xmath4[]5007-matching we see no difference , or can not see , in the sn counts near type-1 and type-2 agn : we find one sn near type-1 hosts and one sn near type-2 hosts . while matching is good , a better way is still to select on @xmath31[]5007 . selecting on @xmath31[]5007 means that all type-1 and type-2 above a certain flux value in @xmath31[]5007 should have the same properties . we try this for two separate lower flux limits : @xmath54 10 or 30 @xmath45 10@xmath50 erg / s/@xmath28 , yielding the following sn counts within @xmath55 100 kpc from the host galaxies in the largest samples : 1 . @xmath54 10 @xmath5610@xmath50 erg / s/@xmath28 . we have 5101 type-1 agn and 47010 type-2 agn . the number of sne near type-1 agn is 1 ( detection fraction @xmath0 1.96 @xmath57 10@xmath47 ) the number of sne near type-2 agn is 41 ( detection fraction @xmath0 8.72 @xmath57 10@xmath47 ) . the estimate of the size of effect is @xmath0 2.8@xmath1 . @xmath54 30 @xmath5610@xmath50 erg / s/@xmath28 : we have 3403 type-1 agn and 19976 type-2 agn . the number of sne near type-1 agn is 1 ( detection fraction @xmath0 2.94 @xmath57 10@xmath47 ) . the number of sne near type-2 agn is 19 ( detection fraction @xmath0 9.51 @xmath57 10@xmath47 ) . the estimate of the size of effect is @xmath0 1.8@xmath1 . the number of sne in the largest type-2 host samples ( 77709 objects ) within @xmath42 100 kpc was 59 sne ( see table [ largest ] ) . considering the smaller size of the flux - restricted type-2 agn samples , the expected new sn counts are : ( a ) @xmath54 10 : 47010/77709 * 59 sne @xmath0 35 , ( b ) 19976/77709 * 59 @xmath0 15 . therefore , we can easily see that the loss of significance in the @xmath31[]5007-selected sample reflects upon the decreased number of objects in the samples . to further probe the relevant nlr kinematics behind the @xmath4[]5007 emission , we analyze the agn themselves . if the same physical mechanisms are causing the [ ] 5007 emission in type-1s and type-2s and isotropically the [ ] 5007 luminosity - normalized line widths @xmath1([]5007 ) must be the same . when we explore the gaussian line widths @xmath1([]5007 ) of the various classes of objects , we use refined samples matched in redshift and @xmath31[]5007 . the normally distributed @xmath58 of the @xmath1([]5007 ) values for estimating means and errors are calculated for @xmath4[]5007-matched , face - on spiral hosts . no significant difference in line width for the type-1s and type-2 objects is found ( @xmath59 0.13 ) . this disagrees with the earlier observation that the nlr has a component of motion giving rise to geometric differences @xcite . a clearer line width difference between the type-2 ( @xmath60(@xmath1)= 0.242 @xmath61 0.003 ) and star - forming objects ( @xmath60(@xmath1)=0.191 @xmath61 0.004 ) . the slighty wider [ ] 5007 line in the type-1 and type-2 agn ( over star - forming galaxies ) support that a significant contribution to the [ ] 5007 flux might arise isotropically distributed in a region close to the agn nucleus where the clouds rotate around the center of the galaxy at higher velocities , causing additional doppler broadening . but also outflows in agn are known to cause broadening of the [ ] 5007 line and can at high redshift @xmath62 2.5 in extreme cases show broadening corresponding to @xmath63 km / s @xcite . perhaps , the slight differences in [ ] 5007 between the agn and star - forming galaxies therefore may indicate the presence of outflows from the nucleus . the results are displayed in table [ matchedzoocorrect ] . c c c + type of match , `` unification '' & @xmath64 & @xmath65(@xmath1 ) + type-1 * & 123 & 0.4713 @xmath61 0.0140 + type-2 & 123 & 0.4377 @xmath61 0.0101 + type of match , `` stars '' & @xmath64 & @xmath65(@xmath1 ) + [ 0.2ex ] type-2 * & 1244 & 0.2415 @xmath61 0.0035 + star - forming & 1244 & 0.1906 @xmath61 0.0036 + [ matchedzoocorrect ] if the obscuration is the dominant factor that separates type-2 from type-1 agn , one may expect that the refined samples of type-1s and type-2s matched in the heated - dust emission from their host galaxy are more or less as luminous in @xmath4[]5007 . for the 137 paired type-1s and type-2s in the refined samples matched in redshift and @xmath30 , we compare the mean @xmath4[]5007 . the mean @xmath60@xmath4[]5007 [ erg / s ] is 40.934 @xmath61 0.0455 for type-1 objects and 39.9738 @xmath61 0.0889 for type-2s . this demonstrates that type-1s are much more luminous ( @xmath0 9.6@xmath1 ) than type-2s in host galaxies with similar dust distributions . the effect is equally convincing if exploring and matching objects with wise bands dominated by star formation @xmath2 . with the simple criterion @xmath66 one can easily identify sources dominated by agn @xcite . in general , @xmath67-selected samples are biased towards more seyfert - like agn with low luminosity @xcite . we do this for our type-1 and type-2 agn and find that majority in the matched samples have @xmath68 0.8 , i.e. have the wise bands dominated by dust heated by star - formation , permitting the comparison using @xmath69 colours . an example are the @xmath4[]5007-matched samples where only 109 out of 123 type-1 agn and 116 out of 123 type-2 agn have their dust dominantly heated by star - formation . the best way to find out if this influenced our conclusions , is by redoing the tests using only objects that fulfil the @xmath70 @xmath23 0.8 condition by assef et al . ( 2012 ) and , or the even the stricter wright et al . ( 2010 ) condition @xmath70 @xmath23 0.5 . using these criteria the objects clearly are dominated by dust emission near the stars and comparison of the @xmath69 colour as well as the @xmath71-matching is valid . these objects we match by either @xmath71 or @xmath4[]5007 . the resulting samples are clearly smaller . the resulting , new @xmath71- and @xmath4[]5007-matched samples show the same average @xmath69 as before . type-2 shows again higher star - formation than type-1 . also here in the @xmath71-matched samples one can see that type-1 are much luminous than type-2 agn . an alternative view can be gained from comparing the @xmath30 in @xmath4[]5007-matched samples : -30.2 + /- 0.1 mag ( type-1 agn ) or -30.8 + /- 0.1 mag ( type-2 agn ) , supporting there is more dust in type-2 agn hosts than in type-1 agn hosts . one may wonder if this particular result has any connection to the receding torus model @xcite , where the opening angle of the torus gets larger with increasing agn luminosity . the increased ratio of type-1/type-2 agn at larger luminosities @xcite supports the idea of a receding torus . however , as it appears that the age of the stellar population differs between type-1 and type-2 agn , it seems more reasonable that the difference in the dust is on host galaxy scale . while our results say that type-1 agn are more luminous than type-2 agn , they do not support a receding torus per se . there are some potential selection biases that can influence our samples . weak lines used in the object classification might be influenced by stellar absorption . in this study , demanding a @xmath25 3 in h@xmath12 emission before selection gives similar results on the sn counts in table [ largest ] & [ supernovaall ] , but with poorer statistics . dust extinction effects due to inclination might potentially influence the emission line strengths @xcite or the sn detection rate . the differences in how type-1 and type-2 agn are classified ( type-1 needs only emission in h@xmath12 , while type-2 agn in three additional lines ) can bias the type-2 agn towards more star - forming hosts . but also might the selection of type-1s and type-2s become rather anisotropic due to the contribution of a non - stellar continuum component to the observed luminosity of type-1 agn . many of these issues where treated in the close neighbours study @xcite , correcting for these typical biases e.g. increasing the @xmath72 ratio of the used emission lines also did not change the results in any way . there , also effects from the removal of liners and the clumpy tori were explored , showing no influence on the outcome . nevertheless , we here present tests dealing with some particular problems related to the current study . to control several biases at the same time , we use the refined , pairwise matched subsamples . the pairwise matching ensures our samples to have the same redshift distribution and also the same distribution in one of the remaining four parameters separately ( only @xmath32 , @xmath4[]5007 , @xmath31(h@xmath12/h@xmath29 ) , @xmath73 ) . the refined sample sizes are shown in table [ wiserefinedzoosamples ] . type of match & @xmath64 type-1 * & @xmath64 type-2 & type-1 , @xmath69 & type-2 , @xmath69 + @xmath32 & 178 & 178 & 2.976 @xmath61 0.02 & 3.487 @xmath61 0.03 + @xmath4[]5007 & 123 & 123 & 2.938 @xmath61 0.030 & 3.489 @xmath61 0.04 + @xmath31(h@xmath12/h@xmath29 ) & 136 & 136 & 2.981 @xmath61 0.030 & 3.496 @xmath61 0.03 + @xmath73 & 176 & 176 & 2.975 @xmath61 0.02 & 3.672 @xmath61 0.03 + + type of match & @xmath64 type-1 * & @xmath64 type-2 & type-1 , @xmath69 & type-2 , @xmath69 + @xmath32 & 2788 & 2788 & 3.509 @xmath61 0.007 & 3.628 @xmath61 0.006 + @xmath4[]5007 & 1244 & 1244 & 3.495 @xmath61 0.001 & 3.648 @xmath61 0.009 + @xmath31(h@xmath12/h@xmath29 ) & 1616 & 1616 & 3.518 @xmath61 0.009 & 3.633 @xmath61 0.008 + @xmath73 & 2714 & 2714 & 3.462 @xmath61 0.006 & 3.668 @xmath61 0.006 + [ wiserefinedzoosamples ] one would wish to explore the sn rate in these samples at @xmath42 10 kpc . however , the currently available sn samples are far too small to allow this kind of investigation . we have to rely on an alternative star - formation indicator . we can use @xmath67 @xmath69 colour to measure the activity of star - formation , assuming that our objects @xmath67 colours are dominated by the same source ( either agn or the stars ) : the higher @xmath69 , the more star - formation @xcite . for the refined samples , we see a strong difference in the @xmath69 colour . the @xmath69 is significantly larger in type-2 agn compared to type-1 agn for all four matchings . an example is the @xmath71-matched type-1 and type-2 refined samples , @xmath69=2.962 @xmath61 0.03 for type-1s and @xmath69=3.674 @xmath61 0.029 for type-2s . the other matchings give very similar results . this strongly supports the observed larger number of sn detections around type-2 agn . also if redoing the entire pairwise matching and analysis only using objects having the infrared wise emission dominated by dust emission near stars using either @xmath70 @xmath23 0.8 @xcite and or @xmath70 @xmath23 0.5 @xcite , the conclusion stays equally true . the galaxy zoo data release 1 morphologies ( for those that have , @xmath0 90 % of all galaxies ) fall into three categories : spiral , elliptical and uncertain . the higher redshift , the more difficulties a galaxy zoo volunteer has to recognize a certain morphology . therefore , it might be difficult for a galaxy zoo - observer to recognize a spiral at higher @xmath32 , especially if there is a strong light from the nucleus . some spirals can fall into the category of `` uncertain '' due to the strong light in type-1 agn . however , this bias can not cause misclassifications in the other direction . while a spiral can be classified as `` uncertain '' it is very unlikely an `` uncertain '' galaxy will be classified as a spiral . therefore , what is in galaxy zoo classified as a spiral , is very likely to be a spiral as voted by hundreds of galaxy zoo volunteers . any redshift - dependent biases as the morphology - classification bias is removed by the use of redshift - matched ( all - morphologies ) samples . they show significant differences between type-1 and type-2 agn . we attempted doing the same analysis for redshift - matched type-1 and type-2 spiral hosts , but given the extremely small statistics it was not possible . we can only hope that future data releases will permit us to do this final , but very important test . this motivated us to use the alternative star - formation indicator as in section [ sec : altsfr ] . as an additional test we also visually classified the de vaucoleurs - buta stage in the de vaucouleurs revised hubble - sandage classification system of the type-1 and type-2 agn in the refined samples to see if any insight could be gained about the relative age of the stellar populations without seeing difference . when matching our agns with the iptf sne , some biases may be introduced which could affect the results . two sources of bias are considered : * on 2014 june 17 , automatic filtering was introduced in the iptf sn scanning software in order to save known agns as so called _ nuclear _ objects before a human scanner could begin vetting the candidates . the agn identification done from 2014 june and onwards was based on sdss dr10 data . in 2015 february , qsos from sdss dr 12 were added ( yi cao , personal communication ) . * the match radius used to tie changes in brightness to a certain transient is @xmath74 . this means that if the angular separation between two transient sources exceeds @xmath74 , they are considered to be different sources . this could lead to a potential skewness in sn detections near agns , arising from confusing agns with sne if they reside @xmath74 or less from each other . we avoid the first bias , throughout this work , by only considering sne discovered _ before _ 2014 june 17 . however , we need to perform a test for the second potential bias . the impact of the second bias can be weakened by only considering sne appearing at angular distances from agn @xmath75 . this is generously larger than what is called for by the @xmath74 matching condition . we will herein present three different tests that probe the two biases . the results support a coherent targeting of sne around type-1s and type-2s . the observed differences in sne counts presented is thus a physical effect . evaluating the detection efficiency and completeness of a sn survey is a complex task @xcite . the untargeted iptf sn search is concentrated on finding sne in the nearby ( @xmath15 ) universe . the lack of completeness can be evaluated by repeating our sn matching to the our galaxy samples using the extra condition e.g. @xmath76 . applying this condition when matching gives 1 sn around 11632 agn type-1s ( detection fraction @xmath0 8.6 @xmath57 10@xmath46 ) and 50 sne around 77708 agn type-2s ( detection fraction @xmath0 6.4 @xmath57 10@xmath47 ) for @xmath77 kpc and the largest host samples . the estimated size of effect is @xmath0 4.1@xmath1 in this case . this indicates that the iptf catalogue of sne used by us is sufficiently complete for this investigation . we remind that all 2190 sne used in our study are spectroscopically classified ( sect . [ sec : iptf ] ) . since not all sn candidates found by iptf with the p48 telescope were eventually spectroscopically classified , we have to examine whether a bias against classification of sne residing in either agn type 1 or type 2 hosts was somehow introduced when sn classification targets were selected . such bias , if present , would adversely affect the usefulness of our sn sample . to examine if there is a bias , we manually vetted all transients found during our 2009 - 2014 period in vicinity ( on the plane of the sky ) of our agn sample galaxies . the search radius around each galaxy was set as the apparent size of 100 kpc at the distance of each galaxy . we noted that , around both types of agns , about 1/3 of the likely sn candidates found were eventually classified spectroscopically . the similarity in fraction of sn candidates spectroscopically classified in and around our agn type-1 and type-2 hosts , respectively , shows that our sn sample is suitable for our study . adding the harsh condition of at least @xmath79 separation when matching our 2190 sne gives in total 1 sn around 11632 type-1s ( detection fraction @xmath0 8.6 @xmath57 10@xmath46 ) and 46 sne around 77708 type-2s ( detection fraction @xmath0 5.9 @xmath57 10@xmath47 ) for @xmath77 kpc and the largest host samples . the estimated size of effect is @xmath0 4.1@xmath1 in this case . visual inspection of iptf discovery images of sne found @xmath80 from an agn show it is generally easy for a scanner to unambiguously tell if a sn candidate is separated from the central region of a galaxy . the demonstration that even such a conservative limit as @xmath78 can maintain a @xmath0 4.1@xmath1 effect suggests that we are seeing a physical effect and not a bias effect . we can also explore possible detection effects by looking at the number of foreground and background sne . if no detection probability effects are at play , the number of background / foreground sne should be the same near type-1 and type-2 agn . we set the redshift criterion @xmath43 @xmath22 0.07 and search for sne within 100 kpc of projected distance on the sky around our redshift - matched samples . we find no sne in apparent vicinity of any agn , in the foreground or background , within projected distance @xmath81 kpc . no significant difference in sne counts are found around the largest host samples either . another way of exploring whether or not there is a different detection rate of sne near type-1s and type-2s , is by counting the number of sne at large projected separations . at sufficiently large separations one expects the agn neither influence their surroundings nor the surroundings to influence the agn significantly . on the other hand , some recent articles ( e.g. * ? ? ? * ) suggest unobscured ( presumably type-1s ) agn reside in less dense large - scale ( mpcs ) environments than obscured ( presumably type-2s ) agn . such an effect would influence the stellar mass and also the sn rate that is expected to be higher in the obscured agn case . we use the @xmath83 0.003 but now select sne within @xmath82 kpc , where effects are expected to disappear . for redshift - matched samples of 10146 type-1 and type-2 agn , we see that for type-1 agn we find 22 sne ( detection fraction @xmath0 2.2 @xmath57 10@xmath84 ) , and for type-2 agn we find 15 sne ( detection fraction @xmath0 1.5 @xmath57 10@xmath84 ) , yielding no significant difference in the large - scale environment . for none of the samples we find significant differences in the detected sne rate for type-1s and type-2s in an annulus of @xmath82 kpc , supporting our claim that the detection of sne near type-1s and type-2s in iptf is not biased in favour of any of the two agn types . a statistical analysis of the results can be done in several different ways . we can get an estimate of the size of effect without assuming anything about the underlying distributions ( section [ sec : unc ] ) . a detection signal from estimating at what level the reported difference between the two samples is just by chance can be obtained by assuming the two agn samples have equal sn rates as the null hypothesis ( section [ sec : nulldist ] ) . but , ultimately hypothesis testing should be used to estimate the probability @xmath6 for the hypothesis to be true and tell us whether or not we can reject the null hypothesis for the nominal value @xmath12=0.05 ( section [ sec : pairwise ] ) . let the random variables @xmath85 and @xmath86 denote the number of sne in galaxy sample no . @xmath87 ( type-1 agn ) and @xmath88 ( type-2 agn ) , respectively , and let @xmath89 , be the total number of galaxies observed in sample no . @xmath90 . here , the parameter @xmath91 denotes the average rate of sne per galaxy in the type-@xmath92 agns , while @xmath93 denotes the number count of type-@xmath92 agns in the observed volume . the random variable @xmath94 then describes the difference between the rates of sne in the two samples . in order to put the observed difference in context , we define the ratio @xmath95 which is a measure of the expected size of the difference in units of the standard deviation of the distribution of @xmath96 . note that , in general , @xmath96 is not normally distributed , wherefore a detection at a level of , say , @xmath97 can not directly be compared with a @xmath97 detection where the distributions of interest are normal . in our case , however , the difference should be relatively small . + since the difference in eq . ( [ eq.z ] ) involves ratios of two random variables , the computation of the expectation and variance of @xmath96 is not straightforward and we will use gauss s approximation formulae ( first order ) to derive an explicit expression of eq . ( [ eq.etos ] ) . we have that @xmath98 where it is assumed that @xmath99 and @xmath100 are independent . now , @xmath101 by the fact that @xmath102 , the law of total expectation gives that @xmath103 . similarly , for the variance we have that @xmath104 while the covariance also amounts to @xmath105 . hence , we have that @xmath106 for the expectation of @xmath96 , we have that @xmath107 . it is noted that also for the second order approximation , @xmath108 thus , we have @xmath109 now , a plug - in estimate of the ratio in eq . ( [ eq.ratio ] ) is given by the expression @xmath110 where @xmath111 are estimated from the observed numbers . for example , for the full samples and @xmath112 kpc , we have that @xmath113 , and @xmath114 ( see table [ largest ] ) . thus , we obtain @xmath115 i.e. , the observed difference is more than @xmath116 away from @xmath117 , as defined by eq . ( [ eq.etos ] ) . + since the sn rates generally are very small , i.e. , @xmath118 , the factor @xmath119 in the expression for @xmath120 and we may take @xmath121 and @xmath122 as fixed . then , @xmath96 reduces to @xmath123 and @xmath124 where @xmath125 is the total number of sne observed in the type-@xmath92 agns . for @xmath126 , and @xmath127 , we have @xmath128 , as the result in for the hypothesis testing discussed in sect . 1.7.3 , we make the assumption that @xmath129 and @xmath130 are fixed . + let s hypothesise that the sn rates of the two agn samples are equal . at what level will the observed difference then just be due to chance ? by taking the null hypothesis to be + @xmath131 versus + @xmath132 , + where @xmath133 is the common sn rate , we have instead that @xmath134 and @xmath135 . consequently , the variance of the random variable @xmath136 becomes @xmath137 the expression of the ratio @xmath29 , defined as @xmath138 where @xmath139 denotes the observed difference @xmath140 , then equals @xmath141 following the steps is sect . [ sec : unc ] , we have that @xmath142 where an estimate of @xmath133 is given by @xmath143 hence , for the full , unmatched samples ( @xmath112 kpc ) we have that @xmath144 and @xmath145 . this is a more conservative measure of detection , i.e. a difference in the sn rates is detected close to the @xmath146-level . + finally , we have that @xmath147 under the assumption that @xmath148 . it is noted that for this approximation , the estimates of @xmath29 as given by eq . ( [ eq.red ] ) and eq . ( [ eq.red2 ] ) are equal for @xmath149 . + we report this estimate of the signal of detection alongside with the size of effect and p - value in the captions to table [ supernovaall ] . the number of sne in the host galaxies of agn should , within the realm of the unification model , not depend on the type of agn . we have performed a statistical test of this hypothesis . recall that the sn rate ( read proportion / success probability ) @xmath5 in a host galaxy population of size @xmath150 should be binomially distributed . however , since @xmath151 in all observed cases , @xmath152 to a very high degree . therefore , we assume that the number of sne believed to be associated with the type-1 and type-2 agn samples are , respectively , observations of the random variables @xmath153 and @xmath154 . here , @xmath155 is the size of the type-@xmath92 agn sample and @xmath91 is the corresponding rate of sne in the type-@xmath92 agn population , uniformly corrected for the biases discussed above . furthermore , it is assumed that @xmath156 and @xmath157 are independent . let the null hypothesis be + @xmath158 , against the alternative + @xmath159 . + the exact conditional test ( c - test ) for comparing two poisson means by przyborowski & wilenski ( 1940 ) is known to be overly conservative , i.e. , the chance of failing to reject a false null hypothesis is higher than the nominal level . we have therefore chosen to perform a test based on estimated p - values instead ( krishnamoorthy & thomson , 2004 ) . the pivot statistics for @xmath117 is given by @xmath160 where @xmath161 is the unbiased variance estimator of the standardised difference @xmath162 . the p - value for the two - sided test is then given by @xmath163 where @xmath164 is the observed value of @xmath165 and @xmath166 and @xmath167 are the observed numbers of sne in the type-1 and type-2 sample , respectively . hence , the p - value is estimated by the expression ( see krishnamoorthy & thomson , 2004 ) @xmath168 where @xmath169 denotes the indicator function such that @xmath170 and @xmath171 is the estimate of @xmath172 ( see eq . 17 ) . note that under the null hypothesis , @xmath173 . also , we have that @xmath174 . the null hypothesis is then rejected at the significance level @xmath12 if @xmath175 . we use the nominal value @xmath176 . we results from the hypothesis test are reported in the table [ supernovaall ] . it might be interesting to speculate about the origin of the obscuration . it seems fairly natural to assume that a larger number of massive , young stars in the galaxy ( dying as core - collapse sne ) leads to a larger production of dust . the first suspicion of that galactic - scale dust is responsible for some of the obscuration in type-2 agn came from the discovery that type-1 agn rarely are found in edge - on systems @xcite . later , it was found that seyfert-2 have more dust lanes and dust patches near their nuclei than do seyfert-1 agn @xcite , in contradiction with the simplest unification where all obscuration is caused by the torus the doughnut - like dust structure that surrounds the agn engine , no larger than a few hundred parsecs . the column density of galactic dust outside the most central 500 pc was estimated to @xmath177 @xmath178 @xmath179 , close to the frequently used limit in column density to separate between obscured and unobscured agn ( unobscured @xmath180 @xmath179 ) . recent works support the idea that the host galaxy has enough dust @xcite to cause obscuration of the broad - line region . dipompeo et al . ( 2016 ) estimate the fraction of ir - selected `` obscured '' agn that are obscured by dust outside the torus to @xmath0 25% , while buchner et al . ( 2016 ) estimate that 40 % of all agn have considerable host galaxy obscuration . but objects with large measured column densities @xmath181 @xmath179 can not be explained with host galaxy obscuration and need heavy obscuration around the nucleus . we include a cartoon of an agn , figure [ unification ] , that shows all components needed to describe the agn unification . if we compare the @xmath4[]5007-matched type-1 galaxies ( @xmath64=6252 ) and type-2 galaxies ( @xmath64=6252 ) used for sn counts , we can get a rough estimate of the fraction of agn that are obscured only by the torus using the nave assumption that there is a one - to - one correspondence in expected properties between the objects within each @xmath4[]5007-matched pair of host galaxies , i.e. same galaxy zoo morphology , and with measured exponential fit scale radius & @xmath30 within 20% error . the fraction of objects fulfilling these criteria shows that 10% can be explained purely by ( parsec - size ) torus obscuration and agree with the predictions from the simplest unification about identical host galaxy properties . ( using the @xmath4[]5007-matched refined samples based on spiral face - on hosts , the corresponding nave fraction is 25% . ) this also means that for majority of type-2 agn , up to 90% , the obscuration must stem from larger scales : host galaxy obscuration and/or large - scale environment in which the host galaxies reside . the finding that @xmath30-matched type-1 agn are significantly more luminous in [ ] 5007 than the type-2 agn in host - matched samples , asks for a physical connection between the agn classes that goes beyond pure obscuration . perhaps the two important differences between type-1 and type-2 agn that we find on a population level the different average agn luminosity and the different stellar ages of the host galaxies can be explained within the framework of an evolutionary scenario @xcite where the type-2 agn are those with obscured by heavy star formation during the initial agn phase after the merger , where the agn becomes stronger and stronger until the `` blow - out phase '' is reached , later leading to a strong , naked `` unobscured '' agn with less dust and gas . the agn environment also supports this scenario through differences in neighbour counts @xcite and neighbour properties @xcite . the torus could be a heavily star - forming molecular disk at @xmath182 pc where the sne keep up the geometrical thickness of the dust disk @xcite and is more likely to be heavily star - forming in gas - rich galaxies with younger stellar populations . the presence of starbursts @xcite and past starbursts @xcite in agn in the most central regions support the view of heavy star formation in the most central 100 pc . but the observed difference in luminosity between the @xmath30-matched seyfert-1 and seyfert-2 galaxies perhaps means we are dealing with a receding molecular disk . continued theoretical efforts into modeling the torus are therefore needed . using sne from the iptf and galaxies from the sdss , we have searched for possible physical differences between type-1 and type-2 agn . simultaneously , we have also carefully matched type-1 and type-2 agn host galaxies to gain information on the agn engine . we find that : 1 . the two agn classes differ in term of sn counts in their hosts . agn hosts have a larger number of sne . this differences appears to originate in a difference between stellar ages ( and not only stellar masses ) and more recent star formation in type-2 agn hosts . star - forming galaxies have more recent star formation than type-2 agn host . 2 . based on equal [ ] 5007 line widths in type-1 and type-2 agn ( significantly wider than in star - forming hosts ) , there are good reasons to accept an isotropy in the [ ] 5007 distribution as long assumed by the simplest unification . type-1 and type-2 agn hosts that are dominated by star formation @xmath2 and are matched in cold dust emission have strong differences in nlr luminosity @xmath4[]5007 type-1 agn being 10 times more luminous . summarizing , we identify two more factors beyond the viewing angle agn luminosity and the age of stellar populations making unification not only a question of obscuration but also one of generation . finally , we conclude the great potential of iptf and surveys like the upcoming zwicky transient facility ( ztf ) , for resolving the question with larger sn samples . the ztf @xcite will use the same telescope as the iptf , but with a larger field of view , and is expected to start in 2017 . [ largest ] [ supernovaall ] b.v . wishes to thank kjell lundgren for early discussions about the work . she also wishes to thank martin gaskell and robert r.j . antonucci and for many constructive and helpful comments , and finally acknowledge a. magnard , c. franck , g. von rivia and j. pianist for inspiring discussions . wishes to thank ariel goobar for useful suggestions . b.v . was funded and supported by the center of interdisciplinary mathematics ( uppsala universitet ) and erik and mrta holmbergs donation from the kungliga fysiografiska sllskapet . supernova research at the oskar klein centre is supported by the swedish research council and by the knut and alice wallenberg foundation . the intermediate palomar transient factory project is a scientific collaboration among the california institute of technology , los alamos national laboratory , the university of wisconsin , milwaukee , the oskar klein centre , the weizmann institute of science , the tango program of the university system of taiwan , and the kavli institute for the physics and mathematics of the universe . this publication makes use of data products from the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration . this research also heavily relies on the sloan digital sky survey ( sdss ) . funding for 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 japanese monbukagakusho , and the max planck society . davies r.i , maciejewski w. , hicks e.k.s , emsellem e. , erwin p. , burtscher l. , dumas g. , lin m. , malkan m.a . , mller - snchez f. , orban de xivry g. , rosario d.j . , schnorr - mller a. & tran a. , 2014 , the astrophysical journal , 792 , 101 hakobyan a.a . , nazaryan , t.a . , adibekyan v. zh . , petrosian a. r. , aramyan l. s. , kunth d. , mamon g.a . , de lapparent v. , bertin e. , gomes j.m & turatto m. , 2014 , monthly notices of the royal astronomical society , 444 , 2428 willett , k.w . , lintott , c.j . , bamford , s.p . , masters , k.l . , simmons , b.d . , casteels , k.r.v . , edmondson , e.m . , fortson , l.f . , kaviraj , s. , keel , w.c . , melvin , t. , nichol , r.c . , raddick , m.j . , schawinski , k. , simpson , r.j . skibba , r.a . , smith , a.m. , thomas , d. , 2013 , monthly notices of the royal astronomical society , 435 , 2835 zakamska , n.l , hamann , f. , paris , i. , brandt , w.n . , greene , j.e . , strauss , m.a . , villforth , c. , wylezalek , d. , alexandroff , r.m . & ross , n.p . , monthly notices of the royal astronomical society , 2016 , 459 , 3144

the spectral energy distribution of the dark cloud ldn 1622 , as measured by finkbeiner using _ wmap _ data , drops above 30 ghz and is suggestive of a boltzmann cutoff in grain rotation frequencies , characteristic of spinning dust emission . ldn 1622 is conspicuous in the 31 ghz image we obtained with the cosmic background imager , which is the first cm - wave resolved image of a dark cloud . the 31 ghz emission follows the emission traced by the four _ iras _ bands . the normalised cross - correlation of the 31 ghz image with the _ iras _ images is higher by 6.6 for the 12 m and 25 m bands than for the 60 m and 100 m bands : and . the mid - ir cm - wave correlation in ldn 1622 is evidence for very small grain ( vsg ) or continuum emission at 2636 ghz from a hot molecular phase . in dark clouds and their photon - dominated regions ( pdrs ) the 12 m and 25 m emission is attributed to stochastic heating of the vsgs . the mid - ir and cm - wave dust emissions arise in a limb - brightened shell coincident with the pdr of ldn 1622 , where the incident uv radiation from the ori ob 1b association heats and charges the grains , as required for spinning dust .

astro-ph0511283

an increasing amount of evidence supports the existence of a new continuum emission mechanism in the diffuse interstellar medium ( ism ) at 1030 ghz , other than free - free , synchrotron , or an hypothetical rayleigh - jeans tail of cold dust grains @xcite . examples of excess emission at cm - wavelengths over known emission mechanisms have been found in the spectral energy distributions ( seds ) of the dark cloud ldn 1622 and the diffuse hii region lph 201.7 + 1.6 @xcite , in the helix planetary nebula @xcite , and in another diffuse hii region in perseus @xcite . at the date of writing , the only morphological evidence for the existence of a new emission mechanism at cm - wavelengths in a specific object is provided by the helix nebula . but a comparative analysis of the cm - wave , mid- and far - ir continua in the helix is hampered by strong line contamination in the short wavelength _ iras _ maps . the cosmic background imager ( cbi ) observations of ldn 1622 provide an opportunity of performing such morphological analysis . as modelled by @xcite a possible candidate mechanism is electric dipole emission from spinning very small grains ( vsgs ) , or ` spinning dust ' . the spectral energy distribution of the dark cloud ldn 1622 ( lynds dark nebula , * ? ? ? * ) is suggestive of spinning dust : it rises over 59.75 ghz @xcite , following dipole emission , and then drops above 30 ghz @xcite , as expected from a boltzmann cutoff in the grain rotation frequencies . the dark cloud ldn 1622 lies within the orion east molecular cloud @xcite , at a distance of @xmath4120 pc @xcite and in the foreground of the orion b cloud . its far - ir linear size is slightly less than 1 pc . it is a conspicuous cs@xmath5 and n@xmath6h@xmath7 `` starless '' ] core ( * ? ? ? * with an h@xmath6 density of @xmath8 @xmath9 ) . ldn 1622 is devoid of hii regions , aside from barnard s loop ( e.g. * ? ? ? * ) , a very diffuse hii region ( * ? ? ? * with electron density of 2 @xmath9 ) separated by @xmath41 deg from ldn 1622 . no free - free emission is expected from ldn 1622 , which is indeed absent from the parkes - mit - nrao survey at 5 ghz noise level in the pmn survey is 5 mjy beam@xmath10 , the free - free emission measure towards ldn 1622 must be less than 10 pc @xmath11 , which for a spherical nebula 10 arcmin in diameter implies electron densities of less than 10 @xmath9 ] ( hereafter pmn survey , * ? ? ? * as presented in _ skyview _ , http://skyview.gsfc.nasa.gov ) . only the h@xmath13 corona of ldn 1622 , outlining its photon - dominated region , is marginally detected in the pmn survey . here we present the first cm - wave continuum image of a dark cloud , and report morphological evidence that supports spinning dust as the mechanism responsible for the anomalous foreground . we first describe data acquisition ( section [ sec : obs ] ) , and image reconstruction ( sec . [ sec : syn ] ) , and then discuss the effects of ground spill over and give flux estimates ( sec . [ sec : ground ] ) . we analyse the 31 ghz data by comparison with the _ iras _ bands used as templates for the emission by cool dust and by vsgs ( or hot dust , sec . [ sec : xcorr ] ) , which leads us to infer a limb - brightened morphology of ldn 1622 at 31 ghz . the comparison with h@xmath13 and 5 ghz templates shows that any free - free contribution at 31 ghz is negligible , and that the 31 ghz emission is interior to the h@xmath13 corona of the cloud ( sec . [ sec : ha ] ) . we discuss the spectral energy distribution of ldn 1622 ( sec . [ sec : spec ] ) , and finally summarise our results ( sec . [ sec : conc ] ) . the cbi @xcite is a planar interferometer array with 13 antennas , each 0.9 m in diameter , mounted on a 6 m tracking platform , which rotates in parallactic angle to provide uniform @xmath14-coverage . the cbi receivers operate in 10 frequency channels , with 1 ghz bandwidth each , giving a total bandwidth of 2636 ghz . it is located in llano de chajnantor , atacama , chile . we observed ldn 1622 ( j2000 ra : 05:54:23.0 , dec : + 01:46:54 ) on 03-nov-2003 , 02-dec-2003 and 21-nov-2004 , for a total 10000 s. the compact configuration of the cbi interferometer results in the @xmath15 coverage shown on fig . [ fig : uvcov ] , where it can be seen that baseline length varies between 100 @xmath16 and 400 @xmath16 , corresponding to spatial scales of 34.4 arcmin and 8.6 arcmin , respectively . each receiver is equipped with phase shifters which allow selecting its polarization mode . we set all receivers to l polarization , so that the visibilities are sensitive to the combination of stokes parameters @xmath17 . in what follows we assume that stokes @xmath18 ( circular polarization ) is negligible in ldn 1622 . approximate cancellation of ground and moon contamination was obtained by differencing with a reference field at the same declination but offset in hour angle by the duration of the on - source integration . we used an on - source integration time of 8 min , with a trailing reference field . for phase calibration purposes we interspersed a 2 min integration on j0607085 between each 16 min cycle of differenced observations . j0607085 was observed with identical telescope settings as ldn 1622 . the data were reduced and edited using a special - purpose package ( cbical , developed by t.j . pearson ) . flux calibration was performed using either saturn or tau a , whose fluxes are in turn calibrated against jupiter ( with a temperature of 146.6 k , * ? ? ? * ) . the flux calibrator is also used as the reference for an initial phase calibration . the phase calibration was subsequently refined by using the calibrator interspersed between each cycle on ldn 1622 . we applied a phase shift to bring j0607085 to the phase center . the magnitude of the offsets by which we had to correct the position of j0607085 varied between 15 and 40 arcsec . in a final stage we combined all available visibilities of ldn 1622 to produce two final datasets , with and without reference field subtraction . since the angular distance of ldn 1622 from the moon was larger than 80 deg for all three nights of observations , the contamination on the shorter baselines is probably entirely due to ground spill over . image reconstruction is difficult for an object such as ldn 1622 , which extends to about half the cbi s primary beam of 45 arcmin fwhm , and is surrounded by diffuse emission . additionally the cbi s synthesized beam obtained with natural weights , @xmath19 arcmin fwhm , is about the size of the object , @xmath20 arcmin . thus in order to perform a morphological analysis we need to extract a finer resolution from the visibilities than that obtained from the restored images . the maximum entropy method ( mem ) fits model images to visibility data . the mem models can potentially recover details on finer angular scales than the synthesized beam . in this section we present the results of our reconstructions . the algorithm and model validation are described in appendices [ sec : memalgo ] and [ sec : modval ] in figure [ fig : cbi]a we present a mem model of our data . the noise of the restored image in fig . [ fig : cbi]b is close to that expected from the instrumental noise . the difmap package @xcite estimates a theoretical noise in the dirty map ( using natural weights ) of 3.2 mjy / beam , which should give 3 @xmath0 deviations of about 10 mjy beam@xmath10 for an optimal reconstruction . the dirty map of the residual visibilities , obtained with difmap using natural weights , has a minimum of @xmath2111 mjy / beam within the half - power contour of the primary beam , consistent with the theoretical noise , at j2000 equatorial coordinates ( 89.15 deg,+2.10 deg ) , which we identify as a 170 mjy point source ( pmn j0604 + 0205 ) offset by 38 arcmin from the phase center in the reference field . this negative point source in the restored image is at @xmath40.5 deg from the phase centre , and is outside the region of interest ] . we also obtained `` clean '' images with difmap , which qualitatively confirm the mem models . we show an overlay of the mem model on a `` clean '' restoration in fig . [ fig : clean]a , obtained with the difmap package and uniform weights . we anticipate from sec . [ sec : xcorr ] the good match between 31 ghz and 12 @xmath1 m emission to test which of the two reconstructions , whether mem or `` clean '' , extracts the most of the data . [ fig : clean]b also shows an overlay of the 31 ghz contours on the _ iras _ 12 @xmath1 m map in grey scale . the _ iras _ 12 @xmath1 m image is from the iras sky survey atlas @xcite , as obtained in _ skyview_. it can be appreciated by inspection of fig . [ fig : clean ] that the mem model recovers low - level details that are absent in the `` clean '' image , such as the 12 @xmath1 m emission peaks at @xmath22 and @xmath23 . there are two features in the mem model that do not seem to have a 12 @xmath1 m counterpart . one is a low - level contour at @xmath24 , which turns out to be the location of the brightest radio point source in the field ( see sec . [ sec : ha ] and fig . [ fig : shassa ] ) . the other is a 31 ghz peak at @xmath25 , which matches an h@xmath13 feature at the outskirts of barnard s loop ( see fig . [ fig : shassa ] ) . we will use mem in what follows because it provides reconstructions that do not depend on user - defined `` clean '' boxes , and because it allows extracting details on fine angular scales while preserving the sensitivity of the dataset . the cbi image can be compared with that available in the first _ wmap _ data release . barnard s loop is the most conspicuous feature in the _ wmap _ ka - band image of the region . but it is apparent that the cbi data on ldn 1622 are much more sensitive , and allow resolving the dark cloud . the cbi image is thus the first at cm - wavelength of a dark nebula , i.e. a cold dust cloud identified by visible - light stellar counts . in order to cross - correlate the cbi data with the comparison templates we compute template visibilities , obtained by a simulation of cbi observations on the template images ( `` cbi - simulated visibilities '' herafter , see appendix [ sec : memalgo ] and [ sec : modval ] ) . the 31 ghz-100 @xmath1 m visibility plot on fig . [ fig : xcorr]a for the undifferenced dataset allows assessing the level of ground and moon contamination in the shorter baselines . the enhanced scatter above 100 @xmath1 m visibilities of @xmath26jy and at @xmath27jy is suppressed in the differenced dataset shown on fig . [ fig : xcorr]b . the enhanced scatter due to ground or moon contamination in the shorter baselines correspond to where the real parts of @xmath28 reach about 300 jy and where the imaginary parts of @xmath28 reach @xmath29 jy . restricting to baselines above 120 @xmath16 retains visibilities devoid of ground contamination , as shown on fig . [ fig : xcorr_uvrange ] . typical ism power spectra are decreasing power - laws@xcite , the ensemble - averaged modulus of the visibility is thus a monotonic function of @xmath14-radius . this is also true for ldn 1622 , for the case of the 100 @xmath1 m cbi - simulated visibilities : the azimuthally averaged power - spectrum is monotonic . we can not recover a power spectrum for the cbi data by simple averaging because the signal is affected by noise , so that the derived spectrum is artificially flat . the 31 ghz flux density measured on the restored image within a circular aperture with 45 arcmin diameter centered on ldn 1622 is [email protected] jy . we caution that the cbi images are heavily affected by flux loss for emission on 45 arcmin scales : because of incomplete sampling in the @xmath14 plane the reconstructed images have missing spatial frequencies , and part of the extended nebular emission is lost . we can infer a flux density corrected for flux loss , of @xmath31 jy , by referring to a template map , for which we use _ iras _ 100@xmath1 m . this flux density is estimated by extracting the flux density in the template map within a 45 arcmin aperture , and scaling by the cbi-_iras _ 100@xmath1 m correlation slope given in table [ table : corr ] for the differenced dataset . template maps which follow closely the 31 ghz emission also allow a cross - check on the pointing accuracy of the cbi . we vary an @xmath32 shift on the coordinates of the reference pixel of the template maps to minimize @xmath33 , where the uncertainties @xmath34 only contain the cbi noise . the optimal shifts we find for each of the _ iras _ maps are ( in arcmin ) : ( @xmath35 , @xmath36 ) , ( @xmath37 , @xmath38 ) , ( @xmath39 , @xmath40 ) , ( @xmath41 , @xmath42 ) for the 12 @xmath1 m , 25 @xmath1 m , 60 @xmath1 m , and 100 @xmath1 m maps , respectively . thus no particular trend is found , although the average values of the shifts is @xmath43 arcmin and significantly different from zero . but in what follows we ignore a possible residual error in telescope pointing because the overlays of the 31 ghz and far - ir images in fig . [ fig : templates ] show a good match , and would not improve by shifting on 0.5 arcmin scales . if dust is reponsible for the 31 ghz emission in ldn 1622 then a tight relationship is expected with the mid- and far - ir emission . here we investigate the consequences of assuming that the emission traced by the cbi scales linearly with the four _ iras _ maps . the infrared emission from dust is discussed in details by , e.g , @xcite . the _ iras _ 100@xmath1 m band traces emission from large grains , with sizes greater than 0.01 @xmath1 m . the large grains are in equilibrium with the interstellar radiation field , with a temperature of order 1020 k depending on environment . continuum emission at shorter wavelengths is due to hot dust , at @xmath4100 k , which is too hot to be maintained in equilibrium with the interstellar uv field . mid - ir emission from classical hot dust is not expected because of the absence of a strong uv source within ldn 1622 , in contrast with compact hii regions or planetary nebulae . thus stochastic heating of vsgs dominates the dust emission in the _ iras _ 12@xmath1 m and 25 @xmath1 m bands . the heat capacity of a vsg is so small that the absorption of a single uv photon increases the particle temperature high enough for it to emit at @xmath44 m . thus , by examing the degree of correlation with the 4 _ iras _ bands , we hope to determine which type of grain , whether the large grains or the vsgs , are responsible for the 31 ghz emission . we caution that from the _ iras _ photometry alone we can not differentiate a 31 ghz link to the vsgs from a link to a hot molecular phase that shines in the h@xmath6 lines . after all the vsgs can also be regarded as large molecules , such as polycyclic aromatic hydrocarbons ( pahs ) . another important source of flux in the _ iras _ 12@xmath1 m and 25@xmath1 m bands are the h@xmath6 ro - vibrational lines , such as h@xmath6(00)s(2 ) 12.3@xmath1 m , and h@xmath6(00)s(0 ) 28.2@xmath1 m . the h@xmath6 line fluxes integrated over the _ iras _ band passes could account for part of the mid- and far - ir morphological differences ( as could be the case in pdrs , with conspicuous h@xmath6 lines , * ? ? ? the 31 ghz mem contours can be compared by inspection with the raw _ iras _ images , as extracted from _ skyview_. the diffuse emission in the mid - ir images is closer to the 31 ghz contours than the far - ir images . in this section we quantify this qualitative result , and show it is not affected by noise or missing spatial frequencies at 31 ghz . is the cm - wave mid - ir correlation detectable directly in the visibility data ? the cross - correlations may be different in the image plane and in the @xmath14 plane because of two reasons . one is the contribution of point sources at 12 @xmath1 m , which are absent at 31 ghz . the fainter point sources at 12 @xmath1 m may be numerous and act as diffuse emissionm image of ldn 1622 in fig . [ fig : isocam ] ] . the subtraction of the brightest point sources may not be accurate enough to retain genuinely diffuse emission at 12 @xmath1 m . another difficulty with a @xmath14-plane analysis of the diffuse emission are the uncertainties in the cbi primary beam ( * ? ? ? * their fig . 1 ) . variations between antennas introduce uncertainties beyond about 40 arcmin from the phase center . ldn 1622 is surrounded by diffuse emission , such as that traced by _ iras_. the outskirts of barnard s loop are within 35 arcmin from the phase center , and it peaks at about 60 arcmin . also in the neighborhood of ldn 1622 is the reflection nebula ngc 2067 , which at 1.9 - 2.7 deg from the phase center falls on a sidelobe of the primary beam at 2.2 deg . barnard s loop or ngc 2067 are bound to enter the side lobes and low level wings of the primary beam , where the uncertainties in the primary beam model used in cbi - simulated visibilities become important . we linearly correlate the cbi visibilities with the cbi - simulated visibilities on the four _ iras _ bands , one template at a time , and after processing as described in appendix [ sec : modval ] . table [ table : corr ] lists reduced-@xmath45 , linear correlation coefficients ( as defined in * ? ? ? * ) , correlation slopes and uncertainties . in table [ table : corr ] we also consider the non - differenced cbi dataset , because its signal to noise ratio ( s / n ) improves by a factor @xmath46 . we restrict the analysis of the non - differenced dataset to @xmath14-radii above 120 @xmath16 . such a baseline range allows minimizing ground spill over or moon contamination . another reason for restricting baseline lengths above a minimum is that a constant background in the template maps affects the simulated visibilities for the shortest baselines . we minimize this effect , which is due to the restricted sky domain available to compute the simulated visibilities , by clipping the templates so that their minimum intensity value is zero . lllrr + @xmath47 : & 1.07 & 1.06 & 1.06 & 1.06 + @xmath48 : & 0.395 & 0.399 & 0.394 & 0.395 + @xmath49 : & @xmath50 & @xmath51 & @xmath52 & @xmath53 + + @xmath47 : & 1.74 & 1.74 & 1.74 & 1.75 + @xmath48 : & 0.243 & 0.241 & 0.246 & 0.242 + @xmath49 : & @xmath54 & @xmath55 & @xmath56 & @xmath57 + but from the visibility correlations alone we can not ascertain which _ iras _ map correlates best with the 31 ghz data . the significance of the results is difficult to assess because the noise in the comparison maps is not known accurately ( and is neglected in this analysis ) , and the confidence level associated to the @xmath45 distribution with @xmath58 degrees of freedom is extremely sharp at @xmath59 . a drawback of analysing the visibility data directly in the fourier - plane is that the effect of the point sources is difficult to isolate , especially at shorter ir wavelengths , where point sources are more frequent . here we compare the cbi data and the _ iras _ templates in the image plane , based on our mem modelling . inspection of fig . [ fig : templates ] reveals that the 12 @xmath1 m and 25 @xmath1 m mem maps are the most similar to the 31 ghz mem model . in fig . [ fig : templates ] the _ iras _ maps are the same as in fig . [ fig : irasmem ] , i.e. they are reconstructed from simulated cbi visibilities , following the algorithm described in appendix [ sec : modval ] . thus the mid - ir cm - wave correlation is not the effect of missing spatial frequencies in the 31 ghz visibility data . the 100 @xmath1 m emission is concentrated in a single maximum , while the 31 ghz , 12 @xmath1 m and 25 @xmath1 m images show two peaks near the phase center , at the origin of coordinates in fig . [ fig : templates ] , which we refer to as the northern and southern peaks . the 60 @xmath1 m image is also double - peaked , but the southern peak is offset relative to the 31 ghz southern peak . the 31 ghz morphology of ldn 1622 is remarkably similar to that in the 12 @xmath1 m _ iras _ band . however there is an interesting feature at 12 @xmath1 m which is absent at 31 ghz . the northen peaks at 31 ghz and 12 @xmath1 m are slightly offset , while the southern peaks are exactly coincident . we explain the shift in the position of the northen peak as being due to a young stellar object ( yso ) , namely l1622 - 10 , whose emission contributes at 12 @xmath1 m but not at 31 ghz . thus the 31 ghz emission is genuinely diffuse , while 12 @xmath1 m includes photospheric emission , or unresolved very hot dust . the point - source flux for l1622 - 10 , as listed in the _ iras _ point source catalog , is subtracted from the processed _ iras _ 12@xmath1 m image used as comparison template ( see appendix [ sec : modval ] ) . but the yso is still present in the processed image , even after subtraction , as can be inferred by comparing the raw _ iras _ 12@xmath1 m image in fig . [ fig : clean ] and the processed image in fig . [ fig : templates ] . the imperfect subtraction of l1622 - 10 is probably due to an inaccurate catalog flux , perhaps due to the uncertainties inherent in deriving a flux density for a point source on top of a compact source , such as the northern peak . a discussion on the properties of this yso is given in appendix [ sec : midirps ] . in order to quantify the similarities that meet the eye when comparing the 31 ghz and the _ iras _ templates , we compute the normalised cross - correlation @xmath60 of the 31 ghz mem images with the _ iras _ models , one at a time : @xmath61 where the sums extend over all pixels in the model images . we can estimate the significance of the results by calculating the scatter of the cross - correlation @xmath60 for each of 90 different realisations of noise on the template visibilities ( see appendix [ sec : modval ] ) . we attempted to assign a @xmath45 value to the comparison between the cbi and the _ iras _ models . but the pixels in the model images are correlated , and the covariance matrix is prohibitively large , with @xmath62 elements , in the case of @xmath63 free - parameters per mem model . the tests we ran to estimate the covariance matrix from the simulations described in sec . [ sec : syn ] showed we need many more than only 90 different noise realisations . we reached a suitable accuracy on the covariance matrix only in the useless case of a model image with @xmath410 free parameters . table [ table : skyxcorr ] lists the cross - correlation results , which are also summarised in fig . [ fig : skyxcorr ] . the weighted average of the 60 and 100 @xmath1 m cross - correlations is worse than that of the 12 @xmath1 m and 25 @xmath1 m cross - correlations by 6.6 @xmath0 . the solid line on fig . [ fig : skyxcorr ] has a slope of @xmath64 , and is thus different from zero at 6.3 @xmath0 . to test the hypothesis that @xmath60 is independent of _ iras _ band we calculate @xmath65 with the data from table [ table : skyxcorr ] , where @xmath66 is the weighted average of the cross - correlations . reduced @xmath45 is 7.7 for 3 degrees of freedom , which discards a constant value of the cross - correlation as a function of _ iras _ wavelength . we also carried out the same simulations but with a different entropy term , @xmath67 ( described in appendix [ sec : memalgo ] ) , obtaining the same results at lower significance . in this case the mid - ir far - ir difference is 3.7 @xmath0 , or 4.4 @xmath0 after subtraction of the yso l1622 - 10 . lrrrr @xmath60 : & 0.782 @xmath30 0.022 & 0.748 @xmath30 0.021 & 0.647 @xmath30 0.015 & 0.624 @xmath30 0.019 + the comparison between the cbi image and the four _ iras _ bands allows us to conclude that the cbi emission is best represented by _ iras _ 12@xmath1 m . the morphology of the mid - ir _ iras _ maps is suggestive of limb - brightening of a vsg - emitting shell coincident with ldn 1622 s pdr , as required by uv excitation of the vsgs . ldn 1622 is a rather peculiar cloud in that its mid - ir emission is limb - brightened . by contrast , ldn 1591 reaches higher 100 @xmath1 m intensities than ldn 1622 , and yet is not detected by @xcite . the facts that ldn 1591 is not limb - brightened and that the 2636 ghz emissivity is enhanced in ldn 1622 lead us to propose that the 2636 ghz emission stems from the photon - dominated region , with abundant uv radiation . this scenario is consistent with `` spinning dust '' , or electric dipole radiation from spinning vsgs exposed to the incident uv radiation and charged by the photo - electric effect . a possible test for this interpretation may derive from the analysis of the cm - wave morphology of other limb - brightened clouds , such as dc300 - 17 in chamaeleon @xcite , which like ldn 1622 also harbors low - mass ysos . could `` magnetic dipole emission '' from large grains , proposed by @xcite , also account for the mid - ir cm - wave correlation ? the fact that the 100 @xmath1 m emission does not trace the 31 ghz double - peaked morphology suggests large grains , with a modified black - body spectrum , do not contribute at 31 ghz . an increased 31 ghz emissivity through a temperature enhancement in the pdr of ldn 1622 would have a concomitant limb - brightened morphology at 100 @xmath1 m . we infer a classical dust temperature map for ldn 1622 , shown on fig . [ fig : tdust ] , from the _ iras _ 60/100 @xmath1 m color map . we adopted a @xmath68 emissivity law , and degraded the two maps to a common resolution guided by the point sources in the field . we solve for the dust temperature using the brent method @xcite , pixel by pixel . it is apparent that the large grain temperature is fairly constant across ldn 1622 , and does not follow the 31 ghz contours . it may be argued that the _ iras _ 12 @xmath1 m band is not well suited to trace vsgs because it is often contaminated by ionic line emission , for instance by [ neii ] 12.8@xmath1 m , which could arise in the pdr at the surface of ldn 1622 . but the similarity of the 12 @xmath1 m and 25 @xmath1 m maps argue against significant line contamination . it would be very contrived to have just the right contribution of flux from lines in both bands ( although the h@xmath6 lines could still contribute to both bands in similar proportions ) . the surface of ldn 1622 is exposed to the interstellar uv field . such ionised corona of ldn 1622 is conspicuous in the shassa image @xcite , shown on fig . [ fig : shassa ] . the v - shaped h@xmath13 corona points towards the orion ob 1b association ( * ? ? ? * their fig we note ldn 1622 corresponds to a minimum in h@xmath13 brightness , therefore it is a foreground object obscuring the diffuse h@xmath13 from the orion - eridanus bubble , consistent with the short distance of @xcite . that the 31 ghz emission is not free - free is apparent from fig . [ fig : shassa ] , where h@xmath13 seems to anti - correlate with the radio continuum , although we did not attempt to correct the h@xmath13 map for extinction . the h@xmath13 and free - free emission both trace electron - ion collisions , so that if the electron temperature is constant , then the unreddenned h@xmath13 intensities are proportional to the radio - continuum specific intensities . but the only correspondence between 31 ghz and h@xmath13 is at ( 88.15,+1.4 ) , and stems from the outskirts of barnard s loop . there is no counterpart of h@xmath13 emission inside the cbi primary beam . from the comparison with the pmn survery in fig . [ fig : shassa ] we further confirm free - free emission is negligible at 31 ghz . there is a hint of a radio counterpart of the h@xmath13 corona , but no 5 ghz emission coexists with the 31 ghz emission . we can further test the bremsstrahlung hypothesis for the 31 ghz emission by extrapolating the observed intensity levels to 5 ghz with a spectral index of @xmath69 , in the optically thin approximation . the restored cbi image on fig . [ fig : cbi]b reaches peak intensities of 0.22 jy beam@xmath10 . since the pmn beam is 3.7 arcmin fwhm units , but its resolution depends on whether the data were acquired with the green bank 300-foot dish or with the parkes 140-foot dish . we calibrated the pmn survey with the 17 brightest point sources in a 6 deg field centered on ldn 1622 , using as reference the fluxes listed by ( * ? ? ? * north 6 cm catalog , also based on the pmn survey ) . we fitted elliptical gaussians to each point source to extract fluxes , and obtained that the beam solid angle used in the intensity units must correspond to a 3.7 arcmin fwhm psf to reproduce the catalog fluxes . the average fwhm of the elliptical gaussians is @xmath70 arcmin , coincident with the chosen intensity units . ] , peak 5 ghz intensities should range from 264 mjy beam@xmath10 for an unresolved source , to 52 mjy beam@xmath10 for a uniformly extended source . the root - mean - square ( rms ) noise in the pmn image is @xmath71mjy beam@xmath10 . the absence of the 31 ghz features from the pmn image therefore allows us to rule out free - free emission at 9 @xmath0 . in order to assess possible contamination at 31 ghz by background sources , we have overplotted on fig . [ fig : shassa ] the entries from the nvss catalog @xcite with flux densities greater than 10 mjy , as well as the entries from the north 6 cm database @xcite . only one source may be present at 31 ghz . this is pmn j0555 + 0116 , or nvss j055516 + 011622 ( j2000 ra : 05:55:16.62 , dec : + 01:16:22.9 ) , which is the source at ( 88.72 , 1.28 ) in fig . [ fig : shassa]b , and well outside the dark cloud and the cbi primary beam . @xcite explain that low spatial frequencies , on scales larger than 30 arcmin in declination , are filtered - out from the pmn survey . but the cbi - pmn comparison is not affected by this filter . the pmn filter corresponds to the very largest angular scales observed by the cbi , and ldn 1622 is a compact object of order 10 arcmin in diameter . the outskirts of barnard s loop , picked up in the mem model at @xmath25 , is probably filtered - out in pmn . for a comparison with @xcite we must consider the consequences of differencing on their chopped observations , which filters - out low spatial frequencies . the flux densities in @xcite are referred to the _ iras _ 100 @xmath1 m map by linear cross - correlation . but in general the radio and ir emissions are bound to have different power - spectra in the ism at large . thus in general the slopes of the straight line fits between radio and ir visibilities depend on baseline length . in the case of ldn 1622 the radio - ir conversion factors given in table [ table : corr ] do show some variation at 12 @xmath1 m when comparing cases a and b. in order to approximately account for the @xmath72arcmin chop throw of @xcite , we restrict our analysis to baselines in excess of @xmath73 @xmath16 , or 286 @xmath16 , and use the non - differenced dataset . in this case , @xmath74 , which is @xmath42 @xmath0 higher than the value of @xmath75 listed in table [ table : corr ] for the full dataset . the tentative detection of spinning dust in ldn 1622 by @xcite is based on a rise in flux density from 5 ghz to 9.75 ghz . since their 5 ghz data were not chopped because of hardware limitations , the rising sed could simply reflect the missing spatial frequencies . we nonetheless confirm the tentative detection of a rising spectrum by @xcite : after scaling units , the value for the dimensionless @xmath76 at 9.75 ghz is @xmath77 , which by comparison with our value for @xmath78 @xmath16 implies a spectral index @xmath79 . estimating a spectral index from the 10 cbi channels is difficult for an extended object such as ldn 1622 because of varying @xmath80coverage . flux loss , due to missing spatial frequencies , is greater in the high - frequency channels than in the low - frequency channels . assuming that the mem model is a good approximation to the sky signal , a single spectral index @xmath13 can be varied to minimize @xmath81 where the sum extends over all baselines and all channels . @xmath82 is the reference frequency used by mockcbi to scale the intensity map by the input spectral index @xmath83 to the frequency of the @xmath84 visibility data point . note that although in this application mockcbi internally uses @xmath83 , the model visibilities @xmath85 still bear a frequency dependence through the @xmath80coverage and the primary beam . we optimize @xmath45 by finding the root of @xmath86 . the entropy term does not depend on @xmath13 since it is calculated on the model image , which is kept constant for all channels in our implementation . yet the inclusion of a regularizing entropy biases the spectral index estimates . in the case of ldn 1622 , pure @xmath45 reconstructions , with @xmath87 , result in noisy model images , while in the absence of data a pure mem reconstruction , with @xmath88 , defaults to a flat image , whose intensity is @xmath89 ( see appendix [ sec : memalgo ] ) . increasing values of @xmath16 result in smoother model images , and the lower frequency channels recover more flux from the model images than the higher frequency channels . we confirmed by simulation that @xmath13 is recovered in pure @xmath45 reconstructions . the highest value of @xmath16 for which the resulting spectral index is not significantly biased is @xmath90 . to obtain this limiting value we simulated the cbi observations on template maps . we fit a model image and a single spectral index to simulated visibilities on the processed _ iras _ 12@xmath1 m and 25@xmath1 m templates . the cbi - simulated visibilities are calculated with mockcbi using @xmath83 . we ran our mem algorithm 90 times , with exactly the same settings as for the cbi models , feeding as input the cbi - simulated visibilities with the addition of 90 different realisations of gaussian noise , as explained in point [ pt : noise ] of appendix [ sec : modval ] . we rejected models that converged early on a local minimum by requiring a minimum number of iterations @xmath91 . for reference the @xmath92 cbi model converged in @xmath93 iterations . the average value of best - fit indices in the simulations , without @xmath91 cutoff , is @xmath94 for _ iras _ 12@xmath1 m and @xmath95 for _ iras _ 25@xmath1 m . @xmath96 decreases with increasing @xmath91 , until it reaches the input value at @xmath97 for both _ iras _ 12@xmath1 m and _ iras _ 25@xmath1 m . the resulting spectral index is @xmath98 for _ iras _ 12@xmath1 m and @xmath99 for _ iras _ 25@xmath1 m , and satisfactorily close to zero , in the sense that the systematic bias due to the smoothness introduced by the entropy term is of order @xmath100 . the 2636 ghz cbi spectral index we obtained from the mem modelling with @xmath92 is @xmath101 . spectral indices are sometimes inferred by reference to a template image , as in sec . [ sec : lowfreqindex ] . the 31 ghz sky image of ldn 1622 is assumed to follow exactly a template image , say _ iras _ 100@xmath1 m , so that the cbi image is a scaled version of the reference image , and @xmath102 , where @xmath103 are the cbi channel frequencies and where @xmath104 are cbi - simulated visibilities . the scaling factor @xmath49 can be obtained as explained in point [ pt : acoef ] of appendix [ sec : modval ] . the spectral behavior of the cbi visibilities is thus cast into the scale coeficients . however this strategy yields inconsistent results because it is difficult to find an ideal reference image . using the four _ iras _ band , and averaging the 10 cbi channels in two frequencies , 28.5 ghz and 33.5 ghz , we obtain spectral indices that depend strongly on the reference template and on baseline range . for the full range of baselines , @xmath13 varies from @xmath105 for _ iras _ 12@xmath1 m to @xmath106 for _ iras _ 60@xmath1 m , the other _ iras _ bands giving intermediate values . for the non - differenced dataset and @xmath14-radii in excess of 120 @xmath16 , we obtain values ranging from @xmath107 for _ iras _ 12@xmath1 m to @xmath108 for _ iras _ 25@xmath1 m . all of these alternative cbi-_iras _ cross - correlations could equally well be used to infer a spectral index . but the difference between the extremal values obtained above is greater than 3 @xmath0 , and is therefore significant . these results are reported here to emphasize the systematic uncertainties involved in determinations of spectral energy distributions inferred by cross - correlations . we extracted fluxes from the _ wmap _ , _ iras _ , and pmn surveys using a circular aperture with a diameter equal to the fwhm of the cbi primary beam at 31 ghz , or 45 arcmin . in order to compare with the cbi measurement , we also subtract a background level given by the flux density in the cbi reference field ( offset by 8 min to the east ) . for all maps the reference field is essentially devoid of emission compared to the object field . to take into account flux loss , the cbi flux density we discuss here is that obtained by scaling the _ iras _ 100@xmath1 m flux density ( see sec . [ sec : ground ] ) . the existing data on the integrated sed of ldn 1622 are summarised in fig . [ fig : l1622_sed ] and table [ table : l1622_sed ] . lrr 4.85 & @xmath109 & pmn + 5.00 & @xmath110 & green bank@xmath111 + 8.25 & @xmath112 & green bank@xmath111 + 9.75 & @xmath113 & green bank@xmath111 + 23.0 & @xmath114 & _ wmap _ + 31.0 & @xmath115 & cbi@xmath111 + 33.0 & @xmath116 & _ wmap _ + 41.0 & @xmath117 & _ wmap _ + 61.0 & @xmath118 & _ wmap _ + 94.0 & @xmath119 & _ wmap _ + 3000 & @xmath120 & _ iras _ + 5000 & @xmath121 & _ iras _ + the spectrum of the emissivity per unit proton column density in ldn 1622 can be fit with the spinning dust emissivities of @xcite , as first shown by @xcite . the data points are fit with a mixture of free - free emission , a modified blackbody representative of traditional dust emission , and the spinning dust emissivities . we require that a 15 k modified blackbody , with a 1.7 emissivity index , crosses the 90 ghz _ wmap _ pointm point and the _ wmap _ w band simultaneously resulted in excessively low emissivity indices , or in an unrealistic sub - mm peak . the bulk of the dust in ldn 1622 is thus characterised by at least 2 modified black bodies . ] . the spinning dust emissivities depend on environment , and we confirm the result of @xcite that the sed is best fit with a mixture of cnm and wnm emissivities ( as defined by * ? ? ? * ) , with a fraction of 37@xmath305% cnm and 63@xmath3011% wnm and a proton colummn averaged over the cbi primary beam of @xmath122 @xmath123 ( somewhat less than @xmath124 @xmath123 , the value used by @xcite referring to the peak extinction value ) . the spectral indices obtained from the _ wmap _ data are @xmath125 and @xmath126 , or @xmath127 , which is within 1 @xmath0 from @xmath101 . combining all measurements gives a 30 ghz index @xmath128 . the cbi observations of ldn 1622 resulted in the first cm - wave continuum image of a dark cloud , at frequencies where traditional emission from dust is not expected . the cbi data follow a tight correlation with the far - ir emission , confirming that the 31 ghz emission is nonetheless related to dust . under visual inspection the 31 ghz map is closer to the _ iras _ 12 @xmath1 m and _ iras _ 25 @xmath1 m maps than to the _ iras _ 100 @xmath1 m map . to quantify the ir radio similarities we calculate the cross - correlation of the 31 ghz images with each of the _ iras _ images . we find a trend for a decreasing cross - correlation with wavelength , such that the 31 ghz12 @xmath1 m comparison has the highest cross - correlation . the mid - ir cm - wave correlation in ldn 1622 indicates that the cm - wave continuum emission arises in a shell coincident with the pdr at the surface of ldn 1622 exposed to the ori ob 1b uv field . the closer match between the 31 ghz and 12 @xmath1 m images can be interpreted as support for spinning dust . alternatively the 31 ghz continuum may stem from a mechanism of molecular continuum emission at 31 ghz or a dense molecular forest spread over 2636 ghz . we suspect the reason why the mid - ir cm - wave correlation was not previously detected in other objects , or in the diffuse ism , is because the _ iras _ 12@xmath1 m maps are contaminated by many more point sources than the _ iras _ 100 @xmath1 m maps . the stellar emission at mid - ir wavelengths has no counterpart in cm - waves , as shown here in the case of ldn 1622 . the 10 cbi channels allow estimating a spectral index @xmath101 . combining all measurements we obtain @xmath129 . this article benefitted from the constructive comments of an anonymous referee that motivated sec . [ sec : sed ] , the discussion on unresolved radiosources , fig . [ fig : ysosed ] and its discussion in appendix c. s.c . acknowledges support from fondecyt grant 1030805 , and from the chilean center for astrophysics fondap 15010003 . we gratefully acknowledge the generous support of maxine and ronald linde , cecil and sally drinkward , barbara and stanely rawn , jr . , fred kavli , and rochus vogt . this work is supported by the national science foundation under grant ast 00 - 98734 . we acknowledge the use of nasa s _ skyview _ facility ( http://skyview.gsfc.nasa.gov ) located at nasa goddard space flight center . the mem algorithm was programmed by us and fits model visibilities , calculated on a model image , to the observed visibilities . the free - parameters of our mem model are the pixels in the model @xmath130 image , @xmath131 . we set to zero all pixels that fall outside a region of the sky where the expected noise is larger than a specified value . in practice , for one pointing as is the case here , this means restricting the number of free pixels to those that fall within a user - supplied radius from the phase center . the relatively small number of visibilities for the cbi ( @xmath132 for each on - off cycle ) allows one to work in the @xmath80plane and fit for the observed visibilities directly , rather than work in the sky plane and deconvolve the synthesized beam . we did not apply any gridding of the visibilities @xcite , which we postpone to a future development of our code . the use of a direct fourier transform in our current implementation is time consuming . the model functional we minimize is @xmath133 , with @xmath134 where the symbol @xmath135 stands for the modulus of a complex number @xmath136 , the sum extends over all visibilities ( i.e. , the sum runs over 10 channels and 78 baselines ) , @xmath34 is the root - mean - square ( rms ) noise on the corresponding visibility , @xmath137 stands for the observed visibilities , and the model visibilities @xmath85 are given by @xmath138 \frac{dx\,dy}{\sqrt{1-x^2-y^2 } } ~ , \label{eq : vmodel}\ ] ] where @xmath139 is the cbi primary beam and @xmath140 and @xmath141 are the direction cosines relative to the phase center in two orthogonal directions on the sky . the model visibilities are calculated using the mockcbi program ( see below ) . we assume a flat spectral index for the model image , i.e. @xmath142ghz ) over the 10 cbi channels . we use the entropy @xmath143 , where @xmath144 is the model image and @xmath145 is a small intensity value taken as the noise estimated by difmap and divided by 10000 . we also investigated an entropy term of the form @xmath146 , where @xmath147 , obtaining essentially the same results . image positivity is enforced by clipping . all intensities below the threshold value of @xmath145 are set equal to @xmath145 . our choice for the image entropy is such that the entropy term minimizes the need for clipping with a diverging derivative at zero intensities . however we caution that the true sky signal in our differenced observations may not be strictly positive : sources in the reference field act as negative signal . the entropy is used as a regularizing term . because the reconstruction is degenerate in the sense that we have more free parameters than data points , pure @xmath45 reconstructions lead to artificially low values of reduced @xmath45 , so that @xmath45 models end up fitting the noise ( i.e. the residual image is artificially flat at the locus of free parameters ) . the parameter @xmath16 was adjusted by hand and kept fixed during the optimization . intermediate values of @xmath16 from infinity to zero recover the sum of object signal and noise in gradually increasing detail . the exact value of @xmath16 is set by trial and error , requiring that @xmath45 ( eq . [ eq : chi2 ] ) is close to its expected value given by approximately twice the number of imaginary data visibilities . a dimensionless value of @xmath148 ( @xmath149 for @xmath67 ) gave good results when reconstructing on test images ( see fig . [ fig : irasmem ] below ) . we obtain a reduced @xmath45 value of 1.04 for the cbi visibilities , with 25726 data points ( i.e. twice the number of complex data points ) . reduced @xmath45 for the mem models of the template _ iras _ images is 0.99 . the slightly larger @xmath45 for the cbi data is probably due to faint sources in the reference field acting as negative sky signal . the positivity requirement precludes modelling such negative signal . convergence is achieved in @xmath150 iterations using the fletcher - reeves conjugate - gradient algorithm from _ numerical recipes _ * nr ) , or @xmath151 if using the _ gnu scientific library _ ( gsl , http://www.gnu.org/software/gsl/ ) . the gsl algorithm is double - precision , but is too slow for our needs as it requires more gradient evaluations per iteration than in nr . thus the models presented in this work use the nr implementation . one reconstruction takes about 30 min using the amd athlon xp3000 processor , or 1h with an intel pentium 4 at 2.80 ghz . to validate our mem model we reconstructed the sky emission from model visibilities , obtained by a simulation of cbi observations on reference images ( `` cbi - simulated visibilities '' ) . simulation of the cbi observations is performed with the mockcbi program ( pearson 2000 , private communication ) , which calculates the visibilities @xmath152 on the input images @xmath153 with the same @xmath14 sampling as a reference visibility dataset ( eq . [ eq : vmodel ] ) . thus mockcbi creates the visibility dataset that would have been obtained had the sky emission followed the template . we used as reference images the maps of ldn 1622 in the four _ iras _ bands , as downloaded from _ skyview _ ( http://skyview.gsfc.nasa.gov ) . the procedure is as follows : 1 . subtract conspicuous mid - ir point - sources in the 12 @xmath1 m and 25 @xmath1 m 3@xmath1543 degree fields . we fit elliptical gaussians on a second order polynomial surface . only one of these point - sources coincides with the object itself , namely l1622 - 10 ( see appendix [ sec : midirps ] ) , but all contribute to the simulated visibilities . l1622 - 10 is an entry in the _ iras _ point source catalog , so it was removed from the 12@xmath1 m template by subtracting a point source with l1622 - 10 s tabulated 12@xmath1 m flux of 1.027 jy , with a psf given by the minimum width of the elliptical gaussian fits to the other point sources ( 5.4 arcmin fwhm ) . we performed tests both with and without subtraction of l1622 - 10 . clip the _ iras _ images so that the minimum intensity value is zero . the processed images are shown on fig . [ fig : irasmem ] . 3 . simulate cbi visibilities on the processed _ iras _ images using mockcbi . 4 . cross - correlate the observed cbi visibilities with the model visibilities to obtain 31 ghz far - ir conversion factors , @xmath49 : @xmath155 , in the complex plane . we fit for @xmath49 by minimising @xmath156 , where the notation is the same as in eq . [ eq : chi2 ] . [ pt : acoef ] 5 . divide the model visibilities by @xmath49 to obtain model visibilities scaled to the 31 ghz values . values for @xmath49 are given in table [ table : corr ] , for case a ( differenced dataset ) . add gaussian noise to the complex model visibilities , ( i.e. we assume the model visibilities have no noise ) , with a dispersion given by the root - mean - square ( rms ) noise on the corresponding cbi visibility . [ pt : noise ] 7 . run the mem reconstruction algorithm with the same parameters as for the observed cbi data . repeat the simulation 90 times with 90 different realisations of noise . average the 90 model images . we tested that the measured scatter in the properties of the simulated reconstructions does not increase when increasing the number of noise realisations from 60 to 90 ( although 30 realisations was not enough ) . we did not take into account the finite resolution of the _ iras _ maps , which is due to the coarse pixelization used in the iras sky survey atlas maps available at _ skyview_. the net effect is that the template resolution is lower than that of the cbi data . the mid - ir point sources allow estimating that the natural - weight synthesized beam is 20% larger for the _ iras _ simulations that for the cbi data . [ fig : irasmem ] shows the average mem models overlaid on the input maps , and allows judging by inspection the level of detail that can be recovered from the cbi visibilities of ldn 1622 . in this case we used the full @xmath14-coverage of the cbi , as in fig . [ fig : cbi ] . although ldn 1622 figures in lists of starless cores @xcite it harbors an entry of the _ iras _ point source catalog , iras 05517 + 0151 , whose presence can be inferred from the _ iras _ 12@xmath1 m image in fig . [ fig : irasmem ] , because the peak of emission at a position of @xmath157 is bright and unresolved , and stands out over the diffuse emission . iras 05517 + 0151 is coincident within the uncertainties with an entry from the 2mass catalog @xcite , 2mass 05542277 + 0152039 , and with the binary pre - main - sequence star l1622 - 10 ( j2000 ra : 05:54:26.8 , dec : + 01:52:16 , * ? ? ? the yso is very clear as a saturated pixel in the isocam 6.7 @xmath1 m image of ldn 1622 presented by @xcite . note that ldn 1622 is curiously listed as ldn 1672 in @xcite , and the orientation of the image is not as that obtained from the _ iso _ archive , and is thus probably wrong . for these reasons we present in fig . [ fig : isocam ] an overlay of the _ iras _ 12@xmath1 m emission in contours on the upright isocam 6.7 @xmath1 m map . the 12 @xmath1 m peak and the saturated region at 6.7 @xmath1 m are coincident . l1622 - 10 is probably a t - tauri binary @xcite . we extracted the b , r , and i , photometry of l1622 - 10 from the usno - b10 catalog @xcite , and constructed the sed shown on fig . [ fig : ysosed ] . an estimate of the iso flux is included at 6.7 @xmath1 m , but is assigned zero weights because the presence of pixel glitches spreading away from l1622 - 10 suggests the detector may be saturated . the _ iras _ 25 @xmath1 m flux is also assigned zero weight because of uncertainties in the nebular contamination . a simple blackbody fit gives a temperature of 1680@xmath3050 k , using conservative uncertainties on the data points , and a linear size of [email protected] au for a distance of 120 pc . the integrated luminosity of the blackbody fit is thus 2.8 l@xmath158 , corresponding to a 1.4 m@xmath158 main - sequence star , which confirms that the yso is a low - mass object . bacmann , a. , andr , p. , puget , j .- l . , abergel , a. , bontemps , s. , ward - thompson , d. , 2000 , , 361 , 555 banday , a.j . , dickinson , c. , davies , r.d . , davis , r.j . , grski , k.m . , 2003 , , 345 , 897 becker , r. h. , white , r. l. , edwards , a. l. 1991 , , 75 , 1 bevington , p.r . , robinson , d.k . , 1992 , in `` data reduction and error analysis for the physical sciences '' , 2@xmath159 edition , mcgraw - hill boumis , p. , dickinson , c. , meaburn , j. , goudis , c. d. , christopoulou , p. e. , lpez , j. a. , bryce , m. , redman , m. p. , 2001 , , 320 , 61 briggs , a.s . , schwab , f.r . , sramek , r.a . , 1999 , in `` synthesis imaging in radio astronomy ii '' , asp conference series 180 , 127 casassus , s . , readhead , a.c.s . , pearson , t.j . , nyman , l. . , shepherd , m.c . , bronfman , l. , 2004 , , 603 , 599 condon , j. j. , cotton , w. d. , greisen , e. w. , yin , q. f. , perley , r. a. , taylor , g. b. , broderick , j. j. , 1998 , , 115 , 1693 condon , j.j . , griffith , m.r . & wright , a.e . , 1993 , , 106 , 1095 cutri , r.m . , et al . , 2003 , `` the 2mass all - sky catalog of point sources '' , university of massachusetts and infrared processing and analysis center ( ipac / california institute of technology ) . de oliveira - costa , a , tegmark , m. , finkbeiner , d.p . , davies , r.d . , gutierrez , c.m . , haffner , l.m . , jones , a.w . , lasenby , a.n . , rebolo , r. , reynolds , r.j . , tufte , s.l . , watson , r.a . , 2002 , , 567 , 363 de oliveira - costa , a. , tegmark , m. , gutirrez , c.m . , jones , a.w . , davies , r.d . , lasenby , a.n . , rebolo , r. , watson , r.a . , 1999 , , 527 , 9 dsert , f .- x . , boulanger , f. , puget , j.l . , 1990 , , 237 , 215 draine , b.t . , lazarian , a. , 1998 , , 494 , l19 draine , b.t . , lazarian , a. , 1998 , , 508 , 157 draine , b.t . , lazarian , a. , 1999 , , 512 , 740 draine , b.t . , li , a. , 2001 , , 551 , 807 elmegreen , b. g. , 2002 , , 564 , 773 finkbeiner , d.p . , 2004 , , 614 , 186 , finkbeiner , d.p . , davis , m. , schlegel , d.j . , 1999 , , 524 , 867 finkbeiner , d.p . , schlegel , d.j . , frank , c. , heiles , c. , 2002 , , 566 , 898 gaustad , j.e . , mccullough , p.r . , rosing , w. , van buren , d. , 2001 , , 113 , 1326 gautier , t.n . iii , boulanger , f. , prault , m. , puget , j.l . , 1992 , , 103 , 1313 heiles , c. , haffner , l.m . , reynolds , r.j . , tufte , s.l . , 2000 , , 536 , 335 li , a. , draine , b.t . , 2001 , , 554 , 778 lagache , g.,2003 , , 405 , 813 laureijs , r.j . , chlewicki , g. , wesselius , p.r . , clark , f. o. , 1989 , , 220 , 226 lee , c.w . , myers , p. c. , tafalla , m. , 2001 , , 136 , 703 lee , c.w . , myers , p. c. , 1999 , , 123 , 233 leitch , e.m . , readhead , a.c.s . , pearson , t.j . , myers , s.t . , 1997 , 486 , l23 lynds , b.t . , 1962 , , 7 , 1 maddalena , r.j . , morris , m. , moscowitz , thaddeus , p. , 1986 , , 303 , 375 monet d.g . , levine s.e . , casian b. , et al . , 2003 , , 125 , 984 . padin , s. , et al , 2002 , , 114 , 83 page , l. , nolta , m. r. , barnes , c. , bennett , c. l. , halpern , m. , hinshaw , g. , jarosik , n. , kogut , a. , limon , m. , meyer , s. s. , peiris , h. v. , spergel , d. n. , tucker , g. s. , wollack , e. , wright , e. l. , 2003 , , 148 , 39 . park , y .- s . , lee , c. w. , myers , p. c. , 2004 , , 152 , 81 pearson , t. j. , mason , b. s. , readhead , a. c. s. , shepherd , m. c. , sievers , j. l. , udomprasert , p. s. , cartwright , j. k. , farmer , a. j. , padin , s. , myers , s. t. , bond , j. r. , contaldi , c. r. , pen , u .- l . , prunet , s. , pogosyan , d. , carlstrom , j. e. , kovac , j. , leitch , e. m. , pryke , c. , halverson , n. w. , holzapfel , w. l. , altamirano , p. , bronfman , l. , casassus , s. , may , j. , joy , m. , 2003 , , 591 , 556 press , w.h . , flannery , b.p . , teukilsky , s.a . , vettering , w. , y. , 1996 , numerical recipes ( cambridge , cambridge univerisity press ) reipurth , b. , zinnecker , h. , 1993 , , 278 , 81 shepherd , m.c . , 1997 , in astronomical data analysis software and systems vi , ed . g hunt & h.e . payne , asp conference series , v125 , 77 - 84 `` difmap : an interactive program for synthesis imaging '' . van dishoeck , e.f . , 2004 , , 42 , 119 watson , r. a. , rebolo , r. , rubio - martn , j. a. , hildebrandt , s. , gutirrez , c. m. , fernndez - cerezo , s. , hoyland , r. j. , battistelli , e. s. , 2005 , , 624 , 89 wheelock , s. , et al . , 1991 , iras sky survey atlas explanatory supplement wilson , b.a . , dame , t.m . , masheder , m.r . , thaddeus , p. , 2005 , , 430 , 523 wright , e.l . , 1998 , , 496 , 1

premelting describes the confluence of phenomena that are responsible for the stable existence of the liquid phase of matter in the solid region of its bulk phase diagram . here we develop a theoretical description of the premelting of water ice contained in a porous matrix , made of a material with a melting temperature substantially larger than ice itself , to predict the amount of liquid water in the matrix at temperatures below its bulk freezing point . our theory combines the interfacial premelting of ice in contact with the matrix , grain boundary melting in the ice , and impurity and curvature induced premelting , the latter occurring in regions which force the ice - liquid interface into a high curvature configuration . these regions are typically found at points where the matrix surface is concave , along contact lines of a grain boundary with the matrix , and in liquid veins . both interfacial premelting and curvature induced premelting depend on the concentration of impurities in the liquid , which , due to the small segregation coefficient of impurities in ice are treated as homogeneously distributed in the premelted liquid . our principal result is an equation for the fraction of liquid in the porous medium as a function of the undercooling , which embodies the combined effects of interfacial premelting , curvature induced premelting , and impurities . the result is analyzed in detail and applied to a range of experimentally relevant settings .

1001.1762

due to advances in our understanding of the premelting of ice and other materials over the last fifteen years @xcite , and an increasing appreciation of the influence and extent of this phenomenon in technological , terrestrial and extraterrestrial settings , it is prudent to revisit the detailed thermodynamic basis of its manifestation in porous media where water is invariably found in adsorbed layers or in bulk . in 1992 cahn , dash , and fu ( cdf ) @xcite developed the original premelting based theory for the liquid fraction @xmath0 of water in a porous medium at sub - freezing temperatures . given experimental and theoretical advances in the interim , we extend their work in a number of respects . cdf considered a matrix of monodisperse spheres arranged either in a simple cubic packing or in a cubic close packing , regarded as lower and upper bounds for the experimental systems . we design our theory for the _ random _ close packing of monodisperse spheres , which may be a more realistic description for most applications , e.g. frozen soils , where even a small degree of polydispersity prevents a periodic packing of the matrix particles . the assumed periodic packing of the matrix particles leads cdf to take the grain size of the interstitial ice to be on the order of the size of the matrix lattice unit cell . given that the degree of polycrystallinity of the interstitial ice has a complicated dependence on the freezing rate ( see e.g. , ref . * and references therein ) and the sample annealing history , the ice grain size may differ substantially from the matrix particle size . whence , polycrystallinity of the interstitial ice enters into our theory as a parameter , namely the density of grain boundary surface relative to the interstitial volume . calculations since cdf showed that the dispersion force contribution to interfacial melting could be repulsive or attractive @xcite and thus favor or suppress melting in a manner and strength that depends on the effective dielectric response of the ice / water / matrix interface . moreover , it has been understood that the great inter - experimental variability in both the interfacial and surface melting of ice was likely due to the sensitivity of the phenomena to impurities which have a rich and complex influence on the melting behavior @xcite . thus , our theory allows for both attractive and repulsive dispersion forces of arbitrary strength ( depending on the matrix material ) and includes the effect of charges at the interface , which give rise to an electrostatic force . finally , most importantly , our work includes the effect of charged impurities in the premelted liquid on both interfacial and curvature induced premelting . the article is structured as follows . in section [ sec_theory ] we lay out the tenets and details of our theory , discussing interfacial premelting and curvature induced premelting in the presence of charged impurities and combine them in order to describe premelting in a porous medium . the final result of the theoretical treatment is an implicit functional equation for the liquid fraction @xmath0 , eq . . the first part of section [ sec_analysis ] is dedicated to the analysis of interfacial premelting as described by eq . , and in the second part eq . for the liquid fraction @xmath0 is analyzed by extracting exact solutions for certain limiting cases and numerical solution in the general case . the porous media which we consider consist of fused quartz , gold , and silicon . we summarize in section [ sec_sumconc ] . in order to calculate the equilibrium thickness @xmath1 of a premelted layer for a given undercooling @xmath2 , where @xmath3 is the bulk melting temperature of ice and @xmath4 is the temperature of the system , we consider a three layer planar system . layer @xmath5 is given by the substrate , i.e. , in the case of interfacial melting , the layer consists of the material which forms the porous medium , or of ice when grain boundary melting is considered . layer @xmath6 is the premelted quasi - liquid layer of thickness @xmath1 and finally layer @xmath7 consists of ice . in what follows we consider the complete free energy of the system and then focus on the individual complications and contributions in turn to show how known asymptotic results emerge from the theory . it is assumed that layer @xmath6 contains a certain amount of impurities . the gibbs free energy of the system per unit area is given by @xmath8 + \mathcal{i}(d ) \ , , \end{aligned}\ ] ] where @xmath4 and @xmath9 are temperature and pressure , @xmath10 , @xmath11 , @xmath12 are the chemical potentials per mole of the solid , the liquid , and the impurity species , respectively , @xmath13 , @xmath14 , @xmath15 are the numbers of mole per unit area of the solid , the liquid , and the impurities , respectively . the term @xmath16 $ ] , where @xmath17 is the gas constant , corresponds to the mixing entropy of the impurities assuming low impurity concentration so that the ideal gas result applies . finally , @xmath18 denotes the total effective interfacial free energy described in more detail below . note that @xmath1 is directly proportional to @xmath14 through @xmath19 , where @xmath20 is the molar density of the liquid which is a constant in the calculation . the solid ( ice ) and the liquid ( water ) parts of the system can exchange material and hence the condition for thermodynamic equilibrium is given by @xmath21 applying this criterion to the free energy given above , and using the fact that @xmath22 where @xmath23 ( @xmath24 ) is the specific entropy of the solid ( liquid ) , we can approximate @xmath25 , where @xmath26 is the latent heat of melting per mole and arrive at @xmath27 in the absence of interfacial forces we have @xmath28 , and the result reduces to @xmath29 , where the molar density of impurities is @xmath30 . this undercooling corresponds precisely to the classical colligative effect the depression of the bulk freezing temperature due to dissolved impurities . obviously , compared to the force free case , repulsive interfacial forces for which @xmath31 result in a larger undercooling @xmath32 thereby enhancing the layer thickness . conversely , attractive interfacial forces wherein @xmath33 suppress the formation of a premelted liquid film . indeed , for sufficiently large @xmath32 , eq . may have no solution showing that the attractive interfacial forces can dominate the colligative effect of the impurities and the premelted liquid layer will vanish . the interfacial energy has two contributions ; @xmath34 , where @xmath35 is due to dispersion forces , acting between the three interfaces defining the system , and @xmath36 captures the effect of immobile charges at the interface . although we compute the strength of the dispersion force interaction from the full frequency dependent theory @xcite , we take the small thickness limit in which only non - retarded forces contribute to @xmath35 which consequently can be written as @xmath37 where @xmath38 is the hamaker constant for the given layered system @xcite . for symmetric systems , such as the ice - liquid - ice configuration , which occurs in the case of grain boundary melting , @xmath38 is _ positive _ and hence dispersion forces are _ attractive_. at the interface between the ice and the porous matrix @xmath38 can assume both signs , depending on the material properties of the porous medium . immobile surface charges with a given surface charge density @xmath39 on both interfaces which bound the quasi - liquid layer are screened by the counterions in the liquid . making use of the debye - hckel theory this leads to the contribution @xmath40 where @xmath41 is the vacuum permittivity , @xmath42 the relative permittivity of liquid water , and @xmath43 is the debye length for monovalent ions , with the electron charge @xmath44 , avogadro s number @xmath45 , and boltzmann s constant @xmath46 . the debye length describes the characteristic decay of the ion field adjacent to the charged surface due to the screening by counter ions . a repulsive force between two charged surfaces originates in the restriction of the entropy of the ions as the surfaces are brought closer . whereas ions of charge opposite to those of the surface are attracted to it , they are repelled from each other , and the increased proximity induced by decreasing the film thickness increases the free energy . note the relatively complex @xmath1-dependence of @xmath36 which is due to the fact that @xmath47 is also a function of @xmath1 . thus , the richness of the role of impurities in the premelting of ice is a consequence of this and the efficient segregation of ions into the film ; since nearly all of the impurities remain in the liquid , as the temperature changes the _ amplitude _ and _ range _ of the repulsive electrostatic interaction varies @xcite . in principle , the situation could be even more complex when for example @xmath39 = @xmath48 . indeed , when @xmath1 is small the debye - hckel expression is less accurate and steric effects become more important . however , to introduce such complications here exceeds our understanding of the situation based on present experimental evidence . using these expressions for @xmath34 in eq . we obtain an implicit equation for the thickness @xmath1 of the premelted layer as a function of the undercooling @xmath32 : @xmath49 e^{-c\sqrt{n_i d}}\ , , \ ] ] where the constant @xmath50 is @xmath51 this result is analyzed in detail in section [ sec_reslayer ] . in order to assess the effect of the ice - liquid interfacial curvature on the freezing temperature , we neglect the known anisotropy of the surface energy of ice @xcite and consider a spherical ice crystal of radius @xmath52 surrounded by undercooled water . as opposed to the planar case , the interfacial area is no longer a constant as it was in the treatment of section [ sec_mechlayer ] . with this new complication we reconsider the gibbs free energy @xmath53 of the _ total _ system . the same holds for the numbers of moles @xmath13 , @xmath14 , and @xmath15 which are now taken for the total system and not per unit surface . for ease of notation we use the same symbols for these new quantities . all other quantities are as defined above . whence , we have @xmath54 + 4\pi r^2 \gamma_{sl } \ , , \end{aligned}\ ] ] where @xmath52 is related to @xmath13 through @xmath55 with the constant molar density @xmath56 . we note here that we understand that @xmath57 , where @xmath58 is the angle that defines the surface normal . thus , the surface energy , itself a gibbs free energy conjugate to area , is known in general to depend on the size of the system , the adsorption of impurities and the crystallographic orientation @xcite but the simplification of treating it as a constant is justified both on the basis of experimental evidence for the first two dependencies and on the level of our analysis . the condition for thermodynamic equilibrium is again @xmath59 and hence a calculation analogous to that in the section [ sec_mechlayer ] yields the result @xmath60 or , for the general case of a surface with principal radii of curvature @xmath61 and @xmath62 , @xmath63 this result is the well known gibbs - thomson effect , i.e. the freezing point depression due to convexity of the ice - liquid interface , calculated here for the case where impurities are present in the liquid water . the larger the undercooling @xmath32 , the smaller the radii of curvature of an ice - liquid interface in equilibrium . at the bulk freezing point , @xmath64 , the radius of curvature diverges and a planar interface is the stable configuration . we use the random close packing ( rcp ) of hard spheres as a model for the porous matrix . apart from choosing the matrix material and hence calculating the properties which enter the determination of the hamaker constant ( see section [ sec_mechlayer ] ) , the only free parameter characterizing the porous matrix which contains the partially premelted ice is the radius @xmath65 of the hard spheres . the properties of rcp hard spheres which are required for our calculation are the sphere packing fraction @xmath66 and the average coordination number @xmath67 . recent estimates of these numbers @xcite are @xmath68 and @xmath69 . it should be noted , however , that the basic concept of rcp is not well defined mathematically , so these numbers vary slightly according to the protocol and algorithm used to determine the exact configuration @xcite . at a given undercooling @xmath32 we are interested in the fraction of the interstitial volume which is occupied by quasi - liquid water @xmath0 which we will more compactly refer to as the liquid fraction . the first contribution we consider comes from interfacial premelting at the surface of the porous matrix . in the limit where the premelted layer is much thinner than the radius of the matrix particles , @xmath70 , this fraction is well approximated by @xmath71 where @xmath72 is the surface area of the matrix particles per unit volume and @xmath1 is obtained from eq . using the hamaker constant @xmath73 for the ice - water - matrix layered system and the appropriate surface charge density @xmath74 . the sphere number density of the matrix is @xmath75 and thus we obtain @xmath76 and @xmath77 the second contribution to the liquid fraction is due to grain boundary melting . its importance depends sensitively on the grain size of the interstitial ice , which is a function of such factors as ( 1 ) the rate of cooling , ( 2 ) the undercooling @xmath32 , and ( 3 ) the time allowed for annealing . even were we to attempt a complete computation of the coarsening rate of the ice grains in the matrix , there is no unique way to specify the initial value problem and hence to arrive at a concrete grain size as a function of the properties of the porous matrix . thus , we introduce as an additional parameter the grain boundary density @xmath78 , which is the surface area of grain boundaries per unit volume of the interstitial ice . we denote @xmath79 as the related dimensionless density . the corresponding liquid fraction follows immediately @xmath80 where @xmath81 is the hamaker constant for the ice - liquid - ice layered system and @xmath82 is the surface charge density for the grain boundary . the thickness @xmath1 of the quasi - liquid layer is again obtained from eq . . we now deal with curvature induced premelting . important contributions to the liquid water fraction come from regions where the ice - liquid interface has a large curvature . in our idealized matrix these are the points where two matrix spheres are in contact . following cdf @xcite there is a liquid pocket forming at the contact point with a volume @xmath83 to lowest order in @xmath52 , which is the radius of curvature of the ice - liquid interface ( fig . [ fig_contactpoint ] ) . taking into account that the second principal radius of curvature is much larger than @xmath52 , we can use eq . to relate @xmath52 to the undercooling @xmath32 : @xmath84 using the number density @xmath85 of the sphere matrix and the matrix coordination number @xmath67 we can express the corresponding fraction of liquid water with respect to the total interstitial volume : @xmath86 ^ 2 r}{1-\eta } \nonumber \\ & = & \frac{3}{2 } \times \frac{\eta}{1-\eta } \times \frac{\mathcal{c}}{2 } \times \left[\frac{r(\delta t , \rho_i)}{r}\right]^2 \ , , \nonumber \\\end{aligned}\ ] ] with @xmath87 from eq . . . the amount of premelted liquid depends on the characteristic radius of curvature @xmath52 of the ice - liquid interface as obtained from eq . . depending on the matrix material there is either always a premelted liquid layer ( repulsive dispersion forces , @xmath88 ) or the premelted layer may collapse under certain conditions ( attractive dispersion forces , @xmath89 ) and the liquid - ice interface forms an angle @xmath90 with the matrix surface . for parsimony of presentation we show the neighboring matrix spheres of different materials while the present theory assumes porous media made up of a single material . ] a further curvature induced contribution to the liquid fraction brought out by cdf comes from the lines where a grain boundary and a matrix sphere meet . in the limit that the matrix particles are small on the scale of the linear scale of the ice grains , but that @xmath65 is much larger than @xmath87 the corresponding liquid fraction can be calculated using elementary geometry ( see appendix [ app_grainbmat ] ) @xmath91 ^ 2 . \quad\end{aligned}\ ] ] finally , curvature induced premelting gives rise to liquid veins along the lines where several grain boundaries meet . with the simplifying assumption that the ice grains are arranged in a cubic packing , the total length of veins per interstitial volume , or the density of veins @xmath92 , is obtained from @xmath93 as @xmath94 . given that the cross - sectional area of a vein in cubic packing is @xmath95 , we can calculate the corresponding contribution to the liquid fraction as @xmath96 ^ 2 \nonumber \\ & = & \frac{1}{3}(4-\pi ) \times \tilde{\rho}_{gb}^2 \times \left(\frac{r(\delta t , \rho_i)}{r}\right)^2 . \qquad \end{aligned}\ ] ] therefore , the total liquid fraction is obtained as the sum of the five individual contributions which we have derived above ; @xmath97 , and @xmath98 . thus far we have taken impurities into account in terms of a molar density @xmath99 or , in the case of the premelted layer , a surface density of impurities @xmath100 . laboratory probes using scattering or related approaches involve relatively small samples with good thermal control and a well defined ( fixed for a given run ) dopant level @xcite . the temperature changes in these systems are quasi - static so that the sample is at coexistence . in the natural environment there are of course temperature gradients . therefore , for applicability of our work there , the typical broad temperature gradients found for example in natural soils imply that a local region within will be nearly isothermal and thus isosaline . clearly , there will be many environmental circumstances where this may be violated , but because the first test of such a theory should be of high precision we focus mainly on the simplest experimental setting and exploit the fact that the impurity segregation coefficient of ice is small @xcite so that the premelted liquid is enriched / diluted when the system is cooled / warmed . hence it is prudent to introduce @xmath101 , the molar density of impurities in the interstitial water when the system is above the bulk melting temperature . thus , because of the validity of the assumption that impurities are contained almost exclusively in the liquid fraction of the interstitial water it follows that @xmath102 thereby properly accounting for the increase in the impurity concentration in the premelted liquid as the liquid fraction decreases . due to the fact that the regions of premelted water are all interconnected , @xmath99 is homogeneous over the entire sample . having now considered the essential physical phenomena we can formulate an equation for the total liquid fraction @xmath103 in the porous medium as a function of the undercooling @xmath32 and the impurity concentration @xmath101 in the completely melted sample . combining all the contributions discussed in section [ sec_premeltporous ] , we obtain @xmath104 which is an implicit function for @xmath105 and in the general case has no analytical solution . moreover , the underlying equation for the thickness @xmath1 of the premelted layer , which is required for @xmath106 and @xmath107 , is itself an implicit function of @xmath1 . this calls for a careful stepwise analysis first of eq . and subsequently of eq . , which we perform in sections [ sec_reslayer ] and [ sec_resliquidfr ] in section [ sec_mechlayer ] we derived eq . for the thickness @xmath1 of the premelted layer as a function of the undercooling @xmath32 , the impurity concentration @xmath15 , the dispersion forces as embodied in the hamaker constant @xmath38 , and the surface density @xmath39 of immobile charges on the solid - liquid interfaces . this expression can not be solved analytically in the general case . hence , presently we analyze eq . analytically , to demonstrate that we can obtain known distinguished limits and facilitate intuition , and numerically , to provide specific ranges of results for particular physical systems of interest . we begin by rewriting this equation in terms of the reduced , or dimensionless , undercooling @xmath108 as follows @xmath109 where the constants @xmath110 , @xmath111 , and @xmath112 . for _ attractive _ dispersion forces which _ suppress _ premelting we have @xmath113 , and for _ repulsive _ dispersion forces which _ promote _ premelting we have @xmath114 . all other quantities in eq . are positive . the simplest case occurs when both dispersion and electrostatic forces are absent , i.e. , @xmath115 and @xmath116 . the thickness @xmath1 of the liquid film is them simply given by @xmath117 representing the colligative effect of the dissolved impurities which , due to the small segregation coefficient of ice , are strongly enriched in the liquid layer . the situation immediately becomes more complex if we take dispersion forces into account . let us first consider materials that have @xmath113 , which means that dispersion forces suppress the formation of a premelted liquid film . assuming @xmath116 for the moment , eq . possesses a solution as long as the undercooling @xmath118 does not exceed a critical value @xmath119 corresponding to a minimum film thickness @xmath120 . if the temperature is reduced such that @xmath121 the attractive dispersion forces dominate the colligative effect and the liquid film collapses to zero thickness . for the most general form of eq . , the maximal undercooling @xmath122 and corresponding minimal layer thickness @xmath123 at which the quasi - liquid layer collapses can not be expressed analytically . we obtain the global maximum @xmath124 by extremizing the right hand side of eq . with respect to @xmath1 . two limiting cases can be distinguished . for large impurity concentrations @xmath15 the contribution due to surface charges becomes negligible and hence we recover the result @xmath125 for the case @xmath116 which we derived above . on the other hand , it can be shown that in the limiting case @xmath126 the values @xmath123 and @xmath122 follow from the surface charge dominated limit ( @xmath127 and @xmath115 ) which has @xmath128 in fig . [ fig_dmin ] we plot the minimal thickness @xmath123 of the quasi - liquid layer corresponding to the maximal undercooling @xmath122 as a function of the impurity concentration @xmath15 for different values of the surface charge @xmath39 . we use a hamaker constant @xmath129 which is the result obtained by wilen et al . @xcite for the ice / liquid / fused quartz layered system . the analytical results for @xmath126 and @xmath130 are characterized by the @xmath15-dependencies of the asymptotic behavior . note that the minimal thickness in the dispersion force dominated limit ( @xmath131 ) is below @xmath132 and therefore equal to the dimension of a single molecule . hence , the notion of a quasi - liquid layer can no longer be applied and for the ice / liquid / fused quartz system layer collapse can only be observed experimentally at low impurity concentrations and it is always triggered by surface charges . this is seen directly from eq . where the contribution of the surface charge for sufficiently small @xmath1 becomes _ negative _ , indicating _ attractive _ forces , which suppress liquid layer formation . in order to observe a collapse of the quasi - liquid layer driven by dispersion forces , the latter have to be more strongly attractive . we have included in fig . [ fig_dmin ] the case of the ice - liquid - gold system which has a hamaker constant of @xmath133 @xcite , and hence dispersion forces are indeed strong enough to yield a liquid - layer collapse at thicknesses of the order of those calculated for the surface charge dominated case . the contrast between the two scenarios will be studied further in the second part of this section ( figs . [ fig_layerthicknessquartz ] and [ fig_layerthicknessgold ] ) . in fig . [ fig_tmax ] the corresponding results for the maximal undercooling @xmath134 are plotted . as we expect from eqs . and , @xmath135 increases with surface charge @xmath39 resulting in a @xmath39-dependent offset for @xmath126 . moreover , for all @xmath39 , @xmath135 increases with impurity concentration @xmath15 . a large hamaker constant , for example of the ice / liquid / gold system , results in a low @xmath135 as shown by eq . , where @xmath136 . there are interesting experimental consequences of fig . [ fig_tmax ] . obviously , measuring @xmath135 for just one given impurity concentration @xmath15 is _ not _ sufficient to determine the surface charge @xmath39 at the interface . this follows from the fact that any two curves intersect , and hence measurements for at least two different impurity concentrations are required . , plotted here as a function of the impurity density @xmath15 for different values of the surface charge @xmath39 , to zero when the temperature is lowered such that the undercooling exceeds @xmath135 ( plotted in fig . [ fig_tmax ] ) . the collapse is caused by the combined effect of immobile surface charges ( dominating as @xmath126 ) and dispersion forces ( dominating as @xmath130 ) . for the ice / liquid / fused quartz system the collapse caused be dispersion forces occurs at thicknesses on the scale of the size of a single molecule where the notion of a quasi - liquid layer no longer applies . in order to observe collapse induced by dispersion forces a larger hamaker constant is required such as provided by the ice / liquid / gold system . ] as a function of the impurity concentration @xmath15 for the same parameters as used for the calculation of @xmath123 in fig . [ fig_dmin ] . the zero slope asymptote for @xmath126 is observed as expected from eq . . the large hamaker constant of the ice - liquid - gold system leads to low values of @xmath135 . ] when @xmath116 , eq . can still be solved analytically but because the resulting expression is lengthy and not particularly instructive we present the solution only in the small @xmath118 limit of moderate undercooling , @xmath137 this result clearly illustrates the fact that the presence of attractive dispersion forces ( @xmath113 ) acting across the system suppresses the formation of a quasi - liquid layer , whereas repulsive dispersion forces ( @xmath114 ) tend to increase the thickness of the premelted layer . when surfaces charges on the interfaces of the quasi - liquid layer with the substrate / ice are involved , eq . becomes transcendental and has no analytical solution . in order to get some insight in the effect of surface charges , we consider several special cases . first , let us assume that for moderate undercooling the system is _ dominated _ by surface charges . this scenario corresponds to setting @xmath115 , @xmath127 in eq . and considering only the leading contributions as @xmath1 approaches infinity . the solution is then @xmath138 ^ 2}{c^2 n_i } \ , .\ ] ] this result reveals an interesting consequence of the presence of charges at the interface . under appropriate conditions the layer thickness @xmath1 _ decreases _ when the impurity concentration _ increases_. this observation is contrary to the naive expectation based on the colligative effect and originates in the fact that for larger @xmath15 the electrostatic repulsion is screened for effectively @xcite . for repulsive dispersion forces ( @xmath114 ) the solution of eq . can be obtained asymptotically in the limit of large undercooling @xmath139 , whereas for attractive dispersion forces ( @xmath113 ) the premelted layer collapses to zero thickness for @xmath118 larger than some finite value . the repulsive case is @xmath140 and seems to contradict eq . for two reasons . first , the negative sign of the term proportional to @xmath141 in eq . indicates that immobile surface charges impede the formation of the premelted layer , whereas in eq . immobile surface charges alone are sufficient to stabilize a premelted layer . second , while eq . implies that @xmath1 decreases when the impurity concentration @xmath15 is increased , the opposite conclusion follows from eq . where , due to the term @xmath142 , the inhibition due to surfaces charges is reduced when @xmath15 increases . the apparent contradiction can be reconciled by the following observations . a direct consequence of eq . is that immobile surface charges do not contribute to the film thickness @xmath1 when the latter equals the debye length , namely @xmath143 , in which case the surface charge term vanishes . the film thickness @xmath144 translates into an undercooling @xmath145 . for an undercooling @xmath146 surface charges enhance the formation of a quasi - liquid layer while for @xmath147 surface charges suppress layer formation . considering that eq . applies to the @xmath148 case while eq . to the @xmath139 case , this explains the essential difference between eqs . and in terms of whether they support or suppress the formation of a liquid film . in order to isolate the dependence of @xmath1 on the impurity concentration @xmath15 we consider eq . in the case where dispersion forces can be neglected ( @xmath115 ) and the immobile interfacial charge density is small ( i.e. @xmath149 ) . to lowest order @xmath1 depends linearly on @xmath141 , and hence we obtain @xmath150 from this result the sign of @xmath151 can be shown to equal the sign of @xmath152 where @xmath153 and @xmath154 . we find @xmath155 for all @xmath156 as long as @xmath157 . therefore if @xmath158 the layer thickness @xmath1 _ increases _ with @xmath15 for any given undercooling @xmath118 . for lower impurity concentrations ( @xmath159 ) there is a finite interval for @xmath160 which contains @xmath161 , where @xmath162 . therefore , at sufficiently low impurity concentration there exists an interval of intermediate undercooling @xmath118 where @xmath1 _ decreases _ when @xmath15 increases . this result implies that for large undercooling @xmath118 the layer thickness @xmath1 must always increase with @xmath15 which was precisely a conclusion obtained from eq . . however , the result can not be applied to the charge dominated case , eq . , because the corresponding limit @xmath163 renders eq . it is possible though to obtain @xmath162 in the related limit @xmath164 , @xmath165 , if we assume that @xmath166 approaches a large constant such that @xmath167 holds . [ [ attractive - dispersion - forces ] ] attractive dispersion forces + + + + + + + + + + + + + + + + + + + + + + + + + + + + the non - monotonic dependence of @xmath1 on @xmath15 is illustrated in fig . [ fig_layerthicknessquartz ] where the full numerical solution of eq . is plotted for the case of the ice - liquid - fused quartz layered system . at low undercooling @xmath32 the system displays characteristic colligative behavior as described by eq . . the effect of immobile surface charges vanishes in the limit @xmath148 as indicated by eq . and the layer thickness @xmath1 increases when the impurity density @xmath15 increases . the layer thickness for small @xmath32 is obtained by requiring that the impurity _ concentration _ @xmath30 equals the impurity concentration which would be necessary in a bulk system to see a freezing point depression equal to the undercooling . when @xmath32 is increased , surfaces charges start to _ increase _ the layer thickness relative to the system without surface charges ( @xmath131 ) . interestingly , the resulting layer thickness for the system with the lower impurity density ( @xmath168 ) becomes even larger than for the system with more impurities ( @xmath169 ) . this illustrates the result from the above analysis displaying the existence of an interval of intermediate undercooling where the layer thickness decreases when the impurity density increases . recall that the reason for this is that at low impurity concentrations the electrostatic force is less effectively screened and hence the repulsion due to surface charges has a stronger effect , leading to a larger layer thickness than for higher impurity concentrations . when the system is undercooled further the collapse of the quasi - liquid layer can be observed in the figure . here , because of the weakness of dispersion forces , the collapse is a consequence of the electrostatic force becoming attractive at low layer thicknesses . collapse for the cases without surface charges can not be observed in fig . [ fig_layerthicknessquartz ] as it occurs at thicknesses below @xmath132 ( c.f . , [ fig_dmin ] ) . in order to see layer collapse at larger layer thicknesses a larger hamaker constant is required , as provided by the ice - liquid - gold system . to see this , we plot solutions of eq . for this system in fig . [ fig_layerthicknessgold ] for the same parameters as considered in fig [ fig_layerthicknessquartz ] for the ice - liquid - fused quartz system . layer collapse is now observed at thicknesses above @xmath132 even for the configurations without surface charges . for the configurations with surface charges we observe that for @xmath168 layer collapse is not altered by the stronger dispersion forces while for @xmath169 the stronger attractive dispersion forces lead to a layer collapse at a larger layer thickness compared to the corresponding configuration considered in fig . [ fig_layerthicknessquartz ] . thus , in this case stronger dispersion forces drive layer collapse . of the quasi - liquid layer for the ice / liquid / fused quartz system ( @xmath129 ) as a function of the undercooling @xmath32 . for moderate undercooling the system is dominated by the colligative effect . in this regime the layer thickness increases with impurity density @xmath15 . at intermediate undercooling the effect of immobile surface charges leads to a thickening of the quasi - liquid layer which is most effective for low impurity concentrations leading to a _ decrease _ of the layer thickness when the impurity density is increased . at large undercooling surface charge induced layer collapse is observed , while layer collapse due to dispersion forces alone occurs at layer thicknesses below @xmath132 . ] ) . in contrast to the ice / liquid / fused quartz system , fig . [ fig_layerthicknessquartz ] , layer collapse occurs at thicknesses above @xmath132 even for the cases without surface charges . this is due to the stronger attractive dispersion forces in the ice / liquid / gold system . layer collapse for the cases with surface charges is altered only for the higher impurity density @xmath169 relative to the ice / liquid / fused quartz system , because at @xmath170 dispersion forces are weak compared to the electrostatic force . ] [ [ repulsive - dispersion - forces ] ] repulsive dispersion forces + + + + + + + + + + + + + + + + + + + + + + + + + + + the result of the analysis following eq . can be tested on the ice / liquid / silicon system which has a hamaker constant of @xmath171 @xcite , the negative sign implying repulsive dispersion forces . as shown in fig . [ fig_layerthicknesssilicon ] the system displays the expected colligative behavior for low undercooling and an enhanced layer thickness for intermediate undercooling where the effect of surface charges enhances the colligative effect , which alone would yield a straight line , @xmath172 . for intermediate undercooling the figure reveals a transition from a regime of low impurity density where the layer thickness _ decreases _ when @xmath15 _ increases _ , to a regime of high impurity density where the colligative effect dominates and accordingly an _ increase _ of @xmath1 is observed when @xmath15 _ increases_. from the figure the locus of the transition can be estimated to be @xmath173 , @xmath174 , and @xmath175 . this is in good agreement with the analytical result obtained from eq . , @xmath176 , @xmath177 , and @xmath178 . it should be noted that the analytical result is exact only in the limit @xmath179 but the complete solution of eq . shows that it applies qualitatively and semi - quantitatively for surface charge densities as high as @xmath180 . for the ice / liquid / silicon system no layer collapse is observed . this results from the fact that dispersion forces , which dominate at low layer thicknesses , are repulsive and stabilize the quasi - liquid layer . however , at impurity densities @xmath15 smaller than those considered in fig . [ fig_layerthicknesssilicon ] , the debye - hckel expression for @xmath181 generates an _ attractive _ force in the ice / liquid / silicon system with @xmath182 at sufficiently small layer thicknesses which results in a discontinuous drop in the layer thickness at a certain undercooling . the layer , however , does not collapse to zero thickness but rather is stabilized at a smaller thickness due to the repulsive dispersion forces which dominate for very thin quasi - liquid layers . in this article we do not investigate this phenomenon of layer discontinuity which occurs in a regime where both the @xmath1-dependence of @xmath39 , the peculiarities of the debye - hckel result , and other effects , have to be accounted for . of the quasi - liquid layer for the ice / liquid / silicon system ( @xmath183 ) as a function of the undercooling @xmath32 . as dispersion forces in this system are repulsive they stabilize thin quasi - liquid layers and hence no layer collapse is observed . the figure illustrates the analytical result obtained from eq . which states that for intermediate undercooling @xmath32 and @xmath15 below a transition value the layer thickness _ decreases _ when the impurity concentration _ increases_. for large @xmath15 the colligative effect dominates and @xmath1 increases when @xmath15 is increased . the transition value @xmath184 as well as the locus @xmath185 and @xmath186 where the interval with @xmath187 appears as derived from the figure are in good agreement with the analytical result ( values given in the figure ) . note the sigmoidal transition to small but finite film thickness for the lowest value of @xmath15 . ] having analyzed premelting in planar layered systems , we now return to the implicit equation ( eq . [ eq_implfl ] ) for the volume fraction of premelted liquid at a given undercooling @xmath118 that we introduced in section [ sec_implliquidfr ] . in addition to being implicit for @xmath0 the equation uses the thickness of the premelted layer , which itself is obtained from an equation which does not have an analytical solution ( eq . [ eq_filmnodim ] ) . accordingly , we do not possess an analytical expression for @xmath0 in the general case . [ [ low - surface - charge ] ] low surface charge + + + + + + + + + + + + + + + + + + however , where immobile surface charges can be neglected ( @xmath188 ) , eq . becomes more tractable . in the case where the matrix material has a negative hamaker constant ( @xmath189 ) , i.e. , dispersion forces are repulsive and support the formation of a premelted layer , eq . becomes @xmath190 where @xmath191 with @xmath192 and where @xmath193 is , up to a constant factor , the density of impurities @xmath101 in the fully melted system . obviously , eq . requires that @xmath194 must hold and thus , because @xmath195 , when @xmath196 eq . is not applicable . the case @xmath196 corresponds to an undercooling which is smaller than the depression of the bulk freezing point due to the colligative effect originating from the presence of the impurities . therefore , we must assume a liquid fraction @xmath197 when @xmath196 . while eq . does not posses a general analytical solution , it can be solved in the limit of low impurity concentration @xmath198 to yield @xmath199 with @xmath200 where @xmath201 note that the term @xmath202 is sufficient to obtain the correct leading contribution due to impurities in both limiting cases @xmath203 and @xmath204 . the factor @xmath205 becomes a constant in the limit @xmath148 where @xmath206 vanishes . the coefficient belonging to @xmath207 vanishes as @xmath208 in the limit @xmath139 and is thus subdominant to @xmath209 . in the case where dispersion forces are attractive ( @xmath210 ) the formation of a premelted layer requires an impurity concentration which is _ higher _ than the one which would be required to prevent the bulk liquid from freezing . hence , in this case all the liquid water in the matrix pores must originate from liquid veins which form due to curvature induced premelting . in contrast to the premelted layer with @xmath210 these veins have formed when @xmath211 , a temperature for which the premelted layers are still collapsed and do not contribute to the liquid fraction . thus , premelted layers do not actually form because at higher temperatures , beginning with the veins , the sample completely liquifies . therefore , the corresponding equation for @xmath0 is @xmath212 which has a simple analytical solution . again , we examine the leading order contributions in the limiting cases @xmath148 and @xmath139 to find @xmath213 where the latter relation is an extension of eq . which is undefined in the limit @xmath139 and @xmath214 . [ [ non - zero - surface - charge ] ] non - zero surface charge + + + + + + + + + + + + + + + + + + + + + + + in the general case with immobile surface charges ( @xmath215 ) we solve eq . numerically . in fig . [ fig_flquartz ] we show results for a matrix of fused quartz with a grain size of @xmath216 , which may serve as a model for frozen soils . the surface charge density is chosen to be @xmath180 which has the correct order of magnitude for mineral / water interfaces at high ph @xcite . the ice which fills the voids of the porous matrix is taken to be polycrystalline with a grain boundary density @xmath217 . a comparison of eqs . and shows that for the given @xmath93 the surface area provided by grain boundaries is by a factor @xmath218 larger than the surface area of the porous matrix . therefore , we expect the impurity concentration in the sample to be very low when grain boundaries are melted and , according to eq . , the maximal undercooling @xmath135 for which quasi - liquid interfacial water at the grain boundaries is stable should be low . an upper limit for @xmath135 can be calculated by assuming that all the impurities in the sample are located at the premelted grain boundaries . for the values of @xmath101 from fig . [ fig_flquartz ] we use @xmath219 to obtain a range @xmath220 and , with the aid of eq . this corresponds to a range in @xmath135 of @xmath221 . the range of @xmath135 uses the hamaker constant @xmath222 computed from complete dispersion force theory @xcite in the limit of thin layers with the data for the dielectric functions from elbaum and schick @xcite . including a typical surface charge density of @xmath223 for the grain boundary @xcite lowers these values to @xmath224 . hence , the grain boundaries are always collapsed to zero film thickness for the range of @xmath32 considered in the figure . a prominent feature of @xmath225 with @xmath226 in the fused quartz matrix is the discontinuity at @xmath227 where @xmath0 drops by almost an order of magnitude . this is the signature of the collapse of the premelted layer at the ice / fused quartz interface described in detail in section [ sec_reslayer ] . in the strongly undercooled , layer collapsed system , the only contribution to @xmath0 comes from curvature induced melting ; @xmath228 ( cf . eq . [ eq_fltot ] ) , which discontinuously increases at the undercooling where the premelted layer collapses . this is because , due to the efficient rejection of impurities by the ice lattice , the impurity concentration in the remaining liquid increases discontinuously when the liquid layers collapse . at low undercooling curvature melting becomes increasingly important exceeding the contribution from interfacial premelting @xmath229 . however , at larger undercooling , close to the location of layer collapse , curvature induced melting yields only a minute contribution to @xmath0 . the importance of curvature induced melting at low undercooling is reflected in eq . where the curvature term @xmath230 dominates for @xmath203 . in the opposite limit of large undercooling the analytical result eq . applies . the leading term @xmath231 originates from the colligative effect and curvature induced melting enters only through the subdominant term . hence , for large undercooling the principal contribution to the liquid fraction is the colligative effect . however , liquid water is located in regions of high curvature ; contact of matrix spheres , contact lines of grain boundaries and matrix spheres and liquid veins . figure [ fig_flquartz ] also shows @xmath0 in the same system with a higher impurity concentration ( @xmath232 ) . here the collapse of the liquid layer is moved to a lower temperature ( @xmath233 ) so that no discontinuity is observed in the range of undercooling considered in the figure . for water in a porous matrix of fused quartz with a grain size @xmath216 as a function of the undercooling @xmath32 . the result for the low impurity density @xmath226 ( with respect to the interstitial volume ) shows a discontinuity where the quasi - liquid layer between the ice and the matrix collapses . this is accompanied by an increase of the impurity concentration in the remaining ( curvature melting induced ) liquid which creates a discontinuous increase of @xmath234 . for the higher impurity density @xmath235 the discontinuity is not visible in the figure because it occurs at large undercooling ( @xmath233 ) . ] in fig . [ fig_flgoldvsquartz ] many of the same conditions as in fig . [ fig_flquartz ] are considered ( @xmath236 , @xmath180 , @xmath232 ) the principal difference being that the pore ice is single - crystalline ( @xmath237 ) . we find that @xmath103 in the fused quartz matrix is qualitatively the same as for the polycrystalline ice ( see fig . [ fig_flquartz ] ) in that it does not show liquid layer collapse on the scale of the plot for @xmath232 . when replacing the matrix with gold , which has stronger attractive dispersion forces at the ice - gold interface , we see a collapse of the premelted layer at @xmath238 . this is accompanied by , as in the case of layer collapse for the fused quartz matrix at @xmath239 ( fig . [ fig_flquartz ] ) , a discontinuous increase of the liquid fraction @xmath234 . as for the fused quartz matrix , for sufficiently low undercooling that the premelted layer has collapsed , the only contribution to @xmath103 comes from curvature induced melting ; @xmath240 . while , on the scale of the plot @xmath103 does not display a discontinuity at the point of layer collapse , there is a discontinuity @xmath241 which can be seen by expanding the full result in the limit of large impurity concentration @xmath198 @xmath242 where @xmath243 is the undercooling at layer collapse and @xmath244 the liquid fraction from interfacial melting at a temperature slightly above collapse @xmath245 , with @xmath246 . clearly , @xmath247 for @xmath248 . an important result seen in fig . [ fig_flgoldvsquartz ] is that although the hamaker constants for gold and fused quartz differ by almost two orders of magnitude , the results for @xmath0 are almost identical over a large range of @xmath32 ; the deviation increases from @xmath249 to @xmath250 at layer collapse and is @xmath251 for @xmath252 . this is because dispersion forces are only relevant for thin premelted layers and hence at temperatures slightly above layer collapse . however , when layer collapse occurs , according to eq . the high impurity concentration , combined with single - crystallinity ( i.e. small @xmath253 ) , insures that the debye length and the discontinuity of @xmath0 are small . therefore , the deviation between the results for the two different matrix materials remains small . the conclusions drawn from fig . [ fig_flgoldvsquartz ] are important to note for experimental determination of @xmath0 in different materials because they imply that even though the underlying microscopic behavior of two samples may fundamentally differ ( interfacial premelting vs. collapsed premelted layer ) this may not be resolved in the integrated / homogenized quantity @xmath0 . spatially resolved measurements may be required to detect characteristic transitions such as the collapse of interfacially premelted layers . for the gold matrix compared to the fused quartz matrix as a function of the undercooling @xmath32 for @xmath235 . importantly , although the hamaker constants for the two materials differ by almost two orders of magnitude , which results in a layer collapse for the gold matrix at @xmath238 , the curves for the total liquid fraction are almost identical . note that the curves for fused quartz have been shifted for clarity . in particular , the discontinuity of @xmath103 at the location of layer collapse is too small to be resolved in the figure , which results from the high impurity concentration , see eq . . the strength of dispersion forces is important only for thin premelted layers , i.e. in the immediate vicinity of layer collapse . the discontinuity for fused quartz is not visible in the figure because it occurs at large undercooling ( @xmath254 ) . ] [ [ effect - of - polycristallinity ] ] effect of polycristallinity + + + + + + + + + + + + + + + + + + + + + + + + + + + finally , we study the influence of the degree of polycrystallinity of the interstitial ice for the case where the porous matrix consists of silicon , which leads to dispersion forces that are repulsive ( @xmath88 ) and prevent the interfacially premelted layer from collapsing . as in the previous examples the grain size of the matrix is @xmath216 and as in fig . [ fig_flquartz ] the grain boundaries are collapsed . thus , fig . [ fig_flsilicon ] shows the curves for @xmath255 , the interfacial melting ( @xmath244 ) and curvature induced melting ( @xmath234 ) contributions for single - crystalline ( @xmath256 ) and polycrystalline ice ( @xmath257 ) . as expected , the polycrystallinity leads to a higher liquid fraction @xmath0 . in particular , for small undercooling @xmath32 , where curvature induced melting is enhanced and hence liquid veins greatly contribute , @xmath0 consists almost entirely of @xmath234 . the values of @xmath234 for the single crystal ice , for which contributions derive only from the contact points between matrix spheres , are always by at least an order of magnitude lower than for the polycrystalline ice case . this is not as obvious as it may first appear . due to the higher liquid fraction in polycrystalline ice the impurity concentration @xmath99 is smaller than for the single crystal case . thus , polycrystalline ice has a smaller ice - water radius of curvature ( see eq . [ eq_rcurvmelt ] ) . consequently , the volume of liquid contained for example at the contact point between two matrix spheres , is smaller for the polycrystalline case than for the single crystal case . it is the grain boundary density @xmath93 , which causes @xmath253 to be larger for the polycrystalline ice , that insures @xmath234 will be also larger , as indicated by eq . . in addition to the cases studied in section [ sec_reslayer ] , another interesting manifestation of the non - monotonic dependence of the premelted layer thickness on the impurity concentration for the systems with interfacial charges can be seen in fig . [ fig_flsilicon ] . although the impurity concentration @xmath99 and the impurity density @xmath15 are always larger in the single - crystal case , the liquid fractions @xmath244 , which from eq . are proportional to the corresponding layer thicknesses , intersect at an undercooling @xmath258 . thus for @xmath259 the larger impurity density of a single crystal sample leads to a smaller layer thickness than the polycrystalline case . it is only for @xmath260 that the usual increase of the layer thickness with increasing impurity concentration is observed . this complex behavior provides further evidence for the observation ( fig . [ fig_flgoldvsquartz ] ) , that measuring the integrated / homogenized quantity @xmath0 alone does not yield satisfactory information about the microscopic structure and behavior of the sample , in particular the rich and interesting temperature dependence of the thickness of the premelted layer in a frozen matrix containing ice with different degrees of polycrystallinity . by parity of reasoning , inclusion of just one or two microscopic phenomena in a theory that is intended to predict the effective medium behavior of a porous medium , will provide irrelevant and misleading predictions unless the proper asymptotic limits of a complete theory are assured . for a matrix consisting of silicon spheres with radius @xmath236 as a function of the undercooling @xmath32 . two different cases concerning the ice , which is contained in the matrix , are considered : polycrystalline ice ( @xmath217 ) and single - crystalline ice ( @xmath237 ) . the polycrystalline sample always has a higher liquid fraction , the difference with the single - crystalline sample being most pronounced at small undercooling . in this regime @xmath0 in the polycrystalline sample is almost entirely made up of @xmath261 , which mainly contains contributions from liquid veins in the ice . an interesting consequence of the non - monotonic dependence of the premelted layer thickness on the impurity density @xmath15 can be observed : although @xmath15 is always larger in the single - crystalline sample than in the polycrystalline sample , the curves for @xmath244 ( which are proportional to the layer thicknesses ) intersect . ] in this work we construct a theory for the premelting of ice in a porous matrix ( section [ sec_theory ] ) , which is modeled as a random close packing of monodisperse spheres . the theory includes premelting at the matrix - ice interface , grain boundary premelting , as well as curvature and impurity induced premelting . the latter includes premelting in liquid veins , at contact lines of the matrix with a grain boundary , and at the contact points between two matrix spheres . the matrix is characterized by the sphere radius @xmath65 and the matrix material , which enters the theory via the hamaker constant @xmath38 for the ice / liquid / matrix layered system . in addition to dispersion forces at the ice - matrix interface , which , depending on the matrix material , either stabilize or destabilize the premelted quasi - liquid layer , we include the effect of electrostatic forces at the interface due to adsorbed surface charges at the interface along the lines of our previous work ref . . the theory includes the effect of charged impurities in the premelted water both on the thickness of the premelted layers ( grain boundaries , ice - matrix interface ) as well as on the curvature induced melting . the main idea of the present approach is to make use of the small segregation coefficient of ice which leads to an enrichment of impurities in the premelted liquid ( making the impurity concentration a function of the fraction of premelted liquid ) and of the fact that regions of premelted liquid in a sample are all interconnected and therefore the impurity concentration is homogeneous . this approach allowed us to derive an implicit equation for the liquid fraction @xmath0 as a function of the undercooling @xmath32 , eq . , which we analyze for general parameters by considering limiting cases where analytical solutions are possible ( section [ sec_analysis ] ) . to understand each contribution we began by performing a careful study of interfacial melting , analyzing an implicit equation for the thickness @xmath1 of the premelted layer as a function of @xmath32 , eq . , for general parameters . in particular , we addressed the discontinuous collapse of the premelted layer under certain conditions and the non - monotonic dependence of @xmath1 on the impurity concentration ( cf . ref . ) . however , in the general case both eq . and eq . , the latter of which makes use of the former , have to be solved simultaneously through a numerical scheme . in so doing we chose physically relevant parameters of interest to a wide community of scientists . for the porous matrix we use @xmath216 and considered two materials ( fused quartz and gold ) that yield attractive dispersion forces at the ice / matrix interface , thereby inhibiting the formation of a premelted quasi - liquid layer , and one material ( silicon ) which enhances the premelted layer . the surface charge at the interface is taken to be @xmath180 . the impurity concentrations lie in the range of @xmath262 defined with respect to the interstitial volume , and we consider both single- and polycrystalline ice . we calculate @xmath0 for @xmath32 in the range of @xmath263 and obtain liquid fractions which range between @xmath264 and @xmath265 of the interstitial volume . for the parameters under consideration the grain boundaries are always of zero thickness ( i.e. , collapsed ) , and hence only the collapse of the quasi - liquid layer at the ice - matrix interface is observed in @xmath0 ( see fig [ fig_flquartz ] for the case of the fused quartz matrix ) . when the temperature of the sample is decreased below the point of layer collapse , only curvature induced premelting contributes to @xmath0 . while dispersion forces are too weak in the fused quartz system to significantly influence the results , in the case of the gold matrix dispersion forces are sufficiently strong to drive layer collapse , as opposed to the surface charge triggered layer collapse in the fused quartz system ( see fig . [ fig_flgoldvsquartz ] for a comparison of the two cases ) . an important conclusion from fig . [ fig_flgoldvsquartz ] is that for high impurity concentration and/or low degree of polycrystallinity the characteristic discontinuity of @xmath0 upon layer collapse is minute and most likely not resolved in experiments . thus , transitions that would be dramatic and easily detected at planar interfaces will be much more challenging to extract from measurements of the integrated / homogenized quantity @xmath0 . finally , we study the influence of the degree of polycrystallinity on @xmath0 for the case of the silicon matrix where , due to the repulsive dispersion forces , layer collapse does not occur ( fig . [ fig_flsilicon ] ) . the results reveal an interesting consequence of the non - monotonic dependence of @xmath1 on the impurity concentration . although the actual impurity concentration is higher in the single - crystalline sample than in the polycrystalline sample at all temperatures , the thickness of the premelted layer can be both larger and smaller in either case , depending on the temperature range . this peculiarity is obscured by the curvature induced contribution to @xmath0 providing further evidence that additional measurements are required @xmath0 to capture the physics controlling the sample microstructure . we view our approach as having provided a thorough theory of premelting in a porous medium which can be applied to a plethora of experimental situations and material systems . for the sake of brevity and clarity , we presented examples of specific materials systems ; gold , fused quartz , and silicon . however , we chose these materials because their dielectric properties , as reflected in their hamaker constants , span all the range of materials as diverse as sapphire , polystyrene , and polyvinylchloride @xcite . we did not explore the effect of varying the matrix sphere radius @xmath65 , its roughness or its surface charge density @xmath39 . the value @xmath216 was chosen with the example of frozen soils in mind , while @xmath182 provides a realistic order of magnitude as inferred from experimental data for mineral - water interfaces at large ph @xcite . however , not much is known experimentally about @xmath39 in the ice / liquid / matrix systems and the simplifying assumptions of symmetry ( same charge at the ice - liquid and liquid - matrix interfaces ) as well as independence of @xmath39 on the impurity concentration and layer thickness are not yet established experimentally on planar systems . nevertheless , given the importance of the impurity concentration dependent screening length for regulating the strength and range of the electrostatic force , the precise value of @xmath39 may be considered secondary . an important question for future research concerns the dynamical evolution of the porous matrix system in response to changes in the control parameters , in particular the sample undercooling @xmath32 . in order to guarantee a homogeneous impurity concentration throughout a sample , @xmath32 must be changed on a time scale that is slow compared to the typical impurity diffusion time . an even more ambitious question concerns impurity diffusion during layer collapse . the situation where the premelted quasi - liquid layer between the ice and the matrix collapses , which forces the impurities to migrate to remaining liquid regions with curvature induced melting constitutes a challenging task for further investigations . the authors have benefitted from discussions with j.g . dash , s.s.l . peppin , m. spannuth , and e. thomson at various stages of this work . we acknowledge the u.s . national science foundation grant no . opp0440841 , the department of energy grant no . de - fg02 - 05er15741 , the helmholtz gemeinschaft alliance `` planetary evolution and life '' , and yale university for generous support of this research . from the sphere center . b ) zoom on the curvature melting induced liquid `` vein '' . as @xmath266 the matrix surface can be considered as flat in the vicinity of the liquid vein . ] we consider a matrix sphere that is embedded at the interface between two ice grains . the grains are assumed to have a planar grain boundary and to be much larger than the matrix sphere . let the distance between the sphere center and the grain boundary be @xmath267 . curvature induced melting leads to a circular `` vein '' along the line where the grain boundary and the sphere intersect ( fig . [ fig_grainbmat]a ) . because @xmath52 is assumed to be much smaller than @xmath65 we can consider the matrix surface as being planar in the vicinity of the vein as shown in fig . [ fig_grainbmat]b . the angle between the grain boundary and the matrix sphere is @xmath268 using elementary geometry we obtain the surface areas @xmath269 and the total surface area @xmath270 accordingly , the volume of premelted water for the configuration in fig . [ fig_grainbmat ] is @xmath271 . taking all possible distances of the matrix sphere to the grain boundary into account yields an average volume @xmath272 of premelted liquid as @xmath273 now consider a random close packing with packing fraction @xmath66 ( the porous medium ) , which contains ice with two different crystallographic orientations being separated by a planar grain boundary . the volume fraction @xmath66 is also the surface fraction of the grain boundary which is cut out in the form of circles with radii between @xmath274 and @xmath65 due to the presence of the matrix spheres . given that the average surface area per intersecting sphere is @xmath275 a sample of volume @xmath276 with a grain boundary perpendicular to the @xmath277-axis contains @xmath278 intersection circles of matrix spheres with the grain boundary , if @xmath279 is sufficiently large . if the sample contains the total grain boundary surface @xmath280 this corresponds to a net surface @xmath281 and hence a density of intersection circles @xmath282 where @xmath93 is the grain boundary density with respect to the interstitial volume . the volume of premelted liquid along the intersection circles ( per unit volume of the sample ) is @xmath283 . or , relative to the interstitial volume : @xmath284

we prove there is no sentence in the monadic second - order language that characterises when a matroid is representable over at least one field , and no sentence that characterises when a matroid is-representable , for any infinite field . by way of contrast , because rota s conjecture is true , there is a sentence that characterises-representable matroids , for any finite field .

1412.8399

a matroid captures the notion of a discrete collection of points in space . sometimes these points can be assigned coordinates in a consistent way , and sometimes they can not . the problem of characterising when a matroid is _ representable _ has been the prime motivating force in matroid research since whitney s founding paper @xcite . plenty of effort has been invested in characterising matroid representability via excluded minors . less attention has been paid to the prospect of characterisating representability via axioms . perhaps this is because of vmos s well - known article @xcite , which has been interpreted as stating that no such characterisation is possible ( see @xcite ) . in @xcite , we pointed out that the possibility of characterising representable matroids in the language of whitney s axioms was still open ; that , in other words , it was still not proved that the missing axiom of matroid theory is lost forever " , _ contra _ vmos s title . we conjectured that in fact there was no such characterisation , and we made some partial progress towards resolving the conjecture by showing that it was impossible to characterise the class of representable matroids , or the class of matroids representable over an infinite field , using a logical language based on the rank function . however , that language imposed quite strong constraints on the form of quantification . in this article , we present a language with no such constraints , and we prove that it is impossible to characterise representability , or representability over an infinite field , in this more natural language . the language that we develop is a form of monadic second - order logic , @xmath0 . as we show in , @xmath0 is expressive enough to state the matroid axioms , and to state when a matroid contains an isomorphic copy of a fixed minor . this means that any minor - closed class of matroids can be characterised with a sentence in @xmath0 , as long as it has a finite number of excluded minors . in particular , since rota s conjecture has been positively resolved by geelen , gerards , and whittle ( see @xcite ) , it follows that the class of @xmath2-representable matroids can be characterised by a sentence in @xmath0 , whenever @xmath2 is a finite field . our main results show that this is not the case for the class of matroids that are representable over at least one field , nor for the class of @xmath1-representable matroids , when @xmath1 is any infinite field . we say that a matroid is _ representable _ if it is representable over at least one field . [ jungle ] there is no sentence , @xmath3 , in @xmath0 , such that a matroid is representable if and only if it satisfies @xmath3 . [ proton ] let @xmath1 be any infinite field . there is no sentence , @xmath4 , in @xmath0 , such that a matroid is @xmath1-representable if and only if it satisfies @xmath4 . these theorems may seem stronger than those in @xcite , but in fact the results are independent of each other . the logical language used in @xcite had constraints on quantification , unlike @xmath0 , but it also had access to the rank function , and to the arithmetic of the integers , while @xmath0 does not . follow easily from the next two lemmas . let @xmath5 be a positive integer . define @xmath6 to be @xmath7 and recursively define @xmath8 to be @xmath9 . let @xmath10 be @xmath11 . thus @xmath10 is @xmath12 where the tower of exponents has height @xmath13 . recall that if @xmath14 and @xmath15 are matroids with disjoint ground sets , then @xmath16 is the _ direct sum _ of @xmath14 and @xmath15 . [ iguana ] let @xmath5 be a positive integer . there is a partition of the set of matroids into blocks , @xmath17 , such that @xmath18 and the following property holds : for any sentence , @xmath3 , in @xmath19 with @xmath5 variables , whenever @xmath14 and @xmath15 are matroids that belong to the same block , then , for any matroid , @xmath20 , either @xmath21 and @xmath22 both satisfy @xmath3 , or neither satisfies @xmath3 . by using , we can easily deduce . assume that there is a sentence , @xmath3 , in @xmath0 , that characterises representable matroids . let @xmath5 be the number of variables in @xmath3 , and let @xmath17 be the partition provided by . because there are infinitely many prime numbers , we can let @xmath23 and @xmath24 be distinct primes such that the projective planes @xmath25 and @xmath26 belong to the same block . now , setting @xmath20 to be @xmath14 , we see that either @xmath27 and @xmath28 both satisfy @xmath3 , or neither satisfies @xmath3 . in either case we have a contradiction , since the first of these direct sums is a representable matroid , and the second is not . let @xmath5 be a positive integer . we say that matroids @xmath14 and @xmath15 are _ equivalent _ if , for any sentence @xmath3 with @xmath5 variables , and any matroid @xmath20 , either both @xmath21 and @xmath22 satisfy @xmath3 , or neither do . shows that this equivalence relation has finite index for any @xmath5 . in this way , the proof of is reminiscent of the myhill - nerode characterisation of regular languages ( see @xcite or ( * ? ? ? * section 6.1 ) ) . by way of contrast , the theorem in @xcite used a proof technique that was essentially an ehrenfeucht - frass game ( see ( * ? ? ? * section 2.2 ) ) . for every @xmath5 , there is a partition of matroids into finitely many blocks , such that two matroids from the same block satisfy exactly the same @xmath5-variable sentences . this follows from by letting @xmath20 be the empty matroid @xmath29 , or from the fact that there are only finitely many _ rank-@xmath5 @xmath30-types _ ( see ( * ? ? ? * section 3.4 ) for an explanation ) . for the next lemma , we construct families of matroids by using gain graphs . loosely , a gain graph is a graph equipped with edge labels that come from a group . for each such graph , there is a gain - graphic matroid , whose ground set is the edge set of the graph . in , we introduce two families of gain graphs . hoop matroid _ is the gain - graphic matroid that corresponds to a gain graph of the form @xmath31 . here @xmath1 is a field , @xmath32 is an integer , and @xmath33 is a non - zero element of @xmath1 with order greater than @xmath34 . the edge labels of the gain graph @xmath31 come from the multiplicative group of @xmath1 . we also introduce a family of gain graphs , @xmath35 . in this case @xmath36 is a field , @xmath37 is an integer , and @xmath38 is a non - zero element with order greater than @xmath39 . loop matroid _ is gain - graphic , corresponding to a gain graph of the form @xmath35 . if @xmath14 is a hoop matroid , corresponding to the gain graph @xmath31 , and @xmath20 is the loop matroid of @xmath35 , then @xmath14 and @xmath20 can be glued together along a rank-@xmath40 flat , to form an amalgam of the two matroids . the resulting matroid is denoted @xmath41 . in the case that @xmath42 and @xmath43 , where the order of @xmath33 is greater than @xmath44 , then both @xmath14 and @xmath20 can be represented over @xmath1 , but @xmath41 can be represented over @xmath1 if and only if @xmath45 . this means that the following quickly leads to a proof of , with hoop matroids playing the same role that projective planes did in the proof of . details of the proof will be left until the end of the paper . if @xmath5 is a positive integer , then @xmath46 is @xmath47 , and @xmath48 is @xmath49 . define @xmath50 to be @xmath51 [ sentry ] let @xmath5 be a positive integer . there is a partition of the set of hoop matroids into blocks , @xmath52 , such that @xmath53 and the following property holds : for any sentence , @xmath3 , in @xmath19 with @xmath5 variables , whenever @xmath14 and @xmath15 are hoop matroids that belong to the same block and @xmath20 is a loop matroid , when @xmath41 and @xmath54 are defined , either both satisfy @xmath3 , or neither satisfies @xmath3 . also implies the following ( unsurprising ) fact : using @xmath0 to characterise increasingly large finite fields requires increasingly large sentences . [ casava ] there is no integer , @xmath55 , such that , for every finite field , @xmath2 , there is a sentence , @xmath56 , in @xmath0containing at most @xmath55 variables , with the property that a matroid is @xmath2-representable if and only if it satisfies @xmath56 . in fact is sufficient to prove both and , since , if @xmath41 is not representable over the field @xmath1 , then it is not representable over any field ( ) . however , we feel that is more intuitive , and also interesting in its own right , so we prefer to prove that lemma , and then note the easy changes required to produce a proof of . for all matroid essentials we refer to oxley @xcite . we construct the logical language @xmath0 for making statements about matroids . it includes an unlimited supply of variables , @xmath57 along with the binary predicate , @xmath58 , the unary predicates , @xmath59 and @xmath60 , as well as the standard connectives @xmath61 and @xmath62 , and the quantifiers @xmath63 and @xmath64 . the variables will be interpreted as representing subsets of a ground set . thus @xmath65 is interpreted as the statement that @xmath66 is a subset of @xmath67 and @xmath68 is interpreted as the statement that @xmath66 has cardinality equal to one . moreover , @xmath69 is interpreted as the statement that @xmath66 is independent in a given matroid . we recursively define formulas in @xmath0 , and simultaneously define their sets of variables . the following statements define expressions known as _ atomic formulas_. 1 . @xmath70 is an atomic formula , for any variables @xmath66 and @xmath67 , and @xmath71 . @xmath68 is an atomic formula , for any variable @xmath66 , and @xmath72 . 3 . @xmath69 is an atomic formula , for any variable @xmath66 , and @xmath73 . a _ formula _ is an expression generated by a finite application of the following rules . every formula has an associated set of variables and _ free variables _ : 1 . every atomic formula , @xmath3 , is a formula , and @xmath74 . if @xmath3 is a formula , then @xmath75 is a formula , and @xmath76 , while @xmath77 . 3 . if @xmath78 and @xmath79 are formulas , and @xmath80 for @xmath81 , then @xmath82 is a formula , and @xmath83 , while @xmath84 . if @xmath3 is a formula and @xmath85 , then @xmath86 and @xmath87 are formulas , and @xmath88 , while @xmath89 . a variable in @xmath90 is _ free _ if it is in @xmath91 , and _ bound _ otherwise . a formula is _ quantifier - free _ if all of its variables are free , and is a _ sentence _ if all its variables are bound . rule ( 3 ) insists that no variable can be free in one of @xmath78 and @xmath79 and bound in the other , if @xmath82 is to be a formula . we can overcome this constraint if necessary by relabelling the bound variables in a formula . we use @xmath92 as shorthand for @xmath93 . we use the shorthand @xmath94 to mean @xmath95 and we use @xmath96 to mean @xmath97 . likewise , we use @xmath98 to mean @xmath99 . next we spend some time illustrating the expressive power of @xmath0 . it is powerful enough to state the axioms for matroids , and to characterise when a matroid contains a fixed minor . if @xmath100 is an integer , we use @xmath101 as shorthand for the statement @xmath102 thus @xmath101 is interpreted as being true if and only if @xmath103 is equal to @xmath104 . ( the variable @xmath105 stands for some variable not equal to any of @xmath106 . ) now we let the predicate @xmath107 stand for @xmath108 thus @xmath107 is interpreted as the statement that @xmath66 is maximal with respect to satisfying the predicate @xmath60 . the family of sets on which the predicate @xmath60 returns ` true ' is the collection of independent sets of a matroid if and only the following sentences are satisfied : 1 . @xmath109 2 . @xmath110 3 . @xmath111 + @xmath112 + @xmath113 next we let @xmath55 be a fixed matroid on the ground set @xmath114 , with @xmath115 as its collection of independent sets . let @xmath116 be the set of dependent subsets of @xmath55 . a matroid has a minor isomorphic to @xmath55 if and only if it contains distinct elements @xmath117 , and an independent set , @xmath118 , such that @xmath119 , and @xmath120 is independent precisely when @xmath121 is an independent set of @xmath55 . in this case , @xmath55 is isomorphic to the minor produced by contracting @xmath118 and restricting to the set @xmath122 . thus we see that a matroid has a minor isomorphic to @xmath55 if and only if it satisfies the following sentence : @xmath123 throughout this section we let @xmath5 be a fixed positive integer . recall that @xmath6 is @xmath124 , and that @xmath8 is @xmath9 , while @xmath10 is @xmath11 . a _ registry _ is a @xmath125 matrix with rows indexed by @xmath60 , @xmath59 , and @xmath126 , and columns indexed by @xmath126 . an entry in row @xmath60 or in row @xmath66 must be ` t ' or ` f ' . an entry in row indexed by @xmath59 is either ` @xmath127 ' , ` @xmath128 ' , or ` @xmath129 ' . it follows that there are at most @xmath130 possible registries . an _ @xmath131-stacked matroid _ is a tuple , @xmath132 , where @xmath14 is a matroid , @xmath133 , and each @xmath134 is a subset of the ground set of @xmath14 . note that each matroid , @xmath14 , is also a @xmath30-stacked matroid , @xmath135 . we define a _ depth-@xmath30 tree _ to be a registry . recursively , a _ depth-@xmath136 tree _ is a non - empty set of depth-@xmath137 trees . if @xmath138 is a depth-@xmath136 tree , then the depth-@xmath137 trees it contains are its _ children_. note that we can use an easy inductive argument to show there are no more than @xmath8 depth-@xmath136 trees , and hence no more than @xmath10 depth-@xmath5 trees . to each @xmath131-stacked matroid , @xmath139 , we associate a depth-@xmath140 tree , @xmath141 . thus each matroid is associated with a depth-@xmath5 tree . a @xmath5-stacked matroid , @xmath142 corresponds to a depth-@xmath30 tree , which is a registry . informally , we think of this registry as containing information about the truth values of the statements @xmath143 , @xmath144 , and @xmath145 . if @xmath139 is an @xmath131-stacked matroid , then the _ children _ of @xmath146 are the @xmath147-stacked matroids of the form @xmath148 , where @xmath149 ranges over all subsets of the ground set of @xmath14 . we are going to define @xmath141 so that its children are exactly the trees corresponding to the children of @xmath146 . so , for example , if @xmath150 , then @xmath141 contains information about the truth values of the atomic formulas for all possible assignments of the variable @xmath151 . it follows inductively that if @xmath152 is a @xmath30-stacked matroid , the depth-@xmath5 tree @xmath141 contains information about the truth value of all possible atomic formulas , for all possible assignments of the variables @xmath126 . formally , we first define @xmath141 in the case that @xmath153 . then @xmath141 is a depth-@xmath30 tree ( a registry ) . the entry of @xmath141 in row @xmath60 and column @xmath66 is ` t ' if and only if @xmath134 is independent in @xmath14 . the entry in row @xmath66 and column @xmath67 is ` t ' if and only if @xmath145 . finally , the entry in row @xmath59 and column @xmath66 is ` @xmath127 ' , ` @xmath128 ' , or ` @xmath129 ' , according to whether @xmath154 is less than , equal to , or greater than one . when @xmath155 , we recursively define @xmath141 to be the set @xmath156 in other words , we let @xmath149 range over all subsets of the ground set @xmath157 , while keeping @xmath158 fixed . this produces a collection of @xmath147-stacked matroids , each of which is thus associated with a depth-@xmath159 tree . we define the set of these trees to be the depth-@xmath140 tree associated with @xmath160 . now let @xmath161 be the set of @xmath125 matrices with rows indexed by @xmath60 , @xmath59 , and @xmath126 , and columns indexed by @xmath126 , where each entry is either ` t ' or ` f ' . thus a matrix in @xmath161 records the truth values of the predicates @xmath69 , @xmath68 , and @xmath70 for a particular set - system . if @xmath162 is a quantifier - free formula in @xmath0 , then @xmath162 is logically equivalent to a boolean expression in the predicates @xmath68 , @xmath69 , and @xmath70 . hence we can consider any such quantifier - free formula to be a function from @xmath161 to @xmath163 . next we define an operation that takes pairs of registries into @xmath161 . the _ sum _ of two registries is produced by simply summing the two matrices , entry by entry , under the following summation rules : @xmath164 notice that we have constructed these summing operations so that , if @xmath165 and @xmath166 , where @xmath14 and @xmath15 have disjoint ground sets , then in the sum of @xmath141 and @xmath167 , the entry in column @xmath67 , and row @xmath60 , @xmath59 , or @xmath66 is ` t ' if and only if , respectively , @xmath168 is independent in @xmath16 , @xmath169 , or @xmath170 . let @xmath3 be a formula in @xmath0 using @xmath5 variables . by relabelling , we can assume that @xmath171 . let @xmath131 be the number of free variables in @xmath3 . by relabelling we can assume that @xmath172 . we can also assume that @xmath3 is in _ prenex normal form_. that is , @xmath173 where each @xmath174 is either @xmath63 or @xmath64 , and @xmath162 is a quantifier - free formula in @xmath0 with @xmath175 . we now define what it means for a pair of depth-@xmath140-trees , @xmath176 and @xmath177 , to be _ compatible _ ( relative to @xmath3 ) . in the case that @xmath153 , then @xmath3 is quantifier - free , and @xmath176 and @xmath177 are a pair of registries . we say that this pair is compatible if and only if their sum is taken to ` t ' by the quantifier - free formula @xmath3 . ( remembering that @xmath3 is a function from @xmath161 to @xmath163 . ) now assume that @xmath155 . thus @xmath176 and @xmath177 are non - empty sets of depth-@xmath159 trees . first consider the case that @xmath178 . then @xmath176 and @xmath177 are compatible if and only if @xmath176 contains a depth-@xmath159 tree , @xmath179 , and @xmath177 contains a depth-@xmath159 tree , @xmath180 , such that @xmath179 and @xmath180 are compatible relative to the formula @xmath181 similarly , if @xmath182 , we say that @xmath176 and @xmath177 are compatible if and only if @xmath179 and @xmath180 are compatible relative to @xmath183 for every depth-@xmath159 tree , @xmath179 , contained in @xmath176 , and every depth-@xmath159 tree , @xmath180 , contained in @xmath177 . an @xmath131-stacked matroid , @xmath132 , _ satisfies _ @xmath3 if @xmath14 satisfies @xmath3 when each free variable , @xmath184 , is interpreted as @xmath134 . [ yogurt ] let @xmath139 and @xmath185 be @xmath131-stacked matroids , where @xmath14 and @xmath15 have disjoint ground sets . then the @xmath131-stacked matroid @xmath186 satisfies @xmath3 if and only if the trees , @xmath141 and @xmath167 , are compatible relative to @xmath3 . note that @xmath141 and @xmath167 are depth-@xmath140 trees . first assume that @xmath153 , so that @xmath3 is a quantifier - free formula . then @xmath141 and @xmath167 are depth-@xmath30 trees , that is , registries . let @xmath187 be the matrix in @xmath161 that is produced by summing @xmath141 and @xmath167 . as we have noted , the entry of @xmath187 in row @xmath66 and column @xmath67 is ` t ' if and only if @xmath145 and @xmath188 ; that is , if and only if @xmath170 . similarly , the entry of @xmath187 in row @xmath60 and column @xmath66 is ` t ' if and only if @xmath134 is independent in @xmath14 and @xmath189 is independent in @xmath15 ; that is , if and only if @xmath190 is independent in @xmath16 . finally , the entry in row @xmath59 and column @xmath66 is ` t ' if and only if @xmath191 . thus @xmath187 encodes the truth values of all the statements @xmath70 , @xmath69 , and @xmath68 , when each @xmath66 is interpreted as the subset @xmath190 of the matroid @xmath16 . now @xmath141 and @xmath167 are compatible relative to @xmath3 if and only if @xmath187 is taken to ` t ' by the quantifier - free formula @xmath3 , which is true if and only if @xmath186 satisfies @xmath3 . this establishes the base case of the proof . next we assume that @xmath155 . consider the case that @xmath178 . then @xmath141 and @xmath167 are compatible relative to @xmath3 if and only if there are depth-@xmath159 trees @xmath192 and @xmath193 such that @xmath179 and @xmath180 are compatible relative to @xmath181 now @xmath179 is equal to @xmath194 for some choice of subset @xmath195 , and @xmath180 is equal to @xmath196 for some choice of @xmath197 . by induction , @xmath179 and @xmath180 are compatible relative to @xmath198 if and only if @xmath199 satisfies @xmath198 . but this is true if and only if there exists a subset @xmath200 , where @xmath195 and @xmath197 , such that @xmath198 is true when each @xmath66 is interpreted as @xmath190 , for @xmath201 . in other words , if and only if @xmath202 is true . since this formula is equal to @xmath3 , this completes the proof when @xmath178 . assume that @xmath182 . then @xmath141 and @xmath167 are compatible relative to @xmath3 if and only if , for all depth-@xmath159 trees @xmath192 and @xmath193 , @xmath179 and @xmath180 are compatible relative to @xmath198 . by induction , this is true if and only if , for every choice of subset @xmath195 , and for every choice of @xmath197 , the @xmath147-stacked matroid , @xmath203 satisfies @xmath198 . this , in turn , is true if and only if @xmath198 is true when each @xmath66 is interpreted as @xmath190 for @xmath201 , for every subset @xmath200 such that @xmath195 and @xmath197 . in other words , if and only if @xmath204 is true . this completes the proof . for any two labelled matroids , @xmath14 and @xmath15 , let @xmath152 and @xmath205 be the corresponding @xmath30-stacked matroids . we define @xmath14 and @xmath15 to be _ equivalent _ if the depth-@xmath5 trees , @xmath141 and @xmath167 , are identical . this equivalence relation partitions labelled matroids into @xmath206 blocks , @xmath17 , where @xmath206 is at most the number of depth-@xmath5 trees . thus @xmath18 . now let @xmath3 be a sentence in @xmath0 using @xmath5 variables . by relabelling and using standard techniques , we assume that @xmath207 . let @xmath14 and @xmath15 be matroids that belong to the same block , and let @xmath20 be an arbitrary matroid . let @xmath208 be the corresponding stacked matroid . now says that @xmath21 satisfies @xmath3 if and only if @xmath209 is compatible with @xmath141 relative to @xmath3 . as @xmath210 , this is true if and only if @xmath209 is compatible with @xmath167 , which is true if and only if @xmath22 satisfies @xmath3 . this completes the proof of . this section is dedicated to proving . the main ideas required here are essentially identical to those in , so we omit many details . we again let @xmath5 be a fixed positive integer ; a registry is again a @xmath125 matrix with rows indexed by @xmath60 , @xmath59 , and @xmath126 , and columns indexed by @xmath126 . as before , an entry in row @xmath66 is either ` t ' or ` f ' , and an entry in row @xmath59 is either ` @xmath127 ' , ` @xmath128 ' , or ` @xmath129 ' . however , in this case , an entry in row @xmath60 is either ` f ' , or an ordered pair from the set @xmath510 it follows that there are at most @xmath511 possible registries . a depth-@xmath30 tree is a registry , and a depth-@xmath136 tree is a non - empty set of depth-@xmath137 trees , so there are no more than @xmath50 depth-@xmath5 trees . assume that @xmath14 is either a hoop matroid , @xmath339 , or a loop matroid , @xmath365 , for some choices of fields , @xmath1 and @xmath36 , integers @xmath448 , and elements @xmath512 and @xmath513 with orders greater than @xmath34 and @xmath514 respectively . in the case that @xmath14 satisfies these conditions , we define an @xmath131-stacked matroid to be a tuple , @xmath132 , where @xmath133 and each @xmath134 is a subset of the ground set of @xmath14 . if @xmath165 is a @xmath5-stacked matroid , then @xmath141 is a registry . the entry in row @xmath66 and column @xmath67 of this registry is ` t ' if and only if @xmath145 . the entry in row @xmath59 and column @xmath66 is ` @xmath127 ' , ` @xmath128 ' , or ` @xmath129 ' , according to whether @xmath154 is less than , equal to , or greater than one . the entries in row @xmath60 are more complicated . let @xmath219 be @xmath377 . consider the entry in row @xmath60 and column @xmath66 . if @xmath134 is dependent in @xmath14 , then the entry is ` f ' . assume that @xmath134 is independent . then the entry is an ordered pair from @xmath515 . the first component in this pair is ` t ' if and only if @xmath516 . the second component is determined by the interaction between @xmath517 and @xmath219 . if @xmath518 , then the entry is @xmath219 . assume that @xmath517 contains at most one element of @xmath219 . if it contains , respectively , @xmath327 , @xmath329 , @xmath278 , @xmath251 , or @xmath334 , then the entry in @xmath141 is , respectively , @xmath327 , @xmath329 , @xmath278 , @xmath251 , or @xmath334 . now assume that @xmath519 . if @xmath520 , then we set the entry in @xmath141 to be @xmath521 . otherwise we set it to be @xmath522 . when @xmath155 , and @xmath139 is an @xmath131-stacked matroid , we define @xmath141 to be @xmath523 . the set of matrices , @xmath161 , is exactly as described in , so that a quantifier - free formula in @xmath5 variables is a function from @xmath161 to @xmath163 . we produce the sum of two registries as before , by summing entry by entry . the rule for summing in a row labelled by @xmath59 or by @xmath66 is the same as before , but we need to define the rule for summing entries in the @xmath60 row . let @xmath524 and @xmath525 be elements of @xmath526 we will describe @xmath187 , the sum of @xmath524 and @xmath525 . if either @xmath524 or @xmath525 is ` f ' , then @xmath187 is ` f ' . assume that @xmath524 and @xmath525 are both ordered pairs from @xmath515 . if the first component of @xmath524 ( respectively , @xmath525 ) is ` t ' , then @xmath187 is ` f ' if and only if the second entry in @xmath525 ( respectively , @xmath524 ) is not equal to ` skew ' . now assume that the first components of @xmath524 and @xmath525 are both equal to ` f ' . then @xmath187 is ` f ' if and only if @xmath524 and @xmath525 have identical second components , and this component is @xmath327 , @xmath329 , @xmath278 , @xmath251 , or @xmath334 . the summing operation we have just described has been designed to align with the conditions in . [ window ] assume that @xmath14 and @xmath15 are , respectively , @xmath339 and @xmath365 , and let @xmath165 and @xmath527 be @xmath5-stacked matroids . then @xmath141 and @xmath167 are registries . in the sum of these registries , the entry in row @xmath60 and column @xmath66 will be if and only if @xmath190 is independent in @xmath372 . this is immediate from the definition of @xmath141 and @xmath167 , the definition of the summing operation , and . now we assume @xmath528 is a formula in prenex form with @xmath5 variables , @xmath131 of which are free . thus @xmath162 is a quantifier - free formula . the definition of two depth-@xmath140 trees being compatible relative to @xmath3 is identical to that in , except that we use the new summing operation for registries when we define the compatability of depth-@xmath30 trees . we can now use to prove the following in exactly the same way that we proved . [ needle ] assume that @xmath14 and @xmath15 are , respectively , @xmath339 and @xmath365 , and let @xmath139 and @xmath529 be @xmath131-stacked matroids . the @xmath131-stacked matroid @xmath530 satisfies @xmath3 if and only if the trees , @xmath141 and @xmath167 , are compatible relative to @xmath3 . follows immediately . we can now complete the proofs of and . let @xmath1 be an infinite field . assume that @xmath4 is a sentence in @xmath0characterising @xmath1-representable matroids . let @xmath5 be @xmath531 . we claim that @xmath1 contains non - zero elements with arbitrarily large order . let @xmath532 be the order of the non - zero element @xmath33 . if @xmath532 is bounded above by the integer @xmath533 , for every non - zero @xmath33 , then every non - zero element of @xmath1 is a root of the polynomial @xmath534 . since there are only finitely many such roots , @xmath1 is finite . this contradiction proves our claim . we let @xmath52 be the blocks provided by . choose the element @xmath512 so that there are at least @xmath535 integers , @xmath34 , satisfying @xmath32 and @xmath536 . then there are two distinct integers , @xmath34 and @xmath357 , satisfying these constraints , such that @xmath366 and @xmath537 are contained in the same block . we let @xmath20 be @xmath376 . then implies that @xmath4 is satisfied by either both of @xmath41 and @xmath54 , or neither . however , the first of these amalgams is @xmath1-representable by , and the second is not representable over any field at all , by . this contradiction completes the proof of the . let us assume that @xmath55 is an integer such that for any finite field , @xmath2 , there is a sentence , @xmath56 , in @xmath0 such that @xmath538 , and a matroid is @xmath2-representable if and only if it satisfies @xmath56 . recall that if @xmath2 is a finite field , then the multiplicative group of @xmath2 has an element of order @xmath539 . choose @xmath2 so that there are least @xmath540 integers , @xmath34 , satisfying @xmath32 and @xmath541 . let @xmath33 be a generator of the multiplicative group of @xmath2 , and let @xmath56 be a sentence with @xmath542 variables that characterises @xmath2-representability . let @xmath52 be the blocks provided by . as @xmath543 , there are distinct integers , @xmath34 and @xmath357 , such that @xmath448 and @xmath544 , and @xmath366 and @xmath537 belong to the same block . now we obtain a contradiction from exactly as before . let @xmath211 and @xmath212 be simple matroids with ground sets @xmath213 and @xmath214 , rank functions @xmath215 and @xmath216 , and closure operators @xmath217 and @xmath218 . let @xmath219 be @xmath220 , where we assume that @xmath221 . a matroid , @xmath14 , on the ground set @xmath222 is an _ amalgam _ of @xmath211 and @xmath212 if @xmath223 and @xmath224 . if @xmath225 is a modular matroid ( meaning @xmath226 for every pair of flats , @xmath227 and @xmath228 , in @xmath225 ) , then ( * ? ? ? * theorem 11.4.10 ) implies that @xmath229 is the rank function of an amalgam of @xmath211 and @xmath212 , known as the _ proper amalgam_. we denote this amalgam by @xmath230 . henceforth , we consider only the case that @xmath231 . since every rank-@xmath40 uniform matroid is modular , this means that @xmath225 is modular , so that @xmath230 is defined . [ jackal ] assume that @xmath232 is a simple matroid with ground set @xmath233 , rank function @xmath234 , and closure operator @xmath235 , for @xmath236 . let @xmath237 , where @xmath221 and @xmath231 . let @xmath105 be a subset of @xmath222 . if @xmath238 is dependent in @xmath211 , or if @xmath239 is dependent in @xmath212 , then @xmath105 is dependent in @xmath230 . if @xmath238 is independent in @xmath211 and @xmath239 is independent in @xmath212 , then @xmath105 is dependent in @xmath230 if and only if 1 . @xmath240 and @xmath241 , 2 . @xmath242 and @xmath243 , or 3 . there is an element @xmath244 such that @xmath245 . if @xmath238 is dependent in @xmath211 , then @xmath238 is dependent in @xmath230 , since @xmath246 . by symmetry , @xmath105 is dependent in @xmath230 if @xmath238 is dependent in @xmath211 or if @xmath239 is dependent in @xmath212 . henceforth we assume that @xmath238 is independent in @xmath211 and @xmath239 is independent in @xmath212 . assume statement ( i ) holds . let @xmath247 be @xmath248 . then @xmath249 so by , the rank of @xmath105 in @xmath230 is less than @xmath250 , as desired . by symmetric arguments , we see that if ( i ) or ( ii ) holds , then @xmath105 is dependent in @xmath230 . next we assume that ( iii ) holds . since @xmath238 contains no circuits of @xmath211 it follows that @xmath251 is not in @xmath105 . if @xmath252 contains distinct elements , @xmath253 and @xmath254 , then by performing circuit exchange on @xmath255 and a circuit contained in @xmath256 that contains @xmath251 , we obtain a circuit of @xmath211 contained in @xmath238 . this contradiction means that @xmath257 . let @xmath247 be @xmath258 . then @xmath259 again we see that @xmath105 is dependent in @xmath230 , and this completes the proof of the ` if ' direction . for the ` only if ' direction , we assume that @xmath105 is dependent in @xmath230 . as @xmath238 is independent in @xmath211 and @xmath239 is independent in @xmath212 , it follows that @xmath105 is contained in neither @xmath213 nor @xmath214 . there is some set @xmath247 such that @xmath260 and @xmath261 . assume that amongst all such sets , @xmath247 has been chosen so that it is as small as possible . if @xmath251 is an element in @xmath262 , then we could replace @xmath247 with @xmath263 . therefore no such element exists . by symmetry it follows that @xmath264 . if @xmath265 , then @xmath266 is independent in @xmath211 and @xmath267 is independent in @xmath212 , so @xmath268 . this contradiction means that there is an element , @xmath251 , in @xmath269 . the minimality of @xmath247 means that @xmath270 it follows that @xmath251 is in @xmath271 and @xmath272 , but not @xmath273 . we combine the observations in this paragraph to deduce that @xmath274 . assume that @xmath275 , so that @xmath276 and @xmath277 . let @xmath278 be the element in @xmath279 . since @xmath280 contains @xmath278 and @xmath251 , it contains @xmath219 . as @xmath251 is in @xmath281 , it follows that @xmath282 . therefore statement ( i ) holds . now we assume that @xmath283 . if @xmath284 , then @xmath277 , and @xmath251 is in @xmath285 so statement ( iii ) holds . therefore we assume that @xmath286 , and hence @xmath287 . earlier statements imply that @xmath288 if @xmath238 is a flat of @xmath211 , and @xmath239 is a flat of @xmath212 , then @xmath289 . similarly , @xmath290 . but this means that @xmath291 which is a contradiction . now , by using symmetry , we can assume that @xmath251 is in @xmath292 . this means that @xmath293 , and hence @xmath219 , is contained in @xmath292 . also , @xmath294 so statement ( i ) holds , and the proof is complete . next we introduce two families of matroids via gain graphs . if @xmath295 is an undirected graph ( possibly containing loops and multiple edges ) with edge set @xmath296 and vertex set @xmath297 , then @xmath298 is the following subset of @xmath299 : @xmath300 a _ gain graph _ ( over the group @xmath301 ) is a pair @xmath302 , where @xmath295 is a graph , and @xmath303 is a function from @xmath298 to @xmath301 , such that @xmath304 for every non - loop edge @xmath305 with end - vertices @xmath253 and @xmath254 . we say that @xmath303 is a _ gain function_. if @xmath306 is a cycle of @xmath295 , where @xmath307 , then @xmath308 is defined to be @xmath309 note that , in general , @xmath301 may be nonabelian , and the value of @xmath308 depends on the choice of starting point and orientation for @xmath310 ; however , if @xmath308 is equal to the identity of @xmath301 , then this equality will hold no matter which starting point and orientation we choose . in this case , we say that @xmath310 is _ balanced_. a cycle that is not balanced is _ unbalanced_. the _ gain - graphic _ matroid @xmath311 has the edge set of @xmath295 as its ground set . the circuits of @xmath311 are exactly the edge sets of balanced cycles , along with the minimal edge sets that induce connected subgraphs containing at least two unbalanced cycles and no balanced cycles . any such subgraph is either a _ theta graph _ , or a _ the former consists of two vertices joined by three internally - disjoint paths ; the latter is two edge - disjoint cycles joined by a single path , which intersects the cycles only in its end - vertices . note that this path may have length zero , in which case the two cycles share a single vertex . assume that @xmath302 is a gain graph , where @xmath303 takes @xmath298 to the multiplicative group of a field , @xmath1 . let @xmath312 and @xmath313 be orderings of the vertex and edge sets of @xmath295 . we define a matrix , @xmath314 , with entries from @xmath1 . the rows of @xmath314 are labelled by @xmath312 and the columns are labelled by @xmath313 . if @xmath315 is a balanced loop , then column @xmath315 contains all zeroes . if @xmath315 is an unbalanced loop , let @xmath316 be the vertex incident with @xmath315 ; then column @xmath315 contains a @xmath317 in row @xmath316 , and zeroes in every other row . now assume that @xmath315 joins @xmath316 and @xmath318 , where @xmath319 . column @xmath315 contains a @xmath317 in row @xmath316 , and the entry in row @xmath318 is @xmath320 . all other entries in the column are zero . [ noodle ] let @xmath302 be a gain graph over the multiplicative group of the field @xmath1 . the matrix @xmath314 represents the matroid @xmath311 over @xmath1 . next we construct two families of gain graphs . let @xmath32 be an integer , and let @xmath1 be a field containing a non - zero element , @xmath33 , with order greater than @xmath34 . the gain graph @xmath321 has vertex set @xmath322 . each vertex @xmath323 in @xmath324 is incident with a loop , @xmath325 . in addition , @xmath326 is incident with the loop @xmath327 , and @xmath328 is incident with the loop @xmath329 . the parallel edges @xmath330 and @xmath331 join @xmath323 and @xmath332 for each @xmath137 in @xmath333 . moreover , the edges @xmath278 , @xmath251 , and @xmath334 join @xmath326 and @xmath328 . the gain function , @xmath303 , will be into the multiplicative group of @xmath1 . we define @xmath303 so that it takes each loop to @xmath33 and each @xmath330 to @xmath317 . furthermore , @xmath335 for each @xmath137 in @xmath333 , while @xmath336 , @xmath337 , and @xmath338 . a gain - graphic matroid of the form @xmath339 is said to be a _ hoop matroid_. now let @xmath37 be an integer . we let @xmath36 be a field , and we let @xmath38 be an element in @xmath340 with order greater than @xmath39 . we construct the gain graph @xmath341 . it has @xmath342 as its vertex set . each vertex @xmath343 is incident with a loop , @xmath344 , while @xmath345 is incident with the loop @xmath327 and @xmath346 is incident with the loop @xmath329 . for each @xmath347 , the edges @xmath315 and @xmath348 join @xmath349 to @xmath350 . the edges @xmath278 , @xmath251 , @xmath334 , and @xmath351 join the vertices @xmath345 and @xmath346 . the gain function takes each loop to @xmath38 , and each edge @xmath315 to @xmath317 . the triple @xmath352 is taken to @xmath353 when @xmath354 , and to @xmath355 when @xmath356 . thus @xmath357 of the edges @xmath358 receive the label @xmath353 , and the other @xmath359 receive the label @xmath355 . the values of @xmath360 , @xmath361 , @xmath362 , and @xmath363 are @xmath317 , @xmath353 , @xmath355 , and @xmath364 , respectively . a gain - graphic matroid of the form @xmath365 is a _ loop matroid_. shows @xmath321 and @xmath341 . the edge labels of loops have been omitted . every edge label corresponds to the orientation of the edge shown in the drawing . and @xmath341 . ] note that whenever @xmath366 and @xmath367 , then @xmath368 . let @xmath219 be this intersection . then @xmath369 and @xmath370 are both isomorphic to @xmath371 , so @xmath372 is defined . if @xmath295 is a graph and @xmath105 is a set of edges , then @xmath373 $ ] denotes the subgraph of @xmath295 containing the edges in @xmath105 and all vertices that are incident with at least one edge in @xmath105 . [ velvet ] let @xmath1 be a field , let @xmath32 be an integer , and let @xmath33 be an element in @xmath374 with order greater than @xmath375 . let @xmath14 be @xmath339 and let @xmath15 be @xmath376 . then @xmath372 is @xmath1-representable . let @xmath219 be @xmath377 . let @xmath302 be the signed graph @xmath321 , and let @xmath378 be @xmath379 . the will follow from if we can prove that @xmath372 is gain - graphic and corresponds to a gain graph over the multiplicative group of @xmath1 . to this end , we construct a gain graph by gluing together @xmath295 and @xmath380 . we identify the vertices @xmath326 and @xmath345 , as well as @xmath328 and @xmath381 . let @xmath301 be the resulting graph . note that @xmath382 is a gain function for @xmath301 . let @xmath55 be the gain - graphic matroid @xmath383 . thus we can prove the by checking that @xmath55 and @xmath384 are equal . we do this by showing that a set , @xmath105 , is dependent in @xmath55 if and only if it is dependent in @xmath384 . note that @xmath55 is obviously an amalgam of @xmath14 and @xmath15 . for the first direction , we assume that @xmath105 is a circuit in @xmath55 . as @xmath55 is an amalgam of @xmath14 and @xmath15 , we assume that @xmath105 is contained in neither @xmath157 nor @xmath385 . we start by considering the case that @xmath105 is a balanced cycle in @xmath386 . if @xmath105 contains an edge joining @xmath326 and @xmath328 , then this edge is @xmath351 , and @xmath387 $ ] contains a path with vertex sequence @xmath388 , for otherwise @xmath105 is contained in @xmath157 or @xmath385 . the product of edge labels along this path is @xmath389 , where @xmath390 . we also require that @xmath391 , since @xmath351 is labelled with @xmath392 , and @xmath105 is a balanced cycle . but @xmath391 can not hold , as @xmath33 has order greater than @xmath375 , and @xmath32 so @xmath393 . therefore we conclude that @xmath105 does not contain any edge between @xmath326 and @xmath328 . let @xmath389 be the product of edge labels along the path in @xmath387 $ ] with vertex sequence @xmath388 . thus @xmath394 . let @xmath395 be the product of edge labels along the path in @xmath387 $ ] with vertex sequence @xmath396 , where @xmath23 and @xmath397 are non - negative integers . thus @xmath398 and @xmath399 , as @xmath105 is balanced . as the order of @xmath33 is greater than @xmath375 , we deduce that @xmath400 , and hence @xmath401 is equal to @xmath30 , @xmath402 , or @xmath34 . in these three cases , @xmath278 , @xmath251 , or @xmath334 is an element in @xmath403 . thus statement ( iii ) of holds , so @xmath105 is dependent in @xmath372 . now we can assume that @xmath105 does not contain a balanced cycle of @xmath386 . thus @xmath387 $ ] is a theta graph or a handcuff . let @xmath404 be @xmath405 . assume that @xmath406 $ ] is a path from @xmath407 to @xmath408 . none of the internal vertices of this path has degree three or more in @xmath387 $ ] . from this , we can see without difficulty that @xmath409 $ ] contains a unbalanced cycle joined by a path to the loop @xmath327 , and a cycle joined by a path to the loop @xmath329 . therefore @xmath410 ( and hence all of @xmath219 ) is contained in @xmath411 . also , @xmath412 $ ] is a handcuff , and hence @xmath413 is a circuit of @xmath212 that spans @xmath219 . this means that @xmath414 . now ( i ) of holds , so @xmath105 is dependent in @xmath372 . we can now assume that neither @xmath415 $ ] nor @xmath416 $ ] is a path from @xmath407 to @xmath408 . assume @xmath415 $ ] is a forest . as @xmath387 $ ] has no vertices of degree one , the forest must be a path , and its end - vertices must be @xmath407 and @xmath408 , contradicting our assumption . by symmetry , it follows that each of @xmath415 $ ] and @xmath416 $ ] contains an unbalanced cycle . since @xmath387 $ ] is connected , either @xmath326 or @xmath328 is on a path from one of these cycles to the other . let us assume the former , since the latter case is identical . now @xmath327 is in a handcuff , and hence in a circuit of @xmath14 that is contained in @xmath417 , and also in a circuit of @xmath15 that is contained in @xmath418 . thus statement ( iii ) of holds , and @xmath105 is dependent in @xmath372 . we have proved that if @xmath105 is dependent in @xmath55 , it is dependent in @xmath372 . for the other direction , we assume that @xmath105 is independent in @xmath55 . this means that @xmath387 $ ] contains no balanced cycles , and any connected component of @xmath387 $ ] contains at most one cycle . let us assume for a contradiction that @xmath105 is dependent in @xmath372 . in fact , we can assume that @xmath105 is a circuit of @xmath372 . as @xmath55 is an amalgam , it follows that neither @xmath419 nor @xmath420 is dependent , so @xmath105 is contained in neither @xmath157 nor @xmath385 . one of the three statements in must hold . we prove the following statements for @xmath14 and @xmath15 simultaneously , by letting @xmath404 be @xmath405 . [ junket ] the subgraph @xmath406 $ ] either contains a connected component that contains both @xmath326 and @xmath328 , or a connected component that contains either @xmath326 or @xmath328 and a cycle . assume is false , so that any component of @xmath406 $ ] contains at most one of @xmath326 and @xmath328 , and any component containing one of these vertices contains no cycle . this means that if @xmath23 and @xmath397 are distinct elements of @xmath219 , then @xmath421 $ ] contains no balanced cycles , and no theta graphs or handcuffs . from this it follows that @xmath422 does not contain any element of @xmath219 , so statement ( iii ) of does not hold . moreover , @xmath423 . as one of the three statements in must hold , it follows that @xmath219 is in @xmath424 and @xmath425 . but this now means that @xmath105 contains at least two elements of @xmath219 , or else @xmath426 . hence @xmath427 , so implies that @xmath428 is dependent in @xmath372 . since @xmath105 is a circuit of @xmath372 , this means that @xmath429 , contrary to hypothesis . therefore holds . [ isobar ] there is no connected component of @xmath406 $ ] that contains both @xmath326 and @xmath328 . assume that @xmath430 is such a component . then @xmath430 is contained in a connected component , @xmath431 , of @xmath432 $ ] . if @xmath431 contains a cycle , then by applying to @xmath433 $ ] , we can deduce that the union of @xmath431 with a component of @xmath433 $ ] contains a theta graph or a handcuff . we have assumed that @xmath105 contains no such subgraph , so this is a contradiction . therefore @xmath431 contains no cycle , from which we deduce that @xmath434 and that @xmath406 $ ] is a path from @xmath326 to @xmath328 . note that @xmath435 $ ] contains no circuit of @xmath212 , so @xmath219 is not in @xmath422 . if there is a component of @xmath433 $ ] that contains @xmath326 and @xmath328 , then by the reasoning in the previous paragraph , @xmath433 $ ] is a path from @xmath326 to @xmath328 , and @xmath436 . therefore @xmath387 $ ] is a hamiltonian cycle , and the only statement in that can hold is statement ( iii ) . we have noted that @xmath327 ( and by symmetry @xmath329 ) is not in @xmath422 , so there is an edge , @xmath437 , joining @xmath326 and @xmath328 , such that @xmath437 is in both @xmath422 and @xmath438 . this means that @xmath439 $ ] and @xmath440 $ ] are both balanced cycles . thus the product of edge labels on the path @xmath433 $ ] is the inverse of the product on the path @xmath406 $ ] ( assuming that we travel in a consistent direction around the hamiltonian cycle @xmath387 $ ] ) . hence @xmath105 is a balanced cycle , a contradiction . therefore no component of @xmath433 $ ] contains @xmath326 and @xmath328 . recall that @xmath441 and neither @xmath327 nor @xmath329 is in @xmath422 . as one of the statements from must hold , either there is an edge between @xmath326 and @xmath328 that is in both @xmath422 and @xmath438 , or @xmath438 contains @xmath219 . therefore in either case we can let @xmath437 be an edge between @xmath326 and @xmath328 that is in @xmath438 . let @xmath310 be a circuit of @xmath211 contained in @xmath442 that contains @xmath437 . no component of @xmath433 $ ] contains @xmath326 and @xmath328 so @xmath443 $ ] is not connected . it follows that @xmath444 $ ] is a handcuff , and @xmath437 is an edge in the path between the two cycles . therefore @xmath433 $ ] contains two distinct components , each containing a cycle and one of @xmath326 and @xmath328 . as @xmath406 $ ] is a path from @xmath326 to @xmath328 , it follows that @xmath387 $ ] is a handcuff , a contradiction . this completes the proof of . we have shown that neither @xmath415 $ ] nor @xmath416 $ ] contains a component that contains @xmath326 and @xmath328 . by using , symmetry , and the fact that @xmath105 contains no handcuffs , we can assume the following : there is a component of @xmath433 $ ] that contains @xmath326 and a cycle , and any component that contains @xmath328 contains no cycle ; similarly , there is a component of @xmath406 $ ] that contains @xmath328 and a cycle , and any component that contains @xmath326 contains no cycle . it follows from this assumption ( and the fact that @xmath387 $ ] contains no handcuffs ) that @xmath441 . notice that @xmath327 is the only element in @xmath445 similarly , @xmath446 . therefore none of the statements in can hold , so we have a contradiction . now it follows that if @xmath105 is independent in @xmath55 it is also independent in @xmath372 , so @xmath447 , exactly as desired . [ alcove ] let @xmath1 be a field and let @xmath34 and @xmath357 be distinct integers satisfying @xmath448 . let @xmath33 be an element in @xmath374 with order greater than @xmath44 . let @xmath14 be @xmath339 and let @xmath15 be @xmath449 . then @xmath372 is not representable over any field . let us assume that the matrix @xmath450 represents @xmath372 over the field @xmath36 . let @xmath451 be the set @xmath452 . thus @xmath451 is the set of all loops in @xmath321 and @xmath453 . it is clear that @xmath454 and @xmath455 are independent in @xmath14 and @xmath15 , and moreover , @xmath456 and @xmath457 . now it follows easily from that @xmath451 can not be dependent in @xmath372 . if @xmath305 is any element of the ground set of @xmath372 that is not in @xmath451 , then @xmath458 contains a circuit of either @xmath14 or @xmath15 , and this circuit has cardinality three . from this it follows that @xmath451 is a basis of @xmath372 . we can assume that the columns of @xmath450 labelled by the elements of @xmath451 form a identity matrix . as the fundamental circuits relative to @xmath451 all have cardinality three , every column of @xmath450 contains either one or two non - zero elements . by scaling , we can assume that the first non - zero entry in each column is @xmath317 . thus @xmath459 , for some gain graph @xmath302 over the field @xmath36 . by examining the fundamental circuits relative to @xmath451 , we see that @xmath295 is the graph in . by scaling rows of @xmath450 , we can assume that @xmath460 for each @xmath461 and @xmath462 for each @xmath463 . for @xmath464 , let @xmath465 be @xmath466 and for @xmath467 , let @xmath468 be @xmath469 . let @xmath470 , @xmath471 , @xmath472 , and @xmath473 be @xmath474 , @xmath475 , @xmath476 , and @xmath477 . . ] because @xmath478 is a balanced cycle in @xmath321 , and hence a circuit in @xmath372 , it follows that it is also a balanced cycle in @xmath302 . this means that @xmath479 . next we notice that @xmath480 is a balanced cycle of @xmath321 and hence of @xmath302 , for any @xmath137 in @xmath333 . the product of edge labels on this cycle is @xmath481 , which implies that @xmath482 for any @xmath483 . this in turn implies that @xmath484 for any @xmath485 and that @xmath486 . as @xmath487 is a balanced cycle , it follows that @xmath488 . next we observe that @xmath489 is a balanced cycle in @xmath453 , and hence in @xmath302 , for any @xmath490 . thus @xmath491 for any such @xmath137 . similarly , @xmath492 is a balanced cycle for any @xmath493 , from which we deduce that @xmath494 . as @xmath495 and @xmath496 are both balanced cycles , it now follows that the products @xmath497 and @xmath498 are both equal to @xmath473 . thus @xmath499 , implying @xmath500 . let @xmath501 be the order of @xmath502 in @xmath36 . since @xmath503 , we know that @xmath504 . but if @xmath505 , then @xmath506 is a balanced cycle in @xmath302 , although it is not a circuit in @xmath14 . therefore @xmath507 . now the product of edge labels on the cycle @xmath508 is @xmath509 , so this is a balanced cycle in @xmath302 , although not a circuit in @xmath15 . this contradiction proves the . we thank noam greenberg for helpful advice . the research in this article was supported by the rutherford discovery fellowship , the marsden fund of new zealand , and nserc canada .

we construct a class of projected models where the gutzwiller constraint of no - double - occupancy is implemented exactly . we introduce the concept of projected symmetry where all static correlation functions are exactly symmetric and discuss the signature of the projected symmetry in dynamical correlation functions . we show that this class of projected models can give a realistic description of the global phase diagram of the high superconductors and account for many of their physical properties .

cond-mat9904142

recently , a unified theory of antiferromagnetism ( af ) and superconductivity ( sc ) has been proposed for the high @xmath1 cuprates@xcite . this theory is based on the @xmath0 symmetry between af and sc , and offers a unified description of the global phase diagram for this class of materials . while the theory was originally proposed as a effective field theory description , it was soon realized that the @xmath0 symmetry could be implemented exactly at a microscopic level@xcite , and it can also be checked numerically in common strongly correlated models such as the @xmath2 model@xcite . while the phase diagram@xcite and collective excitations@xcite in the sc state derived from these @xmath0 models bear strong resemblance with the high @xmath1 cuprates , and a number of novel experimental predictions have been made@xcite , the mott insulating behavior at half - filling is a puzzling aspect which challenges the fundamental validity of the @xmath0 models@xcite . to be more precise , the exact @xmath0 symmetry requires collective charge two excitation at half - filling to have the same mass as the collective spin wave excitations . this condition is clearly violated in a mott insulating system where all charge excitations measured with respect to a particle hole symmetric point have a large energy gap of few @xmath3 , while the spin wave excitations are massless . in the original @xmath0 proposal , it was pointed out that this situation is analogous to a easy - axis antiferromagnet in a external uniform field , and a @xmath0 symmetry breaking term at half - filling was introduced in order to describe this asymmetric behavior between spin and charge . the chemical potential also introduces a @xmath0 symmetry breaking term , however , it was shown that these two terms could compensate each other@xcite so that the _ static potential _ governing the @xmath0 superspin could still be @xmath0 symmetric . since the asymmetry between the charge and spin excitations at half - filling is of the order of the coulomb energy scale @xmath4 , the @xmath0 symmetry breaking terms must also be of that order . since there are various types of symmetry breaking terms , one might hope that their effects could partially cancel each other to arrive at a qualitatively correct picture . however , this type of cancellation is very delicate , and approximate calculations could easily lead to erroneous conclusions . in particular , one is interested in which physical properties could exhibit @xmath0 symmetric properties in the limit when the coulomb gap is taken to infinity . for example , one could ask the following questions : \1 ) one of the hallmarks of the @xmath0 symmetry is not only the degeneracy between the af and sc states at a given chemical potential , but the approximate degeneracy among all mix - states interpolating between af and sc , _ i.e. _ the independence of the ground state energy on the superspin angle . what is the potential barrier separating the af and sc states at their degeneracy point in the limit @xmath5 ? if there is a large energy barrier in this limit , one would argue that the concept of @xmath0 symmetry is not a useful one , at least not for quantitative calculations . on the other hand , if the potential barrier is finite and small in the @xmath5 limit , the concept of a approximate @xmath0 symmetry would be a useful one . \2 ) exact @xmath0 symmetry predicts four massless collective modes . in the half - filled af state , besides the two conventional massless spin wave modes , the exact @xmath0 symmetry predicts a massless doublet of @xmath6 modes , with charge @xmath7 . however , a mott insulator has a large gap to all charge excitations . therefore , it is clear that one of the @xmath6 has to be projected out of the spectrum in the limit @xmath5 , say the @xmath8 mode carrying charge @xmath9 . what happens to the rest of the goldstone modes , the @xmath10 mode carrying charge @xmath11 and the @xmath12 triplet mode of the sc state ? in the @xmath5 limit , can they all be simultaneously massless at the transition point between af and sc ? since the pure sc state can only be reached with a finite doping concentration , is it possible that the gutzwiller projection does not affect the @xmath12 triplet mode of the sc state ? in order to address these questions , it is desirable to construct a low energy effective theory without any parameters of the order of the coulomb scale @xmath4 . in this work , we construct a class of projected @xmath0 models which treat the gutzwiller constraint exactly and locally on every site . we use this model to answer the physical questions posed above and show that the answers are affirmative . in the @xmath5 limit , when the gutzwiller constraint is implemented exactly , the ground state energy can still be @xmath0 symmetric and independent of the superspin direction . after projecting out the @xmath8 mode , all other goldstone modes remain massless at the symmetric point . the dispersion relation of the collective modes bear unique signature of the projected @xmath0 symmetry . furthermore , the @xmath12 triplet modes of the pure sc states are unaffected by the gutzwiller projection . these properties define the concept of a projected @xmath0 symmetry ( @xmath13 ) , whose properties and consequences we shall explore in this paper . the fundamental quantity in the @xmath0 theory is the locally defined five component superspin vector @xmath14 describing the local af and sc order parameters respectively . in the nonlinear @xmath15 model formalism , these are treated as mutually commuting coordinates and their dynamics is given by their conjugate momenta @xmath16 . the charge operator is the angular momentum in the @xmath17 plane : @xmath18 implementing the gutzwiller constraint corresponds to requiring @xmath19 for every local @xmath0 rotor . from equations ( [ charge ] ) and ( [ gutzwiller ] ) and subsequent discussions , we shall see that the gutzwiller projection in the @xmath0 formalism corresponds to going from a fully symmetric @xmath0 rotor model to a _ chiral _ @xmath0 rotor model , where both the static potential of the individual rotors and the coupling between the rotors are still @xmath0 symmetric , but the rotors are constrained to rotate only in one sense in the @xmath17 plane , consistent with ( [ gutzwiller ] ) . this observation reveals a deep connection between the gutzwiller projection and the lowest - landau - level ( lll ) projection in the fractional quantum hall effect@xcite . to be more precise , the gutzwiller projection represented by equations ( [ charge ] ) and ( [ gutzwiller ] ) is analogous to the lll projection , where all states in the lll have a definite sign of angular momentum . the lll projection can be analytically implemented by separating the cyclotron degrees of freedom from the guiding center degrees of freedom , which amounts to changing the commuting property between the @xmath20 and @xmath21 coordinates to a canonically conjugate commutation relation : @xmath22 = i \l_0 \label{guiding - center}\ ] ] where @xmath23 is the landau length . exploiting this analogy , we find that the original @xmath0 model can be fully gutzwiller projected without changing its form , if one imposes the simple quantization condition between the superconducting components of the superspin vector : @xmath24 = i/2 \label{non - commute}\ ] ] in the symmetric @xmath0 model , the wave function of the @xmath0 rotors are functions of the local coordinates @xmath25 and @xmath26 , while the projected @xmath0 model only depends on their holomorphic combination @xmath27 and is independent of their anti - holomorphic combination @xmath28 this way , we arrive at a natural projection of the @xmath0 model where the local gutzwiller constraint is taken into account exactly , and the resulting model is free of the large coulomb @xmath4 parameter . because the functional form of the symmetric @xmath0 model remain the same and only the quantization condition is modified upon projection , many important properties associated with the @xmath0 symmetry remain . the central hypothesis of the @xmath0 theory is that this projected model is quantitatively accurate in describing both the static and dynamic properties of the high @xmath1 cuprates , and we shall compare the properties of this model with the phenomenology of the high @xmath1 systems . we begin with the symmetric @xmath0 hamiltonian defined on a lattice , @xmath29 where @xmath30 denotes the five component superspin vector on a given site , and @xmath31 is the @xmath0 symmetry generator , @xmath32 expressed here in terms of the superspin vector @xmath33 and its canonically conjugate momenta @xmath34 , @xmath35 = i \delta_{ab}\ ] ] this lattice quantum non - linear @xmath15 model can be rigorously derived as the low energy limit of a microscopic @xmath0 ladder model@xcite . on the ladder , the rung @xmath0 singlet state is the vacuum @xmath36 , from which the lowest so(5 ) multiplet @xmath37 is created by a quintet of bose creation operators , which satisfy here , @xmath39 denote the triplet ( magnon ) states , and @xmath40 are the hole and particle pair states . ( see fig . ( [ fig1 ] ) ) . the superspin coordinates are microscopically constructed using these lattice bosons , @xmath41 = 4.0 cm due to their microscopic origin , these bosonic states are hard - core bosons , in the sense that one can not define two of them on the same rung . the @xmath42 term in equation ( [ sigma ] ) describes the gap energy of the magnon and the pair states , the @xmath43 term stands for the hopping and the spontaneous creation / destruction process of these states , and the @xmath44 term describes their nearest - neighbor interaction . this quantum nonlinear @xmath15 model can in principle also be derived in higher dimensions from a microscopic @xmath0 symmetric model@xcite , by introducing a superspin vector as a hubbard - stratonovich decoupling field , and integrate out the fermionic degrees of freedom in a gradient expansion . however , we shall proceed more heuristically here@xcite . for a two dimensional system , one can imagine that the quantum @xmath15 model hamiltonian is obtained from a block spin " type of coarse - graining of the microscopic electron hamiltonian , and is defined on a lattice with twice the lattice constant compared to the microscopic electron model . ( this doubled unit cell is the minimal size needed to define the local af and d wave sc order parameters ) . therefore , each site @xmath45 in the effective model correspond to a plaquette of the microscopic electron model . ( on a ladder , this corresponds to going from the lattice sites to ladder rungs ) . the @xmath0 singlet state @xmath36 corresponds to a rvb " type of singlet state , while the five - fold states @xmath46 describe the triplet magnon states , and the @xmath47-wave hole and particle pair states on a plaquette . unlike the ladder case , the magnon and the @xmath47- wave pair states could condense in the ground state to form af and sc broken symmetry states . in fact , eder@xcite has recently shown that properties of the af states can be described by a coherent state of magnon condensation on top of a uniform spin liquid state . our model therefore describes competition among the rvb " type of singlet vacuum and the two forms of broken symmetry order . while it is reasonable to take @xmath43 and @xmath44 to be approximately equal for magnons and pairs , the gap energy @xmath42 for the neutral magnons and the charged pairs are very different in the insulating state at half - filling . in fact , their difference is of the order of the insulating gap @xmath4 at half - filling . taking into account the hard - core condition and neglecting the nearest - neighbor interaction @xmath44 for now ( it has higher powers of time and space derivatives in the continuum limit ) , we can express the general anisotropic @xmath0 model as @xmath48 in this paper we shall use the convention where @xmath49 denote the superspin indices , @xmath50 denote the spin indices and @xmath51 denote the charge indices , and repeated indices are summed over . the main focus of our paper is to consider the limit where @xmath52 . let us define the charge eigen - operators @xmath53 and @xmath54 as @xmath55 from this definition , it is clear that @xmath56 is the creation operator for a hole pair and @xmath57 is the creation operator for a particle pair . we can introduce a chemical potential term @xmath58 to describe the effects of doping . in the presence of this chemical potential term , the gap energy of the hole and particle pairs are @xmath59 and @xmath60 respectively . a chemical potential of the order of the charge gap @xmath61 is needed to induce a metal - insulator transition in this system . near such a transition point , the gap energy of the hole pair @xmath62 can be comparable to the spin gap @xmath63 , while the gap towards a particle pair excitation is of the order of twice the charge gap , and needs to be projected out of the spectrum in the low energy limit . therefore , within this formalism , the gutzwiller projection is equivalent to restricting ourselves to the projected hilbert space where @xmath64 at every site @xmath45 . within this projected hilbert space , the projected hamiltonian takes the form @xmath65 this hamiltonian has no parameters of the order of @xmath4 , and it is reasonable to expect @xmath66 and @xmath67 . we see that the form of the hamiltonian hardly changes from the unprojected model , but the definition of @xmath25 and @xmath26 is changed from @xmath68 to @xmath69 from equation ( [ before ] ) , we see that @xmath25 and @xmath26 commute with each other before the projection . however , after the projection , they acquire a nontrivial commutation relation , as can be seen from equation ( [ after ] ) : @xmath24 = i/2\ ] ] therefore , the gutzwiller projection can be analytically implemented in the @xmath0 theory by retaining the form of the hamiltonian and change only the quantization condition . the discussions outlined above reveal a deep connection between the gutzwiller projection within the @xmath0 formalism and the projection onto the lowest landau level ( lll ) in the context of the fractional quantum hall effect . consider the problem of a charged particle in a strong magnetic field @xmath70 and a rotationally symmetric potential @xmath71 . in the absence of a magnetic field , all eigenstates form irreducible representations of the two dimensional rotation group @xmath72 , characterized by integral eigenvalues of the angular momentum operator @xmath73 however , in the presence of a strong magnetic field and projected into the lll , only negative eigenvalues of @xmath74 are realized . this is analogous to the situation encountered here . the local charge operator in the @xmath0 theory takes the form of the angular momentum in the @xmath17 plane as given by equation ( [ charge ] ) . when doubly occupied sites are locally projected out , the local charge operator , or the angular momentum in the @xmath17 plane , takes only negative values . since the chemical potential couples directly the angular momentum in the @xmath17 plane , it plays the role of a fictitious magnetic field threading every @xmath0 rotor in the @xmath17 plane . the landau level spacing @xmath75 is analogous to the charge gap @xmath61 encountered here , and both are taken to be infinity in the projected models . after the projection , the hamiltonian in the landau level problem retains its @xmath72 symmetric form , @xmath76 although a new quantization condition is imposed between @xmath20 and @xmath21 , as given by equation ( [ guiding - center ] ) . this is analogous to the observation we made here that the hamiltonian formally retains a @xmath0 symmetric form after the projection ( [ constraint1 ] ) , but the quantum dynamics is changed due to the non - trivial commutator between @xmath25 and @xmath26 . in both cases only a part of the full symmetry multiplets remain after the projection . however , the formal symmetry of the hamiltonian has direct physical manifestations despite the projection . for example , in the lll problems , semi - classical orbits of the guiding center coordinates are still @xmath72 symmetric . in our case , we shall see that the static potential for the superspin vector can still be @xmath0 invariant despite the projection . perhaps the most explicit way to establish the precise connections between these two problems is to consider the constraints on the wave function . in the symmetric gauge of the lll problem , the annihilation operator for the cyclotron coordinates takes the form@xcite @xmath77 where @xmath78 and @xmath79 . projection onto lll requires @xmath80 which determines the form of the lll wave function to be @xmath81 where @xmath82 is a holomorphic function of @xmath83 only . this holomorphic condition also places strong constraints in many - body systems and led to the celebrated laughlin s wave function . our no - double - occupancy constraint ( [ constraint1 ] ) is analogous to the lll constraint ( [ constraint2 ] ) . in fact from equations ( [ p - h ] ) and ( [ before ] ) , we obtain @xmath84 where @xmath85 and @xmath86 . for a single unprojected @xmath0 rotor , the wave function @xmath87 is a function of the superspin coordinates . however , the gutzwiller projection ( [ constraint1 ] ) restricts the wave function to be @xmath88 where @xmath89 is a holomorphic function of @xmath83 . for a collection of @xmath0 rotors , the superspin coordinates are themselves functions of the lattice sites @xmath45 , and @xmath90 $ ] is a functional is the superspin coordinates at each site . for the projected @xmath0 models , this functional is restricted to take the form @xmath91 = f(z(x),n_\alpha(x ) ) \prod_{x } e^{-z\bar z(x)/2 } \label{functional}\ ] ] where @xmath92 is a holomorphic functional of @xmath93 . the formal but precise analogy between the two types of projection allows us to introduce the concept of a chiral @xmath0 rotor . this is a system of rotors with @xmath0 invariant potential and coupling , however , the rotation within the @xmath17 plane is chiral , _ i.e. _ only one sense of the rotation is allowed . such a system of chiral @xmath0 rotors is described by the wave functional in equation ( [ functional ] ) . having discussed the general notions of the projected @xmath0 model , we are now in a position to explore the phase diagram of this model . as we commented earlier , the projected @xmath0 model describes the competition and unification of the spin liquid , af and the sc states . in the original unprojected @xmath0 symmetric model , not only are the af and sc states degenerate in energy , but they are also degenerate with all the intermediate coexistence states . this points out a route from af to sc with no potential barrier , and introduces the concept that the metal - insulator transition in the high @xmath1 systems can be viewed as a smooth rotation of the @xmath0 superspin . one of the key questions to be answered in this work is what happens to the picture in the case of projected @xmath0 symmetry . in anticipation of the competition of the states discussed above , we construct a class of variational wave functions in the coherent state representation : @xmath94 here @xmath95 denotes a local singlet state defined by @xmath96 and @xmath36 is a product state of these local singlets , @xmath97 . @xmath98 is a local variational parameter describing the competition between long range order and quantum disorder . for @xmath99 our variational wave function describe a spin singlet ground state , while a non - zero value of @xmath100 describes a coherent state formed by the local singlet , the magnon or the hole pair state this wave function is a generalization of the coherent state description of a af state in terms of a magnon condensate@xcite . as we shall see from the following equation ( [ superspin_vector ] ) , @xmath101 stands for the length of the @xmath0 superspin vector . @xmath102 and @xmath103 are general complex variational parameters describing the local amplitude for magnons and hole pairs . we notice that this wave function satisfies both the gutzwiller constraint ( [ constraint1 ] ) and the hard - core constraint for magnons and hole pairs exactly . it is easy to see that @xmath104 where @xmath105 and @xmath106 denote the real and imaginary parts of a complex number . the coupling terms in the projected @xmath0 hamiltonian depend only on @xmath107 , @xmath108 and @xmath109 . therefore , the coupling energy depends only on the real part of @xmath102 while it depends on both the real and imaginary parts of @xmath103 . therefore , for discussing the ground state wave functions , we can assume without loss of generality that @xmath102 is real and @xmath110 . the normalization condition @xmath111 can be implemented by the constraint that @xmath112 therefore , we see that although we have completely projected out the particle pair states , the local degrees of freedom can still be represented by a vector on a five dimensional sphere . uniform states are obtained by taking all parameters to be constant . for @xmath113 and @xmath114 , our wave function @xmath115 describes a pure af state with the following properties : @xmath116 where @xmath117 is the number of lattice sites . equation ( [ af - state ] ) describes a half - filled state with a macroscopic neel magnetization , and vanishing uniform magnetization . furthermore , this state is composed as a linear superposition of eigenstates with different values of the total spin , and the fluctuation of the total spin scales like @xmath118 , just as one expects from a standard neel state . on the other hand , for @xmath119 and @xmath114 , @xmath115 describes a pure sc state with the following properties : @xmath120 equation ( [ sc - state ] ) describes a state with a finite doping density and a finite sc order parameter . just as in the standard bcs case , this state is composed as a linear superposition of eigenstates with different values of the total charge , and the fluctuation of the total charge scales like @xmath121 , just as one expects from a standard sc state . however , besides these two _ pure _ states , there is a class of _ mixed _ states which interpolates between the pure af and sc states . taking @xmath122 and @xmath123 , we see that the mixed states have the following property : @xmath124 therefore , we see that there is a continuous family of intermediate mixed states interpolating between the pure af state at half - filling and the pure sc state with finite doping density . as the @xmath0 angle @xmath125 rotates continuously from a pure af state with @xmath126 to a pure sc state with @xmath127 , the hole density of the mixed state interpolates continuously between these two limits . therefore , our wave function gives a unified description of af and sc and points out a precise route from af to sc as the doping level is varied . in order for this route , or small deviations from this route , to be physically realized in the high @xmath1 superconductors , we have to demonstrate that there is no large energy barrier for the intermediate mixed states , or that the ground state energy is approximately independent of the @xmath0 mixing angle @xmath125 . in particular , we have to show that the energy barrier is independent of the hubbard energy @xmath4 , in the limit of large @xmath4 . in the following , we shall investigate this question . the energy functional @xmath128 describes the coupling between these five dimensional vectors @xmath102 , and it is given by @xmath129 this ground state energy functional describes a systems of coupled rotors satisfying the constraint ( [ normalization ] ) . at the point @xmath130 and @xmath131 in parameter space , this rotor model is exactly @xmath0 symmetric . this is a central observation of this work . from this consideration we learn a very important lesson about the compatibility of the mott insulating gap and the idea of a smooth @xmath0 rotation from af to sc . as we have seen , the large asymmetry between the charge and spin gap at half - filling necessitates the removal of the @xmath132 part of the @xmath0 multiplets , therefore , the dynamics close to half - filling has to be modified . but the static potential governing the transition from af to sc can remain @xmath0 symmetric , and in particular , the energy barrier separating these two states can remain small in the limit where the mott insulating gap tends to infinity . we also observe a crucial difference between the projected and the unprojected @xmath0 models . in the unprojected @xmath0 model with the full @xmath0 symmetry , af and sc states are degenerate at half - filling , and the rotation between these two states can be continuously performed without changing the density to going away from half - filling . this case is similar to the well - known degeneracy between the cdw state and the @xmath133 wave sc state for the negative @xmath4 hubbard model at half - filling . in the projected @xmath0 model , where all particle pair states have been locally removed , the @xmath0 rotation from the af to sc states are accompanied by the continuous change of the hole density , and a pure sc state can only be reached at a finite critical hole doping density @xmath134 . while the unprojected @xmath0 symmetry is only valid at half - filling , the projected @xmath0 symmetry can be valid for a range of doping concentration @xmath135 , since all these doping concentrations correspond to the same value of the chemical potential @xmath136 at which the ground state energy ( [ energy ] ) is @xmath0 symmetric . however , it should be pointed out that the projected @xmath0 symmetry at @xmath136 has only been demonstrated within the variational mean field approximation . this corresponds to the semiclassical limit , and becomes exact only in the large @xmath133 limit , where @xmath133 labels the representation of the local @xmath0 group at a given site . quantum fluctuations can be systematically investigated as a @xmath137 expansion . assuming uniform ground states , we have studied the effect of zero point fluctuations in sec . [ fluctuations ] and found that at @xmath136 , the intermediate mixed states have slightly higher energy than the af and sc state . therefore , quantum fluctuation leads to a slight breaking of the projected @xmath0 symmetry . the important point here is that this symmetry breaking effect can be systematically controlled in the semi - classical @xmath137 expansion , and certainly is independent of the coulomb energy scale @xmath4 . this fluctuation would induce a first order transition and predict phase separation of af and sc states at @xmath136 . however , there are also other competing interactions such as nearest - neighbor and next - nearest - neighbor interactions which tend to reduce the barrier , and could also lead to non - uniform states like stripes . due to the complexity of the calculations , we shall defer the detailed studies of these competing effects to future works . in this section we investigate the phase diagram of the projected @xmath0 model within the framework of the variational wave function ( [ variational ] ) . taking uniform values of the variational parameters @xmath100 , @xmath138 , @xmath122 and @xmath139 , the variational energy in ( [ energy ] ) reduces to @xmath140 in the following , we shall mainly study the @xmath0 symmetric case , and take @xmath141 . defining @xmath142 and @xmath143 , and the dimensionless coupling constants @xmath144 , @xmath145 and @xmath146 , we obtain @xmath147 we shall minimize ( [ mean_energy ] ) with respect to @xmath45 and @xmath148 , subject to the condition that @xmath149 . the phase diagram can be plotted in the two dimensional parameter space of @xmath150 and @xmath151 . we notice that @xmath152 depends linearly on @xmath148 , therefore , for @xmath153 we obtain @xmath154 and @xmath155 similarly , for @xmath156 we obtain @xmath157 and @xmath158 from these equation , we can determine the phase diagram as shown in fig . ( [ fig2 ] ) . = 10.0 cm there are seven different phases on this phase diagram . @xmath159 corresponds to a quantum disordered singlet state with no condensed bosons . @xmath160 and @xmath154 corresponds to a quantum disordered state with completely filled hole pairs . @xmath160 and @xmath157 corresponds to a quantum disordered state with completely filled magnons . @xmath161 and @xmath157 describe a pure af phase , while @xmath161 and @xmath154 describes a pure sc phase . when @xmath162 a continuous family of mixed af / sc states labeled by a free superspin angle @xmath163 is realized , while for @xmath164 a continuous family of quantum disordered states labeled by a free superspin angle @xmath163 is obtained . the system traces out a one dimensional trajectory in this two dimensional phase diagram as the chemical potential is increased , as depicted in fig . ( [ fig2 ] ) . increasing the chemical potential decreases the @xmath150 parameter while holding @xmath151 constant . @xmath151 describes the degree of quantum spin fluctuations in the system , since in the af phase , the size of the neel moment @xmath165 decreases with increasing @xmath151 . for @xmath166 , the system goes through a phase transition from af to sc at @xmath167 . at this critical value of the chemical potential , the @xmath100 parameter remains fixed , but the @xmath125 parameter changes continuously from @xmath168 to @xmath169 , and correspondingly , the density changes from @xmath168 to @xmath170 this behavior gives a density versus chemical potential diagram as shown in fig . ( [ fig3 ] ) . = 4.0 cm as we see , for densities in the range @xmath135 , the system is infinitely compressible since @xmath171 . for @xmath172 or @xmath173 , the system has a finite compressibility of @xmath174 . it is interesting to plot both the af and the sc order parameters @xmath175 and @xmath176 as a function of the density for the whole range of @xmath177 . we will restrict to the case of @xmath178 where the undoped state is a af state . we obtain the following doping dependence of the sc order parameter : @xmath179 and the doping dependence of the of the af order parameter : @xmath180 these behaviors are depicted in fig . ( [ fig4 ] ) . = 6.0 cm we note several interesting features of the phase diagram . first of all , we can use the energy functional ( [ energy ] ) as a starting point for a finite temperature classical fluctuation analysis , and estimate the transition temperature due to the classical fluctuations . within such a framework , the 3d af ( @xmath181 ) and sc ( @xmath1 ) transition temperatures are proportional to the stiffness of the spin and the phase fluctuations , which are in turn proportional to @xmath182 and @xmath183 respectively . therefore , fig . ( [ fig4 ] ) gives an approximate estimate of the transition temperatures . for small @xmath184 , we see that there is a sharp drop of @xmath181 and a maximum of @xmath1 at @xmath185 . the system is a af insulator at doping @xmath186 and a pure sc state for @xmath173 . the reason for a maximum of @xmath1 at @xmath185 is due to the strong correlation of the charged bosons . the charge bosons have a hard - core interaction , therefore , they are insulating at both @xmath186 and @xmath187 and have maximal charge stiffness at @xmath185 . we can perform a rough translation of this optimal doping value in our effective model to the microscopic model . since our effective model is defined on a unit cell with twice the lattice spacing of the microscopic model , @xmath185 therefore describes one hole pair per four sites in the microscopic model , or a doping of @xmath188 in the conventional language . this crude argument tends to overestimate the value for optimal doping , since it neglects the effects of unpaired electrons . however , considering the crudeness of the estimate , it is still reasonably close to the optimal doping @xmath189 observed in the lsco family of high @xmath1 superconductors . in the regime of @xmath135 , the system is a coherent mixture of af and sc order . for this entire range of densities , the system has a projected @xmath0 symmetry within the variational approximation discussed above . the projected @xmath0 symmetry manifests itself in terms of a infinite compressibility in the region @xmath135 and , as we shall see in next section , a charge mode with a dispersion relation @xmath190 . since such a state is rather unusual , and maybe highly susceptible to density fluctuations , we would like to discuss more detailed physical properties in this region . first let us comment on the fact that there are familiar physical systems whose uniform ground states are infinitely compressible . the free boson model is certainly such an example , and the density mode also has a @xmath190 dispersion relation . but the infinite compressibility is due to the absence of the interaction , which is not characteristic of the strongly interaction system considered here . a less trivial example is the spin @xmath191 xxz ferromagnetic heisenberg model , given by the hamiltonian , @xmath192 where @xmath193 and the sum extends over nearest neighbor sites of a square lattice . this model can be interpreted as quantum hard - core boson model , where the fully polarized spin down state could be identified with the vacuum of the bosons , the xy part of the hamiltonian describes the hopping of the bosons and the last term describes the nearest neighbor attraction between the bosons if @xmath194 . when @xmath195 , the system is in the ising limit , and the spontaneous breaking of the @xmath196 symmetry implies phase separation of the bosons . on the other hand , when @xmath197 , the system is in the xy limit , and the ground state is a superfluid . therefore , the anisotropy parameter describe the competition between superfluidity and phase separation . at @xmath198 , the system has a @xmath199 symmetry and the dispersion relation becomes quadratic . different directions of the ferromagnetic polarizations are degenerate and can be changed without any energy cost . since the @xmath83 component of the ferromagnetic polarization is identified with the total density of the hard - core bosons , the @xmath199 vacuum degeneracy implies infinite compressibility of the corresponding boson system for the entire range of boson densities @xmath177 . these two examples illustrates that there is nothing intrinsically pathological about having a system with infinite compressibility . the second example is more generic , and shows that uniform states with infinite compressibility can be obtained in systems on the verge of phase separation , and the infinite compressibility can be ensured by symmetry . both of these properties are also shared by the projected @xmath0 model . these models are on the verge of phase separation into af and sc phases , and the infinite compressibility is a result of the projected @xmath0 symmetry . it indicates that small perturbations , such as next nearest neighbor interactions , quantum fluctuations , and quenched disorder will be very important to determine the true ground state . with such perturbations , the ground state is expected to be unstable toward the experimentally reported textures such as the spin glass , stripes and incommensurate spin density waves . next let us investigate the phenomenological consequence of this remarkable property . one of the most puzzling properties of the high @xmath1 superconductors is the constant chemical potential in the underdoped samples . for lsco systems , where the doping level can be varied continuously by the @xmath200 concentration , this effect has been dramatically observed in the arpes experiments and the constant chemical potential persists from the weakly doped insulator to optimally doped superconductor@xcite . in fact numerical calculations on the hubbard model also reveal similar divergent behavior of the compressibility as the metal - insulator transition is approached from the metallic side@xcite . the simplest explanation of the small chemical potential shift is a two phase mixture with different densities at a first order phase transition . if the system globally phase separates into two different spatial regions with different charge densities but the same free energy densities , the added charges only change the proportion of mixture of the two phases and do not change the energy , therefore , @xmath201 . however , this situation of global phase separation can certainly not occur in a system with long ranged coulomb interaction and is ruled out in the real high @xmath1 system . a phenomenon possibly related to the tendency of phase separation is the formation of stripes@xcite . a stripe state can be viewed as microscopic phase separation of af and sc into alternating regions , where each region has different charge density and the same free energy density . however , a crucial difference between the global phase separation and this picture of microscopic phase separation is that the stripe state has infinitely many surfaces between af and sc , and the surface energy makes a finite contribution to the total energy in the thermodynamic limit . in this picture , doping can be accomplished by converting af stripes into sc stripes , thereby creating more surfaces separating af and sc regions . finite doping density therefore leads to a finite density of surfaces and the accumulated surface energy would in general lead to a shift of the chemical potential _ under a generic situation_. additional physical conditions are needed to ensure the constant chemical potential in the stripe phase . therefore , the absence of the chemical potential shift places a very strong constraint on possible theoretical explanations . the projected @xmath0 model proposed in this work offers a possible explanation for the absence of chemical potential shift . at a critical value of the chemical potential where the af and sc states have degenerate energy density , we can have three situations , where the intermediate mixed states have higher , lower or degenerate energy compared to the af and sc states . when the intermediate states have higher energy , the system will go through a first order phase transition at @xmath136 and this will lead to global phase separation into af and sc regions . on the other hand , if the intermediate states have lower energy , there exists a range of chemical potential @xmath202 where the mixed phase has a uniform and continuously varying density . in this case , @xmath203 is obtained . when the region @xmath202 shrinks to zero , we obtain the limiting @xmath0 symmetric case where the system is on the boundary between a first order transition and two second order phase transitions . in this case , @xmath201 for a range of densities @xmath135 . therefore , if we restrict to ground states where the density is not globally inhomogeneous , the absence of the chemical shift directly implies @xmath0 symmetry . within this model , we can therefore define a experimental procedure to measure one of the most crucial parameter of the theory , namely @xmath184 . using the experimental arpes data for lsco system we would identify @xmath184 to be approximately the same as the optimal doping density . we recall that @xmath184 is also a measure of the degree of the quantum spin fluctuation in the system . for the bi - layer materials such as ybco and bisco superconductors , the quantum spin fluctuation are stronger due to the inter - layer spin exchange , and we would predict that @xmath184 should be less than the optimal doping value . having discussed the ground state properties and the phase diagram of the model , we are now in a position to study the collective excitations of the model . we have argued that the ground state energy can remain @xmath0 symmetric despite the projection . however , the projection does affect the collective excitation spectrum near half - filling . nonetheless , as we shall see , there remains a unique signature of the projected @xmath0 symmetry in the collective excitation spectra . in principle , the collective excitation spectra can be obtained straightforwardly by studying the quadratic fluctuations around the mean field minima . the resulting quadratic boson hamiltonian can be simply diagonalized . the main complication in the procedure is the hard - core boson constraint , which requires @xmath204 for every site . there are several ways to implement this constraint rigorously . one is to follow the mapping from the one - component hard - core boson model to the @xmath205 model and generalize it to a multi - component hard - core boson model . one could also convert the above inequality constraint to an equality constraint by introducing a boson creation and annihilation operator for the singlet state . this approach will be implemented in the appendix . for simplicity of presentation , here we shall adopt a less rigorous approach and introduce a on - site boson repulsion term @xmath206 to our hamiltonian ( [ hamiltonian ] ) and convert the hard - core constraint to a soft - core constraint . we shall show later that all results can be expressed in terms of the order parameter , which is implicitly dependent on @xmath207 , but there is no explicit dependence on @xmath207 . we have verified that all three methods give the same long wave length spectra for the collective modes in the limit of low boson density . in the next section , we shall present another calculation based on the continuum effective lagrangian method , which also reproduces the same spectra . to simplify presentation , we shall concentrate on the case where the ground state energy functional ( [ energy ] ) is @xmath0 symmetric , _ i.e. _ for coupling constants @xmath208 and @xmath209 . we choose the direction of spontaneously broken symmetry to be @xmath210 and @xmath211 . the extremal condition can be easily determined to be @xmath212 the combination @xmath213 expresses the fact that the classical minimum is @xmath0 symmetric . we can therefore write @xmath214 and @xmath215 . expanding the boson operators as : @xmath216 we obtain the following quadratic hamiltonian @xmath217 we are in particular interested in the collective mode spectra for the af insulating state with @xmath126 , the mixed states with @xmath163 and the sc state with @xmath127 and how they connect to each other . from this quadratic hamiltonian we can learn a number of important features . first we notice that the @xmath218 and @xmath219 modes are decoupled for all ranges of @xmath220 , but most importantly , their dispersion relations are independent of @xmath125 and given by @xmath221 where @xmath222 and @xmath223 is the lattice constant . this is indeed a very remarkable property . at @xmath126 , @xmath218 and @xmath219 modes are nothing but the transverse af spin wave modes . af spin waves are usually viewed as goldstone modes and their existence is due to the af long range order . however , as @xmath125 changes continuously from @xmath168 to @xmath169 , the af long range order continuously diminishes until it vanishes at @xmath127 . the reason that the properties of the @xmath218 and @xmath219 modes do not change at all is due to the @xmath0 symmetry of this model , since the diminishing af order is compensated by the increasing sc order as the superspin angle @xmath125 is varied . as we shall see , the @xmath224 mode is the af spin amplitude mode at @xmath126 , but it becomes massless and degenerate with the @xmath218 and @xmath219 modes at @xmath127 . these three modes form a massless @xmath225 triplet mode whose existence is purely a consequence of the sc order . therefore , as @xmath125 is continuously varied from @xmath168 to @xmath169 , the transverse af spin wave modes gradually change their character to become the @xmath225 triplet resonance of the sc state . as we shall see , for @xmath172 , the @xmath225 triplet mode becomes massive . at the af point @xmath126 , the spin amplitude mode @xmath224 is decoupled from the sc mode @xmath226 and can be diagonalized separately . the dispersion for the spin amplitude mode has the conventional massive relativistic form for small @xmath227 @xmath228 on the other hand , we have a massless sc goldstone mode @xmath226 with the following dispersion , @xmath229 this mode is an important new prediction of the @xmath0 theory . it is the counterpart of the @xmath225 resonance in the af state . in the unprojected @xmath0 model , there are two such modes , with charge @xmath7 , and they represent gapless fluctuations from af to sc at half - filling . in the projected @xmath0 model , the charge @xmath9 mode is projected out of the spectrum , however , the charge @xmath11 mode remain massless at @xmath136 . it is also a manifestation of the gapless fluctuation from af to sc at half - filling , but the sc fluctuation is hole - like , rather than both hole and particle - like as in the unprojected case . we see again that a large mott - hubbard gap is fully compatible with gapless sc fluctuation at half - filling . experimental detection of this mode could provide a important test of the projected @xmath0 symmetry . the @xmath224 and the @xmath226 modes also decouple in the pure sc state with @xmath127 . however , their physical interpretation change . the @xmath224 mode becomes gapless at this point with the same dispersion as in ( [ spin_wave ] ) . therefore , the three modes @xmath224 , @xmath218 and @xmath219 form a gapless @xmath225 triplet mode of the pure sc state , and represent the gapless fluctuation from sc to af at @xmath136 , but with a finite hole density @xmath184 given in ( [ rho_c ] ) . the dispersion for the @xmath226 mode is given by @xmath230 and has the natural interpretation of a linearly dispersing phase goldstone mode of the sc state . in the intermediate mixed state with @xmath163 , the @xmath224 and the the @xmath226 modes are coupled . diagonalization of these modes gives : @xmath231 the upper massive mode has predominantly spin amplitude character , and we see that the gap diminishes continuously until it reaches zero at @xmath127 to become the massless @xmath225 triplet mode . the lower mode has predominantly sc fluctuation character , and has a gapless @xmath232 dispersion . in the mixed region where both the af and the sc order parameters are non - zero , one would naturally expect a gapless phase mode corresponding to the sc order . however , in a interacting boson system , the phase mode is expected to have linear dispersion on general ground . therefore , what is surpassing and new here is not the gapless nature of the sc mode , but its _ quadratic dispersion_. in order to locate the origin of the quadratic dispersion , we have perturbed the model away from the projected @xmath0 symmetric point so that a uniform mixed state is stabilized as a classical minimum . sc fluctuation around such a non-@xmath0 symmetric point is gapless and has linear dispersion . a quadratic dispersion is only realized at the @xmath0 symmetric point . therefore , the quadratic dispersion is a unique signature of the projected @xmath0 symmetry in the entire range of densities @xmath135 ! to understand the physical origin of this remarkable phenomenon , we notice that a boson system with gapless quadratic dispersion generally has infinite compressibility . this can be directly seen from the compressibility sum rule @xmath233 where @xmath234 is the dynamical density correlation function . because of the quadratic dispersion relation , we can see explicitly that @xmath235 for small @xmath227 , therefore , a infinite compressibility is obtained . on the other hand , a infinite compressibility implies that @xmath236 , _ i.e. _ the chemical potential is independent of doping . but this is exactly the prediction of the projected @xmath0 model ! for @xmath135 , the chemical potential is pinned at the @xmath0 symmetric point @xmath136 , where the superspin vector can point in any direction . to accommodate a long wave length fluctuation of the hole density , the system rotates into another degenerate minimum with a different superspin angle @xmath125 and a different hole density . for this reason , the chemical potential does not change and the system is infinitely compressible . at the mean - field level , the ground - state energy of the hamiltonian ( [ hamiltonian ] ) with ( [ soft - core ] ) depends on the af and sc order parameters @xmath45 and @xmath148 only via their combination @xmath237 reflecting the so(5 ) invariance of the mean - field result . however , the zero - point energy of the bosons in the quadratic hamiltonian gives a correction to the ground - state energy due to quantum fluctuations . ( calculations of the quantum fluctuation effects at the @xmath238 spin - flop transition have been studied in ref . as we will show below , this correction turns out to depend on @xmath239 and @xmath240 separately . for the soft - constraint case , @xmath241 can be expressed as @xmath242 where @xmath243 are the four collective modes described in sec . [ modes ] ( although extended to all values of @xmath227 in the bz ) . in a systematic @xmath137 expansion , this correction to the ground state energy scales like @xmath137 , and is therefore small in the semi - classical limit . alternatively , @xmath241 can be seen as a small correction to the mean - field energy @xmath244 for small values of the parameter @xmath245 . however , contrary to @xmath244 , @xmath241 also depends on the superspin angle @xmath125 and thus produces a small so(5)-symmetry breaking . we now evaluate the @xmath125-dependent part of eq . ( [ egs ] ) in the small-@xmath246 limit . we thus parametrize @xmath247 ( @xmath248 ) , differentiate the expression ( [ egs ] ) with respect to @xmath249 , and expand it up to second order in @xmath246 . by further transforming @xmath250 with @xmath251 , the derivative of @xmath241 can be expressed as the sum of two terms @xmath252 where @xmath253 with @xmath254 @xmath255 and the density of states @xmath256 is defined as @xmath257 \;.\ ] ] unfortunately , the terms of the @xmath246 expansion in @xmath258 diverge when integrated over @xmath259 . it is thus convenient to carry out the transformation @xmath260 in @xmath258 and _ then _ expand in powers of @xmath246 . we obtain @xmath261 for small @xmath246 , the integral in @xmath83 can be extended to @xmath262 and we obtain from eq . ( [ intg ] ) @xmath263 while in evaluating eq . ( [ e1bf ] ) we only need the density of states at @xmath264 , @xmath265 , for the first term @xmath266 one needs @xmath256 in the whole domain @xmath267 however , @xmath266 is independent of @xmath249 and thus it merely fixes the value of the critical chemical potential . a numerical integration yields @xmath268 by integrating over @xmath249 eqs . ( [ e1af ] ) and ( [ e1bf ] ) , we finally obtain the total contribution to the ground - state energy correction . @xmath269 @xmath241 thus lifts the degeneracy as a function of @xmath249 and initially favors the pure superconducting phase ( @xmath270 ) . a small chemical potential term @xmath271 with @xmath272 restores the degeneracy between the pure superconducting ( @xmath270 ) and the pure antiferromagnetic ( @xmath273 ) phases . however , due to the convexity of @xmath241 as a function of @xmath249 there is always a barrier ( @xmath274 ) between the two phases since the mixed phase always has a higher energy . this means that at @xmath275 and for intermediate densities the system prefers to phase separate between the two pure phases rather than choosing the mixed phase . however , the important point is that this barrier remains small in the limit @xmath276 . in view of the symmetry breaking effects of the quantum fluctuations , it would be interesting to see whether there is a limit when the the wave function ( [ variational ] ) and the projected @xmath0 symmetry becomes exact . rokhsar and kotliar@xcite have shown that this type of wave functions are actually exact in the limit of infinite dimensions . therefore , besides the @xmath137 expansion , we could also use a @xmath277 expansion ( where @xmath47 is the space dimension ) to systematically control the @xmath0 symmetry breaking quantum effects . besides quantum fluctuations , there are also other symmetry breaking terms . nearest - neighbor and next - nearest - neighbor coulomb interactions also break the @xmath0 symmetry , however , their corrections to the ground state is concave , _ i.e. _ the energy of the intermediate states are lowered . therefore , they can also lead to uniform mixed states in some region of the phase diagram . the detailed study of all these competing effects will be carried out in subsequent works . while the projected @xmath0 model defined on a lattice enables us to make some connection to the underlying microscopic physics , for most discussions concerning the long wave length and low energy degrees of freedom , it is desirable to have a effective continuum lagrangian . such a formulation can be directly obtained by taking the long wave length limit of the projected bosonic model discussed previously . however , in order to make the connection to the unprojected @xmath0 model clearer , we shall motivate our discussion from the original @xmath0 effective model . the effective lagrangian for a fully @xmath0 symmetric model takes the form of @xmath278 where @xmath249 measures the superspin susceptibility , @xmath279 measures the superspin stiffness , @xmath280 denotes the spatial directions and @xmath281 is a scalar function of the superspin magnitude @xmath282 only . there are three important symmetry breaking effects connected with the presence of a large mott - hubbard gap . first is an asymmetry in the scalar potential , which can be described by a additional term @xmath283 which for positive @xmath284 favors af at half - filling . second is the asymmetry between the spin ( @xmath285 ) and the charge ( @xmath286 ) susceptibilities , which modifies the kinetic energy to @xmath287 the last symmetry breaking effect is due to the chemical potential @xmath288 , which enters the lagrangian as a gauge coupling in the time direction , and modifies the charge part of the kinetic energy to @xmath289 combining these three symmetry breaking terms , we obtain @xmath290 in the presence of a large mott - hubbard gap , all these three symmetry breaking terms are of the order of @xmath4 , _ i.e. _ @xmath291 . therefore , this lagrangian contains high energy degrees of freedom of the order of @xmath4 . however , as already observed in @xcite , at @xmath292 , their effects cancel completely in the time - independent part of the lagrangian , and the static potential is @xmath0 symmetric just as in the original unprojected model . we also observe that near the af / sc transition point where @xmath293 , the first order time derivative term is of the order of one . furthermore , in the spirit of the low frequency and wave vector expansion , we only need to retain the first order time derivative term in the charge sector and can drop the second term in the above lagrangian . combining all these considerations , we obtain the following low energy effective lagrangian near the af / sc transition region , which is free of any parameters of the order of @xmath4 : @xmath294 this is exactly the lagrangian counterpart of the @xmath0 projection procedure discussed previously in the hamiltonian language . dropping the second order time derivative terms removes half of the ( high energy ) degrees of freedom , and redefines the canonical conjugacy of the dynamical variables . in particular , the conjugate variable of @xmath25 is nothing but @xmath26 itself , since @xmath295 . standard quantization procedure requires the canonical commutation relation @xmath296=i$ ] , which in this case just reproduces equation ( [ non - commute ] ) . this confirms the fact that the @xmath0 projection does not change the form of the interaction potential , only the commutation relation between @xmath25 and @xmath26 . it is easy to see that the low energy effective lagrangian ( [ projected_l ] ) produces exactly the same long wave length collective mode spectrum as the projected @xmath0 hamiltonian ( [ hamiltonian ] ) defined on a lattice . to facilitate the comparison , we take the @xmath0 potential to be @xmath297 assuming broken symmetry in the @xmath25 and @xmath298 directions , we find that the @xmath299 and @xmath300 modes always decouple , and they have a linear spin wave dispersion relation with @xmath301 . the euler - lagrangian equations of motion gives the following dispersion relation for the @xmath25 and @xmath298 modes : @xmath302 for the af state with @xmath303 , @xmath304 for the sc state with @xmath305 , and @xmath306 for the mixed state with @xmath307 and @xmath308 . these dispersion relations agree exactly with the lattice model results at the projected @xmath0 symmetric point if we make the following identification @xmath309 the effective lagrangian can be easily used to discuss effects of @xmath0 symmetry breaking . the simplest form of symmetry breaking is increasing the chemical potential beyond the critical value @xmath310 , so that a pure sc state is realized . the chemical potential enters the effective lagrangian through the gauge coupling in the time direction via the following substitution @xmath311 where @xmath312 is the deviation of the chemical potential away from the critical value . in this case , the spin triplet excitations acquire a finite mass gap , with the following dispersion relation : @xmath313 and the mass gap increases with increasing doping in the sc state . we summarize the behavior of the collective modes obtained in the previous two sections in fig . ( [ fig5 ] ) . = 6.0 cm we see that while there are significant modifications of the collective mode spectra in the density regime @xmath135 from the unprojected @xmath0 symmetry , the spectra beyond @xmath184 is essentially identical to the behavior expected from the unprojected @xmath0 symmetry . this should be expected from our general considerations about the gutzwiller projection without much detailed calculations . we argued that the only effect of the gutzwiller projection is to change the quantum commutation relation between the sc components of the superspin @xmath25 and @xmath26 . however , for @xmath173 , the system is in a pure sc phase where these components acquire classical expectation values . in this case , the modification of the quantum commutation relation does not have any significant effect . this argument can also be illustrated by a simple picture of a _ chiral _ @xmath0 sphere , as depicted in fig . ( [ fig6 ] ) . = 8.0 cm in this picture , the north and south poles represent the three af directions , and the equatorial plane represent the sc directions . the sphere is perfectly @xmath0 symmetric . however , the chemical potential along the pole direction acts like a fictitious magnetic field which restricts the sense of the rotation in the sc plane . small oscillations of a vector pointing close to the north pole enclose the fictitious magnetic flux , and can only execute chiral rotations . this amounts to the projection of the particle - pair states at half - filling . on the other hand , small oscillations of a vector pointing anywhere along the equator does not enclose the fictitious magnetic flux , and their dynamics is therefore unaffected by the projection . dynamics of a vector pointing anywhere between the north pole and the equator is also partially affected by the projection , but the symmetry of the static potential bears a unique signature . the main purpose of this paper is to introduce the concept of projected @xmath0 models and discuss the properties of this model in connection with high @xmath1 superconductivity . the projected @xmath0 model describes the low energy and long distance bosonic degrees of the freedom near the af / sc transition . we showed that the gutzwiller projection can be implemented analytically on every site in the @xmath0 theory . in the presence of a infinite mott - hubbard gap , we show that static properties of the model can remain @xmath0 symmetric , while the modification of the dynamics can be completely cast into a non - trivial commutation relation between the two sc components of the @xmath0 superspin . unlike the unprojected @xmath0 models which can only have the full dynamic @xmath0 symmetry at half - filling , the projected @xmath0 model can have static @xmath0 symmetry at a critical value of the chemical potential @xmath310 and for a finite range of doping @xmath314 . at @xmath186 , the system has a af ground state and zero compressibility . in the intermediate regime @xmath135 , the system has mixed af and sc order and infinite compressibility . for @xmath173 , the system has a pure sc ground state , the sc order parameter rises to a maximal value before it decreases with doping . at the projected @xmath0 symmetric point , we can understand precisely the evolution of the collective modes . on the af side , we have two gapless spin wave modes and a gapless charge mode describing the gapless fluctuation from af to sc . in the intermediate density regime @xmath135 , the physical properties of the spin waves remain unchanged , while the massive spin amplitude mode gradually decreases its energy and merges with the two spin wave modes at @xmath315 . the charge mode in the intermediate density regime is gapless , but has quadratic dispersion relation , which is a unique signature of the projected @xmath0 symmetry . for @xmath173 , the charge mode is gapless with linear dispersion relation , and the @xmath225 triplet spin mode becomes massive , and gradually increases its energy with increasing doping . in this regime , the behavior of the collective modes are identical to the unprojected @xmath0 model . this very simple model can form the basis to understand many novel and puzzling properties of the high @xmath1 superconductors in a unified framework . it points out a route from af to sc through a gradual rotation of the superspin angle . at the projected @xmath0 symmetry point the mean field energy is independent of the superspin angle , and therefore it offers an explanation of the absence of the chemical potential shift in the underdoped regime without global phase separation . it predicts a phase diagram which is qualitatively consistent with the observed phase diagram in the high @xmath1 materials . in the underdoped regime of the phase diagram , the system have large af and sc fluctuations , and these fluctuations can be responsible for the pseudogap physics observed in these materials . there are many possible directions to carry out this line of research in the future . the most important issue is to understand the precise nature of the intermediate state in the regime @xmath135 . since the system has infinite compressibility in this regime , different small perturbation may select different ground states . such perturbing effects might include quantum fluctuations and longer ranged interactions . in particular , we would like to investigate the possibility that these perturbations might lead to the formation of incommensurate order or stripes . in this work , we have discussed extensively the collective fluctuations in the long wave length limit . due to the definitions of our effective lattice model , the @xmath316 limit corresponds to the @xmath316 limit in the sc correlation functions and the @xmath317 limit of the af spin correlation functions . within the @xmath0 theory , the @xmath225 resonance in the sc state is viewed as the @xmath0 symmetry partner of the @xmath316 goldstone mode of the sc phase fluctuation . while the commensurate neutron resonance mode is observed in both ybco and bisco superconductors , all high @xmath1 systems also have incommensurate spin fluctuations . how can these features be explained within the current theoretical model ? the fact that lsco and ybco have very different fermi surface shapes and yet have similar incommensurate magnetic peaks strongly suggests that the incommensurate peaks are not sensitive to fermi surface effects and should be explainable within an effective bosonic model . let us recall that the collective mode of a superfluid boson system consists of linearly dispersing phonon branch and another roton branch with a minimum located at the inverse inter - particle spacing . so far , we have only studied the phonon branch of the charged bosons . by analogy , the roton branch should also exist , with a wave vector determined by the density of the charged bosons or doping . within the @xmath0 theory , while the commensurate neutron resonance can be viewed as the @xmath0 partner of the sc phase mode , the incommensurate magnetic peaks can be viewed as the @xmath0 partner of the roton minimum of the charged bosons . a detailed quantitative analysis of this picture will be carried out in the future . however , while the ground state in the doping range @xmath135 may depend sensitively on small perturbation effects , at finite temperature , these perturbation effects should be small and the system should display more universal properties . we have shown that the projected @xmath0 symmetry should be valid for the entire doping range @xmath135 , and we shall quantitatively study the manifestation of this symmetry at finite temperature , and see if the projected @xmath0 symmetry can give a universal explanation of the pseudogap physics . _ note added : _ after completing this work , we received a very interesting paper by coen van duin@xcite , in which he also observed the remnant @xmath0 behavior in the large @xmath4 limit " . we would like to acknowledge useful discussions with d. arovas , j. berlinsky , e. demler , r. eder , c. kallin , s. kivelson and d. scalapino . scz and jph are supported by the nsf under grant numbers dmr-9814289 . wh and ea are supported by forsupra ii , bmbf ( 05 605 wwa 6 ) and by the deutsche forschungsgemeinschaft ( ar 324/1 - 1 ) and ( ha 1537/17 - 1 ) . aa is supported by the israel science foundation . aa , ea and wh would like to acknowledge the support and hospitality of the stanford physics department , where most of this work was carried out . alternatively , one can enforce the hard - core constraint ( [ hard - core ] ) by introducing an additional `` slave '' boson for each lattice site . the presence of this boson ( @xmath318 ) indicates that the lattice site @xmath45 is in the singlet state . the `` less or equal '' hard - core condition ( [ hard - core ] ) is replaced with the equality constraint @xmath319 with @xmath320 . since in physical states one always has one and only one boson per lattice sites , destruction ( creation ) of a boson @xmath321 ( @xmath322 ) must always be accompanied by creation ( destruction ) of the empty boson @xmath323 ( @xmath324 ) . in this way , the _ physical _ operators for creating ( destroying ) a triplet ( @xmath325 ) or a hole pair ( @xmath326 ) acquire the form @xmath327 ( @xmath328 ) . the advantage of this method is that the constraint can be enforced exactly ( at least in principle ) by introducing an additional time - independent field @xmath329 at each lattice site , by adding to the hamiltonian a term @xmath330 and by integrating over the @xmath329 on the imaginary axis . the whole hamiltonian ( [ hamiltonian ] ) thus takes the form ( apart for a constant ) @xmath331 in practice , one starts with a mean - field approximation and expands the boson operators around their mean - field values as in sec . [ modes ] . this expansion can be rigorously controlled by generalizing the constraint ( [ constraint ] ) to large values of @xmath332 , whereby one scales @xmath333 ( cf . @xcite ) . physically , this corresponds to allow for a large number of bosons to be present at each site and thus to have a large value for the total spin , or more precisely for the so(5 ) quantum number @xmath133 , at each site . the mean - field result thus corresponds to the @xmath334 limit , while the quadratic expansion corresponds to the first @xmath335 correction . at the mean - field level , the constraint is fulfilled exactly and indeed one obtains the same result and the same phase diagram as the variational ansatz ( [ variational ] ) discussed in sec . [ phase ] . by expanding the bosons quadratically around the mean - field one obtains the same modes as for the soft - core hamiltonian with a similar dispersion . ( here , we restrict again to the so(5)-symmetric case ) . specifically , in the mixed phase one obtains two spin - wave modes with dispersion @xmath336 one massive spin - amplitude mode @xmath337 and a quadratic mode @xmath338 in the pure superconducting phase , one has a @xmath225-triplet with dispersion @xmath339 and the sc goldstone mode with dispersion @xmath340 these results coincide with the ones of the soft - constraint approximation eqs . ( [ spin_wave ] ) to ( [ mixed_modes ] ) in the limit of small @xmath341 , i. e. , for low boson density .

many cell signaling pathways involve the diffusion of _ messengers _ that bind / unbind to intracellular components . quantifying their net transport rate under different conditions , then requires having separate estimates of their free diffusion coefficient and binding / unbinding rates . in this paper , we show how performing sets of _ fluorescence correlation spectroscopy _ ( fcs ) experiments under different conditions , it is possible to quantify free diffusion coefficients and on and off rates of reaction - diffusion systems . we develop the theory and present a practical implementation for the case of the universal second messenger , calcium ( ca ) and single - wavelength dyes that increase their fluorescence upon ca binding . we validate the approach with experiments performed in aqueous solutions containing ca and fluo4 dextran ( both in its high and low affinity versions ) . performing fcs experiments with tetramethylrhodamine - dextran in _ xenopus laevis _ oocytes , we infer the corresponding free diffusion coefficients in the cytosol of these cells . our approach can be extended to other physiologically relevant reaction - diffusion systems to quantify biophysical parameters that determine the dynamics of various variables of interest .

1609.04007

many cell signaling pathways involve the diffusion of messengers in the cytoplasm . in most cases these substances convey their message by binding to target molecules . furthermore , as they reach their targets they not only diffuse freely but also bind / unbind to other cell components . for long enough times , the net transport that results from the combination of free diffusion and binding / unbinding is described by _ effective diffusion coefficients _ that are weighted averages of the free coefficients of the messenger and of the substance it interacts with that depend on their concentrations and reaction rates . the universal second messenger calcium ( @xmath1 ) provides a prototypical example of this behavior . persistently high cytosolic @xmath1 concentrations lead to cell death . for this reason cells have numerous mechanisms to reduce this concentration , the fastest one of which is _ buffering_. buffers are molecules that bind / unbind to / from @xmath1 ions reducing their free concentration . in doing so they also alter the spatio - temporal cytosolic @xmath1 distribution @xcite , @xcite and the effect of the resulting signals on the eventual end responses @xcite . this means that the time and spatial range of action of @xmath1 as messenger is strongly dependent on the @xmath1 concentration itself . there is a large variety of intracellular @xmath1 signals ranging from those that arise upon the opening of a single @xmath2 channel on the plasma membrane or on the membrane of intracellular stores to those that manifest themselves as @xmath1 waves that travel throughout the whole cell @xcite . this in turn implies that the cytosolic @xmath1 concentration attains very different values depending on the signal type . the resulting @xmath1 effective diffusion coefficient then varies across disparate values depending on the signal type . the range of values was estimated to be @xmath3 [ 5 , 220 ] * @xmath4 * @xcite with increasing values as the cytosolic @xmath1 concentration increased . how fast can @xmath1 diffuse inside cells ? in order to answer this question it is necessary to have separate estimates of the @xmath1 and @xmath1 buffers free diffusion coefficients , their concentrations and reactions rates . ideally , having access to this information one could eventually compute the @xmath1 effective diffusion coefficient as a function of its concentration . one could wonder why having access to this information would be necessary to study @xmath1 signals . after all , they can be observed in intact cells using @xmath1 dyes . the @xmath1 dye fluorescence , however , provides information on the @xmath1-bound dye concentration which distribution depends on the dye kinetic and transport properties . in particular , @xmath1 signals that are evoked via the photo - release of caged compounds with uv light are imaged using single wavelength dyes that increase their fluorescence when bound to @xmath1 @xcite , @xcite . it was shown in bruno et al . , 2010 @xcite that @xmath1 current estimates inferred from images that use such dyes are quite sensitive to uncertainties in the on rate of the @xmath1-dye binding reaction and in the diffusion coefficient of the dye , two parameters that are usually poorly known . having reliable estimates of these parameters is thus unavoidable to extract quantitative information from images of @xmath1 signals and allow the effective interplay between modeling and experiment that is necessary to attain a comprehensive description of the signals . in this paper we describe and implement an approach that shows how performing sets of _ fluorescence correlation spectroscopy _ ( fcs ) @xcite experiments under different conditions and using a reaction - diffusion model to interpret the experimental results it is possible to obtain separate estimates of these key biophysical parameters . in fcs the fluorescence intensity in a small volume is recorded along time and , via an analysis of the temporal autocorrelation of the observed fluctuations , the transport rates of the fluorescent species are , in principle , derived @xcite . fcs has been widely used to determine the diffusion coefficients of fluorescently labeled proteins inside cells @xcite , @xcite , @xcite . when the fluorescent species diffuse freely in mediumthere is an analytic expression for the autocorrelation function of the fluorescent fluctuations ( acf ) that is used to fit the experimental observations and derive diffusion coefficients ( see materials and methods ) . when the fluorescent particles diffuse and react , simple analytic expressions for the acf can only be obtained under certain approximations @xcite , @xcite , @xcite , @xcite . in particular , we have shown in sigaut et al . , 2010 @xcite ( see also @xcite , @xcite ) , that when the reactions occur on a somewhat faster timescale than diffusion the correlation time of the fluctuations is determined by the effective diffusion coefficients mentioned before . our theoretical studies showed that information on reaction rates could also be derived from the fitting @xcite , @xcite . in this paper we present a practical implementation of such an approach in which the experimental conditions are changed so as to maximize the information that can be drawn from the data . more specifically , we do it for the case of @xmath1 and single wavelength @xmath1 dyes . this case shares some common features with the case in which proteins diffuse and react , but it differs slightly since fluctuations are also due to changes in fluorescence intensity associated to the @xmath1-dye binding / unbinding reaction . this requires the development of a new theoretical framework that we introduce in this paper as well . in order to advance with the practical implementation presented here we first study theoretically the behavior of the acf for a case with @xmath1 and a single wavelength dye . we derive an analytic approximation under the assumption that the @xmath1-dye reaction occurs on a fast timescale . we compare this approximated acf with the one without approximations computed numerically and determine the range of parameter values for which the approximated acf can give reasonable estimates of certain parameters . we then show the results of a series of fcs experiments performed in aqueous solutions containing @xmath2 and different amounts of the @xmath1 indicator fluo4 dextran both in its high and low affinity versions ( invitrogen - molecular probes , carlsbad , ca ) . fitting the observed acf by the analytic approximation we corroborate the validity of the approximations and derive diffusion coefficients and the off - rate of the @xmath1-dye binding reaction , in solution . a similar approach can be used to characterize the kinetic properties of other @xmath5 dyes . even if the free diffusion coefficients in solution and in the cytosol are different due to differences in viscosity between both media , we may assume that the ratio between the free diffusion coefficients of any two substances remains the same in both settings . this is particularly relevant , because by solely quantifying the rate of diffusion of a molecule that diffuses freely in the cytosol and in solution we can infer the free diffusion coefficient of @xmath1and the dyes in the cytosol as well . we present such quantification in the appendix . thus , the practical implementation presented in this paper not only highlights the advantages of our approach but it also allows us to derive information that is key to quantify the free @xmath5 distribution that underlies a @xmath1 image . @xmath1 signals are not the only example in which being able to tell apart the contributions of free diffusion and reactions on the net transport rate of labeled substances is relevant . we have recently shown @xcite the necessity of going beyond the description of effective coefficients to interpret correctly the apparently disparate estimates of the protein , bicoid , diffusion coefficient derived from fcs @xcite and fluorescence recovery after photobleaching ( frap ) @xcite experiments . this example also shows that the comprehensive quantifiable description of a physiological process requires having a biophysical model for the dynamics of the relevant concentrations that depends on concentration - independent biophysical parameters . it is via such a model that the response of the system over time in front of different stimuli can be predicted . being able to derive estimates of the concentration - free biophysical parameters _ in situ _ is thus relevant to achieve a meaningful description . the approach presented in this paper can be adapted and applied to other problems therefore , its relevance goes beyond quantifying the biophysical parameters associated to @xmath1 and its dyes . fluorescence correlation spectroscopy ( fcs ) monitors the fluctuation of the fluorescence in a small volume . fluctuations are characterized by the time - averaged autocorrelation function ( acf ) which is defined as : @xmath6 where @xmath7 is the average fluorescence in the sampling volume and @xmath8 is the deviation with respect to this mean at each time , @xmath9 . as explained in the appendix , when the fluorescence comes from a single species , @xmath10 , that diffuses freely with coefficient @xmath11 ( i.e. , does not react ) the acf is of the form : @xmath12 where @xmath13 is the aspect ratio of the sampling volume and @xmath14 and @xmath15 are the sizes of the beam waist along @xmath16 and @xmath17 , with @xmath16 the spatial coordinate along the beam propagation direction and @xmath17 a radial coordinate in the perpendicular plane ; the effective volume is @xmath18 ; @xmath19 is the characteristic time of diffusion of the particles across the sampling volume and @xmath20 , where @xmath21 is the particle concentration . when the dynamics of the fluorescent species is described by a reaction - diffusion model most often there is not a simple analytic expression for the acf . it can always be written as a sum of integrals each one associated to one of the branches of eigenvalues that rule the dynamics of the linearized reaction diffusion equations of the model . each of these integrals is called a `` component '' of the acf . in the case of interest for the present paper there are three relevant species : free @xmath1 ( ca ) , free dye ( @xmath22 ) and @xmath1-bound dye ( @xmath23 ) , which diffuse with free coefficients @xmath24 ( @xmath25 ) , and @xmath26 ( @xmath22 and @xmath23 ) and react according to : @xmath27 with on- and off - rates , @xmath28 and @xmath29 respectively . the corresponding ( spatially uniform ) equilibrium concentrations , @xmath30 , @xmath31 and @xmath32 satisfy : @xmath33 where @xmath34 and @xmath35 is the total dye concentration . there are three branches of eigenvalues for this system and the acf then has three components . simple algebraic expressions can be obtained for the components in certain limits . in particular , in this paper we present the results obtained in the `` fast reaction limit '' that holds when the characteristic time of the reaction eq . ( [ eq : reaction ] ) is shorter than the time it takes for the species to diffuse across the observation volume ( i.e. , if @xmath36 ) . for more details see the appendix where we also compare the `` full '' ( integral expression of the ) acf computed numerically with the analytic approximation derived in the fast reaction limit that is presented in the results section and some of their components separately . aqueous solutions were prepared with different concentrations of the @xmath1 indicator fluo4 dextran high or low affinity ( invitrogen- molecular probes , carlsbad , ca ) , employing the solutions of a @xmath2 calibration buffer kit ( invitrogen- molecular probes ) . each solution contained 4.3 @xmath37 @xmath1 , 100 mm kcl , 30 mm mops , ph 7.2 , and different concentrations of the @xmath1 indicator ranging from 200 @xmath38 to 9@xmath37 and from 400 @xmath38 to 20@xmath37 for the low and high affinity version , respectively . four or five separate experiments were performed for each solution . some of the results were finally discarded as explained later . the solutions that were probed and fitted are listed in table [ tab : soluciones_f4_ca ] . c c c + aqueous solution & @xmath39 & @xmath40 + + 1 & 4285 & 429 + 2 & 4285 & 857 + 3 & 4285 & 1371 + 4 & 4285 & 1886 + 5 & 4285 & 2571 + 6 & 4285 & 4286 + 7 & 4285 & 9000 + 8 & 4285 & 15000 + 9 & 4285 & 19286 + + 10 & 4285 & 214 + 11 & 4285 & 429 + 12 & 4285 & 857 + 13 & 4285 & 1114 + 14 & 4285 & 1371 + 15 & 4285 & 1886 + 16 & 4285 & 2571 + 17 & 4285 & 4286 + 18 & 4285 & 9000 + + + fcs measurements were performed on a spectral confocal scanning microscope fluoview 1000 ( olympus , tokyo , japan ) , employing a 60x , 1.35 n.a . oil - immersion objective ( uplansapo , olympus ) and a pinhole aperture of 115 @xmath41 . single point measurements at a 50 khz sampling rate were performed for a total duration of 167 s ( equivalently , 8365312 data points ) employing a 488 nm line and detecting the fluorescence in the range ( 500 - 600 ) nm . the measurements were performed at @xmath320 @xmath41 from the coverslip . experimental acf s were calculated with a custom - made routine written on the matlab platform @xcite . to this end , each 167 s long record was divided into @xmath42 , 164 ms long segments containing 2@xmath43 points each for the experiments in aqueous solutions . the acf was computed for each of the @xmath42 segments from which the average acf was obtained . based on the theoretical calculations presented in the results section , we fitted the average acf by an expression of the form @xmath44 where @xmath13 is the aspect ratio of the sampling volume , as before , and the various times are related to diffusion coefficients by @xmath45 , @xmath46 , with the beam waist , @xmath15 . only experiments for which the mean fluorescence in the observation volume remained approximately constant during the whole record were fitted . experiments for which the average acf was too noisy were also discarded . in all cases we tried to fit the experiments leaving all 7 parameters of eq . ( [ eq : g3comp ] ) ( @xmath47 , @xmath48 , @xmath49 , @xmath50 , @xmath51 , @xmath52 , @xmath53 ) free to be fitted . in others we set @xmath54 and only derived @xmath47 , @xmath48 , @xmath50 and @xmath51 . thus , we tried a 3-component and a 2-component fit for each experiment . all fitting parameters were determined for each average acf via a nonlinear least squares fit using the matlab built - in function nlinfit . in the figures we show the average of the displayed fitting parameters and the average error computed over all the experiments in a given set . the radial beam waist and the aspect ratio were determined to be @xmath55 0.262 - 0.292 @xmath41 and @xmath56 by measuring the translational three - dimensional diffusion of fluorescein ( sigma , st . louis , mo ) in buffer solution ph 9 , assuming a diffusion coefficient of 425 @xmath57 @xcite . thus , the resulting effective volume was @xmath58 ( 0.59 @xmath59 0.1 ) @xmath60 . proceeding as described in the appendix we determine that , in the fast reaction limit for the case of a solution of @xmath5 and a dye the acf of the fluorescence fluctuations can be approximated by the sum of three components of the form : @xmath61 @xmath62 @xmath63 @xmath64 where @xmath65 and @xmath66 , @xmath67 , with : @xmath68 @xmath69 and @xmath70 and : @xmath71 the weights @xmath72 , @xmath73 and @xmath74 are given by : @xmath75 @xmath76 @xmath77 the sum of all the weights is inversely proportional to the concentration of fluorescent particles : @xmath78 the sum of the two effective diffusion coefficients satisfies : @xmath79 as in sigaut et al . , 2010 @xcite , the approximate analytic expression of the acf given by eqs . ( [ eq : gapprox])-([eq : gef2 ] ) is always valid for large enough @xmath80 . the first term , however , is exact . thus , we can expect to be able to derive @xmath26 from all fcs experiments . the approximations of @xmath73 and @xmath81 are valid for the values of @xmath80 that are relevant to determine @xmath82 and @xmath83 from a fit to the experiments if @xmath84 . in this section we show how we proceed to analyze the experimental data . in particular , we show the results of using eq . ( [ eq : g3comp ] ) to fit the acf s obtained from the set of experiments of table [ tab : soluciones_f4_ca ] performed with fluo4 high and low affinity . the fitting parameters are @xmath47 , @xmath85 , @xmath86 , @xmath53 and the characteristic times @xmath50 , @xmath87 and @xmath88 from which we derive three diffusion coefficients @xmath89 , @xmath90 and @xmath91 as explained before . fig . [ fig : d_efectivos ] shows the diffusion coefficients obtained in this way as a function of the total concentration of the dye used in the solutions , @xmath92 , for high affinity ( fig . [ fig : d_efectivos ] ( a ) ) and low affinity ( fig . [ fig : d_efectivos ] ( b ) ) fluo4 . we also plot in these figures the expected free diffusion coefficient of the dye , @xmath9385 @xmath94 @xcite , and effective diffusion coefficients , @xmath95 and @xmath96 , calculated using eqs . ( [ eq : def1 ] ) and ( [ eq : def2 ] ) , with the dissociation constant given by the manufacturer ( @xmath97 772 nm for high affinity and @xmath972600 nm for low affinity ) , @xmath98760 @xmath94 @xcite , @xmath9385 @xmath94 and the total calcium and dye concentrations employed in the solutions . the identification between the fitting parameters @xmath47 , @xmath89 , @xmath85 , @xmath99 , @xmath86,@xmath91 , @xmath53 , and the seven quantities , @xmath72 , @xmath26 , @xmath73 , @xmath95 , @xmath74 , @xmath96 , @xmath100 , of the theoretical formulas eqs . ( [ eq : gapprox ] ) - ( [ eq : gef2 ] ) is immediate in the case of the last three which correspond to the only component with an exponentially decaying term . for the other quantities it is not difficult to make the correspondence because @xmath101 implies that @xmath102 . furthermore , as may be observed in fig . [ fig : d_efectivos ] , there is one diffusion coefficient obtained from the fitting , @xmath89 , that remains approximately invariant for all the analyzed concentrations . this should correspond to the free diffusion of the dye , @xmath26 , which is concentration independent . in this way we determine that @xmath103 in the case of fluo4 high affinity and @xmath104 in the case of fluo4 low affinity . this is clearer in fig . [ fig : d_efectivos ] ( a ) , and the lowest constant diffusion coefficient can also be identified in fig . [ fig : d_efectivos ] ( b ) . the other two diffusion coefficients obtained from the fitting , @xmath90 and @xmath91 , change with the dye concentration . this means that they are effective diffusion coefficients . making the identifications @xmath105 and @xmath106 we know that their lower and upper limits are the free diffusion coefficients of the dye , @xmath26 , and of calcium , @xmath24 , respectively . in fact , both @xmath90 and @xmath91 are larger than @xmath89 . furthermore , in fig . [ fig : d_efectivos ] ( a ) , @xmath99 decreases with @xmath92 while @xmath91 increases similarly to their theoretical counterparts , @xmath95 and @xmath96 . this shows the validity of the identification between fitting and model parameters . a similar trend can be observed in fig . [ fig : d_efectivos ] ( b ) although not as clear as in fig . [ fig : d_efectivos ] ( a ) . in any case , we do make the identification @xmath107 and @xmath108 also in this case . it is remarkable that the obtained results seem reasonable even outside the range of validity of the fast reaction approximation . [ ! b ] ] diffusion coefficients obtained from the fitting of the experimental data using eq . ( [ eq : g3comp ] ) , @xmath89 ( squares ) , @xmath99(circles ) , and @xmath91(triangles ) , as a function of the total calcium dye concentration of the aqueous solutions , @xmath92 . in solid line , @xmath109 , and in dash lines expected effective diffusion coefficients , @xmath95 ( bold ) and @xmath96 ( light ) given by eqs . ( [ eq : def1 ] ) and ( [ eq : def2 ] ) respectively , with the calcium and dye concentrations employed in the aqueous solutions , @xmath110 760 @xmath111 , @xmath93 85 @xmath94 and the dissociation constant given by the manufacturer , @xmath97 772 nm and 2600 nm for high ( a ) and low ( b ) affinity fluo4 . we test the validity of the theoretical approximation eqs . ( [ eq : gof ] ) and ( [ eq : gotot ] ) in fig . [ fig : gof_gotot_nu ] . fig . [ fig : gof_gotot_nu ] shows plots of the inverse of @xmath72 as a function of the total dye concentration used in the solutions , @xmath92 , with symbols , together with the expected values given by eq.([eq : gof ] ) using the concentrations used in the solutions and the observation volume derived from the calibration , with curves . the results for high and low affinity fluo4 are shown in figs . [ fig : gof_gotot_nu ] ( a ) and [ fig : gof_gotot_nu ] ( d ) , respectively . the logarithmic scale used in the figures highlights the fact that both the experimental and the theoretical results scale similarly with @xmath92 , i.e. , as 1/@xmath92 ( see eq . ( [ eq : gof ] ) ) . if we fit the experimental results using the values , @xmath92 , determined by construction of the solution , the effective volume , @xmath112 , can be obtained from the fitting . considering the inverse of @xmath72 versus @xmath92 , we found expected values ( @xmath58 ( 0.54 @xmath59 0.08 ) @xmath113 and @xmath58 ( 0.56 @xmath59 0.08 ) @xmath114 for high affinity and low affinity fluo4 , respectively ) that are consistent with the one obtained from the calibration ( @xmath58 ( 0.59 @xmath59 0.1 ) @xmath60 ) . another property of the acf is that the sum of all the weights , @xmath115 , is inversely proportional to the concentration of @xmath2-bound dye , @xmath32 ( eq.([eq : gotot ] ) ) . in figs . [ fig : gof_gotot_nu ] ( b ) and [ fig : gof_gotot_nu ] ( e ) we show plots of the values of the inverse of @xmath115 obtained from the fitting of the acf as functions of @xmath32 and the expected values according to eq . ( [ eq : gotot ] ) , for high affinity and low affinity fluo4 , respectively . the linear scaling between both quantities is very good also in this case but there is a mismatch in the ordinate . as before , we can fit the experimental results using the equilibrium values , @xmath32 , derived from the concentrations used in the solutions and the dissociation constant provided by the vendor . considering the inverse of @xmath115 versus @xmath32 and fitting with a linear relation , the effective volume inferred was ( 0.23 @xmath116 0.02 ) @xmath60 for fluo4 high affinity and ( 0.17 @xmath116 0.01 ) @xmath60 for the low affinity version of the dye , which are lower than the one obtained from the calibration ( @xmath117 ( 0.59 @xmath59 0.1 ) @xmath60 ) . finally we show the values of @xmath100 derived from the fitting and the theoretical curve obtained using the fast reaction approximation , eqs . ( [ eq : gapprox])-([eq : gef2 ] ) , as a function of @xmath92 for high affinity ( fig . [ fig : gof_gotot_nu ] ( c ) ) and low affinity ( fig . [ fig : gof_gotot_nu ] ( f ) ) fluo4 . there we observe that the values obtained for low @xmath92 concentrations are the ones that can be associated to the theoretical expression ( eq . ( [ eq : nu ] ) ) from which an estimate of @xmath29 can be derived . in order to estimate @xmath29 however we used all the data available as explained in the discussion . ] parameters derived from the fitting of the experimental acfs ( with symbols ) and theoretical expected values ( solid lines ) . in a logarithmic scale , ( left ) inverse of the @xmath72 as function of the total dye concentration used in the solutions , @xmath92 , ( middle ) inverse of the sum of all the weights , @xmath115 , as function of the @xmath5-bound dye concentration , @xmath32 . @xmath32 was estimated theoretically from eq . ( [ eq : gof ] ) with the @xmath1 and dye concentrations of the aqueous solutions and @xmath118 given by the manufacturer . ( right ) @xmath53 as function of the total dye concentration used in the solutions , @xmath92 . ( a , b , c ) fluo4 high affinity and ( d , e , f ) fluo4 low affinity . being able to identify the parameters of the fitting with those of the theoretical acf , eqs . ( [ eq : gapprox])-([eq : gef2 ] ) , allowed us to go further and to quantify some relevant parameters of the underlying biophysical model for each aqueous solution , such as the free calcium diffusion coefficient . this entails solving an over - determined problem ( 7 equations with 6 unknowns ) . in that sense , we preferred to use the information given by @xmath72 and @xmath115 rather than by @xmath119 and @xmath74 because , as discussed before , these weights carry the largest errors . in particular , knowing @xmath26 , @xmath120 , @xmath96 , @xmath72 , @xmath115 and @xmath100 ( which we identify with the 7 parameters of the fitting ) it is possible to infer the off - rate , @xmath29 , of the @xmath1-dye binding reaction , the total concentration of the calcium dye , @xmath121 , the calcium bound dye concentration in equilibrium , @xmath32 , and the free diffusion coefficients , @xmath24 , @xmath26 . we show in fig . [ fig : dca_df_koff ] the values of @xmath24 , @xmath26 and @xmath29 obtained as a function of the total dye concentration used in the aqueous solutions , @xmath92 , both for the high affinity ( figs . [ fig : dca_df_koff ] ( a ) - ( c ) ) and the low affinity ( figs . [ fig : dca_df_koff ] ( d ) - ( f ) ) versions of the dye . since the solutions only differed in the total amount of dye , all estimated parameter values , with the exception of @xmath92 , should remain approximately constant for all solutions . to estimate the free @xmath1 diffusion coefficient the solutions with effective coefficients with large errors or with @xmath24 far away from the average were discarded ( solutions 3 , 6 , and 10 ) . for fluo4 high affinity we obtained @xmath24= ( 948 @xmath122 110 ) @xmath123 , and if we also discard solution 7 which has also a @xmath24 that is very different from the average , it gives @xmath24= ( 861 @xmath122 79 ) @xmath123 . for fluo4 low affinity we obtained @xmath24= ( 966 @xmath122 76 ) @xmath123 , and if we also discard solutions 12 and 18 , that have large errors , we obtained @xmath24= ( 870 @xmath122 55 ) @xmath123 . the average and standard deviation of all estimated biophysical parameters are presented in table [ tab : parametros ] . ] parameters of the underlying biophysical model derived from the fitting parameters for each aqueous solution , @xmath24 , @xmath26 and @xmath29 , ( mean and standard deviation over 2 - 3 experiments with 1or 2 fits ) and average of the values obtained ( solid line ) . ( a , b , c ) fluo4 high affinity and ( d , e , f ) fluo4 low affinity . in all cases we include the expected values ( dashed line ) based on the total concentrations used in the solutions and on previous estimates ( see text ) . c c c parameter & estimation from the model & previous estimates + + @xmath26 & ( 65 @xmath59 7 ) @xmath94 & 85 @xmath94 @xcite + @xmath24 & ( 948 @xmath59 110 ) @xmath94 & 760 @xmath94 @xcite + @xmath29 & ( 88 @xmath59 19 ) @xmath124 & + + @xmath26 & ( 89 @xmath59 8) @xmath125 & 85 @xmath94 @xcite + @xmath24 & ( 966 @xmath59 76 ) @xmath94 & 760@xmath94 @xcite + @xmath29 & ( 483@xmath59 61 ) @xmath124 & + in this work we have shown how free diffusion coefficients and reaction rates can be quantified in reaction - diffusion systems by performing sets of fluorescence correlation spectroscopy ( fcs ) experiments and using a biophysical model to interpret the experimental results . in particular , we have applied this approach to the case of ca@xmath0 and a single wavelength ca@xmath0 dye . to this end we developed the theory that allowed us to derive an approximation of the autocorrelation function of the fluorescence fluctuations ( acf ) in the limit of fast reactions . we then performed a series of experiments in solutions containing ca@xmath0 and the ca@xmath0 dye fluo4 dextran ( both high and low affinity ) with which validated the approach and established its limitations . the analysis of the experiments also allowed us to quantify the transport and reaction properties of two single wavelength ca@xmath0 dyes : high and low affinity fluo4 . in doing so we also derived the free diffusion coefficient of ca@xmath0 in aqueous solution . although this value is already well known ( @xmath24 @xmath3 ( 750 - 800 ) @xmath125 @xcite , @xcite ) , being able to derive it from the observation of a system in which it is not diffusing freely is quite relevant and provides hints on how to proceed in other settings . addressing fundamental problems in @xmath1 signaling and @xmath1-dependent cell function calls for the use of multiple approaches . the undeniable need to combine experiments and modeling requires that key biophysical parameters such as the @xmath1 diffusion coefficient be quantified _ in situ _ @xcite . optical techniques are ideal to probe intracellular transport with minimum disruption @xcite . measuring intracellular @xmath1 transport in this way , however , is not straightforward because of the multiple interactions of the ions with different cell components @xcite , @xcite and because @xmath2 dyes are also @xmath1 buffers that alter the ions transport rate @xcite . the quantification of diffusion coefficients and reaction constants in such a case requires a careful interpretation of the experimental data in terms of an underlying biophysical model @xcite . the work contained in this paper constitutes a necessary first step to advance in this direction . in this paper , we first focused on the theoretical aspects of the problem . to this end , we derived an analytic approximation for the acf of a system with @xmath5 and a single wavelength dye under the assumption that the @xmath1-dye reaction occurs on a fast timescale , that the free and @xmath1-bound dye molecules diffuse at the same rate and that the former is not fluorescent . the expression obtained , eqs . ( [ eq : gapprox])-([eq : gef2 ] ) , coincides with the one derived in bismuto et al . , 2001 @xcite . in particular , we observed that the first two terms in eq . ( [ eq : gapprox ] ) have the same functional dependence on @xmath80 as the only term of eq . ( [ eq : g1comp ] ) which corresponds to a case with a purely diffusive species . the first term ( eq . ( [ eq : gf ] ) ) gives @xmath126 that corresponds to the dye diffusion time across the sampling volume . this term is exact and involves no approximation . the second term ( eq . ( [ eq : gef1 ] ) ) has the time scale @xmath127 and is associated to an effective diffusion coefficient , @xmath95 , that combines information on diffusion and reactions . @xmath95 corresponds to the `` collective '' diffusion coefficient of pando et al . , 2006 @xcite which in turn coincides with the effective coefficient determined in the rapid buffering approximation @xcite . the last term ( eq . ( [ eq : gef2 ] ) ) does not have the functional form of a purely diffusive case , but has an additional exponential factor . depending on the value of @xmath100 , it could be neglected to determine @xmath128 and @xmath129 . in the appendix we presented the results of a thorough analysis of the limitations of this approximation . in particular we computed numerically the `` full '' acf ( with no approximations ) and determined that it could be correctly described by an acf with the time dependence obtained in the fast reaction limit ( eq . ( [ eq : g3comp ] ) ) . the fast reaction approximation is always valid for large enough @xmath80 , but , as shown in @xcite for the case of ` permanently ' fluorescent molecules , it can still provide a good description of the full acf for all @xmath80 even if the reaction and diffusion times are of the same order . our results also showed that even if the rapid reaction limit may not hold , fitting the full acf with an expression of the form eq . ( [ eq : g3comp ] ) still provides reasonably good estimates of the timescales associated to the free diffusion coefficient of the dye and to the exponentially decaying term . the two effective coefficients given by eqs . ( [ eq : def1 ] ) and ( [ eq : def2 ] ) could also be estimated for certain dye concentrations . the term that corresponds to the free diffusion of the dye ( eq . ( [ eq : gf ] ) ) is exact . thus , we can always assume that the weight that corresponds to this timescale is inversely proportional to the total number of dye molecules in the observation volume . the other two individual weights , however , can be incorrectly assessed if the fast reaction approximation is assumed . the total weight , on the other hand , is always inversely proportional to the mean number of ca@xmath0-bound dye molecules in the observation volume . thus , in our application of the theory to derive biophysical parameters from the experimental observations we used the total weight and the weight of the term that corresponds to the free diffusion of the dye , but not the other two . we then performed a series of experiments in solution using @xmath2 and fluo4 high or low affinity at various concentrations . fitting the acf with an expression of the form of eq . ( [ eq : g3comp ] ) we obtained the correlation times from which we derived the corresponding diffusion coefficients as functions of the dye concentration . as shown in fig . [ fig : d_efectivos ] , one of the coefficients ( or , analogously , the correlation time ) remained the same for all the concentrations . according to the theory , this coefficient is to be associated with the free diffusion coefficient of the dye . we observed that the value derived for the dye in its high or low affinity version is approximately the same ( @xmath130 and @xmath131 , respectively ) . these values are consistent with the value derived in solution for 10kda tetramethylrhodamine - dextran ( tmr - d , @xmath132 ) @xcite . the variation of the other two coefficients with the dye concentration is particularly visible in the case of the high affinity version of the dye ( fig . [ fig : d_efectivos ] ( a ) ) . we then performed a series of self - consistency checks of our approach . we first compared the relationship between the inverse of the weights , @xmath115 and @xmath72 , that we obtained from the experimental fits and the total concentrations of fluo4 and of @xmath5-bound dye that we used in the solutions with the theoretical expression , eqs . ( [ eq : gotot ] ) and ( [ eq : gof ] ) . the results are shown in figs . [ fig : gof_gotot_nu ] ( a ) and [ fig : gof_gotot_nu ] ( d ) with symbols for the former and curves for the latter . we can observe in figs . [ fig : gof_gotot_nu ] ( a ) and [ fig : gof_gotot_nu ] ( d ) that , in the case of the inverse of @xmath72 versus @xmath92 relationship , the experimental points match the theoretical prediction . thus , for these experiments in intact cells we expect to be able to obtain a reliable estimation of the amount of indicator that enters the system . we fitted the experimental points by a linear relationship between the inverse of @xmath72 and @xmath92 . we obtained ( 0.54 @xmath59 0.08 ) @xmath60 for high affinity and ( 0.56 @xmath59 0.08 ) @xmath60 for low affinity fluo4 . we can observe in figs . [ fig : gof_gotot_nu ] ( b ) and [ fig : gof_gotot_nu ] ( e ) that , in the case of the inverse of @xmath115 versus @xmath32 relationship , the experimental points lie below the theoretical prediction , as if the actual concentrations of @xmath5-bound dye were smaller than those that can be derived from eq . ( [ eq : caf_eq ] ) using the ones of the solutions and dissociation constant provided by the vendor . if , as before , we fit the experimental points by a relationship between the inverse of @xmath115 and @xmath32 we obtain ( 0.23 @xmath59 0.02 ) @xmath133 for high affinity and ( 0.17 @xmath59 0.01 ) @xmath133 for low affinity fluo4 . the resulting volumes are smaller than the one determined from the calibration , ( @xmath58 ( 0.59 @xmath59 0.1 ) @xmath60 ) , and the mismatch is slightly larger in the case of low affinity fluo4 . we must point out that this relationship also depends on the dissociation constant of the @xmath1-dye reaction and that using larger @xmath118 values would decrease the mismatch between the experimental points and the theoretical curve . in order to analyze to what extent the results obtained for both dyes agree with what can be expected theoretically we show in fig . [ fig : gotot_caf4 ] the ratio of total weights obtained using each dye ( weight for high over weight for low affinity fluo4 with symbols ) as a function of the total dye concentration for which we had experiments performed with both dyes . we also show in the figure the ratio of @xmath1-bound dye concentrations ( low over high ) computed theoretically using the dissociation constant provided by the vendor . these two ratios should be equal according to eq . ( [ eq : gotot ] ) . we observe that the ratio determined experimentally is larger than the theoretical one in most cases . this implies that either the experimentally estimated value of @xmath32 is underestimated for the high affinity dye or it is overestimated for the low affinity one . we must recall that eq . ( [ eq : gotot ] ) holds provided that the fluorescence coming from the free dye molecules is negligible with respect to the one that comes from the @xmath5-bound molecules . assuming that eq . ( [ eq : gotot ] ) holds in a case in which the free dye molecules contribution to the fluorescence is not completely negligible would lead to an overestimation of @xmath32 . in such a case the overestimation of @xmath32 would be larger for the low affinity than for the high affinity dye . this could explain the difference between the experimental points and the theoretical curve of fig . [ fig : gotot_caf4 ] . this observation together with the fact that the mismatch that can be observed in fig . [ fig : gof_gotot_nu ] is larger for the low than for the high affinity fluo4 makes the latter preferable over the former to study @xmath1 transport in other settings . finally , we also compared the dye concentration dependence of the inverse of the exponential correlation time derived from the experiments ( @xmath53 ) with the one predicted from the theory ( @xmath100 in eq . ( [ eq : gef2 ] ) ) using some estimated parameters as explained before . as expected from the analyses of fig . [ fig : gof_gotot_nu ] , it is for the lowest dye concentrations that we obtained comparable results between theory and experiments . ] ratio between the total weights , @xmath115 ( circles ) , obtained in experiments with high and low affinity fluo4 and ratio @xmath32(low)/ @xmath32(high ) , solid line , determined theoretically as functions of the total dye concentration , @xmath92 . the @xmath5-bound dye concentrations were computed using the dissociation constant provided by the vendor . after having tested the self - consistency of our model , we subsequently used it to derive estimates of some biophysical parameters from the parameters of the fitting . more specifically , we obtained the free @xmath5 diffusion coefficient , @xmath24 , and the off - rate of the @xmath5-dye binding reaction , @xmath29 . for the former we used the sum of the two effective diffusion coefficients ( @xmath134 + @xmath135 ) and subtracted the estimate of the free dye diffusion coefficient , @xmath26 . the values , @xmath24 and @xmath26 obtained for each solution probed are shown in fig . [ fig : dca_df_koff ] . the corresponding average values are within the expected range ( @xmath136(861 @xmath59 79 ) @xmath94 , @xmath26= ( 65 @xmath59 7 ) @xmath94 in the case of fluo4 high affinity and @xmath110(926 @xmath59 92 ) @xmath111 , @xmath26= ( 89 @xmath59 8) @xmath94 in the case of fluo4 low affinity ) . in particular , we obtain consistent values of the free dye diffusion coefficient , @xmath26 , for both the high and low affinity version ( i.e. @xmath26 @xmath3 ( 65 - 90 ) @xmath94 ) that are of the same order of value as the one estimated for a 10kda tmr - d @xcite . the estimated free @xmath5 diffusion are also consistent with what we expected @xcite , @xcite . it is important to note that the values , @xmath24 and @xmath26 , are derived exclusively from the diffusive correlation times . thus , these results are not affected by the differences between the @xmath2-bound dye theoretical concentrations , @xmath32 , and the ones estimated by the ( [ eq : gotot ] ) discussed before . in order to obtain @xmath29 from the inverse of the exponential correlation time , @xmath53 , it is necessary to use concentration estimates . in order to avoid introducing an additional error because of the possible mismatch between the concentrations that we discussed in connection with the differences observed in fig . [ fig : gof_gotot_nu ] , to obtain @xmath29 from @xmath53 we used the estimates of the ratio between the concentrations and the dissociation constant derived from the weight , @xmath72 , of the acf obtained in the experiments . as discussed before , the values of @xmath53 seemed to display the correct behavior only for those solutions with the smallest dye concentrations . in any case , applying the theory to all the experimental results regardless of @xmath92 gave values of @xmath29 within the same order of magnitude ( see figs . [ fig : dca_df_koff ] ( c ) and [ fig : dca_df_koff ] ( e ) ) . using the average of these values we obtained @xmath29 = ( 88 @xmath116 19 ) s@xmath137 and @xmath29 = ( 483@xmath59 61 ) s@xmath137 for the high and low affinity fluo4 , respectively . using the dissociation constant provided by the manufacturer we derived the on rates . for the high and low affinity versions of the dye , we found similar values ( @xmath28 = ( 0.114 @xmath59 0.025 ) @xmath138 and @xmath139(0.186 @xmath59 0.023 ) @xmath138 , respectively ) . this is consistent with the fact that , in bapta ( 1,2-bis(o - aminophenoxy)ethane - n , n , n,n-tetraacetic acid ) based calcium indicators , increasing @xmath118 values results from an increase in the dissociation rate constant and negligible or only modest decreases in the association rate constants @xcite , @xcite . it is important to note that , while concentrations at equilibrium do not depend on @xmath29 and @xmath28 , separately , but on @xmath118 = @xmath140 , their time evolution does . therefore , the values of @xmath29 and @xmath28 affect the behavior of the observed signals and knowing them is absolutely necessary to infer the spatio - temporal distribution of free @xmath1 from the images @xcite , @xcite . knowing the free diffusion coefficients of @xmath1 and its dyes in the cytosol is necessary as well for this purpose . the values derived in the results section , however , correspond to coefficients in aqueous solution . assuming that the differences in the free diffusion coefficients in solution and in the cytosol are due to differences in viscosity between both media we may assume that the ratio between the free diffusion coefficients of any two substances remains the same in both settings . thus , by quantifying the rate of diffusion of a molecule that diffuses freely in the cytosol and in solution we can infer the free diffusion coefficient of @xmath1and the dyes in the cytosol as well . we present in the appendix the results of fcs experiments performed in aqueous solution and in oocytes of _ xenopus laevis _ using tmr - d . the acf can be fitted by an expression of the form eq . ( [ eq : g1comp ] ) , _ i.e_. , with a single , free - diffusing component . from the fits we obtained @xmath141 = ( 27 @xmath142 1 ) @xmath143 in the cytosol considering that the tmr - d diffusion coefficient in solution is @xmath141 = 85 @xmath144 @xcite , we obtained @xmath145(solution)/@xmath145(oocyte ) @xmath3 3 . assuming that @xmath145(solution)/@xmath145(oocyte ) @xmath3 @xmath146 ( solution)/@xmath146(oocyte ) , where @xmath146 stands for free diffusion coefficient of any substance , we can use the free transport rates of @xmath1 and of its dyes in solution to infer their values in the cytosol . we obtain @xmath24 @xmath3 ( 261 - 313 ) @xmath147 and @xmath26 @xmath3 ( 19 - 24 ) @xmath125 starting from the free diffusion coefficients in solution obtained in the experiments performed with high affinity fluo4 . thus , the practical implementation presented in this paper not only highlights the advantages of our approach but also allows us to derive information that is key to quantify the free @xmath1 distribution that underlies a @xmath1 image . the cytosolic @xmath24 values derived with our approach are of the same order of magnitude as the one obtained in cytosolic extracts by allbritton et al . , 1992 @xcite although the latter ( 220 @xmath123 ) is below our lower bound . the analysis of buffered diffusion of pando et al . , 2006 @xcite , showed that the effective diffusion coefficient obtained in the experiments of allbritton et al , 1992 @xcite is the single molecule one and a misinterpretation of its meaning could lead to an underestimation of the actual diffusion rate of @xmath1 . this highlights the need of having an underlying biophysical model to interpret transport rates in experiments that do not probe solely free diffusion @xcite . the theory and experiments of this paper illustrates this very important point . it also shows how by changing the experimental conditions so that the correlation times associated to effective diffusion change it is possible to identify the latter and quantify concentration - independent biophysical parameters . other experimentally accessible parameters such as the observation volume can be modified to change some of the correlation times and , in this way , quantify different biophysical parameters @xcite . in fact , a comparison of fcs results obtained for different observation volumes has recently been used to quantify the binding rates of transcription factors in single cells of developing mouse embryos @xcite . this shows the relevance of performing fcs experiments under different conditions to quantify parameters . the approach presented in this paper can then be extended to address the quantification of transport rates in other biologically relevant reaction - diffusion systems . we are thankful to emiliano perez ipia for having provided the code to compute the full acf and to lucia lopez and estefania piegari for help with some of experiments . this research has been supported by uba ( ubacyt 20020130100480ba ) and anpcyt ( pict 2013 - 1301 ) . l.s . and s.p.d . are members of carrera del investigador cientfico ( conicet ) . when the fluorescence comes solely from a single type of particles , @xmath10 , that diffuse freely with coefficient , @xmath11 , the fluorescence is given by : @xmath148(r , t)d^{{\rm 3 } } r\ ] ] where @xmath149(r , t)$ ] is the particle concentration at time , @xmath9 , and spatial point , @xmath17 , the parameter , @xmath150 , takes into account the detection efficiency , the fluorescence quantum yield and the absorption cross - section at the wavelength of excitation of the fluorescence . the illumination is commonly approximated by a three - dimensional gaussian : @xmath151 with @xmath16 the spatial coordinate along the beam propagation direction , @xmath17 a radial coordinate in the perpendicular plane and @xmath152 and @xmath153 the sizes of the beam waist along @xmath16 and @xmath17 , respectively . in this case there is an analytic expression for the acf which is given by eq . ( [ eq : g1comp ] ) . fitting the acf obtained from experiments by eq . ( [ eq : g1comp ] ) two parameters can be determined : @xmath154 and the characteristic time @xmath155 . a previous calibration of the geometric parameters of the sample volume is required in order to obtain @xmath156 from @xmath155 . this is done performing the same experiments on a sample for which @xmath156 is already known . once @xmath153 and @xmath152 are determined , the unknown @xmath156 can be estimated from the characteristic time @xmath155 and @xmath157 from @xmath154 . the equations that describe the dynamics of @xmath1 and a single wavelength dye , @xmath22 , that react and diffuse as described in sec . ii are : @xmath158}{\partial t } { \rm = } d_{ca } \nabla ^{{\rm 2 } } [ ca]-k_{on } [ ca][f]+k_{off } [ caf]\ ] ] @xmath159}{\partial t } { \rm = } d_{f } \nabla ^{{\rm 2 } } [ caf]+k_{on } [ ca][f]-k_{off } [ caf]\ ] ] @xmath160}{\partial t } { \rm = } d_{f } \nabla ^{{\rm 2 } } [ f]-k_{on } [ ca][f]+k_{off } [ caf]\ ] ] in fcs experiments in aqueous solution containing @xmath1 and @xmath22 it is assumed that both species uniformly are distributed and in equilibrium , so that their mean concentrations are given by the equilibrium concentrations @xmath30 , @xmath31 and @xmath32 , that satisfy eq . ( [ eq : caf_eq ] ) and : @xmath161 @xmath162 @xmath163 in the case in which the calcium indicator is practically non - fluorescent while it is not bound to @xmath1 the fluorescence intensity is given by : @xmath164(r , t)d^{{\rm 3 } } r , \ ] ] with @xmath150 and @xmath165 as before . as done in sigaut et al . 2010 @xcite , we follow krischevsky and bonnet 2002 @xcite to determine the spatio - temporal dependence of the fluorescence fluctuations in this case . namely , the evolution equations ( [ eq : app_dca_dt])-([eq : app_df_dt ] ) are linearized around the equilibrium solution , eq . ( [ eq : caf_eq ] ) . the solution of these linearized equations is then computed in fourier space and written in terms of branches of eigenvalues , @xmath166 , and eigenvectors , @xmath167 , with @xmath168 the variable in fourier space ( conjugate to the spatial vector @xmath169 ) . the fluorescence fluctuations are then obtained as in eq . ( [ eq : app_fluo_intensity ] ) but replacing @xmath170 $ ] by the corresponding component of the solution of the linearized problem , @xmath171 $ ] . the calculation of the acf finally assumes that the correlation length of the concentrations at any given time is much smaller than the inter - molecule distance and that the number of molecules obeys poisson statistics so that its variance and its mean are equal . in this way the acf , @xmath172 , can be written as a sum of as many components as branches of eigenvalues of the linearized problem , in this case : @xmath173 with : @xmath174 @xmath175 @xmath176 where @xmath177 , @xmath178 and @xmath179 the fourier coordinates conjugated to the radial and axial coordinates , @xmath17 and @xmath16 , respectively , @xmath180 , @xmath181 , @xmath182 , @xmath183 and the eigenvalues : @xmath184 @xmath185 although @xmath187 and @xmath188 can be computed numerically , in general there is no analytic algebraic expression for these two components as there is for the one that corresponds to the branch of eigenvalues , @xmath189 , associated to the free diffusion coefficient of the dye , @xmath26 ( see eq . ( [ eq : app_g_lambda_f ] ) ) . as done in sigaut et al . 2010 @xcite , however , an analytic expression for @xmath190 and @xmath191 and , consequently , for the acf can be obtained in the limit of small @xmath168 which is always valid for long enough times , @xmath80 . the approximation is good for almost any value of @xmath80 when the observation volume is such that the characteristic reaction time is of the same order or less than the diffusive time across the volume @xcite . in fact , if we expand the integrands that define @xmath192 and @xmath193 in powers of @xmath168 and keep the expansion up to @xmath194 we obtain eq . ( [ eq : gapprox ] ) this limit is valid provided that the reactions occur on a faster timescale than diffusion across the observation volume , i.e. , if @xmath195 . in order to study when the fast reaction approximation of the acf can be used to estimate different biophysical parameters we computed numerically the full acf , @xmath196,given by eqs . ( [ eq : app_g_tau])-([eq : app_g_lambda_2 ] ) using an adaptive lobatto quadrature algorithm , with the__quadl _ _ function on the matlab platform ( the mathworks , natick , ma ) and the parameters listed in table [ tab : app_parametros ] . we compared the results of these computations with the approximated acf , @xmath197 , given by eqs . ( [ eq : gapprox ] ) - ( [ eq : gef2 ] ) using the same parameters . for the comparison we computed the difference between both functions given by : @xmath198 with _ n _ the total number of data points . for the lowest dye concentrations considered @xmath196 and @xmath199 were indistinguishable . as the concentration of dye was increased , the difference between the full and the approximated acf s first increased , with @xmath199 decaying at an earlier correlation time than @xmath196 . the difference between @xmath196 and @xmath199 reached a maximum at @xmath92 @xmath3 4 @xmath37 . further increments in @xmath92 decreased this difference . this is illustrated in fig . [ fig : acf_sol ] where we show @xmath196 and @xmath199 with solid and dashed lines , respectively , for @xmath92=429 @xmath38 , 7500 @xmath38 , 15 @xmath37 using the parameters of fluo4 high affinity . similar results are obtained for fluo4 low affinity ( data not shown ) . the difference between the two acf s , however , is never significantly large : we obtained 2.31x10 @xmath200 @xmath201 @xmath202 @xmath203 1.17x10@xmath204 for high affinity fluo4 and 1.55x10 @xmath204 @xmath203 @xmath202 @xmath203 4.11x10 @xmath204 for low affinity fluo4 . the differences between the individual components associated to @xmath82 and @xmath83 are much larger . .[tab : app_parametros ] parameters used to compute the full and approximated acfs numerically . for the concentrations of dye we tried the values listed in table [ tab : soluciones_f4_ca ] . [ cols= " < , < , < " , ] ] full ( solid line , eqs . ( [ eq : app_g_tau])-([eq : app_g_lambda_2 ] ) ) and approximated ( dashed line , eqs . ( [ eq : gapprox])-([eq : gef2 ] ) ) acf s for fluo4 high affinity using the parameters listed in table [ tab : app_parametros ] . @xmath92 = 429 nm ( a ) , 7500 nm ( b ) , 15 @xmath37 ( c ) . we then analyzed what correlation times could be derived by fitting the full acf with eq . ( [ eq : g3comp ] ) . we probed two options . first , we fixed the timescales as in the fast reaction approximation and fitted the weights . secondly , we fitted both the weights and the timescales . from the second test we determined that the fitted values obtained for @xmath50 were similar to those prescribed by the fast reaction approximation , @xmath205 , for all dye concentrations the fitted values of @xmath206 were similar to the values of the fast reaction approximation for dye concentrations below 4.826 @xmath37 . for higher dye concentrations the fitted values of @xmath206 for high affinity fluo4 followed the same pattern and stayed within the same order of magnitude as the value of the fast reaction approximation although it got three times the approximated value at @xmath92 = 12 @xmath37 . for low affinity fluo4 the variations of @xmath207 with @xmath92 were slightly different but @xmath207 stayed within the fast reaction approximation values for all dye concentrations becoming between twice and three times smaller at @xmath92 = 15 @xmath37 for all dye concentrations we obtained @xmath208 and @xmath209 and for dye concentration below 8.25 @xmath37 we obtained @xmath210 . these results are illustrated in figs . [ fig : ratios ] ( a ) - ( b ) where we show the ratios @xmath211 , @xmath212 , @xmath213 , between the fitted values and those of the fast reaction approximation and @xmath214 , as a function of @xmath92 for high affinity ( fig . [ fig : ratios ] ( a ) ) and low affinity ( fig . [ fig : ratios ] ( b ) ) fluo4 . from the test we determined that the full acf could be approximated fairly well using the expression given by eq . ( [ eq : g3comp ] ) , with the timescales of the fast reaction approximation but with slightly different weights . this is illustrated in fig . [ fig : ratios ] ( c ) where we have plotted these two acf s for high affinity fluo4 at @xmath92 = 7500 nm . similar figures are obtained for low affinity fluo4 and at other dye concentrations ( data not shown ) . in this case the mismatch , @xmath215 @xmath216 , obtained ranged between 7.2x10@xmath217 and 1.57x10@xmath218 for high affinity and between 9.69x10@xmath219 and 2.46x10@xmath200 for low affinity fluo4 . regarding the individual components of the fitted acf , the weights obtained , @xmath47 , @xmath49 , were of the same order of magnitude as those of the fast reaction approximation , @xmath72 , @xmath220 , and @xmath221was negligible for low dye concentrations . this is illustrated in fig . [ fig : ratios ] ( d ) where we have plotted the ratios between the fitted and the fast reaction approximation weights , @xmath222 , @xmath223 , @xmath224 , as a function of @xmath92 for fluo4 high affinity . similar patterns are observed for fluo4 low affinity ( data not shown ) . ] ratios @xmath211 ( white circles ) , @xmath212 ( black circles ) , @xmath225 ( triangles ) between the fitted values and those of the fast reaction approximation and @xmath226 ( squares ) , as a function of @xmath92 for high affinity ( a ) and low affinity ( b ) fluo4 . ( c ) full acf ( dotted line ) fitted with eq . ( [ eq : g3comp ] ) , fixing the timescales ( solid line ) for @xmath92=7500 nm . ( d ) ratios between the fitted weights with the timescales fixed and the fast reaction approximation weights , @xmath222 ( circles ) , @xmath223 ( triangles ) , @xmath224 ( squares ) , as a function of @xmath92 for fluo4 high affinity . we here present the results of performing fcs experiments with tetramethylrhodamine - dextran ( tmr - d ) in aqueous solution and in _ xenopus laevis _ oocytes . the aim of these experiments is to determine the conversion factor between free diffusion coefficients in the two media . _ x. laevis _ oocytes , previously treated with collagenase and stored in barth s solution , were loaded with 37 @xmath227 of tmr - d at different concentrations . intracellular microinjections were performed using a drummond microinjector . assuming a 1 @xmath228 cytosolic volume , the final concentration of tmr - d was 0.9 , 1.1 , 1.4 or 1.85 @xmath37 . fcs measurements were performed on a spectral confocal scanning microscope fluoview 1000 ( olympus , tokyo , japan ) , employing a 60x , 1.35 n.a . oil - immersion objective ( uplansapo , olympus ) and a pinhole aperture of 115 @xmath41 . single point measurements at a 50 khz sampling rate were performed for a total duration of 167 s ( equivalently , 8365312 data points ) employing a 543 nm line and detecting the fluorescence in the range ( 555 - 655 ) nm . for the aqueous solutions the measurements was performed at @xmath320 @xmath41 from the coverslip and for the oocytes , at the cortical granules region in the animal hemisphere . experimental acf s were calculated with a custom - made routine written on the matlab platform @xcite . to this end , each 167 s long record was divided into _ _ n@xmath229__=1021 , 164 ms long segments containing 2@xmath43 points each for the experiments in aqueous solutions and into _ _ n@xmath230__=510 , 328 ms long segments containing 2@xmath231 points each for the experiments in _ x. laevis _ oocytes . the acf was computed for each of the _ _ n@xmath229__=1021 or _ _ n@xmath230__=510 segments from which the average acf was obtained . as the confocal volume dimensions are wavelength - dependent we used the fcs experiments with tmr - d in solution to estimate the beam waist and aspect ratio at 543 nm . assuming a diffusion coefficient of @xmath232 @xcite we obtained @xmath233 and @xmath234 . the acf was fitted using only one ( diffusive ) component as in eq . ( [ eq : g1comp ] ) . we show in fig . [ fig : sol_ovo ] the acf obtained from fcs experiments performed in _ x. laevis _ oocytes with tmr - d ( fig . [ fig : sol_ovo ] ( a ) ) . using eq . ( [ eq : g1comp ] ) to fit the data of fig . [ fig : sol_ovo ] we obtain @xmath235 . the tmr - d diffusion coefficient in solution is @xmath236 @xcite . thus , it is @xmath145(oocyte)/@xmath145(solution ) @xmath3 3 . ] \(a ) acf obtained from fcs experiments performed in _ x. laevis _ oocytes microinjected with 37 @xmath227 of tmr - d = 30 @xmath37 ( dashed line ) fitted by eq . ( [ eq : g1comp ] ) ( solid line ) . ( b ) as in ( a ) for solution of tmr - d = 50 nm . ( c ) acf s from the fits performed in ( a ) and ( b ) ( solid and dashed line , respectively ) , normalized . natexlab#1#1bibnamefont # 1#1bibfnamefont # 1#1citenamefont # 1#1url # 1`#1`urlprefix[2]#2 [ 2][]#2

lee codes have been intensively studied for more than 40 years . interest in these codes has been triggered by the golomb - welch conjecture on the existence of the perfect error - correcting lee codes . in this paper we deal with the existence and enumeration of diameter perfect lee codes . as main results we determine all for which there exists a linear diameter- perfect lee code of word length over and prove that for each there are uncountable many diameter- perfect lee codes of word length over this is in a strict contrast with perfect error - correcting lee codes of word length over as there is a unique such code for and its is conjectured that this is always the case when is a prime . we produce diameter perfect lee codes by an algebraic construction that is based on a group homomorphism . this will allow us to design an efficient algorithm for their decoding . we hope that this construction will turn out to be useful far beyond the scope of this paper . diameter codes , error - correcting codes , lee metric , perfect codes .

1109.3475

the most common metric in coding theory is the hamming metric . in this paper we deal with another frequently used metric , so called lee metric . this metric was introduced in @xcite and @xcite for transmission of signals . the main motive power for studying codes in lee metric goes back to the seminal paper of golomb and welch @xcite . in this paper we will study the existence and will enumerate diameter perfect lee codes . first we recall some definitions and notation . as usual , let @xmath9 be the set of integers , @xmath10 denote the integers modulo @xmath11 and let @xmath12 be the @xmath2-fold cartesian product of a set @xmath13 throughout the paper we will use @xmath14 and @xmath15 for the abelian ( component - wise ) additive group on @xmath14 and @xmath15 as well . it will always be clear from the context whether we have in mind a set or the group on this set.because of the coding theory context the elements of @xmath14 and @xmath15 will also be called words . the lee distance ( = the manhattan distance , the zig - zag distance ) @xmath16 of two words @xmath17 @xmath18 is given by @xmath19 for @xmath20 and by @xmath21 @xmath22 for @xmath23 by @xmath24 we denote the lee sphere of radius @xmath25 centered at @xmath26 that is , @xmath27 . for @xmath28 with @xmath29 the double lee sphere @xmath30 of radius @xmath25 with its center comprising @xmath31,@xmath32 is the set @xmath33 . perfect error - correcting lee codes can be introduced in several ways . for the purpose of this paper it is convenient to define them through tilings . in order not to have to formulate the same statement / definition twice , once for @xmath15 and then for @xmath34 the symbol @xmath35 @xmath36stands for both of them ; that is , by @xmath35 we mean @xmath15 or @xmath37 we set @xmath38 and @xmath39 where the @xmath40-th coordinate equals @xmath41 let @xmath42 be a subset of @xmath35 . by a copy of @xmath42 we mean an image of @xmath42 under a linear distance preserving bijection on @xmath43 translations of @xmath42 form a special family of copies of @xmath44 we recall that by a translation of @xmath42 we understand the set @xmath45 @xmath46 where @xmath47 . as the sphere @xmath48 is symmetric , each copy of the sphere is its translation . each copy of the double - sphere @xmath49 can be obtained by a translation of the double - sphere @xmath50 for some @xmath51 a collection @xmath52 of copies of @xmath42 constitutes a tiling of @xmath35 by @xmath53 if @xmath54 and @xmath55 for all @xmath56 a tiling @xmath57 of @xmath58 by @xmath42 that consists of translations of @xmath42 can be described as @xmath59 where @xmath60 . @xmath36a tiling @xmath57 by translations of @xmath42 is called periodic if @xmath61 is periodic , and @xmath57 is called a lattice tiling if @xmath61 is a lattice ( = a subgroup ) of @xmath35 . let @xmath62 in @xmath35 such that , if @xmath63 then @xmath64 a set @xmath65 is called a perfect @xmath25-error - correcting lee code if @xmath66 is a tiling of @xmath67 if @xmath68 is a lattice then @xmath68 is called a linear code . for @xmath69 the perfect @xmath25-error - correcting lee code is denoted by @xmath70 while for @xmath71 it is denoted by @xmath72 the long - standing golomb - welch conjecture deals with the existence of @xmath70 codes . although there is a vast literature on the subject the conjecture is still far from being solved . in this paper we focus on diameter-@xmath73 perfect lee codes , which constitute a generalization of perfect error - correcting lee codes . ahlswede et al . , see @xcite , introduced diameter perfect codes for distance regular graphs . let @xmath74 be a metric space . then a set @xmath75 is a diameter-@xmath73 code if @xmath76 for any @xmath77 and a set @xmath78 is an anticode of diameter @xmath73 if @xmath79 for all @xmath80 each graph @xmath81 can be seen as a metric space . it is proved in @xcite that if @xmath81 is a distance regular graph , @xmath82 is a diameter-@xmath73 code in @xmath83 and @xmath84 is an anticode of diameter @xmath85 of the maximum size in @xmath81 then @xmath86 that is , the sphere packing bound applies to distance regular graphs as well . a diameter-@xmath73 code that attains the bound in the above inequality is called diameter-@xmath73 perfect code . this definition can be used for example in hamming scheme as the related graph is distance regular . etzion @xcite showed that the sphere packing bound applies to @xmath14 endowed with lee metric as well although the related graph is not distance regular ; analogically to @xcite he defined the diameter-@xmath73 perfect lee code as a diameter-@xmath73 code that attains the equality in the sphere packing bound . clearly , a definition through the sphere packing bound can not be extended to an infinite space . to deal with the case of @xmath15 etzion proved that if @xmath68 is a lattice in @xmath15 with the minimum distance @xmath73 and a volume @xmath87 then @xmath88 for any anticode of diameter @xmath89 then he called @xmath68 to be the diameter perfect code if @xmath68 attains the equality in the bound . as far as we know there is no formal definition of the diameter perfect lee code in @xmath90 in case when @xmath68 is not a lattice . now we provide a unified definition of a diameter-@xmath73 perfect lee code for both spaces @xmath14 and @xmath15 regardless whether the code is a linear one . let @xmath91 be a family of subsets of an underlying set @xmath92 then a set @xmath93 is called a transversal of @xmath94 if there is a bijection @xmath95 so that @xmath96 equivalently , @xmath97 is a transversal of @xmath98 if @xmath99 for each @xmath100 and @xmath101 for all @xmath102 let @xmath65 . then @xmath68 is a diameter-@xmath73 perfect lee code in @xmath35 if @xmath68 is a diameter-@xmath73 code , and there is a tiling @xmath103 of @xmath104 by the anticode of diameter @xmath85 of maximum size such that @xmath61 is a transversal of @xmath105 the diameter-@xmath73 perfect lee code in @xmath15 will be denoted by @xmath106 and in @xmath14 by @xmath107 . in other words , a diameter-@xmath73 code @xmath68 is perfect if there is a tiling @xmath57 by anticodes of diameter @xmath85 of the maximum size so that each tile in @xmath57 contains exactly one codeword of @xmath68 . it is not difficult to see that for @xmath14 and for lattice case in @xmath15 our definition of diameter perfect lee codes is equivalent to the definition due to etzion . any error - correcting perfect lee code is also a diameter perfect lee code . indeed , it is easy to see that , for @xmath73 even , the anticode of diameter @xmath108 of the maximum size is the lee sphere @xmath48 with @xmath109 thus , for @xmath73 odd , @xmath110 and @xmath111 codes are @xmath70 and @xmath112 codes where @xmath113 , respectively . it was proved in @xcite that , for @xmath73 odd , the anticode of diameter @xmath73 of maximum size is the double - sphere @xmath114 with @xmath115 we point out an advantage of our definition . suppose that we need to show the existence of a diameter-@xmath73 perfect lee code in @xmath116where @xmath73 is even . then all we need to do is to prove that there is a tiling @xmath117 of @xmath35 by the double - sphere @xmath114 of radius @xmath118 . indeed , choose a fixed tile @xmath119 in @xmath105 let @xmath120 be one of the two words forming the center of @xmath121we will prove that the set @xmath122 belongs to the center of @xmath123 and @xmath124 is even , @xmath125 is a distance-@xmath73 perfect lee code . clearly , @xmath126 is well defined as the center of @xmath123 consists of two words at distance @xmath127 that is , one of them at even and the other at odd distance from @xmath128 @xmath126 is a transversal of @xmath129 thus we need only to prove that @xmath130 for any @xmath131 . clearly , the distance of any two words in @xmath126 is even . this is obvious to see for @xmath132 it is not true in general for @xmath133 it is valid only in the case when @xmath0 is even . however , as the size of @xmath114 is even ( see e ) @xmath134 has to be even as well , which in turn implies @xmath0 is even . so for the distance of any two words in @xmath126 we have @xmath135 and since @xmath124 is even , we get @xmath136 we have proved : [ x]if there is a tiling of @xmath35 by double - spheres @xmath114 then there is a distance-@xmath73 perfect lee code in @xmath35 with @xmath137moreover , if @xmath138 is a tiling of @xmath139 by translations of @xmath114 such that the weight @xmath140 of @xmath141 is even for all @xmath142 then @xmath68 is a diameter-@xmath73 perfect lee code . the main goal of this paper is to study the existence and to enumerate diameter perfect lee codes . it turns out that some results on diameter perfect lee codes and on perfect error - correcting lee codes are similar but in some cases the results on the two types of codes are very much different . e.g. , it was proved in @xcite that there is only one , up to isomorphism , tiling of @xmath143 by lee spheres @xmath144 but we show in this paper that there are uncountable many tilings of @xmath143 by double lee sphere @xmath145 . the rest of the paper is organized as follows . golomb - welch conjecture claims that there are no @xmath70 codes for @xmath146 and @xmath147 the current state of this conjecture as well as its extension to @xmath110 codes and a further extension to perfect dominating sets is provided in section 2 . we defined perfect lee codes by means of tilings . in @xcite s. stein introduced an algebraic construction of lattice tilings based on group homomorphisms . he used this construction for tilings by crosses . this construction , and its variations occur in many papers , see e.g. @xcite . in section 3 a generalization of stein s construction is given . let @xmath42 be a subset of @xmath148 we state a necessary and sufficient condition for the existence of a lattice tiling of @xmath15 by @xmath42 in terms of a group homomorphism . we guess that this construction will turn useful outside the scope of this paper . the main results of this paper are stated in section 4 . it is believed , see section 2 , that , with only one exception of @xmath149 code , there are no @xmath110 codes , and also no @xmath111 codes for @xmath150 and @xmath151 all values of @xmath0 for which there exists a linear @xmath152 code ( = a linear @xmath153 code ) were determined in @xcite . we prove an analogous result for linear @xmath154 codes . as to the enumeration of all @xmath155 codes , etzion @xcite proved that there are uncountable many of them if @xmath2 is a power of @xmath156 we show that this is also the case for the other values of @xmath157 the situation with @xmath158 codes is very much different . it was proved in @xcite that there are uncountable many @xmath159 codes if @xmath8 is not a prime . however , on the other hand it was showed there that there is only one , up to isomorphism , @xmath160 code , and the authors conjectured that the same is true for each @xmath2 where @xmath8 is a prime . we note that the existence of uncountable many @xmath155 codes implies that there are uncountable many non - periodic @xmath155 codes as the total number of periodic @xmath155 codes is at most @xmath161 non - periodic @xmath155 codes were constructed by means of non - regular lattice tilings ( i.e. tilings that are not a face - to - face tilings ) of @xmath162 by a cluster of cubes centered at words of a double - sphere @xmath163 this is another example of a well know situation when to solve a problem we need to generalize it first . we note that results in section 4 answer in the affirmative questions ( 4 ) , ( 5 ) , and ( 8) raised by etzion in @xcite . question ( 6 ) from @xcite was answered in @xcite . finally , in section 5 an efficient decoding algorithm for linear perfect lee codes , both diameter and error - correcting , is designed . thanks to the representation of these codes by tilings constructed via a group homomorphism , the computation complexity of this algorithm is @xmath164 for @xmath165 codes , and its complexity is @xmath166 for @xmath167 codes . we recall that @xmath168 codes are believed not to exists for @xmath169 and @xmath151 in this section we provide a short account of the current state of golomb - welch conjecture and some of its extensions . in @xcite it is conjectured that : there is no @xmath70 code for @xmath169 and @xmath147 if true then the conjecture is best possible as the existence of @xmath170 codes for @xmath171 and all @xmath172 , and @xmath169 and @xmath173 was showed by several authors , see e.g. @xcite . it is also proved in @xcite that for each @xmath169 there is @xmath174 @xmath175 not specified , so that there is no @xmath70 code with @xmath176 although there are plenty of results in the literature on this conjecture the non - existence of @xmath70 codes has been proved only in few cases . first , gravier et al . @xcite settled the golomb - welch conjecture for @xmath177 and all @xmath147 later pacapan @xcite , whose proof is computer aided , showed the non - existence of a @xmath70 code for @xmath178 and all @xmath179 . horak @xcite provided an algebraic proof that there is no @xmath70 code for @xmath180 and all @xmath181the only other value of parameters for which the golomb - welch conjecture is known to be true is @xmath182 @xmath183 and @xmath184 see @xcite . the non - existence of @xmath112 codes for some pairs @xmath185 and specific values of @xmath0 depending on @xmath185 was stated in several papers , see e.g. @xcite for a comprehensive account . the best result of this type is due to post @xcite who showed that there is no @xmath112 code for any @xmath186 (= code with a large alphabet),@xmath187 and @xmath188 and for @xmath189 and @xmath190 post s result can also be viewed as follows : for the given pairs of @xmath185 there is no periodic @xmath70 code . in @xcite and hm the first author of this paper quoted incorrectly post s result as he omitted the word _ periodic . _ as mentioned in the introduction , @xmath110 codes are a generalization of @xmath70 codes as , for @xmath73 odd , a @xmath110 code is a @xmath191 code as well . therefore the conjecture stated by etzion in @xcite is an extension of the golomb - welch conjecture . @xcite there is no @xmath110 code for @xmath169 and @xmath192 with the exception of the pair @xmath193 etzion s conjecture , if true , is also best possible . the existence of @xmath167 codes , for @xmath73 odd was stated in @xcite in terms of @xmath194 codes . the existence of @xmath110 codes for @xmath171 and all even @xmath195 and all @xmath169 and @xmath196 was established in @xcite and adh , while the existence of a @xmath149 code follows from a minkowski tiling @xcite . to be able to present an extension of etzion s conjecture we state a generalization of diameter perfect lee codes introduced in @xcite . let @xmath197 be a graph , and @xmath198 be a natural number . a set @xmath199 is called a @xmath198-_perfect distance - dominating set _ in @xmath81 , or a @xmath198- in @xmath81 , if , for each @xmath200 , there is a unique component @xmath201 of @xmath202 $ ] so that @xmath203 and there is in @xmath204 a unique vertex @xmath32 with @xmath205 we recall that the distance @xmath206 of a vertex @xmath200 to @xmath207 is given by @xmath208 and that @xmath202 $ ] stands for the subgraph of @xmath81 induced by @xmath209 to facilitate our discussion , a @xmath198- in @xmath81 all whose components are isomorphic to a graph @xmath210 will be denoted by @xmath198-@xmath211.$ ] let @xmath212 be the integer grid , that is , @xmath212 has @xmath15 as its vertex set with two vertices being adjacent if their distance equals @xmath213 . let @xmath214 stand for a path on @xmath215 vertices . thus @xmath216 is a single vertex and @xmath217 is a pair of adjacent vertices . it is not difficult to see that an @xmath25-@xmath218 $ ] in @xmath212 constitutes a @xmath70 code while an @xmath25-@xmath219$]in @xmath212 is a @xmath220 code@xmath221 in @xcite it is conjectured that : [ g]let @xmath222 then there is no @xmath25-@xmath223 $ ] in @xmath224 for @xmath169 and @xmath179 with the exception of @xmath225-@xmath219 $ ] in @xmath226 as with the golomb - welch and the etzion conjecture , if true , then conjecture [ g ] is best possible . the existence of @xmath198-@xmath227 $ ] in @xmath212 for @xmath228 and for @xmath229 has been proved in @xcite , while @xmath225-@xmath219 $ ] in @xmath230 is the @xmath149 code mentioned above . at the end of this section we note that in @xcite there are other results and a conjecture on the existence of @xmath198-@xmath211 $ ] sets in @xmath212 outside the scope of diameter perfect lee codes . for the purpose of this paper it was convenient to define perfect lee codes by means of tilings . the main reason was that with this definition in hand we will be able to prove the existence of certain codes using a construction of lattice tilings through group homomorphism . we believe that stein in @xcite was the first one to use graph homomorphisms to produce a lattice tiling , in this case a tiling by crosses . variations of stein s construction can be found in many papers , see @xcite , where the authors use stein s approach for lattice tilings by different types of crosses and spheres . in this section a generalization of stein s construction is given . we first describe this construction and then we provide a necessary and sufficient condition that allows one to verify whether there is a lattice tiling of @xmath15 by a given subset . let @xmath68 be a lattice ( = subgroup ) of @xmath15 . consider the factor group @xmath232 . two elements @xmath233 belong to the same coset of @xmath234 ( that is , they are two different representations of the same element @xmath235 ) if @xmath236 . choose one element @xmath237 from each coset of @xmath234 ( = for each element of the factor group @xmath81 ) and put @xmath238 then the collection @xmath239 @xmath240 forms a partition of @xmath15 by translations of @xmath42 . as @xmath61 is a lattice @xmath57 constitutes a lattice tiling of @xmath90 by @xmath44 a different choice of the elements @xmath241 although considering the same lattice @xmath68 , provides tilings of @xmath15 by different sets @xmath44 consider a homomorphism @xmath242 where @xmath243 and @xmath244 . in fig . 1 it is shown that we can choose @xmath42 to be a copy of the set @xmath245 as well as @xmath42 to be a copy of the lee sphere @xmath246 however , we are interested in the `` inverse '' process . given a set @xmath247 find a lattice tiling @xmath248 of @xmath15 by @xmath44 to be able to apply the above construction we need to find a lattice @xmath68 so that it would be possible to choose elements @xmath249 from individual cosets of @xmath250 forming a copy of the set @xmath44 however , to find a lattice with the required properties might be a painstaking process . it is nearly impossible to do it by trying all options as usually there are infinitely many of them . to remedy the problem the following theorem provides a way how to construct the required tiling @xmath57 without knowing the lattice @xmath61 explicitly . it turns out that all what is needed is to find a suitable abelian group @xmath81 of order @xmath42 and a homomorphism @xmath251 satisfying a condition that can be easily verified . [ b]let @xmath42 be a subset of @xmath231 then there is a lattice tiling @xmath252 of @xmath15 by @xmath42 if and only if there is an abelian group @xmath81 of order @xmath253 and a homomorphism @xmath251 so that the restriction of @xmath254 to @xmath42 is a bijection . let @xmath81 be an abelian group and let @xmath255 be a homomorphism so that the restriction of @xmath254 to @xmath42 is a bijection . it is well known that @xmath256 is a subgroup of @xmath15 and @xmath257 . as mentioned above , two elements @xmath258 belong to the same coset of the factor group @xmath234 iff @xmath259 to show that @xmath260 is a tiling of @xmath15 by @xmath42 we need to prove that ( i ) @xmath261 ( ii ) @xmath262 for any @xmath263 assume that there is @xmath264 so that @xmath265 let @xmath266 since the restriction of @xmath254 to @xmath42 is a bijection , there is @xmath267 so that @xmath268 that is , @xmath120 and @xmath269 belong to the same coset of @xmath270 thus , @xmath271 which in turn implies , as @xmath272 that @xmath273 , a contradiction . the proof of ( i ) is complete . assume now that there are @xmath274 such that there is @xmath120 in @xmath15 with @xmath275 . hence @xmath120 can be written as @xmath276 and also as @xmath277 for some @xmath278 we have @xmath279 that is @xmath280 however , @xmath281 implies @xmath282 that is @xmath283 a contradiction as @xmath254 is a bijection on @xmath42 . to prove the necessary part of the condition , assume that @xmath59 is a lattice tiling of @xmath15 by @xmath284 that is , @xmath68 is a subgroup of @xmath231 consider the factor group @xmath285 and the mapping @xmath255 given by @xmath286 $ ] , where @xmath287 $ ] is the coset of the factor group @xmath250 containing @xmath120 . it is well known that @xmath254 is a homomorphism . to finish the proof we need to show that the restriction of @xmath288 to @xmath42 is a bijection . assume that there are @xmath289 with @xmath290 then both @xmath291 belong to the same coset of @xmath292 hence @xmath293 this implies that @xmath294 as @xmath42 contains @xmath295 we arrived at a contradiction that @xmath57 @xmath296 is a tiling because @xmath297 where @xmath298 . the proof is complete . we recall that a set @xmath299 @xmath35 is @xmath300-periodic , @xmath301 if @xmath302 iff @xmath303 for all @xmath304 and @xmath300 is the smallest number with the property . [ c]let @xmath57 be a lattice tiling of @xmath15 by @xmath42 given by the homomorphism @xmath305 @xmath306 then @xmath57 @xmath36is @xmath307-periodic , where @xmath308 where @xmath309 stands for the order of the element @xmath310 in the group @xmath81 . let @xmath311 , where @xmath312 . it is well known that each lattice is periodic . to find the smallest number @xmath300 for which @xmath68 is periodic is equivalent to find the smallest @xmath300 so that @xmath313 for all @xmath314 the smallest number @xmath315 for which @xmath316 equals the smallest number for which @xmath317 ( in the group @xmath81 ) ; i.e. , equals @xmath318 therefore , the smallest number @xmath300 so that @xmath313 for all @xmath304 equals @xmath319 . now we list some properties of a homomorphism @xmath251 in the above theorem that are useful to verify the existence of a required tiling . the first of them claims that @xmath254 is fully determined by images of @xmath320 under @xmath321 [ 1 ] if a mapping @xmath322 is a homomorphism then , for any @xmath323 it is @xmath324 . the proof follows from the fact that @xmath325 for any homomorphism @xmath254 and that @xmath326 can be written as @xmath327 the following property is clearly true for all tilings and not only for lattice ones . it allows to cut down significantly on the number of cases if one wishes to prove the non - existence of a certain tiling . the proof is obvious and therefore left to the reader . [ 2]if @xmath311 is a tiling of @xmath15 by @xmath42 then there is a tiling of @xmath15 of the form @xmath328 for any translation @xmath329 of @xmath42 . based on conditions [ 1 ] and [ 2 ] the following procedure enables one to find a lattice tiling of @xmath15 by a set @xmath42 or to prove the non - existence of such tiling . \(i ) take arbitrary copy @xmath329 of @xmath44 it is convenient to choose @xmath329 so that @xmath330 and @xmath331 for as many @xmath332 as possible . \(ii ) choose an abelian group @xmath81 of order @xmath333 \(iii ) choose a homomorphism @xmath255 ; i.e. , choose @xmath2 elements of @xmath81 as @xmath334 check whether the restriction of @xmath254 to @xmath329 is a bijection . as there are finitely many abelian groups of a fixed finite order the above procedure is finite . the computational complexity depends on the size of @xmath247 and on the number of abelian groups of the given order . we illustrate the procedure by two simple examples . in fig . 1 , it is shown how to choose @xmath335 to get a tiling of @xmath336 into lee spheres @xmath246 consider now the set @xmath42 @xmath337 we will show that there is no lattice tiling of @xmath15 by @xmath338 suppose that there is a lattice tiling by @xmath42 . then there is a homomorphism @xmath255 with required properties . clearly @xmath339 as there is only one abelian group of order @xmath340 to cut down on the number of cases we note that wlog we can set @xmath341 indeed , all elements of @xmath342 are its generators , therefore there exists an automorphism @xmath343 on @xmath342 with @xmath344 for any @xmath345 we are left with two choices , namely @xmath346 and @xmath347 for @xmath348 however , in both case the restriction of @xmath254 to @xmath42 is not a bijection , see fig . 2 . thus , there is no lattice tiling of @xmath15 by @xmath44 to be able to enumerate all - not only lattice - diameter-@xmath1 perfect lee codes in @xmath15 , we will describe a construction of non - lattice tilings of @xmath15 . it turns out that a variation of theorem [ b ] might be even used as a starting point to produce such tilings of @xmath15.@xmath349to do so we need to consider a generalization of the problem to a super - lattice of @xmath15 . this is a quite common situation in mathematics when a problem has to be generalized before we are able to find its solution . to have a geometric interpretation of this generalization , instead of tiling @xmath15 by a finite set of words @xmath42 , we will consider tilings of euclidean @xmath2-space @xmath350 by clusters of unit cubes . in this setting a tiling of @xmath15 by @xmath42 generates a tiling of @xmath350 by a cluster @xmath82 of unit cubes so that each word of @xmath42 is a center of a unit cube in @xmath82 . e.g. , a tiling of @xmath15 by a sphere @xmath48 generates a tiling of @xmath350 by cubistic cross - polytopes , see fig . 3 for @xmath351 and @xmath352 . we note that the cubistic cross polytope with @xmath173 is also called @xmath2-cross . clearly , a tiling @xmath57 of @xmath350 by @xmath82 generated by a tiling of @xmath15 has the following property : let @xmath353 be two tiles in @xmath57 with cubes @xmath354 and @xmath355 . if @xmath82 and @xmath356 have a non - empty @xmath357-dimensional intersection then @xmath82 and @xmath356 are neighbors ; this type of tiling is also called regular or a face - to - face tiling . we recall that two cubes @xmath358 are neighbors if their centers @xmath359 have the following property : there is an index @xmath40 with @xmath360 and @xmath361for @xmath362 . thus , a non - regular tiling @xmath363 of @xmath350 by a cluster of cubes @xmath82 contains two cubes @xmath354 and @xmath364 so that @xmath82 and @xmath356 have a non - empty @xmath365-dimensional intersection but they are not neighbors . see fig . 4 for an example of two unit cubes with a non - empty @xmath366 dimensional intersection that are not neighbors for @xmath367 it turns out that non - regular lattice tilings of @xmath350 by a suitable cluster of cubes will allow us to construct non - lattice , and even non - periodic tilings of @xmath15 by spheres @xmath368 and double - spheres @xmath163 we will use the following straightforward generalization of theorem [ b ] . [ d]let @xmath42 be a subset of a lattice @xmath369 then there exists a lattice tiling @xmath57 of @xmath370 by @xmath42 if and only if there is an abelian group @xmath81 of order @xmath371 and a homomorphism @xmath372 so that the restriction of @xmath254 to @xmath42 is a bijection . theorem [ d ] can be proved by mimicking the proof of theorem [ b ] . to construct a non - regular tiling we will choose @xmath370 to be the lattice generated by vectors @xmath373 then a tiling @xmath57 of @xmath350 by a cluster of cubes with their centers in @xmath370 will be non - regular if and only if there are in @xmath57 a cube @xmath82 with its center at an integral point and a cube @xmath356 with its center at a point whose first coordinate is of form @xmath374 . theorem [ e ] provides a sufficient condition for the existence of such non - regular tiling of @xmath350 by a cluster of cubes . [ e]let @xmath82 be a cluster of cubes centered at @xmath375 and @xmath376 be a lattice generated by vectors @xmath377 then there is a lattice non - regular tiling of @xmath378 by @xmath82 if there is a homomorphism @xmath379 an abelian group of order @xmath380 so that the restriction of @xmath254 to @xmath381 is a bijection and @xmath335 is a generator of @xmath81 for some @xmath382 the homomorphism @xmath254 determines a tiling @xmath383 of @xmath370 by @xmath384 where @xmath385 is a lattice . clearly , @xmath57 generates a tiling of @xmath350 by the cluster @xmath82 of cubes centered in @xmath42 . indeed , split each cube in @xmath82 by a hyperplane orthogonal to @xmath386 then we get a collection of @xmath387 half - cubes centered at points of @xmath388 in @xmath389 thus both a tiling @xmath390 of @xmath90 by @xmath42 and a tiling @xmath391of @xmath370 by @xmath392 they generate a tiling of @xmath350 by the cluster @xmath82 . the only difference between @xmath390 and @xmath393 is that while the centers of cubes of @xmath390 are in @xmath394 the centers of the half - cubes in @xmath395 are in @xmath389 to prove that @xmath57 generates a non - regular tiling of @xmath350 it suffices to show that there are @xmath396 in @xmath68 so that @xmath397 is an integer while @xmath398 is of the form @xmath399 where @xmath397 and @xmath398 are the first coordinates of @xmath141 and @xmath400 respectively . indeed , then the centers of cubes in @xmath401 will be at integer points , while the centers of cubes in @xmath402 will not . since @xmath403 for some @xmath404 is a generator of @xmath81 then there are numbers @xmath405,@xmath406 so that @xmath407 and @xmath408 . thus , the two vectors @xmath409 have the required property . this section contains main results of our paper . we focus on the existence and enumeration of diameter perfect lee codes over @xmath14 and @xmath231 it is conjectured in @xcite and @xcite , see section 2 , that for @xmath169 these codes exist only when @xmath410 with the exception of @xmath411 code constructed by minkowski . we concentrate on linear codes as they are most important from the practical point of view . all values of @xmath0 for which there exists a linear @xmath152 code ( = linear @xmath153 code ) have been found in @xcite . now we determine all values of @xmath412for which there is a linear @xmath154 code . clearly , it suffices to determine all values @xmath0 for which there exists a linear _ non - periodic _ @xmath111 code as all other linear @xmath111 codes can be obtained by a periodic extension of a non - periodic one . the following theorem claims that instead of investigating diameter perfect codes of @xmath14 one can confine himself to codes in @xmath90 . the statement is intuitively clear and in some papers the authors take validity of analogous cases for granted . we will sketch its proof . we recall that a set @xmath299 @xmath35 is @xmath300-periodic , @xmath413 if @xmath302 iff @xmath303 for all @xmath414 and @xmath300 is the smallest positive number with the property . [ a]for @xmath73 is even , there is a linear @xmath111 code if and only if there is a linear @xmath110 @xmath300-periodic code with @xmath415 we start with the following auxiliary statement . _ let @xmath68 be a linear diameter-@xmath73 perfect lee code in @xmath35 and @xmath329 be a double - sphere @xmath114 with @xmath416 then @xmath417 is a lattice tiling of @xmath35 by double - spheres . first we prove this lemma . assume that for @xmath418 there is @xmath47 such that @xmath419 then there are @xmath420 so that @xmath421 this in turn implies @xmath422 which is a contradiction since @xmath423 for any @xmath424 as @xmath68 is a diameter-@xmath73 code while @xmath425 for any @xmath426 because @xmath329 is an anticode of diameter @xmath89 to finish this part of the proof we need to show that @xmath427 this is obvious for @xmath428 as the space is finite and @xmath68 is a transversal of a tiling by double - spheres . if @xmath68 is a linear code then the tiling is a lattice tiling . assume now that @xmath68 is a linear @xmath110 code , and @xmath429 then the volume @xmath430 of the lattice @xmath68 would be strictly bigger than the volume of the double - sphere @xmath114 which is a contradiction as @xmath68 is a transversal of a tiling of @xmath15 by double - spheres . the proof of lemma is complete . let @xmath68 be a linear @xmath111 code . then @xmath431 for all @xmath432 and by the above lemma @xmath433 is a lattice tiling of @xmath434 by double - spheres @xmath435 consider an extension of @xmath68 given by @xmath436 where @xmath437 and @xmath438 for all @xmath439 it is a matter of technical routine to prove that @xmath440 is a lattice as well with @xmath441 for all @xmath442 , of period @xmath11 and @xmath443 is a lattice tiling of @xmath15 . therefore , @xmath440 is a linear @xmath110 code that is @xmath300-periodic for some @xmath415 on the other hand , let @xmath68 be a linear @xmath168 @xmath0-periodic code . then , @xmath444 for all @xmath445 and by our lemma , @xmath433 is a lattice tiling of @xmath231 let @xmath446 be a restriction of @xmath68 to @xmath447 clearly , @xmath431 for every @xmath448 further , as above , it is a matter of routine to verify that @xmath449 is a lattice and @xmath450 is a tiling of @xmath14 by double - spheres @xmath435 thus , @xmath451 is a linear @xmath111 code . the previous theorem claims that instead of showing the existence of a distance-@xmath73 perfect linear lee code over @xmath14 ( = a tiling of @xmath434 by double - spheres ) it suffices to prove the existence of a linear , @xmath300-periodic , @xmath452 distance-@xmath73 perfect lee code over @xmath15 ( = a tiling of @xmath15 by double - spheres)@xmath221 the difference in difficulty of proving the existence of a lattice tiling of @xmath14 and of @xmath15 might seem to be negligible . however , when looking for a tiling of @xmath15 one has at its disposal a powerful construction based on groups homomorphism described in the previous section . we are ready to determine all @xmath0 for which there exists a non - periodic linear @xmath154 code . let @xmath453 be the prime number factorization of @xmath454 where possibly @xmath455 and @xmath456 for @xmath457 set @xmath458 then a linear non - periodic @xmath154 code exists if and only if @xmath459 where @xmath460 and @xmath461 . in particular , the smallest @xmath0 for which there exists a linear non - periodic @xmath154 code equals @xmath462 assume that there is a linear non - periodic @xmath154 code . by theorem a there is a linear @xmath0-periodic @xmath110 code @xmath463 and by the lemma stated in the proof of theorem [ a ] , @xmath464 is a lattice tiling of @xmath15 by double - spheres @xmath163 as a copy @xmath42 of @xmath465 we choose @xmath466 clearly , @xmath467 by theorem [ b ] there is a homomorphism @xmath468 an abelian group of order @xmath469 such that the restriction of @xmath254 to @xmath42 is a bijection . first we show that @xmath470 we note that @xmath471 . indeed , consider the subgroup @xmath210 generated by @xmath472 then it is easy to check that for each @xmath473 , @xmath474 or @xmath475 note that this is true also for @xmath476 thus @xmath477 which in turn implies @xmath478 see corollary [ c]@xmath221 now we prove that @xmath415 let @xmath479 be the factorization of @xmath83 where possibly @xmath480 that is , possibly @xmath81 is a cyclic group . we recall that @xmath481 so @xmath482 we show that , for every @xmath483 there is @xmath484 depending only on @xmath485 so that @xmath486 by corollary [ c ] , this will in turn imply that @xmath415 as @xmath487 there is @xmath488 so that @xmath489 wlog assume @xmath490 let @xmath491 we consider two cases . if @xmath492 then @xmath493 and we are done . otherwise , for @xmath494 consider the element @xmath495 since the restriction of @xmath254 to @xmath42 is a bijection , there is @xmath40 so that @xmath496 if @xmath497 or @xmath498 we are done as @xmath499 . finally we are left with two cases , either @xmath500 or @xmath501 for some @xmath40 . in the former case , @xmath502 in the latter case @xmath503 as @xmath504 and @xmath505 we have @xmath506 that is @xmath507 in both cases . thus we proved that @xmath508 which implies that @xmath415 in aggregate , @xmath470 it is obvious that @xmath509 indeed , by corollary c , @xmath57 is @xmath0-periodic where @xmath510 as the order of each element divides the order of the group also @xmath0 does , so @xmath511 which in aggregate implies that @xmath512 where @xmath513 and @xmath514 . now we show that there is a linear non - periodic @xmath154 code for each @xmath512 where @xmath515 and @xmath514 . by theorems [ x ] and [ a ] , it suffices to construct a @xmath0-periodic , lattice tiling @xmath57 @xmath516 of @xmath15 by double - spheres so that , for any @xmath437 the lee weight @xmath517 of @xmath141 is even . this tiling @xmath57 will be constructed by means of theorem [ b ] . hence it suffices to find a homomorphism @xmath251 , where @xmath81 is an abelian group of order @xmath469 so that the restriction of @xmath254 to a @xmath42 is a bijection . to guarantee that @xmath252 is @xmath0-periodic , we choose @xmath81 so that @xmath518 for each @xmath519 , see corollary [ c ] , and we choose @xmath520 so that @xmath521 to have a group @xmath81 of order @xmath522 with the property @xmath518 for each @xmath235 we set @xmath523 where @xmath524 and @xmath525 where @xmath526 for all @xmath527 the homomorphism @xmath254 is defined by choosing @xmath528 first we set @xmath529 and @xmath530 for @xmath531 it is easy to check that @xmath532 similarly , for @xmath533 , @xmath534 with @xmath535 we set @xmath536 for a suitable fixed @xmath537 clearly , @xmath538 @xmath539 finally , if @xmath533 , @xmath534 with @xmath540 we associate @xmath541 dimensions with the elements @xmath542 and @xmath543 by @xmath544 where @xmath545 is a suitable fixed number . then @xmath546 @xmath547 it is easy to see that in aggregate @xmath548 so the restriction of the mapping @xmath288 to @xmath42 is a bijection . thus we proved that there exists a lattice tiling @xmath57 of @xmath15 by double - spheres that is @xmath0-periodic . denote the lattice in the tiling @xmath57 by @xmath549 we will prove that , for each @xmath437 the lee weight @xmath550 is even , which in turn implies , by theorem [ x ] , that @xmath61 is a linear @xmath110 code . to prove it we will construct a vector basis @xmath551 of @xmath68 whose each element is of even weight . first we set @xmath552 as @xmath0 is even @xmath553 is even , and @xmath554 i.e. @xmath555 further , we set @xmath556 for @xmath557 also in this case we get @xmath558 and @xmath559 now , for each @xmath560 there is @xmath561 so that @xmath562 with @xmath485-th coordinate equal to @xmath41 we set @xmath563 again , as before , @xmath564 and @xmath565 is even . finally , if @xmath566 where @xmath567 then we set @xmath568 we get @xmath569 @xmath570 @xmath571 and @xmath572 which is even as @xmath573 is always odd , see definition of @xmath574 thus we showed that , for all @xmath304 @xmath575 belongs to @xmath576 and @xmath577 is even . it is not difficult to see that @xmath578 are independent vectors . let @xmath84 be a matrix with @xmath575 being its @xmath40-th row . after permuting columns of @xmath84 so that the columns @xmath579 become columns indexed @xmath580 we get a lower triangular matrix with its diagonal equal to @xmath581 therefore @xmath582which in turn implies that @xmath575 s form a basis with the required properties . the proof is complete . a cluster consisting of @xmath8 unit cubes centered at points of lee sphere @xmath368 is called the @xmath2-cross . by the @xmath2-double - cross we understand a cluster of @xmath522 unit cubes centered at points of double - sphere @xmath583 see fig . 5 for @xmath346-double - cross . szabo @xcite proved that there is a non - regular lattice tiling of @xmath584 by @xmath2-crosses if and only if @xmath8 is not a prime . the non - existence of such lattice tiling for @xmath8 being a prime follows from redei s decomposition theorem see @xcite . we show that , for some @xmath454 there is a non - regular lattice tiling of @xmath584 by @xmath2-double - crosses . at the first glance it seems that non - regular tilings are not useful in coding theory . however , we will show that they are instrumental for enumerating @xmath155 codes . [ f]let @xmath585 be a natural number that is not a power of @xmath156 then there exists a non - regular lattice tiling of @xmath584 by @xmath2-double - crosses . as above we use the double sphere @xmath586 by theorem [ e ] it suffices to find an abelian group @xmath81 of order @xmath587 and a homomorphism @xmath372 so that the restriction of @xmath254 to @xmath381 is a bijection and @xmath335 is a generator of @xmath81 for some @xmath588 since @xmath589 is not a power of @xmath590 then @xmath2 can be written in the form @xmath591 as @xmath81 we choose the cyclic group @xmath592 we set @xmath593 and , for @xmath594 @xmath595 this way @xmath596 @xmath597 @xmath598 @xmath599 for @xmath600 we set @xmath601 for @xmath602 where @xmath603 for fixed @xmath485 we have @xmath604 where @xmath605 @xmath606 @xmath607 @xmath608 thus , in aggregate , @xmath548 to finish the proof it is sufficient to notice that @xmath609 , a generator of @xmath610 the proof is complete . as mentioned in the introduction the double - sphere @xmath611 can be seen as the union of two lee spheres @xmath144 with centers at distance @xmath612 therefore it is somewhat surprising that there exists a non - regular tiling @xmath57 of @xmath613 by @xmath346-double - cross although there is no non - regular tiling of @xmath613 by @xmath346-crosses . a detailed description of the tiling @xmath57 will be provided in the proof of the following theorem that enumerates @xmath614 codes for all @xmath615 we recall that two tilings @xmath616 and @xmath617 are called congruent ( two codes @xmath68 and @xmath449 are called isomorphic ) if there exists a linear distance - preserving bijection @xmath618 that maps @xmath619 on @xmath620 for each @xmath621 there are @xmath622 non - isomorphic @xmath623 codes . first of all we show that the total number of @xmath155 codes is at most @xmath624 each @xmath155 code is a subset of @xmath15 of cardinality @xmath625 . this in turn implies that the total number of @xmath623 codes in @xmath15 is at most the number of ways how to choose @xmath625 points in @xmath15 ; thus it is at most @xmath626 therefore , there are at most @xmath622 @xmath155 codes . to prove the statement we will construct @xmath622 non - isomorphic @xmath614 codes . etzion @xcite proved that there are @xmath622 non - isomorphic @xmath155 codes for @xmath2 being a power of @xmath156 so we are left with the case when @xmath2 is not a power of @xmath156 let @xmath627by theorem [ f ] there exists a non - regular lattice tiling @xmath57 @xmath628 of @xmath584 by double - crosses , where @xmath329 is the @xmath2-double - cross comprising unit cubes centered at words in @xmath629 . for the purpose of this proof it is convenient to know a basis of the lattice @xmath630 let @xmath631 consider vectors @xmath551 given by @xmath632 @xmath633 and @xmath634 for @xmath635 . it is easy to check that @xmath636 as @xmath637 and @xmath593 see the proof of theorem [ f ] , i.e. , @xmath638 for all @xmath639 let @xmath640 be a matrix whose @xmath40-th row is the vector @xmath641 transpose the first and the @xmath485-th column and the first and the @xmath485-th row of @xmath642 an element of the resulting matrix @xmath643 is non - zero if and only if it is in the first column or it is a diagonal element of @xmath644 . thus @xmath575 s are independent vectors . moreover , @xmath645 equals the product of its diagonal elements ; thus @xmath646 therefore @xmath647 constitute a basis of @xmath549 consider a relation @xmath648 on @xmath68 given by @xmath649 if for their first coordinates @xmath650 it is @xmath651 it is easy to check that @xmath648 is an equivalence relation on @xmath68 with two equivalence classes @xmath652 and @xmath653 where @xmath397 is the first coordinate of @xmath654 for @xmath655 set @xmath656 clearly , @xmath657 and @xmath658 form a partition of @xmath659 it is well known , see e.g. @xcite , @xcite , that @xmath660 is a sublattice of @xmath68 and that @xmath661 is a coset of @xmath662 thus @xmath663 and also @xmath664 moreover , both @xmath657 and @xmath658 constitute prisms along @xmath665-axis . that is , if a point @xmath666 then the whole line parallel to @xmath665-axis passing through @xmath120 belongs to @xmath667 as a very important consequence we get that tiles in @xmath658 can be shifted along @xmath665-axis independently of tiles in @xmath668 for example , shifting @xmath669 by @xmath670 produces a tiling @xmath98 of @xmath584 by @xmath2-double - crosses with all their cubes centered at integer points . it is easy to show that @xmath98 is not a lattice tiling but it is a periodic tiling . now we show that in fact both @xmath657 and @xmath671 consist of infinitely many connectivity ( in the topological sense ) components . for the reader s convenience we start with @xmath672 then the basis of @xmath68 comprises vectors @xmath673 and @xmath674 note that cubes centered at @xmath675 belong to @xmath657 as @xmath676 further , as @xmath657 is a prism along @xmath665-axis , and @xmath677 for all @xmath678 we have that cubes centered at @xmath679 for all @xmath680 similarly , all cubes centered at @xmath681 as @xmath682 thus all cubes centered at @xmath683 for all @xmath684 see fig . 6 , where the intersection of the @xmath685-plane with @xmath686 is shaded . the rest of the plane belongs to the intersection with @xmath658 as @xmath687 and both @xmath688s are prisms along @xmath665-axis . thus we proved that both @xmath657 and @xmath658 comprise infinitely many components . in particular , if @xmath689 is the collection of all components in @xmath658 then the point @xmath690 . the proof for @xmath691 @xmath2 not a power of @xmath225 , would be very similar although technically more involved . in this case the points @xmath692 belong to distinct connectivity components @xmath693 of @xmath694 we recall that @xmath695 the connectivity components of @xmath658 can be shifted along @xmath665-axis independently of each other and on @xmath668 this property of @xmath658 will be used to construct @xmath696 non - isomorphic @xmath155 codes . let @xmath697 be an irrational number , @xmath698 be its binary representation . a tiling @xmath699 will be obtained from the tiling @xmath57 by translating connectivity components @xmath693 of @xmath658 as follows : if @xmath700 or @xmath701 then the component @xmath693 will be shifted by @xmath702 otherwise it will be shifted by @xmath703 clearly all double - crosses in @xmath699 are centered at integer points . in other words , @xmath704 is a tiling such that all coordinates of any @xmath705 are integers . such tiling will be called a @xmath9-tiling . obviously , for any two irrational numbers @xmath25 and @xmath706 the corresponding tilings @xmath707 and @xmath708 are distinct ; we recall that two tilings @xmath709 and @xmath710 are called distinct if @xmath68 and @xmath711 are distinct . so we have produced a family @xmath712 @xmath25 is an irrational number in @xmath713 of @xmath714 distinct @xmath9-tilings of @xmath584 by double - crosses . each @xmath9-tiling of @xmath584 by double - crosses induces a tiling of @xmath15 by double - spheres . so the family @xmath98 induces a family @xmath715 of @xmath716 distinct tilings of @xmath15 by double - spheres @xmath717 . clearly @xmath718 for each tiling @xmath719 that is @xmath720 for each tiling @xmath57 in @xmath98 . to each tiling @xmath721 in @xmath715 we assign a set @xmath722 is a center of @xmath723 and @xmath724 is even@xmath725 is even@xmath726 is odd@xmath727 where @xmath728 is the lee weight of @xmath729 . it was showed in the proof of theorem [ x ] , that @xmath730 is a @xmath155 code . it is easy to see that if @xmath731 and @xmath732 are distinct tilings in @xmath715 then the induced codes @xmath733 and @xmath734 are distinct as well . thus , we have constructed a family @xmath735 of @xmath622 distinct @xmath155 codes . as each code in @xmath736 is isomorphic to at most @xmath625 codes in @xmath737(we recall that two codes @xmath57 and @xmath732are isomorphic if there exists a linear distance preserving bijection mapping @xmath57 onto @xmath732 ) @xmath736 has to contain a set of @xmath622 non - isomorphic @xmath155 codes . the proof is complete . as an easy consequence we get : for each @xmath621 the number of non - periodic @xmath155 codes is @xmath624 it is easy to see that for each @xmath738 there are at most finitely many @xmath300-periodic @xmath155 codes . therefore , in aggregate , there are at most @xmath739 periodic @xmath155 codes . as there are in total @xmath740 @xmath155 codes , @xmath622 of them have to be non - periodic . we note that szabo @xcite constructed non - regular lattice tilings of @xmath584 by @xmath2-crosses if @xmath8 is a prime . the non - existence of such lattice tiling for @xmath8 being a prime follows from a result of redei @xcite on factorization of abelian groups . we have constructed a lattice non - regular tiling of @xmath162 by @xmath2-double - cross for all @xmath2 that are not a power of @xmath225 . we believe that there are no non - regular tilings of @xmath162 by @xmath2-double - cross but at the moment we are not able to prove it . in this section we design an algorithm for decoding a linear @xmath168 code . the computational complexity of this algorithm is linear in @xmath741 for @xmath410 and its complexity is @xmath166 for @xmath167 codes . we recall that @xmath110 codes are believed not to exists for @xmath169 and @xmath742 we stress that this algorithm can be applied to all perfect lee codes as diameter perfect lee codes include also perfect error - correcting codes . the efficiency of the algorithm is achieved thanks to the representation of linear codes by means of a tiling constructed through a group homomorphism . we design the algorithm for a linear @xmath110 code . in case of a linear @xmath107 code it suffices to realize that , by theorem [ a ] , the code is a restriction of a @xmath110 code and the corresponding algorithm would be nearly identical . let @xmath68 be a linear @xmath110 code . theorem [ b ] implies that there is a homomorphism @xmath743 where @xmath232 is an abelian group of order @xmath744 @xmath329 is the anticode of diameter @xmath85 of maximum size , and that the restriction of @xmath745to @xmath329 is a bijection . denote by @xmath746 the function inverse to the restriction of @xmath254 to @xmath747 assume that we receive a word @xmath748 first we find the element @xmath749 of the group @xmath610 @xmath36if @xmath750 the neutral element in @xmath751 we are done as @xmath405 is a codeword in @xmath68 . in general the corresponding codeword @xmath142 is given by @xmath752 indeed , @xmath753 is a tiling of @xmath15 by @xmath747 thus , each element @xmath754 can be written in a unique way as @xmath755 where @xmath142 is the sought codeword and @xmath756 hence , @xmath757 hence the complexity of the algorithm is in fact the number of operations needed for calculation of @xmath758 we note that the number of operations in the calculation of @xmath759 is linear in @xmath2 since @xmath760 , where the values @xmath761 will be stored . now we bound from above the number of operations needed for the calculation of the values of the function @xmath762 for any element @xmath310 in the group @xmath610 the values of the function @xmath762 need to be calculated only once and therefore time needed for obtaining these values is not included in the complexity of the decoding algorithm . the values of @xmath762 will be stored in a tabular way and we will evaluate time needed to find the value of @xmath763 in the table for given @xmath235 . the size of the table is the order of the group @xmath83 which equals @xmath764 it is known , see @xcite , that the volume of the double - sphere is @xmath765the volume of the sphere @xmath48 is smaller than the volume of @xmath114 therefore we find the bound only for the double - sphere @xmath114 . however , we can find the required value of @xmath762 in the table in an efficient way . as @xmath766 is an abelian group , @xmath81 can be written as a direct product of cyclic groups @xmath767 . we store values of @xmath762 in the lexicographic order ; that is , first comes value of @xmath762 at @xmath768 then value of @xmath762 at @xmath769 then at @xmath770 etc . the rank @xmath25 of an element @xmath771 in this order is @xmath772 hence , the rank @xmath773 can be calculated in time linear with @xmath598 as from ( 2 ) it follows @xmath774 the maximum value of @xmath598 is achieved when @xmath775 thus the maximum value of @xmath776 . for @xmath171 we get @xmath777 thus @xmath778 where @xmath779 for @xmath196 it is @xmath780 thus @xmath781 in aggregate , the complexity of the algorithm for @xmath165 codes is @xmath782 and its complexity is @xmath166 for @xmath167 codes .

this is a review of the latest developments in the theory of superdense nuclear matter , formed in relativistic heavy ion collisions or in the core of collapsed stars , as they were reported and discussed at the quark matter 2005 conference in budapest ( hungary ) .

nucl-th0508062

the motivation for the study of superdense matter in relativistic heavy ion collisions is the prospect of observing a novel state of strongly interacting matter , the quark - gluon plasma . experiments at the cern super proton synchrotron ( sps ) and the relativistic heavy ion collider ( rhic ) at brookhaven have yielded important clues of the characteristic signatures of this new state in recent years . in fact , the rhic experiments and brookhaven national laboratory ( bnl ) have recently announced the discovery of a new form of matter with the properties of a `` perfect fluid '' , i. e. of a strongly interacting plasma - like state of nuclear matter with an extremely low ratio of shear viscosity to entropy . the lhc will soon extend the range of energy densities that can be explored with measurements of hard probes ( jets , heavy quarks , photons ) . in parallel to these large - scale experimental programs at major accelerator facilities , continuing progress in the detection and observation of pulsars holds the promise of increasing insight into the limits of stability of neutron stars and their astrophysical properties . there is good reason for hope that such studies will ultimately lead to new insights into the equation of nuclear matter at low temperature and high baryon density . it is an understatement to say that this field of research is driven by experiments at this time . in comparison to the amazing wealth of new and exciting data from the rhic and sps experiments reported at this meeting , the progress in our theoretical understanding of the central questions relating to the properties of quark matter has been modest . this remark should not be taken to mean that no progress is being made on the theory front i will discuss some important and intriguing new developments below but as we learn more about the complexities of the problem , we realize that definitive theoretical results will require very large investments in time , manpower , and computing resources . all these are occurring at the current time , which leads to the expectation that steady theoretical progress will be made over the next few years . the role of theory in physics is to develop ideas and concepts which allow us to quantitatively interpret the experimental data and to propose new experimental tests and analysis strategies . we need a steady stream of new ideas , both `` good '' and `` bad '' ones , because it is notoriously difficult to correctly judge the value of a new idea at an early stage in its development . we also need sustained diligent work on established approaches , which have been found to be fruitful and valid , and we must sort out and safely dispose of failed ideas . in my discussion of what was presented at this conference i will begin with the exciting new ideas , i will then report on the progress in the established areas of theoretical relativistic heavy ion physics and , finally , i will discuss the status of theoretical insight in various areas of heavy ion phenomenology . in doing so , i will not be able to cover every single theory talk at qm2005 , and i apologize to all those speakers and presenters of posters whose work remains unmentioned in this summary . one of the greatest puzzles posed by the experimental results from rhic has been the apparently almost instantaneous equilibration of the produced matter demanded by the flow observables . perhaps the most exciting theoretical development relevant to this issue and reported at this meeting ( by s. mrowczyski @xcite , m. strickland @xcite , and y. nara@xcite ) is the idea that plasma instabilities in the soft modes of the gluon field may be responsible for the rapid momentum equilibration . the fact that unstable field modes exist whenever the particle momentum distribution is anisotropic has long been known in electromagnetic plasma physics . but only in recent years has the potential relevance of these so - called weibel instabilities @xcite to relativistic heavy ion collisions been recognized . the basic idea , nicely explained in mrowczyski s talk , builds on the insight that the momentum spectrum of the partons , which are deposited by semihard qcd interactions during the initial impact of the two nuclei , quickly becomes locally anisotropic due to the longitudinal expansion of the system . measured in the comoving frame , the width of the longitudinal momentum distribution becomes much narrower than the width of the transverse momentum distribution : @xmath0 . a transverse color current fluctuation with wavelength in the beam ( @xmath1-)direction less than the color screening length in the plasma will then induce a transverse chromomagnetic field , which tends to further amplify the current fluctuation . in a range of modes @xmath2 , this leads to exponential growth of the color field at a rate @xmath3 , where @xmath4 is the effective gluon mass ( proportional to the plasma frequency ) in the medium . over the past year , the dynamical evolution of this instability has been studied in detail by means of numerical simulations of the nonlinear plasma dynamics in the presence of an anisotropic parton distribution @xcite . the numerical studies show that the initially exponential growth of the unstable modes eventually moderates , when the magnetic energy density in the soft modes reaches the value @xmath5 , where @xmath6 is the typical energy carried by the particles . thereafter , the energy in the soft gluon modes grows only linearly , indicating a continuing cascade of field energy into short wavelength modes . at present , it is unclear which mechanism triggers the end of the exponential growth . what is clear , however , is that the strength of the chromomagnetic field ( @xmath7 ) is sufficient to bend the momentum of the hard partons on a time scale @xmath8 , which thus governs the isotropization of the parton distribution . note that this is an expression which does not involve @xmath9 and thus constitutes a classical time scale . it is also amusing to note that @xmath10 is reminiscent of the characteristic growth time of the ( coarse grained ) entropy of classical gauge field , which is given by the lyapunov exponents @xmath11 : @xmath12 @xcite . a detailed dynamical framework of the role of this mechanism in the process of parton thermalization has yet to be developed . although it appears unlikely that the weak - coupling estimate of the time to full thermalization obtained in the `` bottom - up '' scenario @xcite will be significantly modified , it is quite likely that the plasma instabilities can induce a precocious approach to isotropy of the local momentum distribution . this would already explain the rapid transition to a regime , in which fluid dynamics can be applied , as the comparison of hydrodynamic calculations of the elliptic flow @xmath13 with the rhic data suggests . it may even be possible to devise an effective description of this early phase in terms of a type of chromo - magneto - hydrodynamics . the bending effect of the soft gluon fields would have a similar effect as collisions in limiting the rate of momentum transport . because particle propagation in coherent fields conserves entropy , it is not clear whether such an effect can lead to an increase in the shear viscosity @xmath14 of the medium . boldly assuming that it does , the general expression for the viscosity , @xmath15 ( where @xmath16 denotes the average momentum of a particle ) , would be replaced with @xmath17 , reducing the shear viscosity by a factor @xmath18 compared with the usual perturbative result . one may ask whether the strong gluon fields generated by the weibel instability can be experimentally observed . one way may be through the study of event - by - event fluctuations in @xmath13 @xcite ; another way could be via their effect on jets , which penetrate the plasma in the transverse direction . since also the chromomagnetic fields are aligned in the transverse plane , they can only deflect jets in the longitudinal direction . a rough estimate suggests that the deflection could be substantial , leading to a broadening of the jet cone along the beam direction , but no effect on the azimuthal opening angle of the jet . note that this is exactly what is observed experimentally @xcite . another interesting approach to the thermalization is to invoke the analogy to the unruh phenomenon of the thermal vacuum in accelerated reference frames @xcite , presented here by d. kharzeev @xcite . as argued by unruh and first demonstrated in detail by fulling @xcite , the minkowski space vacuum of a free relativistic quantum field theory appears to a constantly accelerated observer exactly like a thermal state . the temperature @xmath19 and the acceleration @xmath20 are related by the formula @xmath21 or @xmath22 in natural units . the phenomenon is closely related to , though more general , than the hawking effect @xcite , by which black holes radiate thermally . a constantly accelerated observer also experiences an event horizon which , in contrast to the case of black holes , is not a topological property of the space - time geometry . if this insight could be applied to heavy ion collisions , where strong color fields lead to the coherent acceleration of many partons , the spontaneous appearance of a nearly thermal system could be neatly explained . one might even , with slight hyperbole , call the idea `` thermalization by black magic '' ! the question is how far the analogy carries . the idea that external fields other than gravity could be interpreted as mediators of a thermal ensemble has been discussed before ( see e. g. @xcite ) . it turns out that the analogy is not perfect , because ordinary electric or chromo - electric fields do not accelerate all particles equally but differentiate between particles with different charges , in particular between particles and antiparticles . in other words , quarks and antiquarks in a given chromo - electric field do not experience the same event horizon , nor do gluons . the equivalence principle only applies to gravity , not to other interactions . furthermore , gluon radiation can be induced not only by the acceleration of color charges , but also by their rotation in su(3 ) space . this aspect , which was nicely demonstrated in kovchegov and rischke s derivation of the gluon spectrum emitted in the collision of two sheets of color glass condensate @xcite , is peculiar to qcd and does not occur in qed . finally , even if this idea could be invoked with benefit for the explanation of a ( approximately ) thermal initial gluon spectrum , it would not explain the formation of a system that can be described by fluid dynamics , because it completely lacks the notion of a mean free path . thus , at best , the concepts discussed by kharzeev could be used to explain what has recently been called `` pre - thermalization '' @xcite . let us now turn to other new ideas . i find the idea of jet - induced mach or cherenkov cones intriguing , because it may help us learn about the transport properties of the matter created in heavy ion collisions . but since this topic was covered in jrg ruppert s focus talk @xcite , i refrain from summarizing it here once more . another new idea was discussed by j. cramer @xcite , who presented his and g. miller s calculation of identical pion correlations using the framework of the distorted wave born approximation ( dwba ) . their calculation uses the optical potential for pions interacting with a dense nuclear medium , which has long been studied in pion - nucleus scattering and pionic atoms . the potential has a large imaginary part , which describes absorption within the medium . it has been argued that such a description is inadequate , because the strongest absorptive channels such as the @xmath23-meson and the @xmath24-baryon are short - lived resonances , which mostly decay again by pion emission . in other words , the absorbed pions often reappear quickly , albeit in different quantum states . however , i believe that this critique is unfounded . the dwba is an approximation to s - matrix theory , and if one includes absorptive effects in the final state , they really describe the time reversed process , i. e. emission . since the source term in the hbt formalism describes the vertex positions of the last inelastic interactions of the emitted pions , the dwba seems to be a valid approach to a quantum mechanical description that includes the effects of the in - medium propagation of the observed pions . the one issue that is a bit worrisome is that the cramer - miller analysis invokes both , a large pion chemical potential and a strongly attractive potential , which is interpreted as onset of chiral symmetry restoration . the combination of these two effects may bring the pion field precariously close to bose condensation . it would be good to check whether this is a problem or not @xcite . it would also be interesting to apply the dwba approach to other hadrons , such as protons and charged kaons . as a last example of new developments , which were discussed at this meeting , i want to mention the treatment of ( soft - hard ) recombination as a special case of parton fragmentation . hwa and yang @xcite , as well as greco , ko and levai @xcite , have argued for some time that this hadronization mode is an important source of hadrons emitted with momenta @xmath25 gev / c . e. wang @xcite showed how soft - hard parton recombination can be formulated as a special case of parton fragmentation in the presence of a medium . this represents an important step toward a unified , qcd - based theory of hadron emission at intermediate and high transverse momenta . one of the best established theoretical approaches in our field is lattice gauge theory , which is unique in its ability to provide model independent predictions for observables beyond the reach of perturbative qcd . it is no exaggeration to say that most of what we believe to know about the structure of hadronic matter at high temperature is based on results from lattice gauge theory . lattice simulations predict that the properties of strongly interacting matter without net baryon surplus change drastically at an energy density of the order of 1 gev/@xmath26 , corresponding to a temperature of about @xmath27 mev . previous calculations also indicated that this transition is not a true phase transition ( at zero net baryon density ) , but at rapid crossover from a low - temperature phase dominated by hadrons into a high - temperature phase of a plasma of interacting quarks and gluons . the transition is seen as a steep rise in the vicinity of @xmath28 in the effective number of degrees of freedom @xmath29 defined as @xmath30 , where @xmath31 is the entropy density as function of temperature . up to now , limitations in the available computer power forced lattice simulations to make a number of unphysical assumptions . for example , the quark masses were too large , the quark actions violated chiral symmetry , and the number of lattice points in the euclidean time direction , @xmath32 was usually quite small ( @xmath33 ) . this required rather large extrapolations of the numerical results to the physical point , leading to substantial systematic uncertainties in the predictions . the recent advent of computers with multi - teraflops capability has changed this . several calculations of qcd at nonzero temperature are presently underway with realistic quark masses , finer mesh sizes , and improved fermion actions . the first , preliminary results of one of these calculations were reported at this conference by s. katz @xcite , who showed a long list of improvements compared to previous calculations , including physical quark masses , exact monte - carlo algorithms , and @xmath34 . it may be important to mention , however , that the calculation does not use one of the best improved fermion actions , nor a fermion action with an explicit chiral symmetry , such as domain wall fermions . the results presented by katz confirm many of the previous findings . the transition between the low- and the high - temperature phase is still a smooth , but rapid crossover . however , the new calculation seems to suggest a considerably higher value for the critical temperature . katz @xcite reported the values : @xmath35 and @xmath36 , which lie significantly above the values reported at earlier quark matter conferences ( @xmath37 and @xmath38 @xcite , see also @xcite ) . if the new values are confirmed , it would force us to rethink the message from the chemical equilibrium analyses of the hadron yields in relativistic heavy ion collisions . the temperature @xmath39 obtained in these fits is consistently lower than 180 mev ( see e. g. florkowski s talk at this conference @xcite , or @xcite ) . a clear separation between @xmath28 as defined by the peak in susceptibilities and @xmath39 would imply that the hadronic system does not chemically freeze out immediately after the deconfinement transition . the experimentally determined value of @xmath39 would then not have a universal meaning , because the chemical freeze - out would not reflect thermodynamic properties of qcd matter , but aspects of the hadrochemical kinetics , which is system dependent . this would , of course , not be against fundamental principles of physics , but it would constitute an unfortunate circumstance , because it would make an experimental determination of @xmath28 much more difficult . given these implications , it is probably wise to wait for the results of other currently ongoing lattice calculations before revising the standard picture . it is interesting to ask whether it is possible to put data points on the equation of state curve @xmath40 . this would require the simultaneous measurement of the energy density @xmath41 ( or the entropy density @xmath42 or the pressure @xmath43 ) and the temperature @xmath19 . the latter may be possible , if thermal photon radiation from the quark - gluon plasma can be detected @xcite . an elegant novel experimental method , measuring low - mass dileptons at high transverse momentum as a proxy , presented here by the phenix collaboration @xcite is a promising step in this direction . as k. rajagopal and i have recently pointed out , an alternative way of determining the thermodynamically active number of degrees of freedom @xmath29 is to simultaneously measure @xmath41 and @xmath42 @xcite . we already have a rather precise determination of the entropy per unit of rapidity ( @xmath44 , corresponding to @xmath45 @xmath46 at @xmath47 fm / c ) . what is needed now is an independent determination of the energy density at an early time . the energy loss of partons may be the best available tool for such a determination ; the report at this meeting of the observation of the `` punch through '' of the away - side jet @xcite represents an important progress toward this goal ( see also below ) . there is steady progress toward a detailed understanding of the physics of quark matter at high density , but low temperature . k. rajagopal @xcite review the general status of the theory of color superconductivity . if the existence of `` gapless '' color superconductors ( referring to the absence of a gap in some pairing channels ) is confirmed by improved calculations , it could have interesting phenomenological consequences in collapsed stars . if the ground state of the matter has a crystalline structure , of the `` loff '' type , it would allow quark stars to exhibit glitches and other phenomena , which are normally thought to require a solid neutron crust . t. schfer @xcite discussed the progress in constructing effective theories for the dynamics near the fermi surface when unscreened qcd forces vitiate the formalism of fermi liquid theory . on the numerical front , s. eijiri @xcite presented new results of lattice calculations for nonzero baryon chemical potential , using the taylor expansion around @xmath48 . the results impressively confirm the validity of the resonance gas model ( in the baryonic sector ) for @xmath49 . the @xmath50-dependence of the color screening mass agrees well with perturbation theory for @xmath51 . the calculations have reached a state where many model predictions can be tested against rigorous predictions of qcd . a well established approach to the treatment of the evolution of the dense matter created in relativistic heavy ion collisions is fluid dynamics . one of the core insights of the rhic program is that fluid dynamics is successful in describing the collective flow of hadrons and hadron spectra up to @xmath52 gev / c . the strong rapidity dependence of many observables has made it clear , however , that fully three - dimensional calculations are needed . significant progress on this front was reported at this conference . y. hama @xcite presented results from the spherio code , which is uses the particle - in - cell method and implements a continuous freeze - out model , where hadrons are decoupled on the basis of somewhat schematic mean - free path considerations . t. hirano @xcite showed results from a 3-d eulerian hydrodynamics code , which allows for deviations from chemical equilibrium in the hadronic phase . hirano , as well as c. nonaka @xcite , whose code is based on lagrangian fluid dynamics , also presented first results from so - called `` hydro+micro '' models , in which the evolution above @xmath28 is described by 3-d hydrodynamics , and the hadronic phase is treated by a cascade simulation . the few glimpses we have seen here raise the expectation that the new codes will yield a much improved understanding of the rapidity dependence of soft observables . these are important developments for our field , and we can look forward with great interest to the results of further systematic comparisons of these improved transport models with the experimental data , similar to the studies by hirano and nara @xcite of the interplay between jet propagation and hydrodynamical evolution . it will be highly desirable to see direct comparisons between the different codes for the same initial conditions , equation of state , and other assumptions . the increasing complexity of the various implementations of what amounts to essentially identical physics makes it crucially important to establish the validity of these codes before we use them to draw firm conclusions . ( for such a study of the performance of the spherio code , see @xcite . ) the immediate goals are the determination from the data of the equation of state and the initial conditions reached at rhic . other important applications will be calculations of the effect of the medium on hard probes , such as jets , photons , and lepton pairs . several speakers reported substantial progress in microscopic approaches to parton transport . d. teaney s @xcite discussed the momentum evolution of charm quarks in a quark - gluon plasma by means of a collisional langevin equation of the form @xmath53 with gaussian white noise @xmath54 and the friction coefficient @xmath55 , where @xmath56 is the diffusion coefficient of charm and @xmath57 the charm quark mass . the diffusion coefficient is related to the shear viscosity @xmath14 by a relation of the form @xmath58 . teaney showed that the phenix data on the suppression coefficient @xmath59 of non - photonic electrons ( presumably originating from semileptonic charm decays ) can be well described by a value @xmath60 , consistent with a very low value of the shear viscosity @xmath61 , near its conjectured lower bound @xcite . continuing with the topic of charm transport , r. rapp presented results from a model calculation assuming that charm quarks can scatter from d - meson like resonances with various hadronic quantum numbers in the plasma slightly above @xmath28 @xcite . he and his collaborators ( see also the talk by m. mannarelli @xcite ) find that this mechanism strongly accelerates the momentum equilibration of @xmath62-quarks , even for a single resonant state and its partners dictated by the heavy quark and chiral symmetries . m. djordjevic showed the results of a complete radiative energy loss calculation for heavy quarks in a quark - gluon plasma @xcite . on the basis of a schematic space - time model for the absorbing medium she concluded that the @xmath59 value for electrons from charm decays can not be lower than twice the @xmath59 value for pions . this limit is incompatible with the preliminary results from star and only marginally compatible with the phenix data . it would be good to explore whether the limit holds up in the context of more realistic models of the space - time evolution . the importance of comparing the effects of energy loss on light and heavy quarks was also emphasized in n. armesto s talk @xcite . there also was news on the front of partonic cascade codes . z. xu @xcite reported results from a new cascade code , which implements @xmath63 and @xmath64 scattering by means of local rates with detailed balance . the inelastic process turns out to be crucial , as argued by s. wong almost a decade ago @xcite , not only for chemical but also for thermal equilibration . it still needs to be understood in more detail , how the @xmath65 process manages to speed up thermalization more than the increased transport cross section from its more isotropic angular distribution would naively suggest . other implementations of radiative gluon processes in parton transport codes did not show a similarly large effect @xcite . several speakers , such as d. molnar @xcite and b. zhang @xcite presented calculations of binary parton cascades with hugely cranked up binary cross sections . the studies may serve some purpose in schematic explorations of the origin of flow and energy loss at the partonic level . however , the question needs to be asked what exactly these studies are telling us , beside the obvious ( that larger partonic cross sections lead to faster equilibration and to increased collective flow ) , and whether their conclusions are not misleading due to the unrealistic assumptions . another established fruitful idea is parton saturation with the asymptotic approach to a universal structure , the color glass consensate ( cgc ) . k. itakura @xcite gave an impressive review of the recent conceptual and formal developments in this field . the question whether this asymptotic physics is relevant to the phenomena observed at forward rapidities in d+au collisions at rhic is still under discussion ( see the talks by g. veres @xcite , d. rhrich @xcite , and j. jalilian - marian @xcite ) . the problem is that the phenomenology is complex , involving a mixture of initial - state and final - state effects . it appears to be difficult to cleanly separate different sources of nuclear modifications from each other . one important new result in the context of the cgc , reported here by t. lappi @xcite , is that quark - pair production in the strong gluons fields generated by the shattering cgc is quite large . for reasonable values of the parameters , lappi finds that several hundred light quark pairs are produced . on the one hand , this result is welcomed , because it suggests that a chemically equilibrated quark - gluon plasma is originally produced . on the other hand , it poses a potential problem for the standard jet quenching picture , because the total number of gluons and quarks is limited by the total entropy observed at freeze - out . more quarks imply fewer gluons . but quarks are much less efficient at causing parton energy loss , which would imply that the slight underestimate of the measured hadron suppression by many calculations ( see e. g. b. cole s lecture @xcite ) would be further enhanced . it is too early to draw firm conclusion , but the result suggests that a detailed analysis of the possible implications for jet quenching of different abundance ratios of quarks and gluons in the medium is needed . practitioners of lattice theory have told us for quite some time that qcd matter at temperatures not far above @xmath28 is _ not _ a perturbative quark - gluon plasma . the impressive work which has been done over the past five years on perturbative resummation techniques for the qcd equation of state has confirmed this insight @xcite . the perturbative expansion based on hard - thermal loop quasiparticles works well for @xmath66 , but it fails in the region near @xmath28 , which is explored by the rhic experiments . if the plasma below @xmath67 is a strongly coupled one ( a `` sqgp '' in the current terminology ) , the question is whether it contains any long - lived quasiparticles and if so , what their structure is . we may look for guidance to strongly coupled electromagnetic plasmas , which have been thoroughly studied experimentally and theoretically . the physics of these plasmas and its possible relevance to our field was superbly reviewed by m. thoma @xcite . the two parameters controlling electromagnetic plasmas are temperature and charge carrier density . a line defined by the condition @xmath68 , where @xmath20 is the average distance between particles and @xmath69 their electric charge , divides the phase diagram into regions of weak ( @xmath70 ) and strong ( @xmath71 ) coupling . the precise translation of this condition to a quark - gluon plasma is not known , but it must roughly have the form @xmath72 , where @xmath73 is the eigenvalue of the quadratic casimir operator for the color charges making up the plasma . if one inserts numbers , one finds that @xmath74 is , indeed , somewhat larger than unity for the conditions prevailing in the range @xmath75 . the experience from electromagnetic physics suggests that a plasma in this range of parameters is no longer perturbative , but still far away from the region where regular short - range correlations or even quasi - crystalline structue are found . still , the analogy to electromagnetic plasmas is very useful , because it suggests phenomena and properties to look for , and maybe even theoretical tools to apply , in theoretical investigations of the quark - gluon plasma near the phase boundary . an conjecture , whose ultimate value is still uncertain , is the dominance of partonic bound states in the quark - gluon plasma in the region near @xmath28 , which was discussed by e. shuryak @xcite . such substructures are difficult to track in lattice simulations , because they are not color singlets . their calculation on the lattice , therefore , requires gauge fixing with all its concomitant pitfalls . it is much easier , although possibly less conclusive , to look for indirect signals of the abundant presence of partonic bound states on the lattice . two such studies were discussed by v. koch @xcite and f. karsch @xcite in their talks . koch and collaborators focus on the correlations between strangeness and baryon number . if all quasiparticles in the plasma have the quantum numbers of quarks or gluons , only quarks carry strangeness . if bound states of quarks and antiquarks exist as quasiparticles , strangeness is also carried by states without baryon number , as in a hadronic gas . the @xmath76 correlator can be expressed in terms of quark susceptibilites , where the presence of @xmath77 bound states should show up as an enhancement of flavor - off - diagonal elements . existing lattice calculations rule these out with great precision . karsch and collaborators have done a similar study for electric charge correlations , which are sensitive to the presence of quasiparticles carrying diquark quantum numbers . again , the lattice results indicate a rapid transition from a hadron resonance gas to a perturbative quark - gluon plasma in the range @xmath78 . while it may be too early to proclaim the demise of the `` bound state quark - gluon plasma '' model , these new results are not encouraging for the model . it is true that the lattice results mentioned above have nothing to say about gluonic bound states . it would be worthwhile to explore generalizations of these tests to gluons . a direct application is impossible , because one can not introduce a gluonic chemical potential , but maybe it would be possible to define a chemical saturation factor @xmath79 , similar to the strangeness saturation factor @xmath80 which is introduced in hadro - chemical equilibrium models . one would never have to do a lattice calculation with @xmath81 , the parameter would only be used to define derivatives for infinitesimal deviations from equilibrium . anyway , model builders will have to abide by the constraints introduced by these and future lattice results . quite generally , it appears to me that a low viscosity ( relative to the entropy density ) of the quark - gluon plasma argues against structure formation , except possibly at the nearest neighbor level . materials with long - distance particle correlations , such as polymers , generally have large viscosity . the reason is that , although transport cross sections can be huge , momentum transfer in these materials is not dominated by particle transport , but by the transport of momentum along molecular chains or via collective modes . the mean free path for momentum exchange is therefore large . in the extreme case of crystals , momentum transfer is caused by phonons , which may travel freely over extremely large distances , even if the atoms are immobilized . a small viscosity is thus indicative of a liquid without strong correlations , except possibly on the scale of the interparticle separation . is there any approach , other than euclidean lattice simulations , which would allows us to explore the strongly coupled quark - gluon plasma ? the experience from electromagnetic plasmas is that molecular dynamics might be a promising method . two speakers at this conference , e. shuryak @xcite and p. hartmann @xcite , presented first results from what i would call `` qcd inspired '' molecular dynamics calculations . as intriguing as their results are , the fact that these models are constructed in the framework of direct two - particle interactions makes them suspect , for several reasons : such models make it very difficult , if not impossible , to implement local gauge invariance ; also , the perturbatively unscreened sector in high - temperature qcd is the chromomagnetic sector , which is not easily encoded in two - body forces . it is perhaps worth mentioning that a numerical approach to parton molecular dynamics already exists , which encodes hard thermal loops in minkowski space via a coupled system of colored partons and a spatial lattice @xcite . in this formulation , which was put to use in the work on color instabilities presented here by nara @xcite , the partons propagate according to wong s equation @xmath82 while the soft classical fields satisfy the yang - mills equation with the particles as sources : @xmath83 at weak coupling , this system of equations ( first proposed by heinz @xcite ) is known @xcite to reproduce the htl effective theory , and this has also been demonstrated numerically @xcite . no one has explored what these equations predict for strong coupling . although it is clear that the heinz model does not reproduce full qcd when the coupling becomes strong , the fact that we know and understand its weak coupling limit provides confidence . note that the heinz model only contains a mean field and no two - body forces , which may become increasingly important at stronger coupling . it would be straightforward and logical to include such short - distance interactions as scatterings involving color exchange among partons contained in the same lattice cell @xcite . an important theoretical playground for those interested in strongly coupled gauge theories are the supersymmetric gauge theories , for which a dual gravity - like theory exists . starinets @xcite reported about the progress in calculating corrections to the results obtained in the extreme strong coupling limit . for strongly coupled @xmath84 supersymmetric yang - mills theory we now know that the asymptotic value of the shear viscosity / entropy density ratio is approached as ( @xmath85 denotes the number of colors ) @xmath86 .\ ] ] note that @xmath87 does not grow for very strong coupling in this theory , presumably because there is no formation of long - range structure . unfortunately , the mapping between the quasiparticles at weak coupling and those at strong coupling is not well understood in these susy models , and it thus remains unclear to what extent these beautiful results apply to real qcd . however , if it were possible to apply molecular dynamics methods developed for strongly coupled qcd to the susy theories , the analytical results could provide rigorous tests for the numerical methods . there is little to add to c. gale s beautiful review of electromagnetic probes of dense matter @xcite . the amazing data presented at this conference hold promise for more to come . the di - lepton data from na60 @xcite demonstrate that excellent resolution and high statistics are crucial prerequisites for progress . the invariant mass spectra of di - muons from in+in shown by na60 exhibit not only a broad source of prompt di - leptons , but even show the narrower peak from @xmath23-meson decays in vacuum . future calculations will thus have to describe the magnitude of this peak . the data are good enough to discriminate between different models . even a superficial glance suggests that any model , which predicts only a mass shift in medium , without broadening , will face great difficulties . it would be desirable , if theorists would not only compare the new data with various specific models of the in - medium modifications of the @xmath23-meson . obviously , these modifications can be broken down into the contributions of various hadronic channels contributing to the changes in the spectral function of the @xmath88 . as interesting as the obtained insights are , they may obscure the answer to more general questions , such as : what is the route to chiral symmetry restoration and deconfinement ? is it the general broadening of hadronic states until they dissolve into a continuum ? or does the transition occur primarily through shifts in hadron masses or mass splittings between chiral partners ? it would be useful to interpret the di - lepton data in terms of such deeper concepts in a model independent way . we have seen a lot of ( by now ) routine phenomenology on jet quenching , but also some interesting new ideas . n. borghini @xcite presented an approach , in which the medium effects are simply encoded in a one - parameter change of the splitting function within the well established modified leading logarithmic approximation ( mlla ) theory of jet fragmentation . the resulting @xmath89 distribution of jet fragments looks just like what is observed in au+au collisions when the threshold for back - to - back coincidences is set low . it will be interesting to see how far this idea can be pushed and whether the new parameter in the splitting function can be microscopically calculated . another promising new development is the theory of di - hadron fragmentation functions , discussed by a. majumder @xcite . there is a sense among theorists now especially after the cu+cu data did not generate a new surprise that jet quenching at rhic is sufficiently well understood to justify detailed predictions for the lhc , as presented in c. salgado s talk @xcite . such efforts should be encouraged , because a detailed set of prediction will allow us to assess the quality and predictiveness of the theory when the lhc data arrive in a few years . but let me inject a word of caution . as first demonstrated by armesto et al . @xcite , collective flow can influence the energy loss of a parton in the quark - gluon plasma . this is especially true , if the longitudinal expansion is not boost invariant , leading to a modified density profile with time . as renk and ruppert recently showed @xcite , the deduced energy loss parameter @xmath90 can differ by a factor of five ( ! ) between a boost invariant scenario and one with longitudinal acceleration and strong transverse flow . the fact that the latter would give a much larger suppression for the same value of @xmath90 could help counteract , e. g. , the effect of a larger initial quark - to - gluon ratio as suggested by lappi s calculations @xcite mentioned above . as the phenomenology of jet quenching becomes increasingly quantitative , it will be important to compare different calculations of parton energy loss for the same space - time evolution model . on a personal note , i found the results on single hadron suppression ( @xmath59 ) and additional back - to - back suppression ( @xmath91 ) shown by the star collaboration @xcite , especially exciting . for the specific cuts set by star , the two factors were equal : @xmath92 . why should this be the case ? as i and others have argued , @xmath59 is a geometric suppression factor which represents the size of the surface layer from which observed leading hadrons are emitted , compared with the total volume in which jets are produced . similarly , @xmath91 can be viewed as an additional reduction factor describing the width of the `` equatorial '' torus from which the observed di - jets are tangentially emitted with an approximately equal and modest energy loss ( see figure ) . i predicted this relationship in a somewhat schematic treatment of jet quenching a few years ago @xcite and am happy to see the prediction substantiated . clearly , the relative value of @xmath59 and @xmath91 will depend on the selected momentum thresholds , but the general argument suggests that the connection could be another tool in the analysis of dense matter effects on hard partons . some of the lectures at this meeting , experimental as well as theoretical ones ( see e. g. @xcite , have conveyed the message that jet quenching in au+au and cu+cu collisions at rhic is quantitatively well understood . how sure are we of this ? besides the uncertainties already mentioned ( influence of flow , gluon dominance of the early quark - gluon plasma ) , there is a glaring discrepancy between different approaches , which has not been resolved . those who use the data to deduce an effective value of the energy loss coefficient @xmath93 @xcite ( effective , because the quoted value approximately corrects for longitudinal expansion ) have noted that this value is significantly larger , by a factor five or so , than the value predicted years ago by baier @xcite on the basis of perturbative qcd . on the other hand , those who apply perturbative calculations of radiative energy loss to calculate @xmath59 in nuclear collisions @xcite find good agreement with the data . there are two possible resolutions to this conundrum . one would be that the standard fits of @xmath90 overestimate the value of this parameter due to unrealistic assumptions , e. g. about the space - time evolution of the energy density , as discussed above . another one would be that baier s estimate of @xmath90 as a function of the energy density represents a serious underestimate of the true effective value of @xmath90 in perturbative qcd . i believe that both sides owe us a careful analysis of the problem , on which everyone can agree . because jet quenching is such a central aspect of the new discoveries by the rhic physics program , it is important that the apparent contradiction be resolved soon . another challenge for theorists is , of course , the development of techniques for a nonperturbative calculation of @xmath90 on the lattice . this does not seem totally hopeless , because ( a ) gluon dominance makes quenched lattice simulations suitable for this property of the medium , and ( b ) the parameter @xmath90 can be formulated in terms of correlators along the light cone @xcite of the form @xmath94 which could perhaps be calculated by means of analytic continuation using maximum entropy techniques . even if this turns out to be a very hard computational problem , it needs to be attacked , because the claim that the quark - gluon plasma seen at rhic is strongly coupled and nonperturbative in view of its low viscosity and the claim that perturbative qcd describes the data on jet quenching are not easily reconciled . at the very least , it would be important to get a quantitative understanding of the relevant qcd scale for @xmath90 , in order get better control of its value in perturbation theory . this would require a next - to - leading order calculation of radiative energy loss in the medium . quite possibly the most challenging set of data for theorists , which was first presented at this conference , are the phenix results on charmonium production in au+au and cu+cu . even with their limited statistics , they rule out any simple extrapolation of the models of comover suppression developed to explain the cern - sps data . two diametrically opposed conclusions have been drawn from the data , as discussed by m. nardi @xcite : one possibility is that the @xmath95 is not `` anomalously '' suppressed , both at the sps and rhic , the other one is that there is a fair amount of regneration of the @xmath95 at rhic by @xmath96 recombination . the first alternative is supported by the recent results of quenched lattice calculations of the @xmath96 spectral functions , which show the survival of a pronounced resonance in the vector meson channel up to @xmath51 @xcite . the second alternative is supported by the growing evidence from the rhic experiments that @xmath62-quarks participate in the collective flow , suggesting that they are also subject to statistical hadronization by coalescence . finally , there is the possibility that what has been called `` anomalous '' suppression at sps energies is caused by the absorption of excited charmonium states , especially the @xmath97 , feeding into the observed @xmath95 yield . if these states are completely absorbed in the most central events at the sps , then no additional suppression due to this mechanism can occur at rhic . the problem , in my view , is that we still do not have a comprehensive and generally accepted theoretical framework , in which to discuss @xmath95 suppression . the reason for this lack is not clear . if the @xmath95 acts as a `` hard '' probe , it should be possible to derive a systematic formulation of its production and destruction in heavy ion collisions within the framework of perturbative qcd and effective field theory ( e. g. , nonrelativistic qcd ) . the heavy quark mass provides a large scale , which can serve to separate nonperturbative physics of the dense medium from the dynamics of the heavy quark . such a formulation might only be marginally valid for @xmath62-quarks , but it would surely work well for @xmath98-quarks , which will soon become the center of attention at the lhc . on the other hand , if such a formalism can not be found , we will need to give up on the notion that the @xmath95 ( or the @xmath99 ! ) is a `` hard '' probe . the ball is clearly in the theorists court , and the new data provide ample motivation to address this issue vigorously . on specific issues , the failure of direct lattice calculations of the @xmath96 spectral functions @xcite and potential models motivated by the lattice results @xcite is puzzling . the spectral functions derived from the potential models do not agree with those obtained by analytic continuation directly from the lattice . i am not sure we fully understand the reason for this discrepancy . the spectral functions derived from the lattice do not indicate the presence of a continuum in the vector meson channel . because the @xmath100 channel can be easily incorporated into potential models , but remains absent in quenched lattice calculations of the spectral functions , the role of dynamical quarks in the survival of heavy quarkonium states at @xmath101 remains an issue of concern . it is unclear , however , why the lattice results do not reveal a gluonic continuum at @xmath102 . the existence of a possibly broadened peak in the @xmath95 spectral function does not necessarily mean that the initially created @xmath95 mesons survive in the medium . the present lattice calculations may not reflect the true width of the state and thus may not give a good estimate of its lifetime inside the quark - gluon plasma . on the topic of regeneration , more detailed calculations of @xmath95 formation by @xmath96 recombination are urgently needed . given the success of recombination models for light hadrons , reliable calculations should be within reach . overall , it is clear that there is much room for theoretical improvements in the area of charmonium suppression . as w. florkowski s @xcite talk aptly demonstrated , the dispute about the relationship between chemical and kinetic ( `` thermal '' ) freeze - out of hadrons at the end of a heavy ion collision remains unresolved . clearly , there exist strong theoretical arguments for a differential freeze - out of various hadrons depending on their mean free paths in an equilibrated hadron gas . hadrons with no or few known resonant excited states , such as the @xmath103-meson or the @xmath104-hyperon should freeze out early . on the other hand , a common general freeze - out of hadron abundances and momentum spectra would require a specific mechanism forcing a sudden termination of efficient equilibration mechanisms . if such a scenario were realized in nature , i can only imagine it to be associated with the hadronization transition itself . the fact that , according to lattice qcd , this transition is smooth and not discontinuous in thermal equilibrium , such a scenario will require strongly off - equilibrium conditions during hadronization ( such as the sudden decay of a deeply supercooled state @xcite ) . one can ask whether the available data already provide conclusive evidence for or against a common freeze - out . the new high resolution di - lepton data from na60 should be sufficient to set a lower bound on the duration of the strongly interacting hadron gas phase . the idea that the @xmath23-meson yield seen through the lepton - pair decay channel measures the lifetime of the hadronic phase is not new @xcite . but only now do we have data from the sps , which tell us how many @xmath88 s decay inside a hadronic medium , which is dense enough to modify the spectral function . this broadening is related to the rate of binary collisions which , in turn , maintain the kinetic ( but not necessarily the chemical ) equilibrium of the hadronic medium . single freeze - out models are thus challenged to describe the observed yield of low - mass di - lepton emission . other data , such as the results from star on the yield of strange resonances as function of centrality ( see s. salur s talk @xcite ) should also provide serious tests for freeze - out models . the proponents of these models need to address these results . let me now turn to strangeness . most fits to the data yield an almost perfect abundance equilibration of strange hadrons ( @xmath105 ) , which was predicted long ago as one of the characteristic signatures of deconfinement and chiral symmetry restoration . however , there exist other high - quality fits of hadron yields @xcite , which invoke a significant oversaturation of strangeness . furthermore , because of the different masses of the carriers of strangeness in the quark - gluon plasma and the hadron gas phase , the value of @xmath80 does not have to remain the same during the transition . this suggests that it might be an interesting intellectual exercise to study the equation of state of qcd as a function of @xmath80 , the overall strangeness abundance . as in the case of @xmath50 , one would probably want to calculate the derivatives with respect to @xmath106 at the equilibrium point @xmath105 . in many respects , this method would correspond to studying the effects of a varying number of flavors on the equation of state , which we know to be nontrivial . the recombination model for hadron production at intermediate transverse momenta continues to work amazingly well . the extended range of particle identification in runs 4 and 5 is finally allowing the data to probe the predicted transition between the region dominated by recombination and the high - momentum region dominated by fragmentation . results shown at this conference @xcite confirm the theoretical predictions of a transition between 4 and 7 gev / c , and theorists are eagerly waiting for more analyzed data . how quickly the experiments move to challenge theoretical predictions was demonstrated by star s presentation of data on the systematic baryon / meson violations of the valence quark number scaling of the @xmath107-dependence of the elliptic flow . including higher fock states in the hadron wave functions , we had recently predicted deviations at the few - percent level @xcite . a second source of such deviations from the scaling law can be caused by the internal momentum distribution of constituent quarks @xcite . how exciting to see that the data confirm the existence of such deviations with the correct sign @xcite ! more work will need to be done to determine whether the deviations from universality provide another confirmation of the recombination model , or whether they constitute a challenge . other theoretical progress reported at this conference were the effects of parton correlations on same - side di - hadron yields @xcite and the possible origin of the pedestal effect in the same data @xcite . as r. lednicky s talk @xcite made clear , the `` hbt puzzle '' of the rhic data is still with us . generally the measured correlations indicate a short overall duration of the reaction , strong transverse flow , and a very brief , almost sudden freeze - out . theory still has great trouble getting the small value @xmath108 right . as the success of hydro - motivated parametrizations of the freeze - out shows @xcite , it is not impossible to reproduce the data . the problem lies in finding a dynamical evolution model , which yields the right freeze - out parameters . there is justified hope that the new 3-d hydro plus cascade codes may change the situation . stay tuned to see whether this venerable problem finds a resolution soon . this conference has been a showcase of significant progress in the theory of superdense hadronic matter and its formation and evolution in relativistic heavy ion collisions . it is certainly a very productive time for theorists in our field , because the experiments provide for a wealth of data which require interpretation and explanation . the data also make it easier than ever before to verify or repudiate theoretical approaches and models . as i have tried to point out , the successes of theory are not evenly distributed . there are some areas , where theory is well on its way toward definite descriptions , and others , where even a comprehensive framework is still lacking and it is time to catch up . above all , however , it is upon us theorists to embrace the offer made by one of the earlier speakers in this session @xcite : `` we look forward to working with the theory community to extract the properties of the matter ( produced in relativistic heavy ion collisions ) . '' the following remarks were motivated by comments received after posting a draft version of my summary talk . the characterization of the matter produced in nuclear collisions at rhic is an experimental issue . on the other hand , the predictions of qcd for the properties of thermal qcd matter in the temperature range @xmath109 are a theoretical issue . as discussed above , it has been widely argued that the quark - gluon plasma in this temperature range is a strongly coupled state of matter a `` sqgp '' @xcite . how compelling are the theoretical arguments for this claim ? it is possible to derive an effective theory for the long - distance dynamics ( momenta @xmath110 ) of thermal gauge theories . if the dynamics at momentum scales @xmath111 is weakly coupled , the effective theory can be derived perturbatively . two forms of this effective theory are known . one is the hard - thermal loop ( htl ) effective theory @xcite , the other is the dimensionally reduced theory based on local interactions of the gauge field @xcite . in both frameworks one can ask how strong the effective coupling @xmath112 is as a function of temperature . this question has recently been studied by laine and schrder @xcite , who calculated the effective action for qcd in the dimensionally reduced theory up to two - loops . they found that the effective coupling constant @xmath113 is well controlled down to @xmath114 and remains remarkably small : @xmath115 . these authors also found that the effective theory agrees very well with full thermal qcd for the spatial string tension , which is an important measure of the long distance dynamics . similar conclusions are reached in the framework of the htl approach @xcite which , however , has not yet been carried out to full two - loop order and thus does not yet permit the extrapolation all the way down to @xmath28 . these studies suggest that the quark - gluon plasma in the temperature region near @xmath28 is well described by an effective theory with a modestly strong coupling constant . this raises the question , where the transition between the weak coupling and strong coupling regimes of thermal qcd occurs : at @xmath28 or at some @xmath101 ? this can not be decided by numerology , because the perturbative expansion parameter is , generally , @xmath116 , while the expansion parameter of the strong coupling expansion of string duals of nonabelian gauge theories is @xmath117 @xcite , and the conditions @xmath118 and @xmath119 are both fulfilled in the vicinity of @xmath28 . turning to physical arguments , it is noteworthy that the perturbative quasiparticles of the effective theory , thermal gluons and plasmons , are short - lived . the collisional width of a gluon / plasmon at rest is given by @xcite @xmath120 on the other hand , the effective mass of a gluon / plasmon at rest is @xmath121 . roughly the same relationship holds for thermal quarks @xcite . in other words , all slowly moving quasiparticles are strongly damped . the characteristic nature of the temperature region near @xmath28 may thus be that the quasiparticles , gluons and quarks above @xmath28 and hadrons ( such as the @xmath23-meson ) below @xmath28 , are strongly collision broadened . such a property would be indicative of a liquid , which is characterized by the absence of long - lived quasiparticles and long - range order . because the effective coupling @xmath122 changes only slowly with temperature , it is by no means clear how for above @xmath28 one needs to go before the widths of the plasma quasiparticles become a higher - order effect compared to their masses . does the apparent nearly `` perfect fluid '' property of the matter observed at rhic ( @xmath123 ) imply that the quark - gluon plasma has , by some measure , a very low shear viscosity or is it simply a signature of a large number of degrees of freedom , i. e. a large entropy density , as hirano and gyulassy @xcite have recently argued ? the general expression for the shear viscosity is @xmath15 , where @xmath23 is the particle density , @xmath124 the average momentum of a partice , and @xmath125 the mean free path . since @xmath126 , where @xmath127 denotes the transport cross section , the density cancels from the expression , and we have @xmath128 . because cross sections are unitarity bounded ( @xmath129 for s - wave scattering ) , we find approximately @xmath130 where we made use of the relation @xmath131 for relativistic matter . it is against this value that the smallness of the viscosity of the quark - gluon plasma should be judged . on the other hand , the entropy density of a relativistic medium is approximately @xmath132 , allowing us to write @xmath133 . the smallness of @xmath87 is thus a measure of the smallness of the mean free path , compared with the thermal wavelength @xmath134 of a particle . as such , it is not directly sensitive to the number of degrees of freedom in the medium . this work was supported in part by the u. s. department of energy under grant de - fg02 - 05er41367 . my special thanks go to the organizers of this conference for making it such a stimulating and productive experience . i am grateful for comments from m. asakawa , t. csrg , i. dremin , d. denterria , t. hirano , c.m . ko , s. mrowczyski , p. petreczky , r. pisarski , r. rapp , s. wong , and w. zajc on an earlier version of this manuscript . i also thank m. strickland for helpful communications . p. arnold , g. d. moore and l. g. yaffe , phys . d * 72 * , 054003 ( 2005 ) ; p. arnold , j. lenaghan , g. d. moore and l. g. yaffe , phys . lett . * 94 * , 072302 ( 2005 ) . a. rebhan , p. romatschke and m. strickland , jhep * 509 * , 41 ( 2005 ) ; phys . * 94 * , 102303 ( 2005 ) . a. dumitru and y. nara , phys . b * 621 * , 89 ( 2005 ) . t. s. bir , s. g. matinyan and b. mller , _ chaos and gauge field theory _ , world sci . lect . notes phys . * 56 * ( 1994 ) 1 . r. baier , a. h. mueller , d. schiff and d. t. son , phys . b * 502 * , 51 ( 2001 ) . s. mrowczyski , arxiv : hep - ph/0506179 . j. adams _ et al . _ [ star collaboration ] , arxiv : nucl - ex/0411003 . w. g. unruh , phys . d * 14 * , 870 ( 1976 ) . d. kharzeev , these proceedings ; d. kharzeev and k. tuchin , nucl . a * 753 * , 316 ( 2005 ) . s. a. fulling , phys . d * 7 * , 2850 ( 1973 ) . s. w. hawking , commun . phys . * 43 * , 199 ( 1975 ) [ erratum - ibid . * 46 * , 206 ( 1976 ) ] . b. mller , w. greiner and j. rafelski , phys . a * 63 * , 181 ( 1977 ) . y. v. kovchegov and d. h. rischke , phys . c * 56 * , 1084 ( 1997 ) [ arxiv : hep - ph/9704201 ] . j. berges , s. borsanyi and c. wetterich , phys . * 93 * , 142002 ( 2004 ) . y. v. kovchegov , these proceedings ; y. v. kovchegov , nucl . a * 762 * , 298 ( 2005 ) . j. ruppert , these proceedings , and references therein . j. g. cramer , these proceedings ; j. g. cramer , g. a. miller , j. m. s. wu and j. h. s. yoon , phys . rev . * 94 * , 102302 ( 2005 ) ; g. a. miller and j. g. cramer , arxiv : nucl - th/0507004 . this may not be a problem for the model parameters required to describe the rhic data ( j. g. cramer , private communication ) . r. c. hwa and c. b. yang , phys . c * 70 * , 054902 ( 2004 ) . v. greco , c. m. ko and p. levai , phys . lett . * 90 * , 202302 ( 2003 ) . e. wang , these proceedings ; a. majumder , e. wang and x. n. wang , arxiv : nucl - th/0506040 . s. katz , these proceedings . f. karsch , nucl . a * 698 * ( 2002 ) 199 . z. fodor and s. d. katz , jhep * 0404 * , 050 ( 2004 ) . w. florkowski , these proceedings ; w. florkowski and w. broniowski , acta phys . b * 35 * , 2895 ( 2004 ) . j. cleymans , b. kampfer , m. kaneta , s. wheaton and n. xu , phys . c * 71 * , 054901 ( 2005 ) ; p. braun - munzinger , d. magestro , k. redlich and j. stachel , phys . b * 518 * , 41 ( 2001 ) . d. denterria and d. peressounko , arxiv : nucl - th/0503054 . s. bathe , these proceedings . t. hirano and y. nara , phys . c * 66 * , 041901 ( 2002 ) ; phys . c * 69 * , 034908 ( 2004 ) . c. e. aguiar , t. kodama , t. osada and y. hama , j. phys . g * 27 * , 75 ( 2001 ) . d. teaney , these proceedings ; g. d. moore and d. teaney , phys . rev . c * 71 * , 064904 ( 2005 ) . r. rapp , these proceedings ; h. van hees and r. rapp , phys . c * 71 * , 034907 ( 2005 ) ; m. mannarelli and r. rapp , arxiv : hep - ph/0505080 . m. mannarelli , these proceedings . s. m. h. wong , phys . c * 54 * , 2588 ( 1996 ) ; phys . c * 56 * , 1075 ( 1997 ) . z. xu , these proceedings ; z. xu and c. greiner , phys . c * 71 * , 064901 ( 2005 ) . g. r. shin and b. mller , j. phys . g * 28 * , 2643 ( 2002 ) ; j. phys . g * 29 * , 2485 ( 2003 ) . d. molnar , these proceedings ; d. molnar , arxiv : nucl - th/0503051 . b. zhang , these proceedings . j. jalilian - marian , these proceedings ; a. dumitru , a. hayashigaki and j. jalilian - marian , arxiv : hep - ph/0506308 . t. lappi , these proceedings ; f. gelis , k. kajantie and t. lappi , arxiv : hep - ph/0508229 . b. cole , these proceedings . e. v. shuryak , these proceedings ; j. liao and e. v. shuryak , arxiv : hep - ph/0508035 . e. v. shuryak and i. zahed , phys . d * 70 * , 054507 ( 2004 ) . v. koch , these proceedings ; v. koch , a. majumder and j. randrup , arxiv : nucl - th/0505052 . f. karsch , these proceedings . c. r. hu and b. mller , phys . b * 409 * , 377 ( 1997 ) ; g. d. moore , c. r. hu and b. mller , phys . d * 58 * , 045001 ( 1998 ) . u. w. heinz , phys . lett . * 51 * , 351 ( 1983 ) . p. f. kelly , q. liu , c. lucchesi and c. manuel , phys . d * 50 * , 4209 ( 1994 ) . s. a. bass , b. mller and w. pschl , j. phys . g * 25 * , l109 ( 1999 ) . a. starinets , these proceedings ; p. kovtun , d. t. son and a. o. starinets , phys . * 94 * , 111601 ( 2005 ) c. gale , these proceedings . n. borghini , these proceedings ; n. borghini and u. a. wiedemann , arxiv : hep - ph/0506218 . a. majumder , these proceedings ; a. majumder and x. n. wang , phys . d * 72 * , 034007 ( 2005 ) . c. salgado , these proceedings . n. armesto , c. a. salgado and u. a. wiedemann , phys . lett . * 93 * , 242301 ( 2004 ) . t. renk and j. ruppert , phys . c * 72 * , 044901 ( 2005 ) . b. mller , phys . c * 67 * , 061901 ( 2003 ) . x. n. wang , these proceedings . a. dainese , c. loizides and g. paic , eur . j. c * 38 * , 461 ( 2005 ) ; k. j. eskola , h. honkanen , c. a. salgado and u. a. wiedemann , nucl . phys . a * 747 * , 511 ( 2005 ) ; r. baier , nucl . a * 715 * , 209 ( 2003 ) . i. vitev , j. phys . g * 30 * , s791 ( 2004 ) . a. kovner and u. a. wiedemann , phys . d * 64 * , 114002 ( 2001 ) . m. nardi , these proceedings t. csrg and l. p. csernai , phys . b * 333 * , 494 ( 1994 ) . j. rafelski and j. letessier , phys . lett . * 85 * , 4695 ( 2000 ) . u. w. heinz and k. s. lee , phys . b * 259 * , 162 ( 1991 ) . s. salur , these proceedings . s. a. bass , these proceedings ; r. j. fries , s. a. bass and b. mller , phys . lett . * 94 * , 122301 ( 2005 ) . r. hwa , these proceedings ; c. b. chiu and r. c. hwa , phys . c * 72 * , 034903 ( 2005 ) . r. lednicky , these proceedings . m. gyulassy and l. mclerran , nucl . a * 750 * , 30 ( 2005 ) . e. braaten and r. d. pisarski , nucl . b * 337 * , 569 ( 1990 ) ; phys . d * 45 * , 1827 ( 1992 ) . k. kajantie , m. laine , k. rummukainen and m. e. shaposhnikov , nucl . b * 458 * , 90 ( 1996 ) . e. braaten and a. nieto , phys . d * 53 * , 3421 ( 1996 ) . m. laine and y. schrder , jhep * 0503 * , 067 ( 2005 ) . see , e. g. : i. r. klebanov , arxiv : hep - ph/0509087 . e. braaten and r. d. pisarski , phys . d * 42 * , 2156 ( 1990 ) . [ note that in this article @xmath79 denotes half the decay width of a quasiparticle . ]

we analyze the time profiles of individual gamma - ray burst ( grb ) pulses , that are longer than 2 s , by modelling them with analytical functions that are based on physical first principles and well - established empirical descriptions of grb spectral evolution . these analytical profiles are independent of the emission mechanism and can be used to model both the rise and decay profiles allowing for the study of the entire pulse light - curve . using this method , we have studied a sample of 77 individual grb pulses , allowing us to examine the fluence , pulse width , asymmetry , and rise and decay power - law distributions . we find that the rise phase is best modelled with a power law of average index and that the average decay phase has an index of . we also find that the ratio between the rise and decay times ( the pulse asymmetry ) exhibited by the grb pulse shape has an average value of 0.47 which varies little from pulse to pulse and is independent of pulse duration or intensity . although this asymmetry is largely uncorrelated to other pulse properties , a statistically significant trend is observed between the pulse asymmetry and the decay power law index , possibly hinting at the underlying physics . we compare these parameters with those predicted to occur if individual pulse shapes are created purely by relativistic curvature effects in the context of the fireball model , a process that makes specific predictions about the shape of grb pulses . the decay index distribution obtained from our sample shows that the average grb pulse fades faster than the value predicted by curvature effects , with only 39 of our sample being consistent with the curvature model . we discuss several refinements of the relativistic curvature scenario that could naturally account for these observed deviations , such as symmetry breaking and varying relative time - scales within individual pulses .

astro-ph0303556

the temporal structure of gamma - ray bursts ( grbs ) varies drastically , with no apparent pattern among bursts . of the @xmath32704 grbs detected by the batse instrument onboard the compton gamma ray observatory ( cgro ) @xcite , less than 10 @xmath2 of the detected light - curve profiles can be categorized as being similar in overall shape or morphology , in that they comprise coherent structures , or pulses of radiation . it is generally believed that these pulses represent the fundamental constituent of grb time profiles ( light curves ) , and appear as asymmetric pulses with a fast rise and a slower decay , often denoted fred `` fast rise and exponential decay '' ( see figure 1 ) . each pulse is assumed to be associated with a separate emission episode , with complex grbs being superpositions of several such episodes , of varying amplitude and intensity . although the duration and amplitude of the fred pulses vary considerably , the shape is the only recurring pattern that can be distinguished among the vast range of complex grb light curves that have been observed . the observed @xmath4-ray pulses are believed to be produced in a highly relativistic outflow , an expanding and collimated fireball , based on the large energies and the short time scales involved . there are several possible mechanisms underlying the emission , although the commonly assumed scenario is that individual pulses are created , when shocks internal to the relativistic outflow , drain the kinetic energy and accelerate leptons which radiate . in this paper we base our discussion on this standard , fireball model , in which the @xmath4-rays arise from the internal shocks at a distance of @xmath5 cm from the initial source @xcite . in the context of the fireball model , the episodic nature of the outflow causes inhomogeneities in the wind to collide and thus create the shocks . the dominant emission mechanisms are most probably non - thermal synchrotron @xcite ) and/or inverse compton emission @xcite , but there have been other suggestions , for instance , thermal , saturated comptonization @xcite . the simplest scenario here is to assume an impulsive heating of the leptons and a subsequent cooling and emission . therefore , the rise phase of the pulse is attributed to the energization of the shell and the decay phase reflects the cooling of the energized particles . as many authors have pointed out , this cooling interpretation does not work well when applied to non - thermal synchrotron and/or inverse compton emission because the radiative cooling timescales alone are typically much too short to explain the pulse durations . for example , if we interpret the average break energy of a grb spectrum ( @xmath6 ) as the characteristic energy of synchrotron self - absorption , then the resulting magnetic field must be extremely high , about @xmath7 to @xmath8 gauss . similar estimates are obtained if we assume equipartition conditions between the lepton energy and the magnetic field . in either case , such a high @xmath9 field would create a synchrotron cooling timescale of the order of @xmath10 seconds in the comoving frame @xcite . one resolution to this problem is to introduce relativistic effects . if the relativistic fireball expands with a lorentz factor of @xmath11 @xmath3 100 , then the geometry of the shell would make radiation emitted off the line of sight delayed and affected by a varying doppler boost ( see figure 2 ) . this would cause the grb spectra to evolve to lower energies and produce decay profiles much longer than the microscopic cooling timescale . these relativistic curvature effects would produce a signature decay profile that can be obtained analytically and searched for in a sample of grb pulses . thus it is of great importance to characterize the individual pulse profiles within grbs and study their parameter distributions . several investigations along the lines of pulse modelling have perviously been made . initially , the pulse profiles were modelled by `` stretched exponential '' functions , both for the rise phase and for the decay phase @xcite : @xmath12 where @xmath13 is the time of the maximum flux , @xmath14 , of the pulse , @xmath15 are the time constants for the rise and the decay phases , respectively , and @xmath16 is the peakedness parameter , equation ( [ stretch ] ) is , strictly speaking , a compressed exponential . ] . such a function is very flexible which makes it possible to describe the whole shape of most pulses , and to quantify the characteristics of the pulses for a statistical analysis , most notably their location , amplitude , width , rise phase , transition phase , and decay phase . @xcite studied a sample of bursts observed by batse and found that the decay generally lies between a pure exponential ( @xmath17 ) and a gaussian ( @xmath18 ) . @xcite studied approximately 2500 pulse structures , in individual energy channels , using the high time resolution batse tts data type . they confirmed the general behavior , namely , that pulses tend to have shorter rise times than decay times . the stretched exponential was introduced because of its very flexible nature , although recently @xcite have proposed that the decay phase would be better described by a power law , which in terms of energy flux is : @xmath19 see [ sec : decay ] for the derivation of equation ( [ fe ] ) and definition of @xmath20 . the motivation for this type of shape is entirely based on empirical relations describing the evolution of the grb spectra during the decay phase of individual pulses . @xcite have recently shown that the form expressed in equation ( [ fe ] ) can be produced through simple relativistic kinematics when applied to a spherical shell expanding at extreme relativistic velocity . the curvature of a relativistic fireball would make the photons emitted off the line of sight ( los ) delayed and affected by a varying doppler boost due to the increasing angle at which the photons were emitted with respect to the observer ( see figure 2 for an illustration ) . they show that this doppler boosting of off - axis photons can reproduce the two well known empirical correlations observed in the grb spectra , namely the hardness - intensity and hardness - fluence correlations ( hic and hfc , respectively ) , in a manner that is highly predictive of the resulting pulse profile . following @xcite , we can derive the expected emission profile from a spherical shell that radiates for an infinitesimal period of time at a peak energy @xmath21 in the comoving frame . the lorentz boosting factor for transformations from the comoving frame to the observers frame of photons emitted from different locations on the surface of a spherical shell is given by @xmath22 where the angle @xmath23 shown in figure 2 . if we define the point where the flow velocity is parallel to the line of sight ( los ) to be @xmath24 , then the difference in light travel time between photons emitted along the los and photons emitted at an angle @xmath25 is given by @xmath26 , which gives @xmath27 . substituting this into the lorentz boost factor , we find @xmath28 for extremely relativistic outflows @xmath29 , and the boost factor becomes @xmath30 the outcome of such a doppler profile is that if the emitted spectra from different parts of the shell are identical , then the observed spectra will be gradually redshifted in time by a factor of @xmath31 as photons from different parts of the shell are received by the observer . the peak energy of the spectra will be observed to evolve as @xmath32 where @xmath33 is the peak energy in the comoving frame . by similar arguments , @xcite show that the resulting bolometric energy flux should evolve as @xmath34 therefore , the decay of the resulting light curve , produced purely by relativistic effects , should exhibit a distinct profile , given by equation ( [ ft_rel ] ) . this result becomes more complicated when one considers the three primary timescales relevant for the shape of the pulse : the shell crossing time , the angular spreading time , and the cooling time . the first two timescales are comparable and dominate over the third and compete with each other to produce the observed results . the authors find that for the case of `` pure '' spherical curvature , where the curvature timescale dominates over the intrinsic light - curve profile ( the shell crossing time ) , the convolved light curves indeed reach the behavior of equation ( [ ft_rel ] ) fairly quickly . however , if the crossing timescale and the curvature timescale are comparable , then the resulting light curve will have some convolved shape and the index value of 2 will only be manifested at late times in the pulse decay . the @xcite treatment assumes that the variations in the low- and high - energy power law indices ( @xmath35 and @xmath36 ) that are commonly used to describe the grb spectra tend to be small over the length of the pulse , which may not necessarily be the case . in order to account for this , @xcite and @xcite used a photon number spectrum which they described by a single power law index equal to the average value of @xmath37 . they examined the resulting light curve and spectral evolution that would be expected if the fred profile were due primarily to relativistic effects . they conclude that the decay phase of fred profiles should be independent of intrinsic spectral variations and should scale roughly as @xmath38 . therefore , one of the primary goals of this paper is to model the fred light curves using an analytic function derived from physical first principles and relations describing the spectral evolution of grbs to obtain parameters that uniquely quantify the shape of smooth pulses and compare those values to the predicted signatures of relativistic kinematics . preliminary results of this analysis have been presented in @xcite and @xcite , where we derived analytic models and tested for a small number of pulses . this paper expands on these previous results to include a much larger sample and a more detailed analysis of the distribution of pulse attributes . in @xmath39 , we review the empirical hardness - intensity and hardness - fluence correlations which are crucial to the derivation of an analytic pulse model . in @xmath40 , we derive an analytic function based on the above - mentioned spectral correlations . in @xmath41 , we define a sample of fred pulses observed by cgro to which we fit our function to obtain a distribution of model parameters , most notably the decay power law index and pulse asymmetries . in @xmath42 , we discuss our measured parameter distributions and compare them to the values predicted by relativistic geometry . we discuss several mechanisms that can skew the distributions away from the expected values . gamma - ray burst continuum spectra have a well - known evolution as the burst proceeds , both over the entire , often complex light curves , and over individual pulses . the latter evolution is often more spectacular . this evolution is generally characterized by an overall softening of the spectra to lower energies with time . two specific correlations between observable quantities have been found that describe this softening in a qualitative manner . the first is the _ hardness - intensity correlation _ or hic , which relates the instantaneous hardness of the spectra and the instantaneous energy flux @xmath43 , within individual pulses . for the decay phase of a pulse the most common behavior of the hic is @xmath44 where @xmath45 and @xmath14 are the initial values of the peak of the @xmath46 spectrum and the energy flux at the beginning of the decay phase of each pulse , respectively . @xmath47 is the power law index for the power law index , which unfortunately is also a notation for the lorentz factor ] . this is a general form for non - thermal energy flux spectra that is largely independent of the emission mechanism which gives @xmath48 an implicit time dependence because the peak energy evolves with time . the original study by @xcite found the power law index to vary between @xmath49 over the whole grb . moreover , @xcite studied a sample of 82 grb pulse decays and found them to be consistent with a power law hic in , at least , 57% of the cases and for these found @xmath50 . several physical emission processes are described by such a relation , with the most prominent example being single particle synchrotron emission with a constant magnetic field and a variable electron lorentz factor @xmath11 . in this particular case , @xmath51 while @xmath52 , so the hic index is @xmath53 . the second correlation is the _ hardness - fluence correlation _ or hfc @xcite which describes the observation that the instantaneous hardness of the spectra decays exponentially as a function of the time - integrated flux , or fluence , of the burst : @xmath54 where @xmath55 is the photon fluence integrated from the start of the burst and @xmath56 is the exponential decay constant . @xcite found the hfc to be valid for 35 of 37 smooth grb pulses . these findings were later confirmed by @xcite who studied a sample of 41 pulses in 26 grbs and found that the spectral evolution in the majority of the pulses could be described by the hfc . the physical interpretation of the hfc is not immediately clear , but upon differentiation of equation ( [ equ : lk96 ] ) , it can be seen that it simply states that the rate of change in the hardness is proportional to the luminosity of the radiating medium ( or , equivalently , to the energy density ) ; @xmath57 here the last equivalency is only approximately true . the energy flux is actually defined as the integral of the photon energy times the photon spectrum , @xmath58 , where in this case the integral would be carried over the batse energy band . we make the assumption that if @xmath59 falls well within that band , then the energy flux is approximately @xmath59 times the instantaneous photon flux , @xmath60 . since @xmath59 is defined to be the peak of the @xmath46 spectrum , it represents the energy of most of the arriving photons and therefore this equivalency is robust as long as @xmath59 does not evolve outside of the batse window . we also make the assumption that the evolution of the low - energy power law index @xmath35 should not effect our results except for pulses with an extreme amount of evolution . this assumption is supported by the findings of @xcite and @xcite who avoided making the assumption that @xmath60 by directly using photon flux to test the hfc and hic relations . using independent methods , they show the approximation to be valid for a large sample of grbs . liang and kargatis originally proposed that the hfc could be the result of a confined plasma with a fixed number of particles cooling via @xmath4-radiation . this type of exponential decay of the break energy with photon fluence that is seen in the hfc is expected if the average energy of the emitted photons is directly proportional to the average emitting particle energy such as in thermal bremsstrahlung or multiple compton scattering , although this interpretation is not unique . we now follow @xcite to find an analytical description of the energy - flux decay - profile . note that in their original description they used the photon flux , and we will here reformulate their results in terms of the energy flux instead , which makes a physical interpretation easier . combining the empirical relations , the hic and hfc , gives the following differential equation governing the spectral evolution @xmath61 the solution gives an expression for the energy flux during the decay phase of a pulse @xmath62 and correspondingly the peak of the * @xmath46 * spectra follow @xmath63 where @xmath64 . note that this gives the possibility to measure @xmath65 directly from the light curve with the knowledge of @xmath45 . introducing @xmath66 ( @xmath20 as in the asymptotic _ d_ecay of the energy flux ) we can describe the @xmath67 and the @xmath68 decays as @xmath69 @xmath70 for the @xmath71 case . note that this representation is completely model - independent and is based solely on the empirical hic and hfc relations that describe grb s spectral evolution . as will be shown in the following sections , specific models , such as relativistic kinematics , make specific predictions regarding the value of the hic power - law index and hence the value of decay constant @xmath20 . the analytic profile derived above describes only the decay phase of a pulse , since that is the portion of the grb time history for which the hic is often a power law ( eq . [ 3 ] ) and the hfc follows equation ( 4 ) . this introduces an ambiguity when fitting equations ( [ e7 ] ) and ( [ e8 ] ) to the data , since the power law index @xmath20 and the time constant @xmath72 are coupled and are therefore not well constrained by fitting . furthermore , choosing the moment when the pulse changes from the rise to the decay phase is somewhat subjective . in order to eliminate these ambiguities , a rise phase must be supplemented to equations ( 9 ) and ( 10 ) in order to produce a complete description of the fred profile . this also allows for additional parameters such as the pulse rise time and asymmetry values to be measured . we first note the fact that in most physical models both the peak of the energy spectrum and the luminosity are proportional to the random lorentz factor of the shocked electrons to some power @xcite : @xmath73 and @xmath74 the functions @xmath75 and @xmath76 parameterize the unknown time dependencies on particle densities , optical depth , magnetic field and kinematics , etc . the correlation between hardness and the energy flux can thus be described as @xmath77 with @xmath78 and @xmath79 . during the decay phase of a pulse the power law relation dominates the hic , which has been elucidated in previous studies . however , the hic during the rise phase will be dominated by the unknown function @xmath80 . we will now assume two simple representations of @xmath80 and recapitulate the analytical discussion in the previous section to find a description for the shape of the entire pulse . the simplest assumption one can make is that the rise phase is dominated by a power - law increase of flux . this leads to a simple prescription of the function @xmath81 where the power - law index , @xmath82 , stands for the _ r_ise . in the case of optically - thin synchrotron emission the energy flux is given by @xmath83 , so the hic exponent in equation ( [ hic ] ) is unity , @xmath84 is the thompson depth , and @xmath9 is the magnetic field . in this case the explicit time dependence arises from the evolution of the thompson depth and the magnetic field strength . our initial assumption gives rise to a general form that should also emerge when considering other non - thermal emission mechanisms . to fully model the fred light curve we have to define the time at which the light curve peaks as @xmath85 or @xmath86 for brevity this allows us to define several new parameters , @xmath87 and @xmath88 and @xmath89 and that @xmath90 . the hic can then be written as : @xmath91 which combined with the hfc now gives the differential equation @xmath92 now , we define the exponent @xmath20 ( for finite @xmath82 ) to describe the asymptotic power law behavior of the light curve , @xmath93 as @xmath94 which gives @xmath95 . we can now better define the peak of the pulse to be at @xmath96 by letting @xmath97 and get the condition that for @xmath98,@xmath99 $ ] . therefore , for a simple power - law rise , the final pulse shape and @xmath59 evolution can be described by : @xmath100 @xmath101 this four parameter model fully describes a single pulsed grb event with direct measurements of an observed light curve yielding the values for @xmath102 . this then gives @xmath103 , which in its turn , gives the product @xmath104 . as @xmath105 can be measured from the @xmath106 decay , @xmath65 can be deduced . example profiles with varying decay parameters for the power - law rise - model are shown in figure 3 . it must be noted that in this scenario ( rise+decay ) , the hic power law index , @xmath107 , will correspond to both the implicit power law behavior , @xmath108 as well as the explicit behavior , @xmath109 . here @xmath107 denotes the power law in the second description . the two descriptions coincide as @xmath82 tends to @xmath110 and as @xmath111 which gives @xmath112 and thus @xmath113 by identification . note that @xmath114 tends to @xmath110 as @xmath82 approaches @xmath110 . it is also readily seen that the asymptotic behaviors as @xmath115 approaches @xmath116 are the same . it is reasonable that the rise phase is connected to some transient process , e.g. , an initial decrease in optical depth , increase in the number of energized particles , or the merging ( or crossing ) of two shells . after this initial phase , the original decay behavior as described by equations ( [ e7 ] ) and ( [ e8 ] ) should emerge . we therefore try the following prescription @xmath117 where the time constant @xmath118 now represents the _ r_ise phase . this corresponds to @xmath119 as @xmath120 . we therefore have @xmath121 which gives the differential equation @xmath122 in the same manner as before we define the decay index @xmath20 by requiring @xmath93 as @xmath123 . this gives @xmath124 . solving the differential equation ( [ equ : diff2 ] ) and combining with equation ( [ equ : f2 ] ) we find @xmath125 @xmath126 where @xmath127 and @xmath128 are analytical functions of @xmath129 and @xmath114 . there is no analytical expression for the peak time @xmath114 , but it can be solved for numerically if the other parameters are known . equation ( [ f2 ] ) describes the pulse by four parameters : @xmath130 and @xmath20 . this description coincides with the decay function ( eqs . [ [ e7 ] ] and [ [ e8 ] ] ) when @xmath131 or @xmath132 with @xmath133 ( and @xmath134 ) . note that in this description there are two explicit time constants @xmath118 and @xmath135 , while in equation ( [ f1 ] ) the decay constant is an expression of @xmath82 and @xmath114 . for a sharp rise phase the explicit time dependence will disappear . this is , however , not the case for equation ( [ f1 ] ) . for the purposes of this study , we utilize 64ms count data provided by the batse instrument onboard the _ cgro _ spacecraft . these data were gathered by batse s large area detectors ( lads ) which provide discriminator rates with 64 ms resolution from 2.048 s before the burst to several minutes after the trigger @xcite . the discriminator rates are gathered in four broad energy channels covering approximately 25 - 50 , 50 - 100 , 100 - 300 , and 300 to about 1800 kev allowing for excellent count statistics since the photons are collected over a wide energy band . for our analysis , we combine the data from all four channels in order to study the `` bolometric '' light curve profile . we chose to study the count flux light curves rather then directly using the energy flux because of the spectral fitting that is trespassory to obtain energy flux measurements . although batse s spectroscopic detectors ( sds ) can gather enough data to produce spectra on a 0.128 s timescale , the data usually has to be integrated to get an acceptable fit to a given spectra model . this reduces the resolution of the energy flux time history to a level that is unacceptable for a four or five parameter fit . furthermore , @xcite have shown that , empirically , there is a power - law relation between the count flux and the energy flux , with an index clustering around @xmath136 , which supports the use of the count light - curves for our study of the pulse shapes . the background is , in general , fitted to a quadratic polynomial using 1.024 s resolution background data that are available from 10 minutes before the trigger to several minutes after the burst . the data along with the background fit coefficients were obtained from the cgro science support center ( grossc ) at goddard space flight center through its public archives . we selected bursts from the entire batse catalog with the criteria that the peak flux be greater than 1.0 photons @xmath137 s@xmath138 on a 256 ms timescale . of these , we selected bursts that exhibited clean , single - peaked events , or in the case of multi - peaked bursts , pulses that were well distinguished and separable from each other . we limited the bursts to events with durations longer than 2 s ( full width half max - fwhm ) , primarily because this study focuses on the properties of long grb events . there are strong indications that short bursts with @xmath139 s have different temporal behaviors compared to long bursts and that they may actually constitute a different class of grbs @xcite . therefore , in this study we are primarily interested in the shape and temporal properties of long grb events . in the search we did not assume any functional description of the pulse profile , i.e. pulses that did not look like freds were also selected , in order to eliminate any preconceived idea of what the fundamental pulse shape should be . we note that we do not know what the clean , `` generic '' pulse actually looks like , and some noise , or substructure , in the comoving light curve can be expected , for instance , if the surface brightness of the shell is non - homogeneous and/or if the density profile of the shell , through which the shock is traversing , has a complicated radial profile . therefore , we adopt the following criteria . if a significant substructure on top of a smoothly fitted pulse is within a time interval of 1/4 the fwhm of the main pulse peak and it contains more than 15 @xmath2 of the total flux in the main pulse , the pulse structure is rejected . if the substructure occurs in a time interval of more than twice the fwhm of the primary pulse , then the pulse is treated as an independent event and is fitted separately . we believe that this limits the possible contamination of the sample with pulses that are actually a composite of overlapping emission episodes . the final sample consisted of 76 pulses within 68 bursts . this sample differs from those examined in previous studies such as @xcite because they primarily represent separable pulses that exhibit long decay tails that can be followed to background . therefore , no attempt was made to perform deconvolution fits of overlapping pulses , for the purpose of obtaining robust measurements of the asymptotic decay of each pulse , which is not possible for overlapping regions of activity . for more details on the sample selection , see f. ryde @xmath140 d. kocevski ( in prep . ) . these bursts are presented in table 1 , and are denoted by both their grb names and batse trigger numbers . the pulses within multi - peaked bursts are distinguished by their @xmath141 value . all of the background - subtracted light - curves were fitted with the power - law rise- ( eq . [ f1 ] ) and the exponential rise- ( eq . [ f2 ] ) functions via @xmath142 minimization . to aid in this , we developed a graphical idl fitting routine which allowed the user to set the initial pulse attributes manually before allowing the fitting routine to converge on the best fit model . the convergence routine utilized the marquardt technique to minimize the difference between the model and the time profile by adjusting the model parameters and tracking the resulting effects on the @xmath142 . in the case of the power - law rise - function , the free parameters were : peak flux @xmath143 , position of peak flux @xmath141 , the rise index @xmath82 , and the decay index @xmath20 . most of the grbs had significant emission prior to the batse trigger time , so we introduced a fifth parameter @xmath144 which measures the offset between the start of the pulse and the trigger time . the initial value of @xmath144 was typically set to where the light curve would rise above 10@xmath2 of the background , helping to eliminate the ambiguity associated with the start of the pulse . in the case of the exponential rise - function , the free parameters were : pulse amplitude @xmath145 , rise time constant @xmath146 , decay time constant @xmath147 , the asymptotic decay power law index d , and the trigger offset @xmath144 . presented in figure 4 are four type - fits to the power - law rise - function ( eq . [ f1 ] ) . these pulses represent a good example of the clean separable fred pulses chosen for this analysis and the degree of acceptability of @xmath142 for our fits . overall the power - law rise - function appears to give a better fit to most of the pulses compared to the exponential rise - function ( eq . [ f2 ] ) . this can be seen in the spread of the respective @xmath148 distribution for our entire sample shown in figure 5 , with the power - law rise - model giving a mean @xmath142 of 1.16 with a standard deviation of 0.19 . this is most likely due to the fact that the power - law function allowed for the modelling of both concave and convex rise profiles , whereas the nature of the exponential function is limited to modelling concave shapes . this is a major constraint for the latter model since many grb pulses that are arguably single emission episodes are seen to have concave rise profiles , at least in the initial stage . the narrow distribution of @xmath142 values also indicates that a power law is sufficient to model the decay portion of the pulse light curve . according to @xcite , this would not be the case if evolution of the low energy spectral index @xmath35 greatly affected the resulting light curve . as discussed in @xmath149 , they show that the decay phase should scale roughly as the power of the low energy spectral index @xmath38 . if @xmath35 exhibited a large range of evolution during the decay phase , then a single power law would no longer be adequate to model the temporal behavior of the light curve . the @xmath142 distributions support the assumption that the evolution of the low energy spectral index does not significantly alter our results . the primary pulse attributes that can be derived from fitting the time profiles are related to the pulse rise and decay rates . in the case of the power - law rise - function , these consist of the asymptotic rise and decay power - law indices @xmath82 and @xmath20 described in @xmath1502.3 . the resulting @xmath82 and @xmath20 parameter distributions for our entire sample are shown in figure 6 . the rise index appears to be the tighter of the two , with a mean value of @xmath151 and a standard deviation of 0.63 . the distribution of the decay index is somewhat broader with a mean value of @xmath152 with a standard deviation of 0.76 . the exponential - rise function has no corresponding rise index , but rather the rise timescale is associated with the time constant @xmath146 in equation ( [ f2 ] ) . the fit results show that the rise timescale covers a very broad range of values with a median value of @xmath153 and a standard deviation of 2.50 . the asymptotic decay index is similar to that found with the power law rise model except that it has a much broader distribution with a median value of @xmath154 with a standard deviation of 1.50 . a measure of the average pulse time - asymmetry can be easily constructed using the fit parameters obtained from both models . several authors have studied burst asymmetry properties in the past @xcite - and found interesting correlations between the pulse asymmetries and other temporal properties . @xcite measured asymmetry values for some 400 fitted pulses and reported a general trend that narrower pulses tend to be more symmetric and have harder spectra . this is often referred to as the grb pulse paradigm . here we define the asymmetry as the fraction of the fwhm belonging to the rise phase of the pulse to the fraction belonging to the decay phase ( see figure 7 for an illustration ) . this description is independent of the model parameters used to fit the pulses and hence allows the asymmetry to be measured for both of the functions utilized in this analysis in a comparative manner . the resulting asymmetry distribution for the power - law rise - model is remarkably tight with a median value of 0.47 @xmath155 0.08 and a standard deviation of 0.09 . the exponential rise model provides similar results yielding a median value of 0.45 @xmath155 0.07 and a standard deviation of 0.07 . alternatively , the asymmetry can be defined using the @xmath156 of the pulse instead of the fwhm . the rise phase would then be defined as the interval from when where 5@xmath2 of the pulse counts are accumulated till the time of peak flux as measured by the model parameter @xmath141 . similarly , the decay phase is simply the time between the peak flux till when 95@xmath2 of the pulse counts are collected . note that here we are using @xmath156 for the individual @xmath157 , not the entire burst which is the common practice . although this description is straightforward , it is somewhat more problematic when dealing with multi - pulsed bursts because subsequent pulses will routinely begin before the signal from the first pulse falls to within 5@xmath2 of the background . it is presumed that the decay phase of the pulse is hidden under the rise phase of the following pulse , therefore in these cases the decay time is only a lower limit to the true length of the decay phase of the pulse . this occurred in 10@xmath2 of the pulses observed , so the effect on the overall distribution should be minimal . the resulting median asymmetry values using the @xmath156 definition for both the power law and exponential models are 0.30 @xmath155 0.04 and 0.21 @xmath155 0.03 respectively . the resulting asymmetry distributions for the power - law rise - function for the fwhm definition is plotted in figure 8 . surprisingly , we find no evidence of a connection between the pulse asymmetry and the pulse duration ( @xmath156 ) for the individual fred pulses . likewise , we find no significant correlations between the pulse asymmetry and the pulse width ( fwhm ) . the lack of any significant correlation can be seen in figure 9 and measured quantitatively by the use of the spearman rank - order statistic which provides a robust and convenient means of evaluating the statistical significance of a given correlation . a spearman rank order analysis on the data shown in figure 9 yields a correlation coefficient of @xmath158 , where @xmath159 represents a 1:1 correspondence . both of these conclusions are contrary to the pulse paradigm referred to earlier and the disagreement is most likely due to the difference in samples used in the two studies . the @xcite study examined complex overlapping bursts for a majority of their data sample , while the majority of our sample consist of single - peaked fred bursts . this leads to the conclusion that most of the simple fred pulses that were examined in our sample are extremely self - similar , independent of their duration and amplitude . we now examine several correlations between temporal properties measured from the pulse profiles such as the fwhm , rise and decay times and fluence values . first , figure 10 shows the pulse rise time versus the pulse width , as measured by the fwhm , for our entire sample . the resulting linear correlation has a slope of 0.323 @xmath155 0.01 with an associated spearman rank order correlation coefficient of @xmath160 and is a direct consequence of the tight asymmetry distribution discussed above . this same effect is also observed between the time constant @xmath146 derived from the exponential rise model and the pulse width . here again , the rise time constant , which is an alternative way of measuring the characteristic rise timescale , is linearly correlated to the fwhm . note that the fwhm values plotted in figure 10 are not normalized to any timescale such as the pulse duration , reasserting that the rise time is a constant fraction of the overall pulse width independent of the duration of the pulse . another correlation born from the data is the relationship between the normalized rise time and the decay power - law index , which is shown in figure 11 and has a correlation coefficient of @xmath161 . here the rise time is normalized to the fwhm of the pulse , so the longer the rise time is , relative to the fwhm , the steeper the resulting power - law decay of the pulse envelope becomes . this is very interesting since the rise time is thought to be proportional to the dynamic or shell - crossing time of the pulse and is generally believed to be independent of the subsequent fall of the fred profile . a similar effect can be seen in figure 12 where the pulse asymmetry data are plotted versus the asymptotic decay power - law index . the degree of asymmetry is seen to be proportional ( @xmath162 ) to the decay index in such a way that pulses with steep decay profiles ( large @xmath20 ) have a larger rise - to - decay ratio . the latter correlation is to be expected since , as @xmath20 tends to 0 , the decay time approaches @xmath116 and hence the asymmetry tends to 0 . a similar correlation is not seen with the rise power - law index . the fit results presented above indicate that the analytic functions derived in 3 are very successful in describing the overall profile of individual grb pulses . these descriptions differ from those used in previous modelling surveys as they are motivated by physical first principles within the general fireball model with the requirement that the previously observed empirical relations for the spectral evolution must hold . these descriptions also build on the original @xcite analysis by including a description of the rise phase of the pulse . of the two assumed rise profiles presented in 2 , a simple , power - law rise seems to better describe the majority of the fred pulses examined . this is one of the simplest assumptions that can be made regarding the dynamic ( or shell crossing ) phase of a grb pulse and may point to the explicit time - dependance of the parameters internal to the shock such as thompson depth or the magnetic field . hence , the measured pulse parameters have the potential to be used to diagnose pulse characteristics , such as bulk lorentz factors , @xmath11 , shell radii , and thickness . the decay power - law index appears to cover a wide range of values with a median centered around 2.39 @xmath155 0.12 , which is 3.25 @xmath163 steeper than the analytically derived value of @xmath164 for pulse profiles created principally by relativistic effects due to spherical symmetric shells . overall , 30 of the 77 pulses that we analyzed had decay indices that were consistent ( within 1@xmath163 ) with the predicted @xmath164 value , constituting 39@xmath2 of our sample . we note that the distribution of hic indices , which were shown in @xmath40 to be directly related to the decay index , found by @xcite also has a broad distribution , with several cases differing substantially from @xmath165 . there are several mechanisms that can cause the decay envelope to deviate from the predicted spherical scenario , but the most obvious would be the breaking of local spherical symmetry . both prolate and oblate shell geometries as seen by the observer would create light curves with resulting power - law indices that differ from the spherical case . in @xmath39 we derived the shape that the light curve is expected to take if the pulse profile were created purely by a spherical shell . in the case of an elliptical geometry , the time delay between the arrival of off axis photons will be given by @xmath166 , where @xmath167 is the radius of the shell at @xmath24 . the radius @xmath168 of the shell is no longer independent of the angle @xmath25 , and is given by @xmath169 where e is the eccentricity of the shell , a is the semi - major axis , and @xmath170 . substituting this back into @xmath171 , we find that @xmath172 solving for @xmath173 and substituting into the doppler boost expression ( equation [ doppler ] ) gives the doppler profile ( doppler boost as a function of time ) for elliptical shells @xmath174 following from the discussion in @xmath39 , the resulting evolution of the spectral break - energy and light curve of the energy - flux should follow @xmath175 and @xmath176 , so the profile of the varying doppler boost should directly give rise to the time dependence of the @xmath59 evolution and the pulse light curve . the doppler profile , expressed in equation [ doppler2 ] , is shown for several different prolate and oblate shock fronts in figure 13 , with the corresponding shell geometry plotted as an inset . it shows that the prolate shells ( solid lines ) have steeper doppler profiles resulting in asymptotic power law slopes which are greater than the spherical case ( thick solid line ) whereas the opposite is true for the oblate geometries ( dashed lines ) . the extreme limit of the oblate scenario is such that the shell becomes a parallel slab , in which case for an unresolved source the doppler profile approaches zero and any pulse evolution is directly due to the shell s intrinsic emission profile . because of this effect , the observed range of power - law decay indices may be explained by a simple distribution of shell geometries , where the median decay index of 2.39 @xmath155 0.12 would indicate that , in the context of relativistic curvature , the majority of the analyzed pulses were produced by shells with a degree of curvature greater than that exhibited by a spherical shell . such conditions are not unreasonable in the context of the fireball model and a simple angular dependance of the bulk lorentz factor can easily produce an elliptical shock front that evolves as the shock propagates . the evolution of the shock geometry , from oblate at early times to prolate at later times , and a variation in the time at which the shell `` ignites '' during this evolution , could account for the distribution of decay indices . @xcite and @xcite provide several other arguments in support of the breaking of local spherical symmetry , the most prominent being that the pulse photon light curves appear to evolve uncoupled to the grb spectra . they find that the grb spectra also evolve faster than the predicted rate of evolution due to the simple boost factor that is expected from a spherically symmetric shell . according to the investigation performed by @xcite , this result is consistent with our findings that the average light curve profile also decays faster than predicted . in the context of their model , a range of relative sizes between the comoving and the curvature timescales defined in @xmath39 can result in a variety of pulse shapes . light curves that exhibit a substantially different shape than that predicted by spherical curvature are attributed to a scenario where the comoving timescale dominates within a burst that already has an intrinsically fast decay rate ( @xmath177 ) of the comoving light curve . this , according to their model , must also be manifested in the spectral evolution in a corresponding manner . therefore , an analysis of the spectral evolution rates and @xmath35 values of the pulses that vary the furthest from the predicted scenario of @xmath164 may test how much these values affect the resulting light curve profiles . this research is currently being pursued ( f. ryde , & d. kocevski in prep ) . the asymmetry results are rather surprising and the reason for the tight asymmetry distribution among such a variety of grb pulses is not immediately clear . it would indicate that the dynamical and angular timescales are not fully independent and that the observed rate at which the shell becomes active is dependant on the decay timescale and hence the curvature of the shell . a situation can be envisioned where a short comoving rise profile is delayed and boosted to the same degree as the shell s intrinsic cooling profile . if both of these timescales are extremely short as compared to the angular spreading timescale , then a relationship can be produced between the observed rise and decay profile . this effect could produce a narrow asymmetry distribution and would also predict a correlation between the rise timescale and decay power - law indices , which is indeed observed in our sample . @xcite have discussed extensively the effect of varying intrinsic emission profiles and angular spreading timescales on the resulting light curve that is observed . they find that the resulting pulse asymmetry relies heavily on the convolution of the intrinsic emission profile and the angular spreading timescale . therefore , future modelling of pulse timescales and shell geometries , with the constraint that they produce the observed asymmetry distribution , may yield clues to the intrinsic profile of the emission episode . we are grateful to the grossc at nasa / gsfc for providing the heasarc online service . we should also like to thank dr . vah petrosian for his valuable discussions and insights . d.k . wishes to express his gratitude to the department of physics at stanford university for hospitality and also acknowledges the nasa gsrp fellowship program for their support . f.r . acknowledges financial support from the swedish foundation for international cooperation in research and higher education ( stint ) and the ludovisi boncomagni , ne bildt , foundation . 910721 & 563 & 1.65 & 2.13 @xmath155 0.29 & 24.8 @xmath155 3.97 & 0.77 @xmath155 0.15 & 1.34 @xmath155 0.07 & 1.34 & 1.34 + 911016 & 907 & 1.47 & 3.74 @xmath155 0.34 & 19.7 @xmath155 2.56 & 0.85 @xmath155 0.11 & 1.46 @xmath155 0.05 & 1.29 & 1.29 + 911022 & 914 & 0.55 & 2.64 @xmath155 0.33 & 6.27 @xmath155 0.76 & 1.81 @xmath155 0.10 & 2.33 @xmath155 0.14 & 0.86 & 0.96 + 911031 & 973 & 2.87 & 5.71 @xmath155 0.41 & 17.9 @xmath155 0.99 & 1.23 @xmath155 0.08 & 3.07 @xmath155 0.10 & 1.34 & 1.35 + 911031 & 973 & 23.9 & 3.17 @xmath155 0.40 & 17.9 @xmath155 2.03 & 1.21 @xmath155 0.23 & 2.15 @xmath155 0.09 & 1.27 & 1.33 + 911104 & 999 & 3.94 & 12.4 @xmath155 0.65 & 1.98 @xmath155 0.20 & 2.40 @xmath155 0.33 & 2.73 @xmath155 0.07 & 1.24 & 3.05 + 911209 & 1157 & 4.83 & 12.2 @xmath155 0.55 & 8.58 @xmath155 1.07 & 1.60 @xmath155 0.58 & 1.62 @xmath155 0.10 & 1.30 & 1.58 + 920216 & 1406 & 3.14 & 2.10 @xmath155 0.27 & 25.1 @xmath155 3.18 & 0.95 @xmath155 0.07 & 2.29 @xmath155 0.11 & 1.06 & 1.07 + 920307 & 1467 & 4.32 & 2.59 @xmath155 0.27 & 18.3 @xmath155 2.05 & 2.53 @xmath155 0.08 & 2.62 @xmath155 0.08 & 1.01 & 1.06 + 920801 & 1733 & 3.30 & 3.20 @xmath155 0.32 & 17.3 @xmath155 1.73 & 1.97 @xmath155 0.08 & 2.05 @xmath155 0.06 & 1.17 & 1.29 + 920830 & 1883 & 1.26 & 5.36 @xmath155 0.37 & 8.19 @xmath155 0.51 & 1.48 @xmath155 0.04 & 2.60 @xmath155 0.08 & 1.06 & 1.07 + 920925 & 1956 & 2.81 & 2.80 @xmath155 0.28 & 10.7 @xmath155 1.04 & 1.31 @xmath155 0.38 & 3.31 @xmath155 0.22 & 1.47 & 1.50 + 921015 & 1989 & 116 & 2.96 @xmath155 0.31 & 18.3 @xmath155 1.83 & 2.00 @xmath155 0.18 & 2.40 @xmath155 0.06 & 1.35 & 1.66 + 921207 & 2083 & 8.59 & 46.6 @xmath155 0.92 & 4.00 @xmath155 0.20 & 2.12 @xmath155 0.05 & 1.85 @xmath155 0.02 & 1.25 & 1.71 + 921218 & 2102 & 1.88 & 1.79 @xmath155 0.26 & 10.9 @xmath155 1.51 & 1.90 @xmath155 1.03 & 2.94 @xmath155 0.29 & 0.89 & 0.95 + 930120 & 2138 & 1.50 & 7.08 @xmath155 0.39 & 9.73 @xmath155 1.27 & 2.00 @xmath155 0.06 & 4.00 @xmath155 0.20 & 1.28 & 1.15 + 930120 & 2138 & 48.9 & 35.6 @xmath155 0.59 & 9.73 @xmath155 1.77 & 1.46 @xmath155 0.47 & 2.23 @xmath155 0.29 & 0.96 & 1.00 + 930120 & 2138 & 86.1 & 9.81 @xmath155 0.10 & 15.0 @xmath155 1.18 & 0.42 @xmath155 0.06 & 2.10 @xmath155 0.11 & 1.29 & 1.35 + 930214 & 2193 & 10.9 & 1.78 @xmath155 0.26 & 70.3 @xmath155 10.3 & 0.80 @xmath155 0.01 & 2.50 @xmath155 0.04 & 1.04 & 1.02 + 930612 & 2387 & 6.57 & 4.11 @xmath155 0.33 & 32.8 @xmath155 2.91 & 1.20 @xmath155 0.04 & 2.87 @xmath155 0.06 & 1.07 & 1.17 + 930807 & 2484 & 1.87 & 1.75 @xmath155 0.27 & 13.5 @xmath155 1.87 & 1.26 @xmath155 0.29 & 3.26 @xmath155 0.43 & 1.10 & 1.17 + 930909 & 2519 & 63.5 & 1.68 @xmath155 0.28 & 15.0 @xmath155 3.21 & 1.69 @xmath155 0.16 & 1.43 @xmath155 0.12 & 1.13 & 1.21 + 930914 & 2530 & 114 & 2.03 @xmath155 0.27 & 25.3 @xmath155 3.28 & 1.12 @xmath155 0.05 & 2.34 @xmath155 0.12 & 1.13 & 1.14 + 931127 & 2662 & 1.10 & 1.82 @xmath155 0.29 & 14.9 @xmath155 2.94 & 1.20 @xmath155 0.05 & 1.30 @xmath155 0.03 & 1.07 & 1.04 + 931128 & 2665 & 1.34 & 2.10 @xmath155 0.31 & 12.8 @xmath155 2.25 & 1.74 @xmath155 0.43 & 1.65 @xmath155 0.10 & 1.00 & 1.16 + 931221 & 2700 & 53.8 & 4.17 @xmath155 0.35 & 13.0 @xmath155 1.13 & 0.75 @xmath155 0.09 & 3.01 @xmath155 0.18 & 1.33 & 1.52 + 940313 & 2880 & 0.4 & 3.20 @xmath155 0.29 & 4.16 @xmath155 0.54 & 1.32 @xmath155 0.27 & 2.13 @xmath155 0.15 & 1.51 & 1.61 + 940410 & 2919 & 0.30 & 6.11 @xmath155 0.39 & 10.6 @xmath155 0.80 & 1.55 @xmath155 0.14 & 1.97 @xmath155 0.04 & 1.28 & 1.55 + 940529 & 3003 & 9.56 & 3.04 @xmath155 0.32 & 28.8 @xmath155 3.13 & 1.51 @xmath155 0.04 & 2.62 @xmath155 0.08 & 1.11 & 1.21 + 940830 & 3143 & 0.63 & 2.69 @xmath155 0.29 & 5.5 @xmath155 0.76 & 1.84 @xmath155 0.68 & 2.40 @xmath155 0.14 & 1.04 & 1.16 + 940904 & 3155 & 0.66 & 1.92 @xmath155 0.26 & 3.39 @xmath155 0.46 & 1.24 @xmath155 0.21 & 3.84 @xmath155 0.60 & 0.88 & 1.01 + 941023 & 3256 & 1.25 & 1.89 @xmath155 0.22 & 14.0 @xmath155 2.68 & 0.80 @xmath155 0.02 & 1.50 @xmath155 0.04 & 1.28 & 1.16 + 941026 & 3257 & 3.27 & 3.21 @xmath155 0.27 & 36.8 @xmath155 5.49 & 0.52 @xmath155 0.04 & 1.57 @xmath155 0.07 & 1.02 & 1.08 + 941026 & 3259 & 4.65 & 2.28 @xmath155 0.23 & 25.0 @xmath155 3.28 & 1.55 @xmath155 0.17 & 2.89 @xmath155 0.16 & 0.96 & 1.15 + 941121 & 3290 & 2.91 & 10.9 @xmath155 0.36 & 1.98 @xmath155 0.47 & 1.47 @xmath155 0.25 & 2.04 @xmath155 0.09 & 0.92 & 1.52 + 950211 & 3415 & 0.25 & 10.3 @xmath155 0.40 & 6.91 @xmath155 1.77 & 2.76 @xmath155 2.14 & 1.68 @xmath155 0.15 & 0.99 & 1.22 + 950211 & 3415 & 11.5 & 2.36 @xmath155 0.11 & 3.78 @xmath155 0.34 & 1.26 @xmath155 0.14 & 2.47 @xmath155 0.17 & 1.48 & 1.36 + 950624 & 3648 & 2.65 & 5.91 @xmath155 0.31 & 22.0 @xmath155 3.97 & 0.84 @xmath155 0.44 & 2.67 @xmath155 0.60 & 0.82 & 0.82 + 950624 & 3648 & 24.1 & 36.6 @xmath155 1.22 & 17.0 @xmath155 3.51 & 1.10 @xmath155 0.04 & 2.40 @xmath155 0.14 & 1.03 & 0.97 + 950624 & 3648 & 40.9 & 1243 @xmath155 3.71 & 11.7 @xmath155 1.16 & 2.54 @xmath155 0.07 & 1.57 @xmath155 0.03 & 1.45 & 1.19 + 950818 & 3765 & 66.1 & 25.9 @xmath155 0.54 & 8.83 @xmath155 0.98 & 2.42 @xmath155 0.07 & 1.84 @xmath155 0.02 & 1.50 & 1.43 + 951016 & 3870 & 0.43 & 14.1 @xmath155 0.46 & 5.76 @xmath155 0.52 & 1.38 @xmath155 0.09 & 1.42 @xmath155 0.02 & 1.52 & 1.78 + 951019 & 3875 & 0.21 & 3.02 @xmath155 0.23 & 4.03 @xmath155 0.42 & 0.96 @xmath155 0.11 & 1.73 @xmath155 0.13 & 1.57 & 1.51 + 951030 & 3886 & 0.20 & 2.27 @xmath155 0.21 & 2.82 @xmath155 0.65 & 1.27 @xmath155 0.12 & 2.45 @xmath155 0.27 & 1.00 & 1.55 + 951102 & 3892 & 0.61 & 1.86 @xmath155 0.20 & 5.31 @xmath155 0.51 & 1.26 @xmath155 0.24 & 2.67 @xmath155 0.29 & 0.97 & 1.03 + 951213 & 3954 & 0.70 & 8.45 @xmath155 0.39 & 10.3 @xmath155 0.70 & 4.31 @xmath155 3.52 & 2.54 @xmath155 0.50 & 1.06 & 1.52 + 951228 & 4157 & 7.67 & 2.32 @xmath155 0.24 & 17.7 @xmath155 2.16 & 1.92 @xmath155 0.09 & 2.53 @xmath155 0.12 & 1.21 & 1.36 + 960113 & 4350 & 0.51 & 3.52 @xmath155 0.25 & 6.66 @xmath155 0.83 & 2.41 @xmath155 0.16 & 1.95 @xmath155 0.07 & 1.29 & 1.86 + 960113 & 4350 & 14.3 & 1.65 @xmath155 0.27 & 6.66 @xmath155 0.75 & 1.00 @xmath155 0.02 & 2.00 @xmath155 0.03 & 1.39 & 1.22 + 960113 & 4350 & 34.1 & 8.71 @xmath155 0.45 & 6.66 @xmath155 0.98 & 1.70 @xmath155 0.34 & 2.66 @xmath155 0.23 & 1.30 & 1.47 + 960114 & 4368 & 2.45 & 58.6 @xmath155 0.83 & 10.0 @xmath155 0.75 & 2.87 @xmath155 0.68 & 2.53 @xmath155 0.07 & 1.26 & 2.01 + 960530 & 5478 & 1.98 & 3.10 @xmath155 0.24 & 14.3 @xmath155 1.43 & 0.98 @xmath155 0.08 & 2.54 @xmath155 0.17 & 0.97 & 0.98 + 960530 & 5478 & 261 & 7.69 @xmath155 1.70 & 37.2 @xmath155 6.22 & 1.20 @xmath155 0.02 & 3.00 @xmath155 0.08 & 1.12 & 1.16 + 960613 & 5495 & 0.23 & 2.39 @xmath155 0.23 & 5.44 @xmath155 1.66 & 0.63 @xmath155 0.34 & 1.65 @xmath155 0.16 & 1.16 & 1.14 + 960624 & 5517 & 0.83 & 2.06 @xmath155 0.23 & 7.42 @xmath155 0.79 & 1.72 @xmath155 0.11 & 3.61 @xmath155 0.37 & 1.04 & 1.05 + 960628 & 5523 & 1.03 & 3.73 @xmath155 0.28 & 6.98 @xmath155 0.71 & 1.92 @xmath155 0.07 & 3.17 @xmath155 0.15 & 1.19 & 1.34 + 960715 & 5541 & 1.32 & 1.84 @xmath155 0.23 & 12.6 @xmath155 1.68 & 1.51 @xmath155 0.47 & 2.96 @xmath155 0.39 & 1.13 & 1.16 + 960912 & 5601 & 7.55 & 4.94 @xmath155 0.29 & 9.00 @xmath155 0.74 & 1.30 @xmath155 0.26 & 2.64 @xmath155 0.16 & 1.29 & 1.38 + 970405 & 6159 & 3.20 & 2.16 @xmath155 0.25 & 12.0 @xmath155 1.79 & 1.82 @xmath155 0.13 & 1.30 @xmath155 0.11 & 1.18 & 1.28 + 970815 & 6335 & 98.3 & 4.29 @xmath155 0.29 & 21.3 @xmath155 3.02 & 1.29 @xmath155 0.08 & 3.52 @xmath155 0.40 & 1.10 & 1.17 + 970925 & 6397 & 3.30 & 6.02 @xmath155 0.30 & 20.4 @xmath155 1.44 & 1.25 @xmath155 0.07 & 2.25 @xmath155 0.05 & 1.61 & 1.53 + 971127 & 6504 & 3.12 & 2.79 @xmath155 0.24 & 22.1 @xmath155 3.00 & 1.18 @xmath155 0.12 & 1.78 @xmath155 0.08 & 1.00 & 1.09 + 980301 & 6621 & 32.5 & 6.85 @xmath155 0.34 & 7.36 @xmath155 0.47 & 1.51 @xmath155 0.05 & 1.65 @xmath155 0.03 & 1.21 & 1.59 + 980302 & 6625 & 5.12 & 2.04 @xmath155 0.26 & 31.4 @xmath155 4.19 & 1.32 @xmath155 0.05 & 3.12 @xmath155 0.19 & 1.04 & 1.16 + 980401 & 6672 & 6.85 & 5.60 @xmath155 0.29 & 3.80 @xmath155 0.41 & 1.17 @xmath155 0.07 & 2.34 @xmath155 0.17 & 0.78 & 0.84 + 980718 & 6930 & 31.7 & 5.76 @xmath155 0.36 & 6.34 @xmath155 0.30 & 1.51 @xmath155 0.04 & 3.70 @xmath155 0.16 & 1.15 & 1.10 + 990102 & 7293 & 3.53 & 3.06 @xmath155 0.23 & 25.2 @xmath155 2.92 & 0.96 @xmath155 0.07 & 1.99 @xmath155 0.07 & 1.20 & 1.20 + 990102 & 7295 & 2.14 & 3.50 @xmath155 0.35 & 8.50 @xmath155 1.05 & 1.20 @xmath155 0.42 & 2.08 @xmath155 0.35 & 0.97 & 0.98 + 990316 & 7475 & 8.54 & 3.87 @xmath155 0.29 & 34.4 @xmath155 2.96 & 0.99 @xmath155 0.07 & 4.03 @xmath155 0.19 & 1.12 & 1.05 + 990505 & 7548 & 3.7 & 3.14 @xmath155 0.25 & 10.6 @xmath155 1.02 & 1.92 @xmath155 0.46 & 2.94 @xmath155 0.17 & 0.82 & 1.04 + 990528 & 7588 & 2.76 & 2.32 @xmath155 0.24 & 15.5 @xmath155 1.77 & 1.15 @xmath155 0.13 & 3.11 @xmath155 0.25 & 0.83 & 0.92 + 990707 & 7638 & 1.12 & 1.92 @xmath155 0.22 & 15.0 @xmath155 2.68 & 1.27 @xmath155 0.30 & 1.22 @xmath155 0.06 & 1.30 & 1.33 + 990712 & 7648 & 4.45 & 1.78 @xmath155 0.21 & 24.7 @xmath155 3.45 & 1.22 @xmath155 0.13 & 3.79 @xmath155 0.41 & 1.05 & 1.17 + 990816 & 7711 & 1.86 & 3.97 @xmath155 0.26 & 13.2 @xmath155 1.32 & 0.85 @xmath155 0.04 & 3.26 @xmath155 0.24 & 1.34 & 1.48 + 323 & 8049 & 30.5 & 1.77 @xmath155 0.19 & 36.6 @xmath155 4.23 & 1.13 @xmath155 0.04 & 4.97 @xmath155 0.42 & 1.04 & 1.24 + 519 & 8111 & 4.98 & 4.33 @xmath155 0.28 & 13.8 @xmath155 1.68 & 1.69 @xmath155 0.03 & 1.48 @xmath155 0.02 & 1.36 & 1.27 +

in this paper , we derive key - rate expressions for several different quantum key distribution protocols . our key - rate equations utilize multiple channel statistics , including those gathered from mismatched measurement bases - i.e. , when alice and bob choose incompatible bases . in particular , we will consider a limited - resource form of bb84 , an extended b92 , and a two - way semi - quantum protocol . for the first protocol , we will show it has the same tolerance to noise as the full bb84 ( a result already known , however we provide an alternative , more information theoretic , proof ) . for the last two protocols , we demonstrate that their tolerance to noise is higher than previously thought . along the way , we will also consider an optimal qkd protocol for various quantum channels . finally , all the key - rate expressions which we derive in this paper are applicable to any arbitrary , not necessarily symmetric , quantum channel .

1608.07728

quantum key distribution protocols allow two users , alice ( @xmath0 ) and bob ( @xmath1 ) , to establish a shared secret key secure against an all - powerful adversary eve ( @xmath2 ) who is bounded only by the laws of physics - an end unattainable through classical means alone . several such protocols have been developed since the original bb84 @xcite ( the reader is referred to @xcite for a general survey ) and many of them include rigorous proofs of unconditional security . such a proof of security generally involves determining a bound on the protocol s key - rate ( to be defined shortly , though roughly speaking , it is the ratio of secret key bits to qubits sent ) as a function of the observed noise in the quantum channel . in this paper , we consider several qkd protocols , and derive key - rate expressions based on multiple channel statistics . furthermore , our key - rate bounds will utilize statistics from mismatched measurements ; that is to say , those measurement outcomes where a and b s choice of bases are incompatible - events which are typically discarded by the protocol specification ( there are exceptions as we mention next section ) . in fact , by using these mismatched measurement results , the key - rate bounds we derive demonstrate that many of the protocols we consider here can actually tolerate higher levels of noise than previously thought . thus , the primary contributions of this paper are two - fold . first , we derive a general approach to deriving key - rate expressions , in the asymptotic scenario , for a wide - range of discrete variable qkd protocol utilizing all possible channel statistics , including mismatched measurement results . secondly , by applying this technique to several , very different protocols ( including a limited - resource bb84 @xcite , an extended b92 @xcite , and a two - way semi - quantum protocol @xcite ) , we not only derive new key - rate expressions applicable to arbitrary , possibly asymmetric quantum channels , but , in many cases , our new bounds are substantial improvements over previous work with these protocols . along the way , we will also use our method to investigate optimal qkd protocols for asymmetric channels . we are not the first to consider the use of mismatched measurement outcomes for quantum key distribution . indeed , in the 1990 s barnett et al . @xcite showed that mismatched measurement results may be used to better detect an eavesdropper using an intercept - and - resend attack . in @xcite , mismatched measurement bases were applied to the four - state and six - state bb84 protocols @xcite . this method was shown to improve the key rate ( as determined by the devetak - winter equation in @xcite ) for certain quantum channels , namely the amplitude damping channel and rotation channel . in @xcite , mismatched measurement results were actually used to distill a raw key ( as opposed to being used only for channel tomography ) - a modified bb84 protocol was adopted and this method was shown to improve the key rate for certain channels . in @xcite , a modified four - state , two - basis bb84 was used where the first basis was the standard computational @xmath3 basis ( @xmath4 ) , while the second consisted of states @xmath5 and @xmath6 where @xmath7 . the authors of that work showed that for small @xmath8 , mismatched measurement bases can still be used to gain good channel estimates while also allowing @xmath0 and @xmath1 to use mismatched measurement bases to distill their key ( since , for @xmath8 small , even with differing bases , their measurement results will be nearly correlated ) . mismatched measurements were used in @xcite in order to get better channel statistics for a single - state semi - quantum protocol first introduced in @xcite . though single - state semi - quantum protocols utilize two - way quantum channels , they admit many simplifications which ease their security analysis . in this paper , we consider a multi - state semi - quantum protocol ( which are more difficult to analyze ) and show mismatched measurements improve its key - rate . in @xcite , it was proven , using mismatched measurement bases , that the three - state bb84 protocol from @xcite has a key rate equal to that of the full four state bb84 protocol assuming a symmetric attack . also a four - state protocol using three bases has a key rate equal to that of the full six - state bb84 protocol . in this paper , we will arrive at the same conclusion , though using a more information theoretic argument . however , we will also analyze other protocols , adapt our approach to two - way quantum channels , consider an optimized qkd protocol , and , for all considered protocols , derive new key - rate expressions suitable for asymmetric channels . our method , building off of our conference paper in @xcite ( where we only considered a three - state bb84 protocol ) , is very general and applicable to multiple qkd protocols , both one - way and , as we will demonstrate , two - way ( those which utilize a two - way quantum channel allowing a qubit to travel from @xmath0 to @xmath1 , then return to @xmath0 thus passing through the adversary twice ) . after an introduction to our notation , we will first explain the parameter estimation method and our technique . we will then apply it to variants of the bb84 protocol , confirming the results of @xcite mentioned above ( though also deriving key - rate expressions for arbitrary channels ) . we will consider the extended b92 protocol @xcite and derive an improved key - rate bound for it . we will then use our method to consider an `` optimal '' qkd protocol . finally , we will analyze a multi - state semi - quantum protocol from @xcite which relies on a two - way quantum channel . this new proof of security will derive a more optimistic bound on the key rate expression than the one previously constructed in @xcite ( the latter did not use mismatched measurement bases ) . we now introduce some notation we will use . let @xmath9 be the shannon entropy function , namely : @xmath10 where all logarithms in this paper are base two unless otherwise specified . we will occasionally use the notation @xmath11 to mean @xmath12 . we denote by @xmath13 the binary shannon entropy function , namely : @xmath14 . we write @xmath15 to mean the von neumann entropy of the density operator @xmath16 . if @xmath16 is finite dimensional ( and all systems in this paper are finite dimensional ) , then let @xmath17 be its eigenvalues . in this case @xmath18 . if @xmath16 acts on a bipartite hilbert space @xmath19 we will often write @xmath20 . when we write @xmath21 we mean the result of tracing out @xmath1 s portion of @xmath20 ( that is , @xmath22 ) . similarly for @xmath23 and for systems with three or more subspaces . given density operator @xmath20 we will write @xmath24 to mean @xmath25 and @xmath26 to mean @xmath27 . we denote by @xmath28 to be the conditional von neumman entropy defined as @xmath29 . if the context is clear , we will forgo writing the `` @xmath16 '' subscript . when we talk about qubits , we will often refer to the @xmath3 , @xmath30 , and @xmath31 bases , the states of which we denote : @xmath32 , @xmath33 , and @xmath34 , where : @xmath35 we will now describe the parameter estimation method and what may be gleaned from it . this extends the preliminary work we did in our conference paper @xcite where we only considered three states from two bases . let @xmath36 be a unitary operator acting on the finite dimensional hilbert space @xmath37 where @xmath38 and @xmath39 ( @xmath36 will model eve s attack operation ) ; @xmath40 models the qubit `` transit '' space , while @xmath41 will model the adversary @xmath2 s private quantum memory . without loss of generality , we may describe @xmath36 s action on states of the form @xmath42 , where @xmath43 is some arbitrary , normalized state in @xmath41 , as follows : @xmath44 where the @xmath45 are states in @xmath41 which are not necessarily normalized nor orthogonal . unitarity of @xmath36 imposes certain obvious restrictions on these states which will become important momentarily . let @xmath32 be the computational @xmath3 basis . let @xmath46 $ ] and denote by @xmath47 and @xmath48 . we will always assume the choice of @xmath49 and @xmath50 is public knowledge and , once chosen , remains fixed . define @xmath51 and @xmath52 , where : @xmath53 clearly @xmath54 and @xmath55 are both orthonormal bases of @xmath40 . note that when @xmath56 we have @xmath57 , the hadamard @xmath30 basis , and @xmath58 , the @xmath31 basis . the @xmath3 and @xmath30 bases are customarily used in the bb84 protocol , while the @xmath3 , @xmath30 , and @xmath31 bases are used in the six - state bb84 protocol @xcite . note also that when @xmath59 , we have @xmath60 . as it will be important later , we note that : @xmath61 where , due to linearity of @xmath36 , we have : @xmath62 as the operator @xmath36 will be used to model @xmath2 s attack , we are interested in determining what @xmath0 and @xmath1 can learn about it ( in particular , the @xmath45 states ) after performing the parameter estimation protocol described in protocol [ alg : pe ] . let @xmath63 be fixed and public knowledge . let @xmath64 also be fixed and public knowledge . finally fix an attack operator @xmath36 known only to @xmath2 . repeat the following procedure for @xmath65 : 1 . @xmath0 chooses a random state @xmath66 and sends a qubit prepared in this state to @xmath1 . 2 . @xmath2 captures the qubit before it arrives at @xmath1 s lab and probes it using her attack operator @xmath36 . we assume this operator acts on the qubit and a freshly prepared ancilla @xmath41 prepared in a state @xmath43 known only to @xmath2 , but constant each iteration . that is , the system evolves from @xmath67 to @xmath68 . the qubit ( in @xmath40 ) is sent to @xmath1 . 3 . @xmath1 chooses a basis @xmath3 , @xmath54 , or @xmath55 ( assuming a state from one of these bases appears in @xmath69 ) and measures the qubit in that basis . 4 . @xmath0 and @xmath1 disclose their choice of basis . @xmath0 discloses the state she prepared and @xmath1 discloses his measurement result . this communication is done via an authenticated classical channel . note that they do not discard any measurement results ( i.e. , they do not discard iterations where they chose different bases ) . we will consider two cases for the choice of @xmath69 in this paper ( others may easily follow from our computations ) : @xmath70 that is , in the first case , @xmath0 is limited to only sending three possible states from the four available in the @xmath3 and @xmath71 bases ; in the second she can only send four possible states from the six available in the @xmath3 , @xmath71 , and @xmath72 bases . note that , on step 3 of the parameter estimation protocol above , @xmath1 will measure in either the @xmath3 or @xmath71 bases in the first case ; in the second case he will measure in the @xmath3 , @xmath71 , or @xmath72 bases ( his outcomes could be any of the four states or six states respectively ) . thus , @xmath0 and @xmath1 can , for instance , estimate the probability of @xmath36 flipping a @xmath73 to a @xmath74 but they can not directly observe the probability of @xmath36 flipping a @xmath74 to a @xmath73 ( as @xmath0 can never prepare the state @xmath74 ) . note that in this paper , we assume @xmath75 may be set large enough so as to attain arbitrarily high levels of accuracy in all estimates . this is clearly without loss of generality in the asymptotic scenario in qkd security ; it remains an open question to analyze this scenario in the finite key setting where imprecision will affect the final key rate computation . clearly @xmath0 and @xmath1 may estimate @xmath76 by using those iterations where @xmath77 and @xmath1 measures in the @xmath3 bases . now , denote by @xmath78 , where @xmath79 and @xmath80 , to be the probability that @xmath1 measures @xmath81 if @xmath0 initially sends @xmath82 in step 1 of the protocol , conditioning on the event both parties choose the correct basis for such an outcome . also , let @xmath83 and @xmath84 denote the real part of @xmath85 and the imaginary part of @xmath85 respectively . now , consider @xmath86 . if @xmath0 sends @xmath87 , after evolution by @xmath36 , the state becomes : @xmath88 from which it is clear that @xmath89 . similarly , the following expressions are easily derived : @xmath90 @xmath91 as already mentioned , if @xmath92 , then each @xmath93 can be estimated by @xmath0 and @xmath1 . if @xmath94 , then @xmath95 and @xmath96 are statistics that are observable by @xmath0 and @xmath1 . if @xmath97 , then this provides @xmath0 and @xmath1 with estimates of the quantities @xmath98 , @xmath99 , and @xmath100 . since @xmath36 is unitary and so @xmath101 , this also provides @xmath0 and @xmath1 with an estimate of @xmath102 ( i.e. , @xmath103 ) . finally , using these , along with the cauchy - schwarz inequality to bound @xmath104 ( both @xmath105 and @xmath106 should be small as they represent the @xmath3 basis error rate ) , @xmath0 and @xmath1 may bound @xmath107 using @xmath96 . we will show a particular example shortly when we consider a symmetric attack . if @xmath108 , then @xmath0 and @xmath1 may estimate all of the above , in addition to @xmath109 , and @xmath110 . this , combined with the above , allows @xmath0 and @xmath1 to estimate : @xmath111 ( i.e. , they may estimate both the real parts - using @xmath71 - and the imaginary parts - using @xmath72 ) . finally , the quantity @xmath110 , together with @xmath96 may be used to compute a better estimate of @xmath112 and @xmath107 . to illustrate , assume @xmath56 ( other values may be used , though the algebra becomes tedious ) . then , the following identities are easy to derive : @xmath113 ( note , in the above , we made use of the fact that @xmath114 and @xmath115 due to unitarity of @xmath36 . ) the above analysis applies to any arbitrary channel . however , to illustrate the use of the above expressions , let us consider a special case that @xmath2 s attack operator @xmath36 is symmetric ; in particular , @xmath2 s attack is such that its action on @xmath40 may be modeled as a depolarization channel with parameter @xmath116 . that is , for any two - dimensional density operator @xmath16 , we have , after @xmath2 s attack : @xmath117 where @xmath118 is the two - dimensional identity operator . this assumption is common in the qkd literature . however , we comment that , in the following , we will only utilize statistics that @xmath0 and @xmath1 are capable of observing directly ; our use of a depolarization channel is simply to demonstrate our technique and to put concrete numbers to these many parameters . furthermore , since only observable statistics are used , this depolarization channel assumption is entirely enforceable by @xmath0 and @xmath1 . let us first consider @xmath119 ( the same set that was considered in our conference paper @xcite ) , with @xmath120 . if @xmath0 sends a @xmath3 state @xmath82 for @xmath121 , then the qubit s state when it arrives at @xmath1 s lab is : @xmath122 from this , it is clear that @xmath123 . thus @xmath124 and @xmath125 . now , if @xmath0 sends @xmath87 , it is clear that @xmath126 . from this , along with equation [ eq : p0a ] , we conclude @xmath127 . ( the reader will observe that , when @xmath128 , and so @xmath129 , we have @xmath130 as expected . ) similarly , we conclude that @xmath131 . substituting this into equation [ eq : paa ] , and solving for @xmath107 , we find : @xmath132 where we used the fact that @xmath133 and so @xmath134 . if @xmath135 ( which it would be in this depolarization channel scenario ) , then the above simplifies to : @xmath136 an expression with no dependence on @xmath49 ( other than @xmath137 ) . by the cauchy - schwarz inequality , we have @xmath138 $ ] thus implying @xmath139 $ ] . now , if @xmath108 , for @xmath140 , then , in addition to the above , we also find @xmath141 ( using a similar method as described above , but with statistics @xmath142 , @xmath143 , and @xmath144 ) . by realizing that @xmath145 , we also have @xmath146 . thus , in this setting , it holds that @xmath147 . clearly , @xmath49 and @xmath50 do not appear in the symmetric case ; let us now consider the choice of these parameters for arbitrary channels . denote by @xmath83 to be , as before , @xmath148 ( similarly @xmath84 ) . this is the actual value of the inner product of these two states . denote by @xmath149 and @xmath150 to be @xmath0 and @xmath1 s estimate of these quantities after parameter estimation . we now determine how the choice of @xmath49 and @xmath50 affect these estimates , assuming we have perfect parameter estimation of the required @xmath78 quantities ( in the asymptotic scenario , which we focus on in this paper , this is not a problem ; we leave the non - perfect case as future study ) . clearly , from equations [ eq : p0a ] , [ eq : p1a ] , and [ eq : pa0 ] , we have @xmath151 , @xmath152 , @xmath153 and @xmath154 . from equations [ eq : p0b ] , [ eq : p1b ] , and [ eq : pb0 ] , we get the same result for the imaginary parts of these quantities . the choice of @xmath49 and @xmath50 , therefore , do not affect these estimates . what remains is to consider how these two parameters affect @xmath155 and @xmath156 . using equation [ eq : paa ] , @xmath0 and @xmath1 will estimate these two quantities using the identity : @xmath157 using the fact that the estimates for all real parts on the right - hand side of the above expression are , indeed , the actual values , the above implies ( substituting in @xmath158 for the actual values @xmath83 ) : @xmath159 from which we see there is no dependence on @xmath49 . of course this parameter may affect the key rate for certain protocols ( as we see later ) , it does not affect the parameter estimation process assuming perfect estimates of @xmath78 . when we consider also the @xmath72 basis , we additionally get the following identity ( using @xmath110 ) : @xmath160 combining this with the above , we find @xmath161 and @xmath162 . thus , in this case , we get exact estimates of the actual values of @xmath2 s attack which , again , do not depend on @xmath49 or @xmath50 . the above all assumed perfect accuracy in our estimation of the quantities @xmath78 . in a realistic scenario , this will not be the case . in such a setting , the choice of @xmath49 and @xmath50 will no doubt alter the estimation of the quantities @xmath83 and @xmath84 . we leave this study as future work . in this section , we will apply the above estimation method to analyze four different qkd protocols . first , a generalized bb84-style protocol with only three or four states being sent . secondly , we will analyze the extended b92 protocol first introduced in @xcite . after this , we will consider an `` optimized '' qkd protocol . finally , we will apply our technique to a protocol relying on a two - way quantum channel ( thus allowing the attacker two opportunities to interact with the qubit each iteration ) . for all these protocols , we will derive key - rate expressions as functions of the various @xmath83 quantities discussed in the last section from which protocol [ alg : pe ] may be used to discover bounds on . a qkd protocol operates by first performing several iterations of a _ quantum communication stage _ , the result of which is an @xmath163-bit _ raw key _ : a bit - string that is partially secret and partially correlated ( @xmath0 and @xmath1 each have their own version of the raw key ) . following this , assuming @xmath0 and @xmath1 do not abort , they perform an error correction protocol and a privacy amplification protocol ( see @xcite ) , resulting in a secret key of size @xmath164-bits ( possibly @xmath165 if the noise rate is too high - i.e. , if @xmath2 has potentially too much information on the raw key ) . the goal is to determine @xmath164 as a function of the noise rate ( in our case , as functions of the various @xmath78 values considered in the previous section ) . actually , we are interested in computing a protocol s _ key rate _ in the _ asymptotic scenario _ , defined to be : @xmath166 we will first consider collective attacks : these are attacks where @xmath2 applies the same attack operation each iteration of the protocol . thus each iteration is independent of all others . later we will consider _ general attacks _ where no such restrictions are placed on the adversary . assuming collective attacks , it was shown in @xcite that the key rate is : @xmath167 where the infimum is over all collective attacks which induce the observed statistics . the computation of @xmath168 is trivial given values @xmath78 for @xmath169 . the challenge is to compute @xmath170 . we will use the following lemma and theorem to compute the conditional von neumann entropy @xmath170 of the quantum systems which will appear later in our analysis of these various qkd protocols . [ lemma : entropy - computation ] let @xmath171 be a finite dimensional hilbert space with @xmath172 an orthonormal basis of @xmath173 . consider the following density operator : @xmath174 where @xmath175 , @xmath176 , and each @xmath177 a hermitian positive semi - definite operator of unit trace acting on @xmath178 . then the von neumann entropy of @xmath16 is : @xmath179 the proof of this is straight - forward . see , for instance , @xcite . [ thm : entropy ] let @xmath180 be a finite dimensional hilbert space . consider the following density operator : @xmath181 + { \ket{1}\bra{1}}_a \otimes \left[\sum_{i=1}^m{\ket{g_i^1}\bra{g_i^1}}\right]\right),\ ] ] where @xmath182 is a normalization term , @xmath183 ( note , this @xmath75 has no relation to the @xmath75 from the parameter estimation procedure described last section ) , and each @xmath184 ( these are not necessarily normalized , nor orthogonal , states ; also it might be that @xmath185 for some @xmath186 and @xmath187 ) . let @xmath188 . then : @xmath189 where : @xmath190 and : @xmath191 note that @xmath192 if and only if @xmath185 . also note that , since @xmath193 is a density operator , and thus unit trace , it must be that @xmath194 . assume , first , that all @xmath195 . we may write @xmath193 in the form : @xmath196 where : @xmath197 note that each @xmath198 is a hermitian positive semi - definite operator of unit trace . let @xmath199 be the @xmath75-dimensional hilbert space spanned by orthonormal basis @xmath200 , and let @xmath201 be the following density operator : @xmath202 clearly , @xmath203 . due to the strong sub additivity of von neumann entropy , we have : @xmath204 by definition of conditional entropy , we have @xmath205 . applying lemma [ lemma : entropy - computation ] twice , we have : @xmath206 from which it is clear that : @xmath207 we now compute @xmath208 . it is easy to see that : @xmath209 thus , we need only to compute @xmath210 . without loss of generality , we may write : @xmath211 where @xmath212 , @xmath213 , and @xmath214 . we also have the following identities : @xmath215 writing @xmath216 in this @xmath217 basis , we find : @xmath218 the eigenvalues of which are : @xmath219 since @xmath220 , and letting @xmath221 , we find : @xmath222 where , for the last equality , we used identity [ eq : ident - ev-3 ] . thus , we have @xmath223 . to complete this case of the proof , observe that @xmath224 and that , as @xmath225 increases ( in particular , as @xmath226 increases ) , @xmath210 decreases . the goal of this theorem is to lower - bound the quantity @xmath227 . thus , we upper - bound @xmath228 . since @xmath229 , if we define @xmath230 as given in the theorem statement ( i.e. , equation [ eq : thm : lambda ] ) , then : @xmath231 therefore , we may conclude : @xmath232 setting @xmath233 equal to the above expression completes the proof for the case @xmath195 for all @xmath186 and @xmath187 . if @xmath234 for some @xmath186 , it holds that @xmath235 . thus , this term does not appear in @xmath193 and so does not contribute to the entropy computation , thus justifying setting @xmath236 in this case . if @xmath237 and @xmath238 , then @xmath239 and so does not show up in @xmath193 . thus , the @xmath186th term in our decomposition of @xmath193 given by equation [ eq : thm : rho - exp ] may be written : @xmath240 the conditional entropy of such a system is simply @xmath241 . the case when @xmath242 and @xmath243 is similar , thus completing the proof . note that the decomposition of a density matrix to the form in equation [ eq : thm:1 ] may not be unique ( indeed , one may permute the various @xmath244 states ) . different decompositions lead to potentially different , though correct , lower - bounds on @xmath170 . when using the above theorem to prove security of a qkd protocol , one must choose the correct ordering so as to produce the highest possible lower - bound ( thus improving the key - rate expression ) . let us first consider a generalized , limited - resource bb84 protocol which we denote by @xmath245 where @xmath69 will denote the set of qubit states that @xmath0 may prepare and send to @xmath1 ( we say `` limited - resource '' as @xmath69 may not contain all states in all bases ) . this generalized bb84 is described in protocol [ alg : bb84 ] . * input * : let @xmath64 be the set of possible states that @xmath0 may send to @xmath1 under the restrictions that @xmath246 . * quantum communication stage * : the quantum communication stage of the protocol repeats the following process : 1 . @xmath0 will send a qubit state @xmath247 , choosing randomly according to some publicly known distribution ( we assume @xmath87 and @xmath248 are chosen with equal probability and that all states in @xmath69 have non - zero probability of being chosen ) . 2 . @xmath1 will choose a random basis to measure in . 3 . @xmath0 and @xmath1 will disclose their choice of basis . on certain , randomly chosen iterations , @xmath0 and @xmath1 will also disclose their measurement results in order to run the parameter estimation protocol described last section . this disclosure is done using the authenticated classical channel . if @xmath0 and @xmath1 both chose the @xmath3 basis , and if this iteration is not used for parameter estimation , this iteration is used to contribute towards their raw key . it is clear that when @xmath249 , then this is the `` standard '' four - state bb84 protocol as originally described in @xcite which can withstand up to @xmath250 error rate assuming a symmetric channel @xcite . it is also clear that when @xmath251 ( where @xmath252 ) the protocol is the standard six - state bb84 protocol @xcite which can withstand up to @xmath253 error ( without preprocessing ) @xcite . in this section , we will consider the three state bb84 protocol @xmath254-bb84 ( first introduced in @xcite ) and a four state , three basis version @xmath255-bb84 ( where @xmath254 and @xmath255 are defined in equations [ eq : psi3 ] and [ eq : psi4 ] ) . for symmetric channels , we will show that the three - state version can suffer up to @xmath250 error rate , while the four - state , three - basis version can tolerate up to @xmath253 error ( thus confirming , using our alternative , more information theoretic approach , the results from @xcite ) . we will also derive a more general key - rate expression which may be evaluated assuming any arbitrary channel . we first consider collective attacks . to compute the key rate , we must first describe the quantum system after a successful iteration of the protocol ( successful meaning this iteration was used to contribute towards the raw key ) . since only @xmath3 basis states are used for this purpose ( other states being used only for parameter estimation ) , this system is easily derived , regardless of @xmath69 , and shown in equation [ eq : bb84-joint - system ] . @xmath256 ( note that we have kept the notation of @xmath2 s quantum probe used in section [ section : pe ] from equation [ eq : u - states ] . ) tracing out @xmath1 s system yields : @xmath257 as before , let @xmath78 ( for @xmath258 ) be the probability that , if @xmath0 sends @xmath82 , then @xmath1 measures @xmath81 , conditioning on the event both users choose the correct basis for such an outcome . applying theorem [ thm : entropy ] ( with @xmath259 , @xmath260 , @xmath261 , and @xmath262 ) , we find the key rate @xmath263 to be lower bounded by : @xmath264\notag\\ & + \frac{p_{0,1 } + p_{1,0}}{2 } \left [ h\left ( \frac{p_{0,1}}{p_{0,1 } + p_{1,0}}\right ) - h(\lambda_2)\right]\notag\\ & - h(a|b),\label{eq : bb84-general}\end{aligned}\ ] ] where : @xmath265 ( above , we waived our hand over the issue when one or more of the @xmath266 , however in that case the summand containing that term goes to zero as described in theorem [ thm : entropy ] . ) the quantity @xmath168 is easily computed by the definition of conditional shannon entropy simply by using the observed statistics @xmath78 ( for @xmath169 ) . this is enough to compute bounds on the key - rate of the bb84 protocol over any arbitrary channel ( bounds on @xmath267 and @xmath268 are found using the parameter estimation process described in section [ section : pe ] ) . to illustrate , however , we note that when @xmath2 s attack is symmetric ( thus @xmath269 and @xmath123 ) which is entirely enforceable , the above expression takes on the much simpler form : @xmath270 where we used the easily shown fact that @xmath271 in this case . if @xmath94 , we may evaluate the above expression by optimizing over all @xmath138 $ ] ( while also utilizing the work performed in section [ section : sym ] ) . doing so yields a result shown in figure [ fig : bb84-keyrate1 ] - in particular , the key rate drops to zero when @xmath272 . when @xmath273 , the key rate drops to zero at @xmath253 as also shown in figure [ fig : bb84-keyrate1 ] . this latter is exactly the noise tolerance of the full six - state bb84 protocol as shown in @xcite ( without preprocessing ) . note that the value of @xmath49 and @xmath50 do not appear in these expression ( so long as they are both in the interval @xmath274 ) . since we are also assuming perfect parameter estimation , they also do not affect that process . -bb84 protocol under a symmetric attack . solid line is the key - rate of @xmath254-bb84 and can tolerate up to @xmath250 noise ( same as the four - state bb84 protocol ) while the dashed line shows the key - rate of @xmath255-bb84 , in which case the protocol withstands up to @xmath253 error ( same as the full six - state bb84 without preprocessing ) . ] this result confirms , independently , that found in @xcite where it was also shown , using mismatched measurement bases , though using a different analytical technique , that three states were sufficient to achieve the noise tolerance of the usual four state bb84 and four states from three bases was enough to attain the noise tolerance of the six - state protocol . in the next sections we show how our method easily extends to other protocols which do not necessarily follow the bb84-style encoding and which may utilize a two - way quantum communication channel . we will use the more general , non - symmetric , expression ( equation [ eq : bb84-general ] ) in a later subsection . we will also comment on general attacks at the end of this section . in this section , we apply our method to the analysis of the extended b92 protocol introduced in @xcite . this protocol , like the standard b92 @xcite , uses two non - orthogonal states to encode the raw key bits , however it extends the protocol by allowing @xmath0 to send other qubit states ( beyond the two only allowed by standard b92 ) for parameter estimation purposes . we will use our technique to derive more optimistic noise tolerances for this protocol than prior work in @xcite . this also serves to show the generality of our technique . we begin by introducing the protocol using our terminology . denote by @xmath69-b92 the protocol shown in protocol [ alg : b92 ] . * input * : let @xmath64 be the set of possible states that @xmath0 may send to @xmath1 under the restrictions that @xmath275 . * quantum communication stage * : the quantum communication stage of the protocol repeats the following process : 1 . @xmath0 will send a qubit state @xmath247 , choosing randomly according to some publicly known distribution ( we assume @xmath87 and @xmath73 are chosen with equal probability and that all states in @xmath69 have non - zero probability of being chosen ) . 2 . @xmath1 chooses a random basis and measures the qubit in this basis . if @xmath0 chose to send @xmath87 or @xmath73 , she sets her raw key bit to be @xmath241 or @xmath276 respectively . 4 . if @xmath1 observes a @xmath74 or @xmath248 , he sets his key bit to be @xmath241 or @xmath276 respectively . 5 . @xmath0 and @xmath1 announce , over the authenticated classical channel , whether this is a _ successful _ iteration : namely , whether @xmath0 choose @xmath87 or @xmath73 and whether @xmath1 observed @xmath74 or @xmath248 ( of course they do not disclose their actual preparations or observations ) . all other iterations , along with a suitable , randomly chosen , subset of successful iterations , are used for parameter estimation as described in section [ section : pe ] . to compute the key rate of this extended b92 protocol , we must first , as before , describe the joint quantum system held by @xmath0 , @xmath1 , and @xmath2 conditioning on `` successful '' iterations . recall that @xmath277 ( thus @xmath278 ) . we also define @xmath279 , where @xmath280 is the conjugate transpose of @xmath281 . it is not difficult to show that this quantum system is : @xmath282\\ & + \frac{1}{n ' } \left [ \frac{1}{2}{\ket{01}\bra{01}}_{ab } \otimes { \ket{e_1}\bra{e_1 } } + \frac{1}{2}{\ket{10}\bra{10}}_{ab } \otimes { \ket{f_1}\bra{f_1}}\right],\end{aligned}\ ] ] where we have adopted the same notation for @xmath2 s attack as in equations [ eq : u - states ] and [ eq : u - states - f ] , and @xmath283 is a normalization term to be discussed shortly . tracing out @xmath1 s system and using the fact that , from equation [ eq : u - states - f2 ] , we find @xmath284 , we have : @xmath285\\ & + \frac{1}{n'}\left [ \frac{1}{2}{\ket{0}\bra{0}}_a \otimes { \ket{e_1}\bra{e_1 } } + \frac{1}{2}{\ket{1}\bra{1}}_a\otimes{\ket{f_1}\bra{f_1}}\right].\end{aligned}\ ] ] in order to apply theorem [ thm : entropy ] , we write the above state in the following form : @xmath286 where we defined : @xmath287 from this , theorem [ thm : entropy ] may be directly applied to compute the key rate of this extended b92 protocol , providing us with a key - rate expression for any asymmetric channel ( computing @xmath168 is , as before , trivial ) . writing out the expression , however , is not enlightening as it would entail simply copying equation [ eq : thm : entropy ] from theorem [ thm : entropy ] . instead , we will illustrate by writing out the case for a symmetric attack ( again , an enforceable assumption ) allowing us to take advantage of certain simplifications in the expressions . furthermore , it will allow us to compare with prior work to demonstrate the advantage to using mismatched measurement bases for this protocol . we stress , however , that symmetry is not required at this point . let @xmath116 denote the error rate of the channel as before . assuming a symmetric attack ( and thus @xmath288 ; see section [ section : sym ] ) , the normalization term simplifies to : @xmath289\notag\\ & = \bar{\alpha}^2(1-q ) + \alpha^2q + q\notag\\ & = 1 - \alpha^2(1 - 2q).\label{eq : b92-n}\end{aligned}\ ] ] where , to derive the last equality , we used the fact that @xmath290 . let @xmath291 ( the two are equal when faced with a symmetric attack ) and @xmath292 . by theorem [ thm : entropy ] , we have : @xmath293 where : @xmath294 @xmath295 may be evaluated directly : @xmath296 ( again we use the notation @xmath83 to mean @xmath297 . ) furthermore , we have : @xmath298 bounds on @xmath107 and @xmath112 are found as described in section [ section : sym ] . all that remains is to compute @xmath168 . but this is simply : @xmath299 to evaluate the key - rate assuming this symmetric attack , we must optimize over all @xmath300 $ ] if @xmath301 . otherwise , if we are using @xmath255 , we have exact values for these parameters and evaluate the expression directly . unlike the bb84 case , where @xmath49 and @xmath50 did not appear in the key - rate expression , here , since @xmath73 is being used to encode a key - bit value , the value of @xmath49 does affect the resulting key rate ( @xmath50 , being used only for parameter estimation , does not ) . the resulting key - rate for various values of @xmath49 and states @xmath69 is shown in table [ table : b92 ] . in that same table , we compare our new tolerated bounds with those found in @xcite ; our new key rate suffers a higher tolerated error than original work in @xcite ( which did not utilize mismatched measurement outcomes ) for all @xmath302 . .comparing our new key rate bound for the extended b92 protocol with the one from @xcite . in particular , we compare the maximally tolerated error rates of our bound with the one from @xcite for a depolarization channel and for various values of @xmath303 ( where @xmath87 and @xmath73 are used to encode the classical value of @xmath241 and @xmath276 respectively ) . [ cols=">,^,^,^,^,^ " , ] we now turn our attention to a qkd protocol reliant on a two - way quantum channel ; that is , a channel which allows a qubit to travel from @xmath0 to @xmath1 and then back to @xmath0 . such a scenario provides an attacker two opportunities to interact with the qubit . as before , we will consider first collective attacks , an assumption which allows us to model @xmath2 s attack as two unitary operators @xmath304 applied in the forward direction ( when the qubit travels from @xmath0 to @xmath1 ) and @xmath305 applied in the reverse channel ( when the qubit travels from @xmath1 to @xmath0 ) . both operators act on the qubit and @xmath2 s private quantum ancilla @xmath41 . see figure [ fig : two - way ] . while we assume @xmath2 s probe state is cleared to some zero state when the iteration starts , we do not make that assumption between rounds - i.e. , when @xmath304 acts , @xmath2 s ancilla is in a @xmath43 state ; when @xmath305 acts , however , this is not the case ( it acts on the same system @xmath304 operated on ) . we will show how mismatched measurement bases can be applied here to gather statistics on each of the operators @xmath304 , @xmath305 and the joint operator @xmath306 leading to more optimistic key rate bounds than previously considered for the protocol in question . begins each iteration by sending a qubit to @xmath1 . @xmath2 probes this qubit using unitary operator @xmath304 , acting on the qubit and @xmath2 s ancilla @xmath41 . later , when the qubit returns from @xmath1 to @xmath0 , @xmath2 attacks with a second unitary operator @xmath305 which also acts on the qubit and the same ancilla space @xmath41 . ] the protocol we chose to analyze is the one described by boyer et al . , in @xcite . it is actually a _ semi - quantum _ protocol as it makes the assumption that , while the user @xmath0 is _ fully quantum _ ( i.e. , she may prepare and measure qubits in any basis she likes ) , the other user @xmath1 is _ semi - quantum _ or `` classical '' in nature in that he is limited only to measuring and preparing qubits in the computational @xmath3 basis . let us more formally describe the capabilities of the parties involved . @xmath0 is permitted to send any qubit chosen from a set @xmath69 ( as in the previous section we will consider two cases : @xmath94 and @xmath108 ) . this qubit is sent to the `` classical '' user @xmath1 who may chose to do one of two things : 1 . * measure and resend * : @xmath1 will perform a @xmath3 basis measurement on the incoming qubit resulting in outcome @xmath307 for @xmath308 ( possibly random if @xmath0 sent a non-@xmath3 basis state ) . he will then prepare a new qubit in this state and send it to @xmath0 ; i.e. , he will send @xmath307 to @xmath0 . 2 . * reflect * : @xmath1 will `` reflect '' the qubit back to @xmath0 . that is , the qubit will pass through @xmath1 s lab undisturbed and return to @xmath0 . in this event @xmath1 learns nothing about the state of the qubit . we chose to analyze this particular protocol for several reasons . firstly , semi - quantum cryptography is very interesting from a theoretical stand - point as it attempts to answer the question `` how quantum does a protocol need to be to gain an advantage over a classical one ? '' thus , finding good bounds on this protocol s key rate is an important question . secondly , a lower bound on its key - rate was derived in @xcite which did not make use of mismatched measurement bases . thus this gives us an additional comparison case showing the advantage of this method of parameter estimation . thirdly , it is a two - way protocol which also admits the possibility of mismatched measurements ( not all two - way protocols permit such a possibility ) and as such allows us to determine how mismatched measurement bases may be used to collect various statistics on the separate unitary operators under @xmath2 s employ : work which may extend to the analysis of other two - way qkd protocols . finally , it shows our technique is strong enough to provide good results even when one of the two users is limited in his capabilities ( i.e. , @xmath1 can never measure in the @xmath71 or @xmath72 bases ) . let us now describe the protocol we consider . it is a generalized version of the one introduced in @xcite . in their protocol , the fully quantum user @xmath0 prepared qubits randomly in either the @xmath3 or @xmath30 basis . here we consider the @xmath3 , @xmath71 , and potentially @xmath72 basis . of course the classical user may only measure in the @xmath3 basis . we will denote this generalized protocol by @xmath69-sqkd , where @xmath69 is the set of states that @xmath0 is allowed to prepare . this sqkd protocol is described in protocol [ alg : sqkd ] . * input * : let @xmath309 be the set of possible states that @xmath0 may send to @xmath1 under the restrictions that @xmath246 . * quantum communication stage * : the quantum communication stage of the protocol repeats the following process : 1 . @xmath0 will send a qubit state @xmath247 , choosing randomly according to some publicly known distribution ( we assume @xmath87 and @xmath248 are chosen with equal probability and that all states in @xmath69 have non - zero probability of being chosen ) . @xmath1 will , with probability @xmath310 perform the * measure and resend * operation saving his measurement result . otherwise , he will * reflect * the qubit . 3 . with probability @xmath311 , @xmath0 will measure the returning qubit in the @xmath3 basis ; otherwise she will measure in the @xmath71 or @xmath72 basis ( assuming a state from such a basis appears in @xmath69 ) . 4 . @xmath0 will disclose her choice of bases and @xmath1 will disclose his choice of operation . on certain , randomly chosen , iterations , @xmath0 and @xmath1 will also disclose their measurement results in order to run parameter estimation . this disclosure is done using the authenticated classical channel . 5 . if this iteration is not used for parameter estimation , and if @xmath0 chose the @xmath3 basis in steps 1 and 3 , and if @xmath1 chose to measure and resend , they will use this iteration to contribute towards their raw key . @xmath1 will use his measurement result as his key bit ; @xmath0 will use her measurement result from step 3 as her key bit ( an alternative construction would be to use her preparation choice ) . note that we altered step 3 of the original boyer et al . @xcite , protocol so as to allow the chance for mismatched measurements to occur on reflection iterations . furthermore , on step 5 ( which is also a slight alteration from the original ) , the reasoning for only using those iterations where @xmath0 chose @xmath3 basis states in both steps ( as opposed to only the last ) is simply to allow us to compare with prior work . similar computations may be performed without this stipulation . note that , as with bb84 , one may set @xmath310 and @xmath311 arbitrarily close to one so as to improve the efficiency of the protocol @xcite . when @xmath312 , this protocol is exactly that described in @xcite ( excepting for those minor alterations we mentioned above ) . we will actually analyze the case when @xmath94 and @xmath108 . for the former , when @xmath128 , the protocol is actually one considered in @xcite ( though in that paper , the protocol was only proven _ robust _ , a far weaker definition of security than the one we consider in this paper ) . the latter case has not been considered . regardless , it is clear that , if we develop a lower bound on the key rate when @xmath313 , we will also have a lower - bound on the original sqkd protocol ( i.e. , we will have a lower - bound on the case when @xmath312 as dictated by the original protocol of boyer et al.@xcite ) . as in the previous subsections , to compute the key - rate , we must first describe the joint quantum system held between @xmath0 , @xmath1 , and @xmath2 after one successful iteration of the protocol ( where , as before , we say successful to imply it was used to contribute towards the actual raw key , and not only for parameter estimation ) . let @xmath304 be the unitary attack operator employed by @xmath2 in the forward channel and @xmath305 that employed by @xmath2 in the reverse ( again , we are first assuming collective attacks ) . we may also assume , without loss of generality , that at the start of the iteration @xmath2 s ancilla is cleared to some @xmath43 state . thus , we may write @xmath304 s action as follows : @xmath314 ( as we did in section [ section : pe ] . ) we will write , without loss of generality , @xmath305 s action on states of the form @xmath315 as : @xmath316 conditioning on a successful iteration , @xmath0 will send either @xmath87 or @xmath248 choosing each with probability @xmath317 . @xmath2 will attack with operator @xmath304 and @xmath1 will measure in the @xmath3 basis , save his result in a private register , and resend his result . this qubit then passes through @xmath2 a second time . at this point ( when @xmath2 captures the returning qubit but before applying @xmath305 ) , the joint system is easily found to be : @xmath318 @xmath2 will now attack with her second operator @xmath305 and forward the transit qubit to @xmath0 who subsequently measures in the @xmath3 basis . the outcome of her measurement is saved in a private register ( it is to be her raw key bit for this iteration ) . the joint system is now of the form : @xmath319 we will now compute @xmath320 , the key rate when reverse reconciliation is used . as described in @xcite this seems the more natural choice for this protocol ; furthermore , it will allow us to immediately compare our new key rate bound with the old one from @xcite . writing @xmath321 to mean @xmath322 and tracing out @xmath0 s system leaves us with the state : @xmath323 ( in the above state , we also defined those states @xmath324 for use when later applying theorem [ thm : entropy ] . ) at this point , let us pause to determine some of the statistics which may be gathered through channel tomography . it is clear that @xmath0 and @xmath1 may learn @xmath93 ; that is , the @xmath3 basis noise in the forward channel ( note that , while @xmath1 may measure in the @xmath3 basis , he can not measure in the @xmath71 basis , so , for instance , he can not estimate @xmath86 ) . they may also learn @xmath325 and @xmath326 for @xmath327 . indeed , consider @xmath328 . if @xmath0 sends @xmath87 and @xmath1 measures @xmath87 ( after @xmath2 s first attack ) , the joint state becomes : @xmath329 now , @xmath1 forwards the qubit to @xmath0 , however before it arrives , @xmath2 attacks with @xmath305 evolving the state to : @xmath330 thus , the probability of @xmath0 measuring @xmath248 is simply @xmath331 , thus providing @xmath0 and @xmath1 with the value @xmath328 . similarly , @xmath0 and @xmath1 may directly observe the other @xmath325 and @xmath326 for @xmath327 . note that , if we consider a symmetric attack ( which imposes certain simplifications ) , we may denote by @xmath332 the @xmath3 basis error in the forward channel and @xmath333 the @xmath3 basis error in the reverse channel . thus : @xmath334 similarly , we have : @xmath335 of course , as with the other protocols considered , we do not need any symmetry assumption - it only helps to illustrate the resulting bounds and allows us to compare with prior work . determining bounds on all combinations of @xmath336 , needed to employ theorem [ thm : entropy ] , proved difficult . we are , however , able to determine a bound on @xmath337 as we soon discuss . using the following , easily proven , lemma , however , will allow us to use only this one term to bound the key rate expression . improving this to incorporate the other , missing terms , could make for interesting future work : a suitable improvement to the parameter estimation process for two - way channels would be required . [ lemma : positive ] let @xmath338 be non - negative numbers such that @xmath339 . then : @xmath340 where : @xmath341 for all @xmath342 $ ] . case 1 : @xmath343 which , of course , implies that @xmath344 . in this case , equation [ eq : lemma : positive - ent ] holds if and only if @xmath345 . but : @xmath346 the case when @xmath347 is similar ( in that case , we have @xmath348 ) . note that , we require @xmath349 so that @xmath350 remains in the domain of @xmath351 . of course , since we are working with quantum states , the cauchy - schwarz inequality will always guarantee this condition for the equations we are interested in . the above lemma implies that each @xmath233 term from theorem [ thm : entropy ] ( equation [ eq : thm : cond - entropy ] ) is non - negative and , since this is a lower - bound , they may be removed if needed . thus , applying theorem [ thm : entropy ] and lemma [ lemma : positive ] , we compute a lower - bound on the conditional entropy of the state @xmath352 , given by equation [ eq : sqkd - state ] , as : @xmath353,\ ] ] where @xmath354 , @xmath355 , and : @xmath356 in the case the attack is symmetric , this expression becomes : @xmath357,\ ] ] where : @xmath358 we will soon see that , even though we discarded several terms which might have increased the conditional entropy ( thus increasing the key rate of the protocol ) , we will still produce a very optimistic lower - bound . we leave as potential future study , the problem of computing the entropy of those ignored systems . our goal now is to determine a bound on @xmath359 . in particular we must lower - bound @xmath360 . to do so , we again utilize the parameter estimation process described in section [ section : pe ] . notice that , if @xmath1 reflects , his operation is , essentially , the identity operator . thus , conditioning on @xmath1 s choice to reflect , the two - way quantum channel becomes , in essence , a one - way channel with a qubit leaving @xmath0 s lab , @xmath2 attacking it via the unitary operator @xmath361 , and the qubit returning to @xmath0 . let us first consider this operator @xmath362 . we may write its action on basis states as follows ( again @xmath2 s lab is cleared to some state `` @xmath43 '' at the start of the iteration ) : @xmath363 due to the linearity of @xmath305 and @xmath304 , these states are : @xmath364 now , @xmath0 may perform parameter estimation on this operator @xmath362 using the process described in section [ section : pe ] in order to learn statistics on the quantities @xmath365 . to evaluate the key - rate bound , @xmath0 and @xmath1 require a bound on @xmath366 . to do so , they may first determine bounds on @xmath367 using the process described in section [ section : pe ] . from this , a bound on the desired quantity may be found by expanding @xmath367 : @xmath368 bounds on the `` @xmath369 '' quantities which appear in the right - hand side of the above expression may be found using the cauchy - schwarz inequality . as before , we will illustrate assuming a symmetric attack allowing us to boil down the above expressions to something simpler to parameterize . if we assume @xmath2 s attack @xmath362 is symmetric ( an assumption that , again , may be enforced ) , and if @xmath313 , we find : @xmath370 . if @xmath108 , then we additionally find @xmath371 . now , consider the probability that @xmath0 measures @xmath74 if she initially sent @xmath73 and @xmath1 reflects . it is clear that this quantity is identical to equation [ eq : paa ] , replacing `` @xmath369 '' states with `` @xmath372 '' states . call this probability @xmath373 . using the above , this becomes : @xmath374 now , let @xmath375 be the @xmath3 basis error rate when @xmath1 reflects ( i.e. , the @xmath3 basis error when a qubit travels forwards to @xmath1 , then backwards to @xmath0 , through the two - way channel ) . it is clear that this is a statistic which @xmath0 and @xmath1 may estimate . furthermore , @xmath376 . substituting equation [ eq : sqkd : g03 ] into equation [ eq : qa ] and solving for @xmath366 yields : @xmath377 ( note that above we used the fact that @xmath134 . ) at this point , we already have enough to find reasonable estimates of @xmath359 . indeed , the cauchy - schwarz inequality enforces the condition that @xmath378 ( though if @xmath255 is used for parameter estimation , we get @xmath371 ) . also , using equation [ eq : prbounds ] and the cauchy - schwarz inequality we have : @xmath379 . the second scenario involves the two channels being correlated in that @xmath380 . for each channel scenario , we will consider two cases : @xmath94 and @xmath108 . for the independent channel case , when @xmath301 we see our key rate remains positive for all @xmath381 . when @xmath382 it remains positive for all @xmath383 . for the correlated channels , when @xmath301 the key rate remains positive for all @xmath384 . when @xmath382 it remains positive for all @xmath385 . these are improvements from the original bound from @xcite ( which did not use mismatched measurement outcomes ) . see table [ table : sqkd ] for a summary of this data . we have thus shown not only that our methods may be extended to two - way quantum channels , but also that this semi - quantum protocol can actually withstand a much higher level of noise than previously thought . improvements in the parameter estimation process for two - way channels may lead to even better bounds ( recall , we were forced to drop certain @xmath233 terms which could only have improved the result ) . r|cc & independent & correlated + old bound from @xcite & @xmath386 & @xmath387 + + new bound using @xmath254 & @xmath388 & @xmath389 + new bound using @xmath255 & @xmath390 & @xmath391 in the previous sections , we considered only collective attacks . however , all protocols considered in this paper may be made permutation invariant in the usual way @xcite . thus , the results of @xcite apply : namely , proving security against collective attacks is sufficient to show security against arbitrary , general attacks . in the asymptotic scenario , which we considered here , the key - rate expressions will remain the same . we have shown , using information theoretic arguments , that the use of mismatched measurement results can be used to improve the key rate bounds and noise tolerances for various protocols , both one - way and two - way . we applied this technique to a general , limited - resource bb84 , an extended b92 , an optimized qkd protocol , and a semi - quantum protocol . for the bb84 protocol , we confirmed through alternative means , the results of @xcite : namely that three states is sufficient to obtain full bb84 security , while four states in three bases is enough to obtain the full six - state bb84 level of security . our new key - rate bounds for the extended b92 protocol show it has a higher tolerance to noise than previously thought . similarly we derived improved key - rate bounds for the semi - quantum protocol of boyer et al.@xcite in all cases , we did not require any symmetry assumptions ( we evaluated our key - rate bounds using a symmetric attack for illustrative and comparative purposes only ) . one might attempt to use this technique on other two - way protocols beyond the class of semi - quantum ones . we did consider this - however the primary advantage of many two - way , fully quantum protocols ( i.e. , not semi - quantum that we considered here ) is that there are no mismatched measurement basis choices . thus a modification to the protocols would be required - an improved key rate may be found in this case , but one must then ask what the resulting advantage would be to the modified protocol . we leave this as an open question . we also leave as future study improving our parameter estimation method for two - way protocols . when deriving the key - rate bound for the sqkd protocol , we were forced to drop several terms which could only have increased the noise tolerance of the protocol . finally , and very importantly , would be to study the performance of this method in finite key settings and when imperfect parameter estimation occurs . all results in this paper assumed @xmath0 and @xmath1 could perform enough iterations so as to derive arbitrarily precise estimates of various statistics . in a practical setting , there will always be some error . taking this into account , and deriving key - rate expressions in the finite key setting is important future work . charles h bennett and gilles brassard . quantum cryptography : public key distribution and coin tossing . in _ proceedings of ieee international conference on computers , systems and signal processing _ , volume 175 . new york , 1984 . walter o. krawec . asymptotic analysis of a three state quantum cryptographic protocol . in _ ieee international symposium on information theory , isit 2016 , barcelona , july 10 - 15 , 2016 _ , pages 24892493 , 2016 .

bremsstrahlung of photons from highly relativistic electrons is investigated . the cross section of the processes , which is suppressed due to a multiple scattering of an emitting electron in dense media ( lpm effect ) and due to photon interaction with electrons of a medium , is calculated with an accuracy up to `` next to leading logarithm '' and with the coulomb corrections taken into account . making allowances for a multiple scattering and a polarization of a medium an analysis of radiation on a target boundary is carried out . the method of consideration of radiation in a thin target under influence of the lpm effect is developed . interrelation with the recent experiment is discussed . = 14.4 cm = -1 cm # 1 .25 cm .5 cm budker institute of nuclear physics , 630090 novosibirsk , russia 2.0 cm

hep-ph9709214

when a high - energy electron emits a soft photon via bremsstrahlung , the process occurs over a rather long distance , known as the formation length . if anything happens to an electron or a photon while traveling this distance , the emission can be disrupted . landau and pomeranchuk were the first who showed that if the formation length of bremsstrahlung becomes comparable to the distance over which a multiple scattering becomes important , the bremsstrahlung will be suppressed @xcite . migdal @xcite , @xcite developed a quantitative theory of this phenomenon . side by side with the multiple scattering of emitting electron one has to take into account also an influence of a medium on radiated electromagnetic field . since long distances are essential in the problem under consideration this can be done by introducing dielectric constant @xmath0 . this effect leads also to suppression of the soft photon emission ( ter - mikaelian effect , see in @xcite ) . a clear qualitative analysis of different mechanisms of suppression is presented in @xcite,@xcite . more simple derivation of the migdal s results is given in @xcite . the next step in a quantitative theory of lpm effect was made in @xcite . this theory is based on the quasiclassical operator method in qed developed by authors @xcite , @xcite . one of the basic equations ( obtained with use of kinetic equations describing a motion of electron in a medium in the presence of external field ) is the schrdinger equation in external field with imaginary potential ( eq.(3.3),@xcite ) . the same equation ( without external field ) was rederived recently in @xcite . the last derivation is based on the approach results of which is coincide basically with the operator quasiclassical method . in @xcite a new calculation approach is developed where multiple scattering is described with the path integral treatment . new activity with the theory of lpm effect is connected with a very successful series of experiments @xcite - @xcite performed at slac during last years ( see in this connection @xcite ) . in these experiments the cross section of bremsstrahlung of soft photons with energy from 200 kev to 500 mev from electrons with energy 8 gev and 25 gev is measured with an accuracy of the order of a few percent . both lpm and dielectric suppression is observed and investigated . these experiments are the challenge for the theory since in all the mentioned papers calculations are performed to logarithmic accuracy which is not enough for description of the new experiment . the contribution of the coulomb corrections ( at least for heavy elements ) is larger then experimental errors and these corrections should be taken into account . in the present paper we calculated the cross section of bremsstrahlung process with term @xmath1 , where @xmath2 is characteristic logarithm of the problem , and with the coulomb corrections taken into account ( section 2 and appendix a ) . this cross section is valid for very high energies when the lpm effect manifest itself for a photon energy of the order of an energy of the initial electron . in the photon energy region , where the lpm effect is `` turned off '' , our cross section gives the exact bethe - heitler cross section ( within power accuracy ) with coulomb corrections . this important feature was absent in the previous calculations . the polarization of a medium is incorporated into this approach ( section 3 ) . the considerable contribution into the soft part of the investigated in the experiment spectrum of radiation gives a photon emission on the boundaries of a target . we calculated this contribution taking into account the multiple scattering and polarization of a medium for the case when a target is much thicker than the formation length of the radiation ( section 4 ) . in section 5 we considered a case when a target is much thinner than the formation length . in this case the cross section has multiplicative form ( probability of radiation times cross section of scattering for the given impact parameter ) . in section 6 a case of an intermediate thickness of a target ( between cases of a thick and a thin target ) is analyzed , polarization of a medium is not included . in section 7 a qualitative picture of a spectral curve ( an effective thickness of a target , position of a minimum ) is discussed . in section 8 we compare the theoretical curve for the intensity spectrum with the data . although agreement between experiment and theory is rather satisfactory , an additional analysis should be done to obtain information about an accuracy of agreement between experimental data and theory . as well known ( see , e.g. @xcite , sec.93 ) the formation length of radiation is ( in this paper the system @xmath3 is used ) @xmath4 where @xmath5 is the energy of the initial electron , @xmath6 is the energy of radiated photon , @xmath7 is the angle between momenta of the photon and the initial electron . we consider first the case when the formation length is much shorter than thickness of a target @xmath8 . in this case the spectral distribution of the probability of radiation per unit time is given by expression ( 2.18 ) , @xcite ( see also @xcite , section 7.4 ) @xmath9 , \label{2}\ ] ] where @xmath10 , functions @xmath11 satisfy an equation @xmath12 with the initial conditions @xmath13 here @xmath14 is the number density of atoms in the medium , @xmath15 is the coordinate in two - dimensional space conjugated to the space ( two - dimensional ) of radiation angle @xmath16 , @xmath17 is the fourier transform of the scattering cross section : @xmath18 for a screened coulomb potential we have @xmath19 where @xmath20 , @xmath21 is the screening radius ( @xmath22 - the bohr radius ) , @xmath23 is the modified bessel function . as we will show below , the main contribution to the probability is given by @xmath24 , where @xmath25 is the electron compton wavelength . expanding @xmath26 as a power series in @xmath27 and introducing new variables @xmath28 we obtain for the spectral distribution of the probability of radiation @xmath29 , \label{8}\ ] ] where @xmath30 , and the functions @xmath31 now satisfy an equation @xmath32 with the initial conditions @xmath33 , the functions @xmath34 and @xmath35 in ( [ 8 ] ) are rescaled according with the initial conditions ( factors @xmath36 and @xmath37 , correspondingly ) . note , that it is implied that in formulae ( [ 2]),([8 ] ) subtraction at @xmath38 is made . the potential @xmath39 ( [ 9 ] ) corresponds to consideration of scattering in the born approximation . the difference of exact as a function of @xmath40 potential @xmath39 and taken in the born approximation is computed in appendix a. the potential @xmath39 with the coulomb corrections taken into account is @xmath41 where @xmath42 , the function @xmath43 see in ( [ a10 ] ) . in above formulae @xmath44 is space of the impact parameters measured in the compton wavelengths @xmath45 , which is conjugate to space of the transverse momentum transfers measured in the electron mass @xmath46 . an operator form of a solution of eq . ( [ 9 ] ) is @xmath47 where we introduce the dirac state vectors : @xmath48 is the state vector of coordinate @xmath44 , @xmath49 . substituting ( [ 10 ] ) into ( [ 8 ] ) and taking integral over @xmath50 we obtain for the spectral distribution of the probability of radiation @xmath51 where @xmath52 here and below we consider an expression @xmath53 as a limit : @xmath54 of @xmath55 . now we estimate effective impact parameters @xmath56 which give the main contribution into radiation probability . since characteristic values of @xmath56 will be found straightforwardly at calculation of ( [ 11 ] ) , we estimate characteristic angles @xmath57 connected with @xmath56 by an equality @xmath58 . the mean square scattering angle of a particle on the formation length of a photon @xmath59 ( [ 1 ] ) has the form @xmath60 when @xmath61 the contribution in the probability of radiation gives a region where @xmath62 , in this case @xmath63 . when @xmath64 the characteristic angle of radiation is determined by self - consistency arguments : @xmath65 it should be noted that when characteristic impact parameter @xmath56 becomes smaller than a radius of nucleus @xmath66 , the potential @xmath39 acquires an oscillator form ( see appendix b , eq.([b3 ] ) ) @xmath67 allowing for estimates ( [ 14 ] ) we present the potential @xmath39 ( [ 9 ] ) in the following form @xmath68 the inclusion of the coulomb corrections @xmath69 and -1 into @xmath70 diminishes effectively the correction @xmath71 to the potential @xmath72 . in accordance with such division of the potential we present propagators in expression ( [ 11 ] ) as @xmath73 where @xmath74 this representation of the propagator @xmath75 permits one to expand it over `` perturbation '' @xmath76 . indeed , with an increase of @xmath77 the relative value of the perturbation is diminished @xmath78 since effective impact parameters diminish and , correspondingly , the value of logarithm @xmath2 in ( [ 16 ] ) increases . the maximal value of @xmath2 is determined by a size of a nucleus @xmath66 @xmath79 where @xmath80 . so , one can to redefine the parameters @xmath21 and @xmath81 to include the coulomb corrections . the matrix elements of the operator @xmath82 could be calculated explicitly . the exponential parametrization of the propagator is @xmath83 below we will use matrix elements of the operator @xmath84 @xmath85 the function @xmath86 satisfies the schrdinger equation ( [ 9 ] ) over each of two ( symmetrical ) variables @xmath87 and @xmath88 with @xmath89 and the initial condition @xmath90 we will seek a solution in the form ( see also @xcite ) @xmath91.\ ] ] substituting this expression into ( [ 9 ] ) we find a set of equations for @xmath92 @xmath93 the initial conditions for this set follows from definition ( [ 20 ] ) : @xmath94 where @xmath95 . from ( [ 20a ] ) one has the initial conditions at @xmath96 @xmath97 the solution of the set ( [ 22 ] ) satisfying these initial conditions is @xmath98 where @xmath99 . as a result , we obtain the following expression for the sought function @xmath100 \right\}. \label{24}\ ] ] substituting formulae ( [ 19 ] ) and ( [ 24 ] ) in the expression for the spectral distribution of the probability of radiation ( [ 11 ] ) we have @xmath101 } , \end{array } \label{25}\ ] ] where @xmath102 . this formula gives the spectral distribution of the probability of radiation derived by migdal @xcite . however , here coulomb corrections are included into parameter @xmath103 in contrast to @xcite . we now expand the expression @xmath104 over powers of @xmath76 @xmath105 substituting this expansion in ( [ 17 ] ) and then in ( [ 11 ] ) we obtain decomposition of the probability of radiation.let us note that for @xmath106 the sum of the probability of radiation @xmath107 ( [ 25 ] ) and the first term of the expansion ( [ 26 ] ) gives the bethe - heitler spectrum of radiation , see below ( [ 35 ] ) . at @xmath108 the expansion ( [ 26 ] ) is a series over powers of @xmath109 . it is important that variation of the parameter @xmath56 by a factor order of 1 has an influence on the dropped terms in ( [ 26 ] ) only . in accordance with ( [ 17 ] ) and ( [ 26 ] ) we present the probability of radiation in the form @xmath110 the probability of radiation @xmath107 is defined by eq.([25 ] ) . in formula ( [ 11 ] ) with allowance for ( [ 17 ] ) there is expression @xmath111 where the matrix element @xmath112 is defined by ( [ 24 ] ) . the term @xmath113 in ( [ 27 ] ) corresponds to the first term ( linear in @xmath76 ) in ( [ 28 ] ) . substituting ( [ 24 ] ) we have @xmath114 \left[r_1+\frac{4q^2\varrho^2}{\nu^2\sinh \nu t_1\sinh \nu t_2 } r_2 \right ] } , \end{array } \label{29}\ ] ] where @xmath99 . substituting in ( [ 29 ] ) the explicit expression for @xmath71 and integrating over @xmath115 and @xmath116 we obtain the following formula for the first correction to the probability of radiation @xmath117},\\ \displaystyle{f_1(z)=\left(\ln \varrho_c^2+\ln \frac{\nu}{i } -\ln \sinh z - c\right)g(z ) - 2\cosh z g(z)},\\ \displaystyle{f_2(z ) = \frac{\nu}{\sinh z } \left(f_1(z)-\frac{g(z)}{2 } \right),\quad g(z)=z\cosh z - \sinh z},\\ \displaystyle{g(z)=\int_{0}^{z}\left(1-y\coth y\right)dy,\quad t = t_1+t_2,~z=\nu t } \end{array } \label{30}\ ] ] as it was said above ( see ( [ 14 ] ) , ( [ 18 ] ) ) , @xmath63 at @xmath118 . if the parameter @xmath119 , the value of @xmath56 is defined from the equation ( [ 14 ] ) , where @xmath120 , up to @xmath121 . then one has @xmath122 it follows from ( [ 31 ] ) that expression ( [ 14 ] ) for @xmath123 , which we chose a priori , corresponds to the mean value of @xmath124 . from the above analysis we have that the factor at @xmath125 in expression for @xmath126 in ( [ 30 ] ) can be written in the form @xmath127 where @xmath128 @xmath129 is the heaviside step function . when a scattering is weak ( @xmath130 ) , the main contribution in ( [ 30 ] ) gives a region where @xmath131 . then @xmath132 the corresponding asymptotes of the function @xmath133 ( [ 25 ] ) is @xmath134 combining the results obtained ( [ 33 ] ) and ( [ 34 ] ) we obtain the spectral distribution of the probability of radiation in the case when scattering is weak @xmath135 @xmath136}\\ \displaystyle{=\frac{\alpha}{2\pi \gamma^2}\frac{2q}{3}\left[r_1\left ( l_1-\frac{1}{3 } \right)+2r_2\left(l_1+\frac{1}{6 } \right ) \right]}\\ \displaystyle{=\frac{4z^2\alpha^3n}{3m^2\omega}\bigg[\frac{\omega^2 } { \varepsilon^2 } \left(\ln \left(183z^{-1/3 } \right)-\frac{1}{6 } -f(z\alpha)\right)}\\ \displaystyle{+2\left(1+\frac{\varepsilon'^2}{\varepsilon^2 } \right ) \left(\ln \left(183z^{-1/3 } \right)+\frac{1}{12}-f(z\alpha ) \right ) \bigg ] } , \end{array } \label{35}\ ] ] where @xmath137 is defined in ( [ 18 ] ) . this expression coincide with the known bethe - heitler formula for probability of bremsstrahlung from high - energy electrons in the case of complete screening ( if one neglects the contribution of atomic electrons ) written down within power accuracy ( omitted terms are of the order of powers of @xmath138 ) with the coulomb corrections , see e.g. eq.(18.30 ) in @xcite , or eq.(3.83 ) in @xcite . the integral in the function @xmath139 ( [ 30 ] ) which defines the first correction to the probability of radiation ( [ 30 ] ) can be transformed into the another form containing the real functions only @xmath140,\quad d_2(\nu_0)=\int_{0}^{\infty}\frac{dz e^{-sz}}{\sinh^3 z}}\\ \displaystyle{\times \left\{\left[d(z)- \frac{1}{2}g(z)\right]\left(\sin sz+\cos sz\right ) + \frac{\pi}{4}g(z)\left(\cos sz - \sin sz\right)\right\}},\\\\ \displaystyle{d(z)=(\ln \nu_0 \vartheta(1-\nu_0)-\ln \sinh z -c)g(z ) -2\cosh z g(z ) } , \end{array } \label{36}\ ] ] where the functions @xmath125 and @xmath141 are defined in ( [ 30 ] ) . the form ( [ 36 ] ) is convenient for numerical calculations . note , that parameter @xmath142 in ( [ 36 ] ) is two times larger than used by migdal @xcite . at @xmath143 the function @xmath144 ( see ( [ 30 ] ) and ( [ 32 ] ) ) can be written in the form @xmath145}. \label{37}\ ] ] integrating over @xmath146 we obtain @xmath147 under the same conditions ( @xmath143 ) the function @xmath148 ( [ 25 ] ) is @xmath149 thus , at @xmath143 the relative contribution of the first correction @xmath113 is defined by @xmath150 where @xmath151 . in the above analysis we did not consider an inelastic scattering of a projectile on atomic electrons . the potential @xmath152 connected with this process can be found from formula ( [ 9a ] ) by substitution @xmath153 ( an analysis of an inelastic scattering on atomic electrons as well as the parameter @xmath154 can be found in @xcite ) . the summary potential including both an elastic and an inelastic scattering is @xmath155}\\ \displaystyle{=-q_{ef}\mbox{\boldmath$\varrho$}^2 \big(\ln \gamma^2 \vartheta_{ef}^2 + \ln \frac{\mbox{\boldmath$\varrho$}^2}{4}+2c \big ) } , \end{array } \label{40a}\ ] ] where @xmath156}.\ ] ] when one considers bremsstrahlung of enough soft photons @xmath157 , one has to take into account the effect of a polarization of the medium . in a dense medium the velocity of a photon propagation differs from the light velocity in the vacuum since the index of refraction @xmath158 @xmath159 because of this the formation length diminishes as well as the probability of radiation ( see @xcite , the qualitative discussion may be found in @xcite ) . for analysis we use the general expression for the probability of radiation , see eq.(2.1 ) , @xcite . the factor in front of exponent in this expression ( see eq.(2.2 ) , @xcite ) contains two terms @xmath160 and @xmath161 , the term @xmath160 is not changed and the term @xmath161 contains combination @xmath162 and its dependence on @xmath163 ( term of the order @xmath36 ) may be neglected also . with regard for the polarization of a medium the formation length ( [ 1 ] ) acquires a form @xmath164^{-1}. \label{42a}\ ] ] so , the dependence on @xmath165 manifests itself in the exponent of eq.(2.1 ) , @xcite and respectively in the exponent of ( [ 2 ] ) only : @xmath166 performing the substitution @xmath167 in formula ( [ 7 ] ) we obtain for the potential ( [ 16 ] ) @xmath168 the substitution ( [ 43 ] ) in the expression for the probability of radiation ( [ 8 ] ) gives @xmath169 the value of the parameter @xmath170 in ( [ 43a ] ) is determined by equation ( compare with eq.([31 ] ) ) @xmath171 in the opposite case @xmath172 and this is possible in two intervals of the photon energy @xmath6 : 1 . for @xmath173 when the multiple scattering and effects of the polarization of a medium are weak ; 2 . for @xmath174 when effects of the polarization of a medium become stronger then effects of the multiple scattering ( @xmath175 ) . in an intermediate region we substitute @xmath176 in eq.([44a ] ) . after it we obtain the equation for @xmath56 which coincides with eq.([31 ] ) , see also ( [ 32a ] ) : @xmath177 thus , for @xmath178 we have @xmath179 while for @xmath180 we have @xmath181 the spectral distribution of the probability of radiation ( [ 35 ] ) with allowance for polarization of a medium have the form @xmath182 } , \label{45}\ ] ] where @xmath183 we consider now the case when an influence the polarization of a medium manifests itself in the conditions of the strong lpm effect @xmath184 . this influence becomes essential for low energy photons , when the mean square angle of the multiple scattering ( [ 14 ] ) on the formation length of a photon becomes smaller than @xmath185 . indeed , in the case @xmath186 one can use asymptotes of functions @xmath133 and @xmath144 at @xmath143 ( see ( [ 37 ] ) , ( [ 39 ] ) ) , we have @xmath187 in the opposite case @xmath188 , the characteristic momentum transfer in the used units ( @xmath189 ) are defined by value @xmath190 , one can use asymptotic expansions ( [ 33 ] ) and ( [ 34 ] ) and we have for the spectral distribution of the probability of radiation @xmath191 where @xmath69 is defined in ( [ a10 ] ) , @xmath192 . the results obtained agree with given in @xcite where calculations are fulfilled within logarithmic accuracy and without coulomb corrections . it is seen that a dependence of spectral distribution on photon energy ( @xmath193 ) differs essentially from the bethe - heitler one ( @xmath194 ) , the probability is independent on density @xmath14 . the formula ( [ 47 ] ) is applicable only up to value @xmath195 or if @xmath196 , where @xmath197 for example , for electrons with energy @xmath198 and gold target ( @xmath199 ) one has @xmath200 . for @xmath201 one has take into account the form factor of a nucleus ( see appendix b ) . in this case the argument of the logarithm in ( [ 48 ] ) ceases its dependence on photon energy @xmath6 . in the limit @xmath202 the spectral distribution of the probability of radiation is @xmath203 in the case when a finiteness of a target is essential the probability of radiation is defined not only by the relative time @xmath204 as in section 2 . the used radiation theory is formulated in terms of two times ( see eqs.(2.1 ) - ( 2.3 ) of @xcite ) . proceeding from this formulation we can obtain more general expression which takes into account boundary effects . with allowance for polarization of a medium we have for the spectral distribution of the probability of radiation @xmath205 } , \end{array } \label{50}\ ] ] where @xmath206 here @xmath207 is the thickness of a target , @xmath163 is defined in ( [ 42 ] ) . so , we split time interval ( in the used units ) into three parts : before target ( @xmath208 ) , after target ( @xmath209 ) and inside target ( @xmath210 ) . the functions @xmath211 satisfy the equation ( [ 9 ] ) , but now the potential @xmath212 depends on time @xmath213 using an operator form of a solution of eq . ( [ 52 ] ) ( compare with ( [ 10 ] ) ) we can present the probability ( [ 50 ] ) in the form @xmath214 } , \end{array } \label{53}\ ] ] where the symbol @xmath215 means the chronological product . note , that in ( [ 50 ] ) and ( [ 53 ] ) it is implied that subtraction is made at @xmath38 , @xmath216 . integrals over time in ( [ 53 ] ) we present as integrals over four domains : 1 . @xmath217 ; 2 . @xmath218 ; 3 . @xmath219 ; 4 . @xmath220 ; in two more domains @xmath221 and @xmath222 an electron is moving entirely free and there is no radiation . we consider in this section the case , when the thickness of a target @xmath2 is much larger than formation length @xmath223 ( [ 42a ] ) or @xmath224 . in this case domain 4 ) does nt contribute . the contributions of other domains are @xmath225 where @xmath226 . the term in @xmath227 : @xmath228 describes the probability of radiation considered in previous sections . all other terms define the probability of radiation of boundary photons . so , making mentioned subtraction we have for the spectral distribution of the probability of radiation of boundary photons @xmath229 for a convenience here we made the subtraction in two stages : first ( in @xmath230 ) we subtracted terms with @xmath38 and second ( in @xmath231 ) we subtracted terms with both @xmath38 and @xmath232 . we consider important case when both the lpm effect and the polarization of a medium are essential . we will calculate the main term with @xmath233 , see ( [ 16 ] ) . needed combinations are @xmath234 } , \end{array } \label{56}\ ] ] where the functions @xmath235 and @xmath236 are defined in ( [ 20a ] ) and ( [ 24 ] ) . substituting into ( [ 56 ] ) the explicit expressions for these functions , calculating the vector derivatives as indicated in ( [ 55 ] ) we have for contribution of the first term in ( [ 56 ] ) @xmath237}\\ \displaystyle{=-\frac{2\alpha}{\pi \omega}r_2~{\rm im}~ \nu \int_{0}^{\infty } dt_1 \int_{0}^{\infty } dt_2 \exp \left(-i(t_1+t_2 ) \right ) \left[\frac{1}{\tanh \tilde{\nu}t_2+\nu t_1}- \frac{1}{\tilde{\nu}t_2+\nu t_1 } \right ] } , \end{array } \label{57}\ ] ] where @xmath238 , the second term in the square brackets is the subtraction term in accordance with ( [ 55 ] ) ( the term @xmath239 ) . for practical use it is convenient to write the probability ( [ 57 ] ) using real variables . after some transformations it can be written as @xmath240 } , \end{array } \label{59}\ ] ] where @xmath241 , parameter @xmath242 is defined in ( [ 32a ] ) . repeating the same operations with the second term in ( [ 56 ] ) ( this is the contribution of the term @xmath243 in ( [ 55 ] ) ) we have @xmath244}\\ \displaystyle{=\frac{\alpha}{\pi \omega}r_2~ { \rm re}~ \int_{0}^{\infty } dz \exp \left(-i\frac{z}{\tilde{\nu}}\right ) \left[\frac{z}{\sinh^2 z}-\frac{1}{z } \right]}\\ \displaystyle{=\frac{\alpha}{\pi \omega}r_2 \int_{0}^{\infty}dz \exp \left(-\tilde{s}z \right ) \cos \tilde{s}z \left[\frac{z}{\sinh^2 z}-\frac{1}{z } \right ] } , \end{array } \label{60}\ ] ] where @xmath245 . the contribution of the term @xmath231 in ( [ 55 ] ) is calculated in ( [ 56 ] ) @xmath246\right\}}. \label{61}\ ] ] the complete expression for the spectral distribution of the probability of radiation of boundary photons , in the case when both the lpm effect and the polarization of a medium are taken into account , is @xmath247 we consider now the limiting case when lpm effect is very strong ( @xmath248 ) . in this case we find for probabilities in formulae ( [ 59 ] ) and ( [ 60 ] ) @xmath249},\\ \displaystyle{\frac{dw_b^{(2)}}{d\omega}=\frac{\alpha}{\pi \omega}r_2 \left[1-\ln 2\tilde{\nu}_0 + c-\frac{\pi^2}{6\sqrt{2}\tilde{\nu}_0 } \right]}. \end{array } \label{63}\ ] ] substituting asymptotes obtained and ( [ 61 ] ) into ( [ 62 ] ) we have @xmath250\right\}}. \label{64}\ ] ] as one can expect , the probability of radiation at @xmath251 depends on the polarization of a medium in the term @xmath252 only . in the opposite case @xmath253 , the probabilities @xmath254 and probability of radiation of boundary photons is determined by the polarization of a medium . just in this case radiation of boundary photons is known as the transition radiation : @xmath255\right\}}. \label{65}\ ] ] in the case of weak lpm effect @xmath256 we have @xmath257 in this case what we calculated as the boundary photons contribution is actually correction ( very small ) to the probability @xmath258 ( [ 35 ] ) which in this case has additional ( suppression ) factor @xmath259 which follows from the decomposition of the function i m @xmath260 . the lpm effect for the case of structured targets ( with many boundaries ) was analyzed recently in @xcite . the radiation of the boundary photons with regard for the multiple scattering was considered in @xcite ( for @xmath261 ) , the polarization of a medium was added in @xcite and @xcite . our results , which are consistent with obtained @xcite , are presented in more convenient for application form and the coulomb corrections are included . in these papers the probability of radiation of boundary photons ( under condition of applicability of eq.([64 ] ) ) was analyzed also to within the logarithmic accuracy ( see eq.(20 ) in @xcite and eq.(15 in @xcite ) ) . this accuracy is insufficient for parameters connected with experiment @xcite-@xcite . for example , for @xmath198 and heavy elements the value @xmath242 equates @xmath262 for @xmath263 . one can see from eq.([64 ] ) that in this case @xmath264 is nearly completely compensated by constant terms . finally we consider a situation when the formation length of radiation is much larger than the thickness @xmath207 of a target ( a thin target , @xmath265 ) . in this case the radiated photon is propagating in the vacuum and one can neglect the polarization of a medium . operator @xmath266 ( [ 53 ] ) we present in the form @xmath267= \exp \left(-i h_0 t_2 \right ) { \cal l}(t_2 , t_1 ) \exp \left(ih_0 t_1 \right)};\\ \displaystyle{{\cal l}(t_2 , t_1)=\exp \left(ih_0 t_2 \right ) s(t_2 , t_1 ) \exp \left(-ih_0 t_1 \right)}. \end{array } \label{66}\ ] ] differentiating the operator @xmath268 over the first of arguments we obtain @xmath269 where @xmath270 ( see ( [ 9 ] ) , ( [ 52 ] ) ) . the formal solution of this equation with the initial condition @xmath271 has the form @xmath272 , \label{68}\ ] ] where @xmath215 means the chronological product . this solution is exact . now we take into account that we are considering a short characteristic time contributing into integral ( [ 68 ] ) , or more precisely @xmath273 where @xmath274 are defined in ( [ 1 ] ) . since the main contribution give @xmath275 , where @xmath276 is characteristic mean value of operator @xmath277 , one can neglect by the term @xmath278 in ( [ 68 ] ) , so that @xmath279 . \label{70}\ ] ] in the probability of radiation enters the expression ( cp ( [ 11 ] ) , ( [ 53 ] ) ) @xmath280 using an explicit form ( [ 20a ] ) of the matrix element @xmath281 and neglecting terms of the order @xmath282 one obtains starting from ( [ 53 ] ) for the spectral distribution of the probability of radiation @xmath283\left(\exp ( -vt ) -1 \right)}\\ \displaystyle{=\frac{\alpha}{\pi^2 \omega } \int_{}^ { } d^2\varrho \left[r_1 k_0 ^ 2(\varrho)+r_2 k_1 ^ 2(\varrho ) \right ] \left(1-\exp ( -vt ) \right ) } , \end{array } \label{72}\ ] ] where @xmath284 is the operator @xmath285 acting on the function of @xmath286 , @xmath287 is the modified bessel function . here we took into account that in our case contribute domain @xmath288 and @xmath289 since in domains @xmath221 and @xmath222 an electron is moving entirely free and there is no radiation . in implicit form the factorization contained in ( [ 72 ] ) is presented in @xcite . if @xmath290 one can expand the exponent ( the contribution of the region @xmath291 is exponentially damped because in this region @xmath292 ) . in the first order over @xmath293 using the explicit expression for the potential ( [ 9a ] ) we have to calculate following integrals : @xmath294 substituting these integrals one obtains in this case the bethe - heitler formula with the coulomb corrections ( [ 35 ] ) . we analyze now the opposite case when the multiple scattering of a particle traversing a target is strong ( @xmath295 , the mean square of multiple scattering angle @xmath296 ) . we present the function @xmath297 ( see ( [ 9 ] ) , ( [ 9a ] ) and ( [ 18 ] ) ) as @xmath298 where @xmath299 is the lower boundary of values contributing into the integral over @xmath300 . substituting this expression into ( [ 72 ] ) we have the integral @xmath301 \right\ } \equiv \pi j}. \label{74}\ ] ] in this integral we expand the exponent in the integrand over @xmath302 keeping the first term of the expansion . we find @xmath303 \varrho d\varrho}\\ \displaystyle{=2k\int_{0}^{\infty}d\varrho \varrho^3\left[k_0(\varrho ) k_2(\varrho)-k_1 ^ 2(\varrho ) \right ] \exp \left(-k\varrho^2\right)},\\ \displaystyle{j_2=-\frac{2k}{l_t}\int_{0}^{\infty } k_1 ^ 2(\varrho ) \exp \left(-k\varrho^2\right)\ln \frac{\varrho^2}{\varrho_t^2 } \varrho^3 d\varrho } \end{array } \label{75}\ ] ] in the integral @xmath304 we performed an integration by parts . in the integrals in ( [ 75 ] ) it is convenient to substitute @xmath305 then @xmath306\exp(-z ) zdz},\\ \displaystyle{j_2=-\frac{1}{kl_t}\int_{0}^{\infty } k_1 ^ 2\left(\sqrt{\frac{z}{k}}\right ) \exp(-z ) \ln z zdz}. \end{array } \label{76}\ ] ] expanding the modified bessel functions @xmath307 at @xmath308 and taking the integrals in the last expression we have @xmath309 in the term with @xmath310 in ( [ 72 ] ) the region @xmath311 contributes . so we have @xmath312 substituting found @xmath313 and @xmath314 into ( [ 72 ] ) we obtain for the spectral distribution of the probability of radiation in a thin target at conditions of the strong multiple scattering @xmath315 the logarithmic term in this formula is well known in theory of the collinear photons radiation at scattering of a radiating particle on angle much larger than characteristic angles of radiation @xmath316 . it is described with logarithmic accuracy in a quasi - real electron approximation ( see @xcite , appendix b2 ) . the formula ( [ 72 ] ) presents the probability of radiation in the case when the formation length @xmath265 . it is known , see e.g. @xcite , that in this case a process of scattering of a particle is independent of a radiation process and a differential probability of radiation at scattering with the momentum transfer @xmath317 can be presented in the form @xmath318 where @xmath319 is the differential probability of scattering with the momentum transfer @xmath317 which depends on properties of a target . the function @xmath320 is the probability of radiation of a photon with a momentum @xmath321 when an emitting electron acquires the momentum transfer @xmath317 . this probability has a universal form which is independent of properties of a target . for an electron traversing an amorphous medium this fact is reflected in formula ( [ 72 ] ) . indeed , passing on to a momentum space we have @xmath322 \left(1-\exp ( -i{\bf q}\mbox{\boldmath$\varrho$ } ) \right)}\\ \displaystyle{=\frac{\alpha d\omega}{\pi \omega}\left[r_1 f_1\left ( \frac{q}{2}\right)+r_2 f_2\left(\frac{q}{2}\right ) \right]};\\ \displaystyle{f_1(x)=1-\frac{\ln \left(x+\sqrt{1+x^2 } \right)}{x\sqrt{1+x^2 } } , \quad f_2(x)=\frac{2x^2 + 1}{x\sqrt{1+x^2}}\ln \left(x+\sqrt{1+x^2 } \right)-1}. \end{array } \label{81}\ ] ] remind that @xmath77 is measured in electron mass . the probability of radiation in this form was found in @xcite . for a differential probability of scattering ( here we consider the multiple scattering ) there is a known formula ( cp ( [ 5 ] ) , ( [ 6 ] ) and ( [ 9 ] ) ) @xmath323 where @xmath324 is the cross section of single scattering . using the formula ( [ 81 ] ) one can easily obtain to within logarithmic accuracy expressions ( [ 79]),([64 ] ) . both a radiation of boundary photons and a radiation in a thin target may be considered as a radiation of collinear photons ( see e.g. @xcite ) in the case when an emitting particle deviates at large angle ( @xmath325 ) . using ( [ 81 ] ) at @xmath326 we find @xmath327};\\ \displaystyle{\int_{}^{}d^2q dw_r(q ) f_s({\bf q } ) \simeq \frac{\alpha d\omega}{\pi \omega } \left[r_1+r_2\left(\ln \overline{q^2 } -1 \right ) \right]}. \end{array } \label{83}\ ] ] for a thin target value of @xmath328 is defined by mean square of multiple scattering angle on a thickness of a target @xmath207 , and for boundary photons is the same but on the formation length @xmath223 . however , if we one intends to perform computation beyond a logarithmic accuracy , the method given in this section has advantage since there is no necessity to calculate @xmath329 and in our approach a problem of calculation of the coulomb corrections is solved in a rather simple way . it appears that used in section 4 approach permits one to consider an important case when @xmath330 . according to the partition of integrals over time in formula ( [ 53 ] ) into four domains we can write the probability of radiation as @xmath331 the integrals in @xmath332 we compute on the assumption : @xmath333 . since integrals in @xmath334 do nt contain the logarithmic divergence , only the domain 4 contributes . in the domains 1 - 3 one of the integrals in @xmath334 contains an integration over an interval @xmath210 and due to this reason @xmath335 . so , we consider @xmath336 @xmath337 k_0(\mbox{\boldmath$\varrho$}_2 , 0 , t_2)}. \end{array } \label{85}\ ] ] here a calculation of integrals over @xmath87 and @xmath88 may be performed e.g. in a such way : * an integral over relative angle between @xmath87 and @xmath88 gives @xmath338 where @xmath339 is the bessel function , @xmath340 and @xmath341 for the first and second terms in the square brackets in the right - hand side of ( [ 85 ] ) , * the remaining integrals over @xmath342 and @xmath343 can be found in tables . so , we have @xmath344},\\ \displaystyle{n(t_1 , t_2)=\frac{\nu}{(1+\nu^2t_1t_2)\sinh \nu t + \nu ( t_1+t_2)\cosh \nu t}}. \end{array } \label{86}\ ] ] for @xmath143 the contribution into integral the term with @xmath345 is of the order of @xmath346 and this term may be neglected . with allowance for @xmath347 we find @xmath348 where @xmath349 . the contribution of the domain 4 into the term with @xmath350 ( @xmath351 in ( [ 84 ] ) ) contain two additional operators @xmath352 ( see ( [ 53 ] ) ) which result additional factor @xmath353 in the integrand . integration over the relative angle between @xmath87 and @xmath88 gives here @xmath354 and subsequent evaluation of integrals is similar to those for ( [ 86 ] ) . we find @xmath355}. \label{88}\ ] ] the contribution into the integral with @xmath356 gives a domain @xmath357 . since @xmath347 as well , we can put an exponent in this integral equal to 1 . so the integral is @xmath358 where @xmath359 . the second term in ( [ 88 ] ) is calculated as ( see ( [ 87 ] ) ) @xmath360 where @xmath361 . putting together ( [ 89 ] ) and ( [ 90 ] ) we have @xmath362 for computation of @xmath363 we will use eq.([57 ] ) @xmath364 } \label{92}\ ] ] integrating by parts over @xmath365 with regard for @xmath366 we have @xmath367}\\ \displaystyle{+\frac{\nu}{4\pi i}\int_{0}^{\infty}dt_1\int_{0}^{t}dt_2 \exp ( -it_1 ) \left[\frac{1}{\nu(t_1+t_2)}- \frac{1}{\cosh \nu t_2(\sinh \nu t_2+\nu t_1\cosh \nu t_2 ) } \right]}. \end{array } \label{93}\ ] ] the second of these integrals is proportional to @xmath368 and can be neglected . the first integral gives @xmath369 for a calculation of @xmath370 we use formulae ( [ 54 ] ) and ( [ 60 ] ) @xmath371}\\ \displaystyle{=\frac{\nu}{4\pi } \int_{0}^{t}dt\left[\coth \nu t- \frac{1}{\nu t}\right]=\frac{1}{4\pi}\ln \frac{\sinh \nu t}{\nu t}}. \end{array } \label{95}\ ] ] combining all the contributions of four domains we obtain finally @xmath372}. \label{96}\ ] ] in the used units ( @xmath373 ) the formation length ( [ 1 ] ) is ( see also ( [ 5 ] ) and ( [ 14 ] ) ) @xmath374 in the case of thick target ( @xmath375 ) we have from ( [ 96 ] ) @xmath376+\frac{\alpha t}{\pi \omega } r_2 \frac{\nu _ 0}{\sqrt{2}},~\frac{\alpha t}{\pi \omega } = \frac{\alpha al}{2\pi \omega}=\frac{\alpha}{2\pi \gamma^2 } \frac{\varepsilon}{\varepsilon'}l}. \label{98}\ ] ] this formula gives the probability of radiation at @xmath143 ( see ( [ 25 ] ) , ( [ 39 ] ) ) where the contribution of boundary photons ( [ 64 ] ) is included . in the case of thin target @xmath377 but when @xmath378 we have from ( [ 96 ] ) the probability ( [ 79 ] ) without term @xmath379 . so we have @xmath380 @xmath381}. \label{99}\ ] ] note , that when the value of the parameter @xmath242 is not very large , the accuracy of the formulae ( [ 98 ] ) and ( [ 99 ] ) may be insufficient . in this case one have to compute the next terms of the expansion , as it was done in sections 4 and 5 ( see ( [ 64 ] ) and ( [ 79 ] ) ) . the same is true for ( [ 96 ] ) . a detailed analysis of the probability of radiation in the targets of an intermediate thickness will be carry out elsewhere . we consider the spectral intensity of radiation for the energy of the initial electrons when the lpm suppression of the intensity of radiation takes place for relatively soft energies of photons : @xmath382 : @xmath383 see eqs.([9 ] ) , ( [ 13 ] ) , ( [ 14 ] ) and ( [ 32a ] ) . this situation corresponds to the experimental conditions . a ratio of a thickness of a target and the formation length of radiation ( [ 1 ] ) is an important characteristics of the process . if we take into account the multiple scattering and the polarization of a medium then the formation length ( [ 42a ] ) has the form @xmath384^{-1 } , \label{100a}\ ] ] this ratio may be written as @xmath385},\\ \displaystyle{t=\frac{l \omega}{2\gamma^2},\quad \omega_p=\omega_0 \gamma,\quad t_c \equiv t(\omega_c ) \simeq \frac{2\pi}{\alpha}\frac{l}{l_{rad } } } , \end{array } \label{101}\ ] ] where we put that @xmath386 . below we assume that @xmath387 which is true under the experimental conditions . if @xmath388 then at @xmath389 a target is thin and the bethe - heitler spectrum of radiation , which is valid at @xmath390 will be also valid at @xmath391 in accordance with eqs.([72 ] ) and ( [ 72a ] ) since @xmath392 . this behavior of the spectral curve will continue with @xmath6 decrease until photon energies where a contribution of the transition radiation become essential . if @xmath393 then at @xmath394 a target is thick and one has the lpm suppression for @xmath395 . there are two opportunities depending on the minimal value of the parameter @xmath396 . @xmath397 if @xmath398 then for photon energies @xmath399 it will be @xmath400 such that @xmath401 and for @xmath402 the thickness of a target becomes smaller than the formation length of radiation so that for @xmath403 the spectral distribution of the radiation intensity is described by formulae of section 5 . under these conditions for @xmath404 the spectral curve has a plateau @xmath405 in accordance with ( [ 77 ] ) . such behavior of the spectral curve ( first discussed in @xcite ) will continue until photon energies where one has to take into account the polarization of a medium and connected with it a contribution of the transition radiation . at @xmath406 a target remains thick for all photon energies and radiation is described by formulae of sections 2 and 3 . in this case at @xmath407 and @xmath408 the spectral intensity of radiation formed inside a target is given by exp.([35 ] ) and ( [ 39 ] ) and the contribution of the boundary photons is given by ( [ 64 ] ) . since a contribution into the spectral intensity of radiation from a passage of an electron inside of a target ( @xmath409 ) is diminishing and a contribution of the boundary photons is increasing with @xmath6 decrease , the spectral curve has a minimum at @xmath410 . the value of @xmath411 may be estimated from equation @xmath412 when a value of @xmath413 is high enough , the solution of eq.([104 ] ) does nt satisfy the condition @xmath414 and in this case the equation ( [ 104 ] ) ceases to be valid . for determination of @xmath411 in this case we use the behavior of the spectral intensity of radiation at @xmath415 . in this case a contribution into radiation from inside passage of a target is described by ( [ 47 ] ) whilst the radiation of the boundary photons reduces to the transition radiation and its contribution is given by ( [ 61 ] ) . leaving the dominant terms ( @xmath416 is @xmath6-independent ) we have @xmath417 since the value @xmath418 is of the order of unity , the solution of ( [ 104 ] ) at @xmath419 differs only slightly from @xmath411 . because of this , if the condition @xmath420 is fulfilled , the position of the minimum is defined by eq.([104 ] ) . now we consider the experimental data @xcite-@xcite from a point of view of the above analysis . it is shown that the mechanism of radiation depends strongly on the thickness of a target . so , we start with an estimate of thickness of used targets in terms of the formation length of radiation . from eq.([101 ] ) we have that @xmath421 the minimum value of the ratio of a thickness of a target to the formation length of radiation is given by eq.([102 ] ) ( @xmath422 ) . for defined value of @xmath413 this ratio is least of all for the heavy elements . indeed , the value of @xmath423 depends weakly on nucleus charge @xmath424 ( @xmath425 ) , while @xmath426 . furthermore , the ratio @xmath427 decreases with energy increase . thus , among all targets with thickness @xmath428 the minimal value of @xmath429 is attained for the heavy elements ( w , au , u ) at the initial electron energy @xmath198 . in this case one has @xmath430 . since the parameter @xmath413 is energy independent and the ratio @xmath431 , the minimal value @xmath432 is attained at the initial electron energy @xmath433 for all targets with thickness @xmath428 which can be considered as thick targets . as an example of obtained results we calculated the spectrum of the intensity of radiation in the tungsten target with thickness @xmath434 at the initial electron energy @xmath433 and @xmath198 . the characteristic parameters of the radiation process for this case are given in the table . we calculated the main ( migdal ) term ( eq.([25 ] ) ) , the correction term ( eqs.([30]),([36 ] ) ) taking into account an influence of the polarization of a medium according to ( eq.([45 ] ) ) , as well as coulomb corrections entering the parameters @xmath242 ( eq.([9a ] ) ) and @xmath435 ( eq.([32 ] ) ) . the contribution of an inelastic scattering of a projectile on atomic electrons ( quite small for the heavy elements ) is not included although this could be done using eq.([40a ] ) . we calculated also the contribution of the boundary photons eq.([62 ] ) . here in the soft part of the spectrum @xmath436 for @xmath198 ) the transition radiation term ( [ 61 ] ) dominates in ( [ 62 ] ) , whilst in the harder part of the boundary photon spectrum @xmath437 the terms depending on both the multiple scattering and the polarization of a medium ( [ 59 ] ) and ( [ 60 ] ) give the main contribution ; for @xmath433 we have @xmath438 . it is seen that we have for the boundary photons spectrum a smooth curve which eliminate difficulties mentioned in @xcite . all these results presented separately in fig.2 as well as their sum ( curve 5 ) . note , that for energy @xmath198 in the region of the minimum of the spectral curve 5 where the ratio of the target thickness to the formation length is minimal ( @xmath439 , see table ) it may be that the target is not thick enough to use the formulae for a thick target . for a comparison with experiment we extract some data from fig.7 of @xcite . the theoretical curve gives the spectral distribution of the intensity of radiation ( in units @xmath440 ) without adjusting parameters . data from @xcite were recalculated according with procedure given in it . one can see that agreement between the experiment and theory is rather satisfactory but far from being perfect . however , one has to take into account that the theory of lpm effect in all previous papers had the logarithmic accuracy and did not contain coulomb corrections . these shortcomings did not permit to pass to the bethe - heitler formula with acceptable accuracy and led to some difficulties in data processing . both these shortcomings are overcome in the present paper . so , in our opinion , it is quite desirable to handle the experimental data using the formulae of this paper . the measurements in @xcite were made also using gold target with thickness @xmath441 . for this case one has @xmath442 , so we have here a target of an intermediate thickness ( see section 6 ) . we want to stress once more that for estimation of an effective thickness one have to use the formation length with regard for the multiple scattering and the polarization of a medium ( see ( [ 42a ] ) and ( [ 100a ] ) ) . a detailed calculation for this case will be published elsewhere . 0.6 cm it is well known , that for heavy elements the coulomb correction to the cross section of bremsstrahlung of high energy particles ( correction to the born approximation ) is quite sizable , see e.g. eq.(18.30 ) in @xcite . the coulomb correction ( order of one for heavy elements ) is subtracted from the `` large '' logarithm and if an accuracy of calculation goes beyond logarithmic one , one has to take into account this correction . for tungsten @xmath443 , gold @xmath444 and uranium @xmath445 in the case of complete screening the relative coulomb corrections to the standard bethe - heitler cross section are respectively -7.5% , -8.3% and -10.7% . we consider the problem using eikonal approximation ( see e.g.appendix e in @xcite ) . an amplitude @xmath446 and a cross section of scattering in this approximation have the form : @xmath447 where @xmath448 are the longitudinal and transverse coordinates respectively , @xmath449 is the potential . repeating a derivation made in section 2 ( eqs . ( [ 3])-([9 ] ) ) but with the cross section ( [ a1 ] ) we find for the potential @xmath39 @xmath450 since the potential @xmath39 was calculated above in the born approximation , we can calculate here the difference of the potentials calculated in the born approximation @xmath451 and in the eikonal approximation @xmath39 @xmath452 -1 \\ + \frac{1}{2}\left[\chi(\mbox{\boldmath$\varrho$}+{\bf x})- \chi(x ) \right]^2 \big\};\\ \displaystyle{\chi(x)=\int_{-\infty}^{\infty } dz \frac{z\alpha}{r } \exp \left(-\frac{r}{a_s}\right ) = 2z\alpha k_0\left(\frac{x}{a_s}\right),\quad r=\sqrt{z^2+x^2 } } , \end{array } \label{a3}\ ] ] where @xmath453 is the modified bessel function . because the eikonal phase enters ( [ a3 ] ) only in the combination @xmath454 in the interesting for us region @xmath455 . since @xmath456 is large only if @xmath457 it is evident that main contribution into integral ( [ a3 ] ) gives the region @xmath458 . in this region one has @xmath459 so , in the expression for @xmath460 enters only one dimensional parameter @xmath461 , where @xmath462 is the unit vector . after substitution of variables @xmath463 we have @xmath464}. \label{a5}\ ] ] changing variables @xmath465 and then @xmath466 we have @xmath467}. \label{a6}\ ] ] integration over azimuthal angle gives @xmath468 changing the variable @xmath469 , splitting the integration interval into two parts : @xmath470 and @xmath471 changing in the second interval @xmath472 we obtain @xmath473 integrating by parts and changing once more variable @xmath474 we find @xmath475}. \label{a9}\ ] ] integrating once more by parts and using the standard ( gauss ) representation of the euler @xmath476-function we have finally @xmath477 = \xi^2 \sum_{n=1}^{\infty}\frac{1}{n(n^2+\xi^2)}. } \label{a10}\ ] ] the obtained function @xmath478 is the known coulomb correction to the bethe - heitler cross section of bremsstrahlung , see e.g. @xcite , sections 17,18 . 0.6 cm when @xmath479 ( see ( [ 14 ] ) , ( [ 15 ] ) ) one can not consider the potential of a nucleus as a potential of a point charge . a contribution into the multiple scattering gives a momentum transfer @xmath480 . because of the same reason the phase @xmath481 in expression ( [ a2 ] ) for the potential @xmath39 is small @xmath482 and one can expand the potential . as a result we obtain @xmath483 where @xmath484 is the classical momentum transfer on a straight - line trajectory with an impact parameter @xmath15 . as one can see from ( [ b1 ] ) , the mean square of the momentum transfer is the same the in eikonal approximation , in the born approximation and in the classical theory . the coulomb correction in this case vanishes . considering a nucleus as an uniformly charged sphere with the radius @xmath66 , we have @xmath485},\\ \displaystyle{\varphi(x)=1-\sqrt{1-x}+x \sqrt{1-x}}. \end{array } \label{b2}\ ] ] substituting the expression obtained into ( [ b1 ] ) we find the potential @xmath486 under conditions considered @xmath487 = 4\pi \xi^2 \bigg[\bigg(\ln \left(\frac{2a_s}{r_n } \right)^2}\\ \displaystyle{-1 - 2c \bigg)+\bigg(\frac{7}{2}-4 \ln 2 \bigg ) \bigg ] = 4\pi z^2\alpha^2 \left[\ln \frac{a_s^2}{r_n^2}+ \frac{5}{2 } -2(c + \ln 2)\right]};\\ \displaystyle{v(\varrho)=\pi \xi^2 n \varrho^2\left[\ln \left ( \frac{a_s^2}{r_n^2 } \right)-0.0407 \right]}. \end{array } \label{b3}\ ] ] if one uses standard representation of nuclear form factor ( see e.g. @xcite ) @xmath488 then one obtains @xmath489 \simeq 4\pi z^2\alpha^2 \left[\ln \frac{a_s^2}{r_n^2}-0.208 \right]}. \label{b5}\ ] ] taking into account that @xmath490 we see that the difference between different models of nucleus is less than 1 % . 99 l. d. landau and i. ya . pomeranchuk , dokl.akad.nauk sssr * 92 * ( 1953 ) 535 , 735 . see in english in _ the collected paper of l. d. landau _ , pergamon press , 1965 . a. b. migdal , phys . rev . * 103 * ( 1956 ) 1811 . a. b. migdal , sov . jetp * 5 * ( 1957 ) 527 . m. l. ter - mikaelian , _ high energy electromagnetic processes in condensed media _ , john wiley & sons , 1972 . e. l. feinberg and i. ya . pomeranchuk , nuovo cimento , supplement to vol . * 3 * ( 1956 ) 652 . v. m. galitsky and i. i. gurevich , nuovo cimento * 32 * ( 1964 ) 396 . v. n. baier , v. m. katkov and v. s fadin , _ radiation from relativistic electrons _ ( in russian ) atomizdat , moscow , 1973 . v.n.baier , v.m.katkov and v.m.strakhovenko , sov . jetp * 67 * ( 1988 ) 70 . v.n.baier , v.m.katkov and v.m.strakhovenko , _ electromagnetic processes at high energies in oriented single crystals _ , world scientific publishing co , singapore , 1997 . b. g. zakharov , pisma v zhetf * 63 * ( 1996 ) 906 . r.blancenbeckler and s. d. drell , phys.rev . * d53 * ( 1996 ) 6265 . p. l. anthony , r. becker - szendy , p. e. bosted _ et al _ , phys . * 75 * ( 1995 ) 1949 . p. l. anthony , r. becker - szendy , p. e. bosted _ et al _ , phys . * 76 * ( 1996 ) 3550 . p. l. anthony , r. becker - szendy , p. e. bosted _ et al _ , preprint slac - pub-4713 , february 1997 ( submitted to phys.rev.d ) m. p. perl in _ proc . 1994 les rencontres de physique de la vallee daoste _ , ( editions frontieres , gif - sur - yvette , france , 1994 ) . greco , p . b. berestetskii , e. m. lifshitz and l. p. pitaevskii , _ quantum electrodynamics _ pergamon press , oxford , 1982 . s . tsai , rev . * 46 * ( 1974 ) , 815 . v.n.baier , v.m.katkov and v.m.strakhovenko , nucl . phys . * b328 * ( 1989 ) 387 . r.blankenbecler phys.rev * d55 * ( 1997 ) 190 . i. i. goldman sov . jetp * 11 * ( 1960 ) 1341 . f. f. ternovskii sov . jetp * 12 * ( 1960 ) 123 . v. e. pafomov sov . jetp * 20 * ( 1965 ) 253 . b. g. zakharov pisma v zhetf * 64 * ( 1996 ) 737 . v. n. baier , v. s. fadin , v. a. khoze and e.a . kuraev phys . rep . * 78 * ( 1981 ) 293 . * * fig.1 * the functions @xmath491 ( eq.([36 ] ) ) vs parameter @xmath242 . * * fig.2 * the intensity of radiation @xmath492 in tungsten with thickness @xmath493 in units @xmath494 , ( ( a ) is for the initial electrons energy @xmath198 and ( b ) is for @xmath433 ) . the coulomb corrections and the polarization of a medium are included . * * curve 1 is the contribution of the main term ( [ 25 ] ) ; * * curve 2 is the correction ( [ 30 ] ) , ( [ 36 ] ) ; * * curve 3 is the sum of the previous contributions ; * * curve 4 is the contribution of the boundary photons ( [ 62 ] ) ; * * curve 5 is the total prediction for the intensity of radiation .

in this note we show that the lagrangian luttinger surgery preserves the symplectic kodaira dimension . some constraints on lagrangian tori in symplectic four manifolds with non - positive kodaira dimension are also derived .

1108.0479

let @xmath0 be a symplectic 4-manifold with a lagrangian torus @xmath1 . it was discovered by luttinger in @xcite that there is a family of surgeries along @xmath1 that produce symplectic @xmath2-manifolds . this family is countable and indexed by the pairs @xmath3 , k)$ ] , where @xmath4 $ ] is an isotopy class of simple closed curves on @xmath1 and @xmath5 is an integer . when @xmath6 and @xmath7 is the standard symplectic form @xmath8 , he also applied gromov s celebrated work in @xcite to show that , for any lagrangian torus @xmath1 , all the resulting symplectic manifolds are symplectomorphic to @xmath9 . this does not occur in general ; a luttinger surgery often fails even to preserve homology . as a matter of fact , many new exotic small manifolds are constructed via this surgery . in this note , we observe that the luttinger surgery preserves one basic invariant : [ kod ] the luttinger surgery preserves the symplectic kodaira dimension . the symplectic kodaira dimension of a symplectic 4-manifold @xmath10 is defined by the products @xmath11 and @xmath12 $ ] , where @xmath13 is the symplectic canonical class ; if @xmath10 is minimal , then @xmath14 < 0\\ 0 & k_{\omega}^2=0\ and\ k_{\omega}\cdot [ \omega]=0 \\ 1 & k_{\omega}^2=0\ and\ k_{\omega}\cdot [ \omega]>0\\ 2 & k_{\omega}^2>0\ and\ k_{\omega}\cdot [ \omega ] > 0 \end{array}\right.\ ] ] for a general symplectic 4-manifold , the kodaira dimension is defined as the kodaira dimension of any of its minimal models . according to @xcite , @xmath15 is independent of the choice of symplectic form @xmath7 and hence is denoted by @xmath16 . theorem [ kod ] is related to a question of auroux in @xcite ( see remarks [ omega ] and [ 44 ] ) . furthermore , together with the elementary analysis of the homology change , the invariance of @xmath17 implies that [ int2 ] let @xmath0 be a symplectic 4-manifold with @xmath18 and @xmath19 be constructed from @xmath0 via a luttinger surgery . then @xmath20 is symplectomorphic to @xmath10 . for minimal symplectic manifolds of kodaira dimension zero , i.e. , symplectic calabi - yau surfaces , we conclude that the luttinger surgery is a symplectic cy surgery . moreover , together with the homology classification of such manifolds in @xcite , we have [ int3 ] suppose @xmath0 is a symplectic 4-manifold with @xmath21 and @xmath22 . if @xmath19 is constructed from @xmath0 under a luttinger surgery , then @xmath23 and @xmath24 have the same integral homology type . in fact , we conjecture that @xmath24 and @xmath23 in theorem [ int3 ] are diffeomorphic to each other . for symplectic cy surfaces with @xmath25 , the only known examples are torus bundles over torus . we conjecture that they all can be obtained from @xmath26 via lutttinger surgeries ( conjecture [ torus ] ) . theorems [ int2 ] and [ int3 ] provide topological constraints , phrased in terms of topological preferred framings ( see definition [ tprefer ] ) , on the existence of exotic lagrangian tori in such manifolds . [ int4 ] let l be a lagrangian torus in @xmath0 . if @xmath27 , or @xmath1 is null - homologous , @xmath21 and @xmath28 , then the lagrangian framing of @xmath1 is topological preferred . in particular , the invariant @xmath29 in @xcite vanishes whenever it is defined . the organization of this paper is as follows . in section 2 , the construction of the luttinger surgery is reviewed . we also discuss the lagrangian fibrations as the first application of this surgery . in section 3 , we establish the invariance of the symplectic kodaira dimension , which is the main result of this note . in section 4 , we prove theorems [ int2 ] and [ int3 ] . in section 5 , we apply these two theorems to derive constraints on framings of lagrangian tori in symplectic 4-manifolds with non - positive kodaira dimension . we are grateful to the referee for many useful suggestions which improve the exposition . the first author would like to thank the following scholars for their insightful discussions and suggestions : anar akhmedov , inanc baykur , joel gomez , robert gompf , conan leung , weiwei wu and weiyi zhang . the second author is grateful to the support of nsf . in this section , we describe the luttinger surgery following @xcite . applications to lagrangian fibrations are also discussed . we assume all manifolds are oriented . topologically , luttinger surgery is a framed torus surgery . we start with a general description of framed torus surgeries . let @xmath23 be a smooth 4-manifold and @xmath30 an embedded 2-torus with trivial normal bundle . then let @xmath31 be a tubular neighborhood of @xmath1 . if we assume @xmath32 is the complement of @xmath31 , @xmath33 and @xmath34 is a diffeomorphism , a new manifold @xmath24 can be constructed by cutting @xmath31 out of @xmath23 and gluing it back to @xmath35 along @xmath36 via @xmath37 : @xmath38 such surgery is called a @xmath39 @xmath40 . it is often more explicit to describe this process via a framing of @xmath1 . let @xmath23 , @xmath1 , @xmath31 and @xmath36 be given as above . a diffeomorphism @xmath41 is called a framing of @xmath1 if @xmath42 . let @xmath43 be the projection . for any @xmath44 and @xmath45 , the lift @xmath46 of @xmath47 in @xmath36 is called a longitudinal curve of @xmath48 . let @xmath49 be the induced map . two framings @xmath50 are smoothly isotopic to each other if the map @xmath51 is homotopic to the identity map . @xmath52 induces a @xmath53-bundle structure on @xmath36 . a positive oriented fiber @xmath54 of @xmath36 is called a @xmath55 of @xmath1 . for @xmath24 in , we will use @xmath56 to denote the torus @xmath57 . notice that @xmath56 also inherits a framing @xmath58 and its meridian @xmath59 satisfies @xmath60=p[\mu]+k[\gamma_\varphi],\ ] ] in @xmath61 . here @xmath62 is a longitudinal curve of @xmath48 and @xmath63 are coprime integers . the diffeomorphism type of @xmath24 only depends on the class @xmath64 $ ] . it is called a _ generalized logarithmic transform _ of @xmath23 along @xmath65 with multiplicity @xmath66 and auxiliary multiplicity @xmath5 , or of type @xmath67 ( see @xcite ) , and denoted as @xmath68 . for brevity , we will call it a @xmath69-surgery . if @xmath23 is a symplectic 4-manifold , weinstein s theorem states that there is a canonical framing for any lagrangian torus of @xmath23 . let @xmath23 be a symplectic 4-manifoldand and @xmath1 a lagrangian torus of @xmath23 . a framing @xmath48 of @xmath1 is called a lagrangian framing if @xmath70 is a lagrangian submanifold of @xmath23 for any @xmath71 . topologically , a luttinger surgery is a @xmath72-surgery with respect to a lagrangian framing . in order to deal with the sympelctic structure , it is more convenient to use a square neighborhood rather than the disk neighborhood of @xmath1 as above . express the cotangent bundle @xmath73 as @xmath74 equipped with the canonical 2-form @xmath75 and let @xmath76 there exists a tubular neighborhood @xmath31 of @xmath1 and a symplectomorphism @xmath77 for small @xmath78 which satisfies @xmath79 in addition , given a simple closed curve @xmath47 on @xmath1 , we can choose the coordinates @xmath80 of @xmath81 such that @xmath82 let @xmath83 be an annular region and @xmath84 $ ] be a smooth increasing function such that @xmath85 for @xmath86 and @xmath87 for @xmath88 . for any integer @xmath5 , we can define a diffeomorphism @xmath89 : @xmath90 observe that @xmath91 which follows from the relation @xmath92 for @xmath93 . let @xmath94 and define @xmath95 via the following diagram @xmath96 then we can construct a new smooth manifold @xmath97 notice that , by , we have @xmath98 . thus @xmath24 carries a symplectic form @xmath99 induced by @xmath7 . this process is called a _ luttinger surgery _ ( along the lagrangian torus @xmath1 ) . we know that [ framing]@xcite any two lagrangian framings of a lagrangian torus are smoothly isotopic to each other . hence the symplectomorphism type of @xmath100 only depends on the lagrangian isotopy class of @xmath1 , the isotopy class of @xmath47 in @xmath1 , and the integer @xmath5 . therefore , @xmath101 is also denoted as @xmath102 . it is worth mentioning that a luttinger surgery can be reversed . let @xmath103 be the subsets @xmath104 and @xmath105 of @xmath24 . we can apply the luttinger surgery to @xmath106 with coefficient @xmath107 to recover @xmath23 . one natural source of lagrangian tori is smooth fibers of lagrangian fibrations . let @xmath10 be a symplectic 4-manifold , and let @xmath108 be a 2-manifold ( with boundary or vertices ) . a smooth map @xmath109 is called a lagrangian fibration if there exists an open dense subset @xmath110 such that @xmath111 is a compact lagrangian submanifold of @xmath23 for any @xmath112 . @xmath23 is called lagrangian fibered if such a structure exists . it is easy to see that any smooth fiber of a lagrangian fibration must be a torus . moreover , we have [ fiber ] a luttinger surgery along a lagrangian fiber preserves the lagrangian fibration structure . let @xmath113 be a lagrangian fibration and @xmath114 a generic fiber . using notations from section 2.1 , it is shown in @xcite that there is a neighborhood @xmath115 of @xmath116 and @xmath117 with local charts @xmath77 and @xmath118 such that the diagram @xmath119 commutes . here @xmath120 is the projection @xmath121 . if @xmath122 is obtained by performing luttinger surgery along @xmath1 ( indexed by @xmath44 and @xmath123 ) , we can define a map @xmath124 as @xmath125 . since @xmath126 , we have @xmath127 so @xmath128 is well - defined . it is clear that @xmath128 is lagrangian and @xmath24 also possesses a lagrangian fibration structure . lagrangian fibrations appear widely in toric geometry , integral systems and mirror symmetry . we will discuss almost toric fibration introduced by symington in some detail . an almost toric fibration of a symplectic 4-manifold @xmath0 is a lagrangian fibration @xmath113 with the following properties : for any critical point @xmath129 of @xmath130 , there exists a local coordinate @xmath131 near @xmath129 such that @xmath132 , @xmath133 , and @xmath130 has one of the forms @xmath134 an almost toric 4-manifold is a symplectic 4-manifold equipped with an almost toric fibration . the base @xmath108 of an almost toric fibration has an affine structure with boundary and vertices . moreover , these three types of critical points project to vertices , edges and interior of @xmath108 respectively . almost toric fibrations are classified by leung and symington : [ almost]@xcite let @xmath10 be a closed almost toric 4-manifold . there are seven types of almost toric fibrations according to the homeomorphism type of the base @xmath108 . 1 . @xmath135 or @xmath136 , @xmath108 is ( homeomorphic to ) a disk ; 2 . @xmath137 or @xmath138 , @xmath108 is a cylinder ; 3 . @xmath137 or @xmath138 , @xmath108 is a mbius band ; 4 . the @xmath139 surface , @xmath108 is a sphere ; 5 . the enriques surface , @xmath108 is @xmath140 ; 6 . a torus bundle over torus with monodromy @xmath141 @xmath108 is a torus ; 7 . a torus bundle over the klein bottle with monodromy @xmath142 @xmath108 is a klein bottle . an immediate consequence of this classification is the calculation of the symplectic kodaira dimension . [ al ] if @xmath143 is an almost toric fibration , then @xmath144 . moreover , @xmath21 if and only if the base @xmath108 is closed . the effect of luttinger surgeries on almost toric fibrations is also easy to describe . [ k3 ] suppose @xmath145 is an almost toric fibration and @xmath20 is obtained from @xmath0 by performing a luttinger surgery along a smooth fiber @xmath1 , then @xmath146 retains an almost toric fibration structure with the same base . moreover , @xmath24 is diffeomorphic to @xmath23 if @xmath147 . the first statement is given by lemma [ fiber ] . if @xmath147 , @xmath23 and @xmath24 are in one of the types ( 1)-(5 ) in theorem [ almost ] . in each of them , the list of manifolds are distinguished by the type of intersection forms and euler numbers . so the second result for types ( 1)-(3 ) follows from proposition [ compare ] and the fact that homology classes of lagrangian tori in manifolds with @xmath148 are torsion . it is clear for ( 4 ) and ( 5 ) from the classification . propositions [ al ] and [ k3 ] provide examples of luttinger surgeries preserving the symplectic kodaira dimension . in the next section , we will show that it is true for any luttinger surgery . in this section , we prove theorem [ kod ] . to proceed , we must first prove the invariance of minimality under luttinger surgery . a symplectic ( smooth ) @xmath149 class is a degree 2 homology class represented by an embedded symplectic ( smooth ) sphere with self - intersection @xmath149 . a symplectic 4-manifold is called @xmath150 @xmath151 @xmath152 if it does not have any symplectic ( smooth ) @xmath149 class . the symplectic minimality is actually equivalent to smooth minimality . [ minimality ] the luttinger surgery preserves the minimality . since a luttinger surgery can be reversed and the reverse operation is also a luttinger surgery , it suffices to show that , if we start with a non - minimal symplectic 4-manifold , then after a luttinger surgery , the resulting symplectic manifold is still non - minimal . but this is a direct consequence of the following fact in @xcite : [ disjoint ] given a lagrangian torus @xmath1 and a symplectic @xmath149 class , there is an embedded symplectic @xmath149 sphere in that class which is disjoint from @xmath1 . now , we analyze the effect of luttinger surgery on the symplectic canonical class @xmath13 and the symplectic class @xmath153 $ ] . recall that @xmath154 is an open submanifold of both @xmath23 and @xmath101 , and let @xmath155 and @xmath156 be the inclusions . to prepare for the following lemma , we use the notations from section [ construction ] . for the sake of simplicity , we will identify any object in @xmath23 with their image of @xmath48 and @xmath157 , @xmath158 will denote the coordinates of @xmath159 on @xmath154 and @xmath31 respectively . [ thom ] there exists a 2-dimensional submanifold @xmath160 such that @xmath161)=pd(k_{\omega})\in h_2(x)$ ] and @xmath162)=pd(k_{\tilde{\omega}})\in h_2(\tilde{x})$ ] . let @xmath163 be a @xmath7-tamed almost complex structure in @xmath23 which induces a complex structure on @xmath164 as @xmath165 assume @xmath166 $ ] is a continuous increasing function satisfying @xmath167 another almost complex structure @xmath168 in @xmath164 is defined as @xmath169 it is easy to check that @xmath168 is @xmath7-tamed and @xmath170 in @xmath171 . let @xmath172 and @xmath173 be the canonical bundles of @xmath23 and @xmath24 , respectively , and let @xmath174 and @xmath175 denote the corresponding embeddings of zero sections . since @xmath176 is trivial on @xmath31 , we can find a global section @xmath177 of @xmath176 and a thom class @xmath178 such that @xmath179 in @xmath171 and @xmath180 in @xmath181 . another nonzero @xmath182-form in @xmath31 is constructed as @xmath183 in @xmath171 , we have @xmath184 @xmath177 and @xmath185 give two local trivializations of @xmath186 with transition function @xmath187 . since @xmath188 arg@xmath189 , we can normalize the frame of @xmath190 such that @xmath191 . hence @xmath192 can be extended to @xmath193 via constant function and form a thom class @xmath194 satisfying 1 . @xmath195 in @xmath196 . @xmath194 is independent of the coordinates @xmath158 in @xmath190 . in particular , @xmath197 in @xmath198 . it is clear that these 2-forms @xmath199 and @xmath200 are equivalent in @xmath154 and vanish in @xmath201 and @xmath202 respectively . using these representations , we can find a 2-submanifold @xmath203 supp@xmath204 which is poincar dual to @xmath13 in @xmath23 , and dual to @xmath205 in @xmath24 . the main theorem can be proved now . suppose @xmath24 is obtained from @xmath23 by applying a luttinger surgery along @xmath1 . let us first consider the case in which @xmath23 is minimal . by proposition [ minimality ] , @xmath206 is also minimal . let @xmath207 and @xmath208 denote the canonical classes of @xmath23 and @xmath24 respectively . by lemma [ thom ] , there exists a submanifold @xmath160 such that @xmath161)=pd(k_{\omega})$ ] and @xmath162)=pd ( k_{\tilde{\omega}})$ ] . we also know that @xmath209 in @xmath154 . so @xmath210 @xmath211 = \int_s\omega=\int_s\tilde{\omega}=k_{\tilde{\omega}}\cdot [ \tilde{\omega}]\ ] ] thus the kodaira dimensions of @xmath23 and @xmath24 coincide . if @xmath23 is not minimal , we can blow down @xmath23 along symplectic @xmath149 spheres disjoint from @xmath1 to a minimal model . these spheres are contained in @xmath154 and the same procedure can be applied to @xmath24 , so we can argue as above . theorem [ kod ] can be used to distinguish non - diffeomorphic manifolds . in @xcite , several symplectic manifolds homeomorphic but not diffeomorphic to non - minimal rational surfaces are constructed . with @xmath212 for the building blocks , it also easily follows from theorem [ kod ] that they are exotic . [ omega ] 1 . the main theorem is proved based on the invariance of @xmath213 $ ] and @xmath214 . actually , the class @xmath153 ^ 2 $ ] is also preserved since the volume is invariant under a luttinger surgery . theorem [ kod ] is expected , in light of auroux s question 2.6 in @xcite : + + _ let @xmath215 be two integral compact symplectic 4-manifolds with the smae @xmath216 , [ \omega]^2)$ ] . is it always possible to obtain @xmath217 from @xmath218 by a sequence of luttinger surgeries ? _ + 2 . it is well known that the dolgachev surfaces @xmath219 obtained by performing two logarithmic transforms with multiplicities @xmath220 to @xmath221 have @xmath222 . so a generalized logarithmic transform may not preserve @xmath17 ( see a related discussion in @xcite ) . in this section we apply theorem [ kod ] to study the effect of luttinger surgeries on symplectic 4-manifolds with @xmath223 . we start by analyzing how homology changes under a general torus surgery . suppose @xmath23 is a smooth 4-manifold and @xmath30 is an embedded 2-torus with trivial normal bundle . moreover , @xmath224 are defined as in section [ construction ] . [ [ section ] ] to compare the homology of @xmath23 and @xmath24 , we need to compare both of them with the homology of @xmath35 . the inclusion @xmath225 induces homomorphisms @xmath226 and @xmath227 in homology . we often use @xmath228 to denote @xmath229 and @xmath230 to denote @xmath231 . we also use @xmath232 to denote the dimension of any @xmath233-vector space @xmath234 . the following lemma is a well know fact , for which we offer a geometric argument . [ mul ] @xmath235\in \ker i_1 $ ] if and only if @xmath236\ne 0 $ ] in @xmath237 . suppose @xmath238=0 $ ] in @xmath239 , i.e. @xmath240=0 $ ] in @xmath241 for some positive integer @xmath242 . thus @xmath242 copies of @xmath54 bounds an oriented surface @xmath234 in @xmath35 . extend @xmath234 by @xmath242 normal disks inside the tubular neighborhood to obtain a closed oriented surface @xmath243 intersecting with @xmath1 at @xmath242 points with the same sign . this implies in particular that @xmath236\ne 0 $ ] in @xmath237 . conversely , suppose @xmath236\ne 0 $ ] in @xmath237 , then there exists a closed oriented surface @xmath108 in @xmath23 intersecting @xmath1 with nonzero algebraic intersection numbers , say @xmath242 . we may assume that the intersection is transverse with @xmath244 positive intersection points and @xmath116 negative intersection points . we can further assume that @xmath108 intersects the closure of @xmath31 at @xmath245 normal disks , @xmath244 of those having positive orientations , the remaining @xmath116 disks having negative orientations . this implies that the complement of those disks in @xmath108 is an oriented surface in @xmath35 , whose boundary is homologous to @xmath246 , and thus @xmath238 $ ] is zero . when we consider the integral homology , lemma [ mul ] immediately implies [ mulz ] if @xmath236=0 $ ] in @xmath247 , then @xmath248 $ ] is a non - torsion class in @xmath249 . consider the mayer - vietoris sequence @xmath250 where @xmath251 and @xmath252 with @xmath253 induced by inclusions . [ meridian ] 1 . @xmath254 and @xmath255 is surjective . 2 . @xmath256\neq 0\\ 2 & \hbox{if } i_1[\mu ] = 0 \end{array}\right.\ ] ] and @xmath257 is injective if and only if @xmath235\notin \ker i_1 $ ] . 3 . @xmath258>\ ] ] 4 . if @xmath236=0\in h_2(x)$ ] , then @xmath259 is surjective . 1 . it is clear that any class @xmath260 in @xmath261 can be represented by a 1-cycle @xmath262 disjoint from @xmath1 . @xmath262 is also disjoint from @xmath36 if the neighborhood @xmath31 is small enough . so @xmath263=0 $ ] . @xmath264 implies that @xmath265 is surjective . 2 . we know @xmath266 . since @xmath267>$ ] , @xmath268\neq 0 \\ < [ \mu ] > & \hbox{if } i_1[\mu ] = 0 \end{array}\right.\ ] ] and @xmath257 is injective if and only if @xmath238\neq 0 $ ] . the rank of im@xmath269 is given from @xmath270im@xmath271 . 3 . the sequence induces a short exact sequence @xmath272 @xmath254 implies that @xmath273 because @xmath265 is surjective , we also have @xmath274 if @xmath275 , \gamma_1 , \gamma_2\}$ ] is a basis of @xmath61 , then @xmath276 , 0),(\gamma_1,\gamma_1 ) , ( \gamma_2,\gamma_2)>$ ] and @xmath277 . for @xmath278 , @xmath279 if and only if @xmath280im@xmath257 , or @xmath281 $ ] for some @xmath123 . so @xmath282>$ ] and @xmath283>\ ] ] 4 . @xmath284=0\in h_2(x ) & \leftrightarrow & [ \mu]\neq 0\in h_1(y ) & ( \hbox{by lemma \ref{mul } } ) \\ & \leftrightarrow & \rho_1\hbox { injective } & ( \hbox{by part(2)})\\ & \leftrightarrow & \partial_2=0 & ( \hbox{exactness})\\ & \leftrightarrow & \nu_2\hbox { surjective } & ( \hbox{exactness } ) \end{array}\ ] ] since @xmath236=0 $ ] also implies @xmath285 , @xmath286 has to be surjective . all the results hold if we replace @xmath23 , @xmath1 , @xmath54 by @xmath287 and @xmath288 . now , we are ready to compare @xmath23 and @xmath24 . lemma [ meridian ] , applied to torus surgeries , gives [ compare ] if @xmath24 is obtained from @xmath23 via a torus surgery , then 1 . @xmath291 , @xmath292 . 2 . @xmath293=0=i_1[\tilde \mu]$ or $ i_1[\mu]\neq 0 \neq i_1[\tilde \mu]$}\\ -1 & \hbox{if $ i_1[\mu]=0 $ and $ i_1[\tilde \mu]\ne 0$}\\ 1 & \hbox{if $ i_1[\mu]\ne 0 $ and $ i_1[\tilde \mu]= 0 $ } \end{array}\right . \ ] ] 3 . @xmath294 and @xmath295 . 1 . obvious . 2 . since @xmath254 , we can conclude that @xmath296\neq 0 \\ b_1(y ) & \hbox{if } i_1[\mu ] = 0 . \end{array}\right.\ ] ] the same is true for @xmath297 with @xmath238 $ ] replaced by @xmath298 $ ] . the proof is finished by comparing @xmath299 and @xmath297 . the first inequality is given by part ( 2 ) . the second inequality follows from part ( 1 ) and the first inequality . the next result concerns with the intersection forms . [ comparez ] suppose @xmath23 and @xmath24 are defined as above . if @xmath236 $ ] is a torsion class in @xmath247 and the intersection form @xmath300 is odd , then @xmath301 is odd as well . in particular , if both @xmath236 $ ] and @xmath302 $ ] are torsion , then @xmath24 and @xmath23 have the same intersection form . since @xmath300 is odd , there exists a closed oriented surface @xmath303 in @xmath23 such that @xmath304 is odd . by lemma [ meridian](4 ) , @xmath305\in \hbox{im}\nu_2'$ ] and @xmath303 can be chosen such that @xmath306 . thus , @xmath303 is contained in @xmath24 , and hence , @xmath301 is also odd . by proposition [ k3 ] , if a symplectic manifold @xmath0 with @xmath27 has an almost toric structure @xmath308 and if we apply a luttinger surgery along a smooth fiber of @xmath130 , the new manifold @xmath19 is diffeomorphic to @xmath0 . such phenomenon is still true for any 4-manifold with @xmath307 . moreover , we have the stronger theorem [ int2 ] . proposition [ minimality ] allows us to reduce to the case where @xmath0 is minimal . we first show that @xmath101 is diffeomorphic to @xmath23 . observe that the diffeomorphism types of minimal manifolds with @xmath307 are distinguished by their euler numbers and intersection forms . since such manifolds have @xmath148 , the homology classes of lagrangian tori are torsion . thus , both quantities are preserved by proposition [ compare ] and [ comparez ] . to show further that @xmath100 and @xmath0 are symplectomorphic to each other , it is enough to show that @xmath7 is cohomologous to @xmath99 ( @xcite ) . if @xmath23 is diffeomorpic to @xmath309 , the symplectic structure is determined by the volume @xmath153 ^ 2 $ ] , which is preserved by remark [ omega](1 ) . when @xmath23 is ruled , @xmath310 is either generated by @xmath207 and the poincar dual to the homology class of a fiber @xmath311 , or by @xmath13 and @xmath153 $ ] . hence the class of @xmath7 is determined by @xmath12 $ ] , @xmath153 ^ 2 $ ] and @xmath153(f)$ ] . as mentioned above , the first two quantities are preserved . by @xcite the fiber sphere can be chosen to be disjoint from @xmath1 , so it follows that the last quantity is also preserved . a symplectic cy surface is a symplectic 4-manifold with torsion canonical class , or equivalently , a minimal symplectic 4-manifold with @xmath312 . by theorem [ kod ] and proposition [ minimality ] , we have [ cy1 ] a luttinger surgery is a symplectic cy surgery in dimension four . there is a homological classification of symplectic cy surfaces in @xcite and @xcite . [ cy2 ] a symplectic cy surface is an integral homology k3 , an integral homology enriques surface or a rational homology torus bundle over torus . the following table lists possible rational homological invariants of symplectic cy surfaces @xcite : @xmath313 [ cols="^,^,^,^,^,^",options="header " , ] @xmath313 it follows from propositions [ compare ] , [ cy1 ] , theorem [ cy2 ] and the table above . it is also speculated that a symplectic cy surface is diffeomorphic to the k3 surface , the enriques surface or a torus bundle over torus . thus we make the following [ 00 ] if @xmath23 is a k3 surface , or an enriques surface , then under a luttinger surgery along any embedded lagrangian torus , @xmath101 is diffeomorphic to x. as for torus bundles over torus , we have [ torus ] any smooth oriented torus bundle @xmath23 over torus possesses a symplectic structure @xmath7 such that @xmath10 can be obtained by applying luttinger surgeries to @xmath314 . in the list of torus bundles over torus in @xcite , any manifold in classes ( a ) , ( b ) and ( d ) has a lagrangian bundle structure . for any such manifold , it is not hard to verify conjecture [ torus ] via luttinger surgery along lagrangian fibers . [ 44 ] 1 . conjectures [ 00 ] and [ torus ] are clearly related to question 2.6 in @xcite . 2 . there is another symplectic cy surgery in dimension six , the symplectic conifold transition . if @xmath315 with @xmath316 contains disjoint lagrangian spheres @xmath317 with homology relations generated by @xmath318=0 $ ] with all @xmath319 , smith , thomas , yau ( @xcite ) construct from @xmath315 a new symplectic manifold @xmath320 with @xmath321 and smaller @xmath322 . we notice that there is a parametrized luttinger surgery in higher dimension and believe it also should be a symplectic cy surgery . this will be discussed elsewhere . in this section , we will introduce topological preferred framings and compare them with the lagrangian framing for lagrangian tori in @xmath223 symplectic 4-manifolds . suppose @xmath23 is a smooth 4-manifold and @xmath30 is an embedded 2-torus with trivial normal bundle . recall that a framing is a diffeomorphism @xmath41 for a tubular neighborhood @xmath31 of @xmath1 such that @xmath42 and a longitudinal curve of @xmath48 is a lift @xmath323 of some simple closed curve @xmath44 in @xmath36 . let @xmath324|\gamma_\varphi : \hbox { longitudinal curve of } \varphi>\ ] ] @xmath325 is a subgroup of @xmath326 and it induces a decomposition of @xmath326 : @xmath327>\oplus h_{1,\varphi}.\ ] ] conversely , any rank 2 subgroup @xmath328 of @xmath61 such that @xmath235 $ ] and @xmath328 generate @xmath61 corresponds to a framing of @xmath1 . in @xcite , luttinger introduced a version of topological preferred framings of lagrangian tori in @xmath329 . it requires that @xmath325 is in the kernel of @xmath330 . on the other hand , fintushel and stern ( @xcite ) defined null - homologous framings for a null - homologous torus via @xmath331 ( seemingly , under the assumption that @xmath261 vanishes , though not explicitly mentioned ) . [ tprefer ] suppose @xmath1 is null - homologous , i.e. , @xmath236=0 $ ] in @xmath247 . a framing @xmath48 is called a topological preferred framing if @xmath332\in \ker i_2^{\mathbb{z}}$ ] . here , @xmath333 is a longitudinal torus of @xmath48 given by @xmath70 , @xmath45 . in order to compare various preferred framings and the lagrangian framing , we need to investigate the relation of the maps @xmath336 and @xmath337 given by and . let @xmath35 be a smooth oriented 4-manifold with boundary @xmath338 . 1 . consider the exact sequence @xmath346 it induces a short exact sequence @xmath347 by lefschetz duality and universal coefficient theorem , @xmath348 and @xmath349 . so implies that @xmath350 which is ( 1 ) . 2 . consider the dual pairing @xmath351 + because the maps @xmath336 and @xmath352 are induced by embedding and restriction , this pairing is natural , i.e. , for @xmath353 and @xmath354 , @xmath355 + there is an isomorphism of long exact sequences induced naturally by lefschetz and poincar dualities : @xmath356 & & \downarrow \cap [ y ] & & \downarrow \cap [ z ] & & \downarrow \cap [ y ] & \\ \cdots \stackrel{i_{3}}{\longrightarrow } & h_3(y ) & \stackrel{\delta_3}{\longrightarrow } & h_3(y , z ) & \stackrel{\partial_{2}}{\longrightarrow } & h_2(z ) & \stackrel{i_{2}}{\longrightarrow } & h_2(y ) & \stackrel{\delta_2}{\longrightarrow } \cdots \end{array}\ ] ] + using this diagram , the dual pairing induces the intersection pairing : @xmath357 + given @xmath358 , there exists @xmath359 such that @xmath360 . let @xmath361 be the lefschetz dual of @xmath362 in @xmath363 . for any @xmath364 , @xmath365 it shows that @xmath366 . from part ( 1 ) , @xmath367 so @xmath368 . similar argument shows that @xmath344 . 3 . suppose @xmath369 , then @xmath370 and @xmath337 is the zero map by part ( 1 ) . let @xmath371 be two nonisotopic embedded tori in @xmath36 intersecting in a curve @xmath262 transversely . since @xmath338 , @xmath372\cap [ t_2]=[c]\neq 0 $ ] in @xmath373 . meanwhile , each @xmath374 bounds a 3-manifold @xmath375 in @xmath35 . @xmath376 is a 2-cycle whose boundary is @xmath262 and @xmath377 $ ] is in the kernel of @xmath336 , which contradicts the assumption that @xmath336 is injective . here is a geometric interpretation of this lemma . assume @xmath378 is an integral class of @xmath335 and @xmath262 is a closed curve in @xmath36 such that @xmath377\cdot z_2\neq 0 $ ] in @xmath36 . there exists a relative 3-cycle @xmath379 in @xmath380 such that @xmath381=z_2 $ ] . in particular , we can assume that @xmath379 intersects @xmath36 transversely and @xmath382 intersects @xmath262 transversely at @xmath383 and @xmath384 in @xmath36 with positive and negative intersections respectively . furthermore , we can give a collar structure @xmath385 near @xmath36 and assume @xmath386 . if we push @xmath262 to @xmath387 in the interior of @xmath35 , then @xmath388 and @xmath379 intersect transversely at @xmath389 and @xmath390 with positive and negative intersections respectively . hence @xmath391=i_1([c])$ ] and @xmath392\cdot [ w]=[c]\cdot [ \partial w]=p - n=[c]\cdot z_2\neq 0\ ] ] so @xmath377 $ ] can not be in @xmath334 . 1 . part ( 1 ) of lemma [ kernel ] is still true in arbitrary dimension . if @xmath35 is a @xmath393-dimensional manifold with connected boundary @xmath36 and @xmath394 denotes the homomorphism induced by the inclusion @xmath395 , then @xmath396 for @xmath397 . 2 . part ( 3 ) of lemma [ kernel ] is pointed out by robert gompf . 1 . let @xmath399 be the trivial knot in @xmath400 and @xmath401 . the complement of the torus @xmath402 is @xmath403 if @xmath404 denote the isotopy classes of these two @xmath53 and @xmath405 , then @xmath406 , [ m ] , [ l]>$ ] and @xmath334 has rank 1 which is generated by @xmath407 $ ] . on the other hand , @xmath335 is generated by @xmath408 , [ l\times m]$ ] and has rank 2 . + in general , if @xmath409 is any knot in @xmath400 and @xmath303 is a seifert surface with boundary @xmath409 , we can define @xmath410 and @xmath411 as above and choose @xmath242 as the push - off of @xmath409 in @xmath303 . then @xmath35 and @xmath412 have isomorphic homology groups and @xmath334 is still generated by @xmath242 , which bounds the surface @xmath303 . similarly , @xmath335 has rank 2 and is generated by @xmath408 , [ l\times m]$ ] . they bound @xmath413 and @xmath414 respectively . in @xcite , fintushel and stern use these manifolds as building blocks to define knot surgery in 4-manifolds . + in the next example , the results of lemma [ kernel ] are not obvious . let @xmath415 be a ruled surface and the loop @xmath416 be a lift of a loop in @xmath417 . we can construct a torus @xmath1 in @xmath23 as the product of @xmath47 and some circle @xmath116 in the fiber . if @xmath418 is a meridian of @xmath1 and @xmath419 is nontrivial in @xmath420 , it is easy to show that @xmath334 is generated by a push - off of @xmath116 . but @xmath335 is not obvious even when @xmath421 is the trivial bundle . by lemma [ kernel ] , we know that @xmath335 has rank 2 and is generated by @xmath422 $ ] and @xmath423 $ ] . 2 . let @xmath424 be the clifford torus embedded in the rational manifold @xmath425 . the group @xmath373 is generated by @xmath426 , [ b]$ ] and the meridian @xmath235 $ ] . it is easy to show that @xmath427 , [ b]>$ ] has rank 2 and @xmath428>$ ] . + in general , if @xmath23 is simply connected and @xmath30 is a torus with trivial normal bundle , then @xmath429 if and only if @xmath236=0 $ ] in @xmath237 . if @xmath430 , it follows from lemma [ meridian ] that such surgery will not change the homology for any torus surgery . @xmath444\in \ker i_2^{\mathbb{q}}\\ & \leftrightarrow & < [ l_{\varphi}]>\subset \ker i_2^{\mathbb{q}}\\ & \leftrightarrow & \mathbf{ann}(<[l_{\varphi}]>)\supset \mathbf{ann}(\ker i_2^{\mathbb{q}})\\ & \leftrightarrow & \ker i_1^{\mathbb{q}}\subset h_{1,\varphi}\otimes { \mathbb{q}}. \ \hbox{(lemma \ref{kernel})}\end{aligned}\ ] ] 1 . since @xmath236=0 $ ] in @xmath247 , there exists a 3-chain @xmath379 such that @xmath446 . we can assume that @xmath379 intersects @xmath36 transversely . in fact , we can choose a framing @xmath41 such that @xmath447 , where @xmath448 . then @xmath449 is a longitudinal torus of @xmath48 and @xmath450 is a relative 3-cycle of @xmath380 with @xmath451 . so @xmath452\in \ker i_2^{\mathbb{z}}$ ] and @xmath48 is a topological preferred framing . @xmath332\in \ker i_2^{\mathbb{z}}$ ] implies that @xmath453)\supset \mathbf{ann}_{\mathbb{z}}(\ker i_2^{\mathbb{z}})$ ] . it is easy to observe that @xmath454>)=h_{1,\varphi}$ ] . by lemma [ zker ] , @xmath455)=h_{1,\varphi}.\ ] ] if @xmath236 $ ] is torsion in @xmath23 , proposition [ exist](1 ) may fail in two situations . first , there may exist @xmath456 such that @xmath457 but @xmath458 for some nonzero integer @xmath5 . so we can only define rational topological preferred framings . second , @xmath235 $ ] and @xmath435 might not generate the group @xmath326 . in this case , rational topological preferred framings also do not exist . 1 . in knot theory , the notion of preferred framings is similar to that of definition [ tprefer ] . let @xmath459 be an integral homology 3-sphere and @xmath460 be a knot . if @xmath328 is a tubular neighborhood of @xmath409 , a diffeomorphism @xmath461 satisfying @xmath462 is called a framing of @xmath409 . furthermore , @xmath463 is called a preferred framing if @xmath464 is homologically trivial in @xmath465 . for any knot @xmath409 in @xmath459 , preferred framings exist and are unique up to isotopy ( @xcite ) . it is easy to see from proposition [ exist ] that luttinger s definition coincides with definition [ tprefer ] when @xmath6 . 3 . in @xcite , an invariant @xmath466 is defined when @xmath236=0 $ ] and @xmath1 has a unique topological preferred framing @xmath467 . assume @xmath468 is the lagrangian preferred framing . then @xmath469 if and only if @xmath470 . otherwise , @xmath466 is the smallest positive integer @xmath5 such that @xmath471+[\gamma_\varphi]\in h_{1,\varphi_{lag}}$ ] for some @xmath472\in h_{1,\varphi_0}$ ] . [ r1k ] suppose @xmath23 is a smooth 4-manifold and @xmath30 is a torus with trivial normal bundle such that @xmath236=0 $ ] in @xmath442 and @xmath48 is a framing of @xmath1 . let @xmath474 be constructed from @xmath23 via @xmath473-surgery along @xmath475 . 1 . if @xmath48 is a rational topological preferred framing of @xmath1 , then @xmath24 satisfies @xmath476 for any @xmath47 and @xmath5 . if @xmath477 for any @xmath47 and @xmath5 , then @xmath48 is a rational topological preferred framing of @xmath1 . 1 . by lemma [ meridian](3 ) , we have @xmath478\neq 0 \\ r(h_1(y ) ) & \hbox{if } i_1[\mu]=0 . \end{array}\right.\ ] ] lemma [ mul ] implies that @xmath238\neq 0 $ ] in @xmath479 . since @xmath48 is a rational topological preferred framing , proposition [ rexist ] implies that @xmath235+k[\gamma_\varphi]\notin \ker i_1 $ ] for any integer @xmath5 and @xmath472\in h_{1,\varphi}$ ] . so @xmath480 if @xmath24 is given via @xmath72-surgery . we first prove that any @xmath481 lies in @xmath325 . let @xmath482+t[\gamma_\varphi]$ ] for some @xmath483 and @xmath44 ( recall that @xmath62 is a lift of @xmath47 ) . for @xmath484 , the meridian of @xmath56 in @xmath24 satisfies @xmath485=[\mu]+kt[\gamma_\varphi]=[\mu]+k(a - s[\mu])=(1-sk)[\mu]+ka\ ] ] so + @xmath486+ka ) > & $ ] ( by ) @xmath487 ) > & ( a\in \ker i_1^{\mathbb{z}})\\ & \cong & h_1(y;{\mathbb{z}})/<i_1^{\mathbb{z}}[\mu ] > \end{array}$ ] + because @xmath235 $ ] is essential in @xmath35 and @xmath5 is arbitrary , @xmath488 should be zero and @xmath489 . otherwise , @xmath261 has infinitely many torsion classes with different orders . + tensoring with @xmath233 , we have @xmath490 by proposition [ rexist ] , @xmath48 is a rational topological preferred framing . [ 1k ] suppose @xmath261 has no torsion and @xmath1 is null - homologous . then a framing @xmath48 of @xmath1 is a topological preferred framing if and only if @xmath477 for any @xmath474 obtained from @xmath23 via @xmath473-surgery . assume @xmath48 is a topological preferred framing . consider the 3-chain @xmath379 given in the proof of proposition [ exist ] . it is clear that a meridian @xmath54 intersects @xmath379 at one point . so @xmath248\cdot [ w]=\pm 1 $ ] and @xmath248 $ ] is a primitive class . similarly , the simple closed curve @xmath288 has class @xmath235+k[\gamma_\varphi]$ ] and is homotopic to a curve intersecting @xmath379 at one point . hence @xmath491 $ ] is also a primitive class for any @xmath492 . since @xmath249 is a free abelian group , we have @xmath493>\cong h_1(y,{\mathbb{z}})/<i_1^{\mathbb{z}}[\tilde{\mu}]>$ ] . so @xmath494 by lemma [ meridian](3 ) . conversely , if @xmath494 for any @xmath24 , the proof of proposition [ r1k](2 ) shows that @xmath48 is a rational topological preferred framing . the assumption that @xmath247 has no torsion implies that @xmath48 is actually a topological preferred framing . the knot surgery in @xcite is an example of @xmath473-surgeries . suppose @xmath495 , @xmath409 is a knot in @xmath400 and @xmath496 . let @xmath54 be the meridian of @xmath1 , @xmath260 the longitude of the preferred framing of @xmath409 and @xmath497 . then @xmath498>$ ] and @xmath499 , [ a\times b]>$ ] . consider the framings @xmath500 of @xmath1 where @xmath501 is generated by @xmath502+[a]$ ] and @xmath503 $ ] . it is clear that @xmath500 is a topological preferred framing of @xmath1 exactly when @xmath504 . if @xmath409 is the trivial knot and @xmath4=p[\mu]+[a]$ ] , the resulting manifold of @xmath72-surgery is @xmath505 and @xmath506 . if @xmath507 , @xmath508 has rank @xmath509 for any @xmath510 , but the torsion subgroups vary . propositions [ 1k ] and [ r1k ] show that @xmath500 is not a rational topological preferred framing actually , such framings do not exist for @xmath1 . here we provide topological constraints on the isotopy classes of lagrangian tori in many symplectic manifolds with non - positive kodaira dimension . in particular , they imply that the invariant @xmath466 of fintushel and stern ( remark [ lambda](3 ) ) is zero if the manifold has non - positive kodaira dimension and vanishing integral @xmath511 . recall that the lagrangian framing for lagrangian tori is defined in section [ construction ] . 1 . if @xmath247 is torsion free and @xmath1 is null - homologous , then the lagrangian framing of @xmath1 is a topological preferred framing . if @xmath236 $ ] is torsion then the lagrangian framing of @xmath1 is a rational topological preferred framing . since @xmath512 and @xmath247 has no torsion , @xmath1 is null - homologous . for any @xmath101 given from @xmath23 by applying luttinger surgery along @xmath1 , @xmath101 is diffeomorphic to @xmath23 by theorem [ int2 ] . now the claim follows from proposition [ prefer ] . when @xmath1 is essential , @xmath235\in\ker i_1^{\mathbb{q}}$ ] by lemma [ mul ] . if @xmath514 , there exists @xmath44 such that @xmath515)\neq 0 $ ] . in the manifold @xmath516 , the class @xmath64=[\mu]+[\gamma_\varphi]$ ] is nonzero in @xmath239 . by proposition [ compare ] , @xmath517 , which contradicts theorem [ int3 ] . so @xmath518 and @xmath1 is completely essential . 999 a. akhmedov , s. baldridge , r. i. baykur , p. kirk , d. park , _ simply connected minimal symplectic 4-manifolds with signature less than @xmath149 _ , j. eur . ( jems ) 12 ( 2010 ) , no . 1 , 133161 . a. akhmedov , i. baykur , b.d . park , _ constructing infinitely many smooth structures on small 4-manifolds _ , journal of topology 1 ( 2008 ) no . 2 , 409428 . a. akhmedov , d. park , _ exotic smooth structures on small 4manifolds _ , invent . math . 173 ( 2008 ) , no . 1 , 209223 . a. akhmedov , d. park , _ exotic smooth structures on small 4manifolds with odd signatures _ , invent . math 181(2010 ) , no . 3 , 577603 . d. auroux , _ some open questions about symplectic 4-manifolds , singular plane curves and braid group factorizations _ , european congress of mathematics , 2340 , eur . zrich , 2005 . d. auroux , s. donaldson , l. katzarkov , _ luttinger surgery along lagrangian tori and non - isotopy for singular symplectic plane curves _ , math . 326 ( 2003 ) , no . 1 , 185203 . s. bauer , _ almost complex @xmath2-manifolds with vanishing first chern class _ , j. diff . 79 ( 2008 ) , 2532 . s. baldridge , p. kirk , _ a symplectic manifold homeomorphic but not diffeomorphic to @xmath519 _ , geom . 12 ( 2008 ) , no . 2 , 919940 . s. baldridge , p. kirk , _ constructions of small symplectic 4manifolds using luttinger surgery _ , j. differential geom . 82 ( 2009 ) , no . 2 , 317361 . r. i. baykur , n. sunukjian , round handles , logarithmic transforms , and smooth 4-manifolds , arxiv:1009.0514v2 . t. etgu , d. mckinnon , d. park , _ lagrangian tori in homotopy elliptic surfaces _ , trans . ( 2005 ) , 37573774 . r. fintushel , r. stern , _ knots , links , and 4-manifolds _ , invent . 134(1998 ) 363400 . r. fintushel , r. stern , _ invariants for lagrangian tori _ , geometry and topology vol.8 ( 2004 ) , 947968 . r. fintushel , b. park , r. stern , _ reverse engineering small 4-manifolds _ , algebr . ( 2007 ) , 21032116 . h. geiges , _ symplectic structures on @xmath520-bundles over @xmath520 _ , duke math . j. 67 ( 1992 ) , 539555 . in mathematics 20 american mathematical society , providence , ri , 1999 . r. gompf , _ a new construction of symplectic manifolds _ , ann . of math . ( 2 ) 142 ( 1995 ) , 527595 . r. gompf , a. stipsicz , _ manifolds and kirby calculus _ , graduate studies in mathematics 20 american mathematical society , providence , ri , 1999 . m. gromov , _ pseudoholomorphic curves in symplectic manifolds _ , invent . 82 ( 1985 ) , no . 2 , 307347 . v. guillemin , s. sternberg , _ birational equivalence in the symplectic category _ 97(1989 ) , no . 3 , 485522 . a. ivrii , _ lagrangian unknottedness of tori in certain symplectic 4-manifolds _ , phd thesis . t. j. li , _ the kodaira dimensions of symplectic 4-manifolds _ , proc . of the clay inst . 2004 summer school on floer homology , gauge theory and low dim . , renyi inst . math , hungary , 249261 . t. j. li , _ symplectic 4-manifolds with kodaira dimension zero _ , j. diff . ( 2 ) 74 ( 2006 ) , 321352 . t. j. li , _ quaternionic bundles and betti numbers of symplectic 4-manifolds with kodaira dimension zero _ , int . notices , 2006(2006 ) , 128 . n. c. leung , m. symington , _ almost toric symplectic four - manifolds _ , to appear in j. of symp . geom . , 8(2010 ) , no . 2 , 143187 . k.m . luttinger , _ lagrangian tori in @xmath521 _ , j. differential geom . 42 ( 1995 ) , no . 2 , 220228 . f. lalonde , d. mcduff , _ @xmath163-curves and the classification of rational and ruled symplectic @xmath2-manifolds _ , contact and symplectic geometry ( cambridge , 1994 ) , 342 , publ . newton inst . , 8 , cambridge univ . press , cambridge , 1996 . d. mcduff , d. salamon , _ introduction to symplectic topology _ , 2nd ed . , oxford mathematics monographs , 1998 . k. mishachev , _ the classification of lagrangian bundles over surfaces _ , differential geom . appl . 6 ( 1996 ) , no . 4 , 301320 . j. park , _ simply connected symplectic 4-manifolds with @xmath522 and @xmath523 _ , invent . 159 ( 2005 ) , no . 3 , 657-667 . d. rolfsen , _ knots and links _ , math lecture series 7 , 1990 . i. smith , _ torus fibrations on symplectic four manifolds _ , turk . j. math . 25(2001 ) , 6995 . i. smith , r. thomas , s.t . yau , _ symplectic conifold transitions _ , j. diff . 62 ( 2002 ) , no . 2 , 209242 . m. symington , _ four dimensions from two in symplectic topology _ , in proceedings of the 2001 georgia international topology conference , proceedings of symposia in pure mathematics , 153208 , 2003 . m. usher , _ minimality and symplectic sums _ , res . not . 2006 , art . i d 49857 , 17 pp . m. usher , _ kodaira dimension and symplectic sums _ , comment . helv . 84 ( 2009 ) , no . 1 , 5785 . s. vidussi , _ lagrangian surfaces in a fixed homology class : existence of knotted lagrangian tori _ , j. differential geom . 74 ( 2006 ) , no . 3 , 507522 . j. welschinger , _ effective classes and lagrangian tori in symplectic four - manifolds _ , j. symp . geom . 5(2007 ) , no .

we construct the equation of state ( eos ) of dense matter covering a wide range of temperature , proton fraction , and density for the use of core - collapse supernova simulations . the study is based on the relativistic mean - field ( rmf ) theory , which can provide an excellent description of nuclear matter and finite nuclei . the thomas fermi approximation in combination with assumed nucleon distribution functions and a free energy minimization is adopted to describe the non - uniform matter , which is composed of a lattice of heavy nuclei . we treat the uniform matter and non - uniform matter consistently using the same rmf theory . we present two sets of eos tables , namely eos2 and eos3 . eos2 is an update of our earlier work published in 1998 ( eos1 ) , where only the nucleon degree of freedom is taken into account . eos3 includes additional contributions from hyperons . the effect of hyperons on the eos is negligible in the low - temperature and low - density region , whereas it tends to soften the eos at high density . in comparison with eos1 , eos2 and eos3 have an improved design of ranges and grids , which covers the temperature range mev with the logarithmic grid spacing})=0.04 $ ] ( 92 points including ) , the proton fraction range with the linear grid spacing ( 66 points ) , and the density range with the logarithmic grid spacing } ) = 0.1 $ ] ( 110 points ) .

1105.1666

the equation of state ( eos ) of dense matter plays an important role in various astrophysical phenomena such as supernova explosions and the formation of neutron stars and black holes @xcite . simulations of core - collapse supernovae cover a wide range of thermodynamic conditions , and extremely high density and temperature may be achieved when black holes are formed by failed supernovae . the temperature may vary from @xmath11 to more than @xmath12 mev , the proton fraction changes from @xmath11 to around @xmath13 , and the density can vary from @xmath14 to more than @xmath15 . clearly , it is very difficult to construct a complete eos over such a wide range of parameters . the information of matter under extreme conditions is far beyond our knowledge of nuclear physics from laboratory experiments . therefore , it is necessary to perform a large extrapolation based on a theoretical model that is supported by microscopic theory and consistent with available experimental data . during the past few decades , great efforts have been made to study the eos of nuclear matter @xcite . however , most of the investigations focused on detailed aspects of nuclear matter , which were often restricted to the case of zero temperature or high density with uniform distribution of particles . this kind of eos is generally not applicable for use in supernova simulations . so far , there exist only two realistic eoss which are commonly used in simulations of core - collapse supernovae , namely the one by @xcite and the one by @xcite . the lattimer swesty eos is based on a compressible liquid - drop model with a skyrme force . the shen eos is based on a relativistic mean - field ( rmf ) model and uses the thomas fermi approximation with assumed nucleon distribution functions in a wigner seitz cell for the description of non - uniform matter . recently , a hartree mean - field calculation @xcite was performed for the wigner seitz cell instead of the thomas fermi approximation used by @xcite . the hartree calculation can incorporate nuclear shell effects , but it requires much more computational resources . in our earlier work @xcite , we constructed the relativistic eos for supernova simulations , which has been widely used in astrophysical simulations over the past decade @xcite . the eos is based on the rmf theory combined with the thomas fermi approximation . the rmf theory with nonlinear @xmath16 and @xmath17 terms is able to reproduce nuclear matter saturation properties and provide a good description for both stable and unstable nuclei @xcite . the thomas fermi approximation in combination with assumed nucleon distribution functions and a free energy minimization is adopted to describe the non - uniform matter which is modeled as a mixture of a single species of heavy nuclei , alpha particles , and free nucleons that exist outside of nuclei , while the leptons can be treated as uniform non - interacting particles separately . the rmf results are taken as input in the thomas fermi calculation , so the treatments of non - uniform matter and uniform matter in this eos are sufficient to obtain the table in a consistent manner . it would be preferable to treat the mixture of nuclei as recently done in @xcite , @xcite , and @xcite for the detailed treatment of electron captures on nuclei in supernova core . however , it is beyond the scope of the current update of the shen eos table , which has been routinely used in astrophysical simulations . toward a more efficient and effective use of the shen eos table , we are required to make improvements in the eos given in @xcite , hereafter referred to as eos1 . the main demand is to increase the number of temperature points which is crucial in the simulation of core - collapse supernovae . in some cases , the temperature may reach more than @xmath12 mev , so it is encouraged to provide results at some higher temperatures although it is probably beyond the applicability of the rmf theory . it is generally believed that at sufficiently high temperature and/or density nuclear matter undergoes a phase transition to quark gluon plasma ( qgp ) . based on experimental data from high - energy heavy ion collisions and lattice quantum chromodynamics ( qcd ) calculations , it is known that the critical temperature for the qcd phase transition is around @xmath18 mev for zero baryon density @xcite . therefore , we note that the nuclear eos at extremely high temperature and/or density is not reliable due to the qcd phase transition , although results of the rmf theory are provided for practical use in astrophysical simulations . another suggestion is to use a linear grid for the proton fraction @xmath19 , instead of the logarithmic @xmath19 grid used in eos1 , which can add more points in the important region @xmath20@xmath21 for supernova simulations and save memory space by reducing the number of @xmath19 points at @xmath22 . furthermore , a proton - rich matter may be involved in astrophysical simulations @xcite , and therefore the information at @xmath23 is needed . in eos1 , the density grid spacing is only approximately equal at high density , it is now possible to perform the calculation with equal grid spacing in the whole range of density . since it is desirable and feasible to make these improvements , we work out a new version of the eos table , hereafter referred to as eos2 , which contains the same degrees of freedom as eos1 . in recent years , there has been extensive discussion in the literature on the influence of non - nucleonic degrees of freedom in dense matter @xcite . it is generally believed that hyperons appear around twice normal nuclear matter density in cold neutron star matter @xcite . the first hyperon to appear is @xmath0 that is the lightest one with an attractive potential in nuclear matter @xcite . @xmath24 hyperons are now considered to appear at a higher density than @xmath0 , because @xmath24 hyperons feel a repulsive potential in nuclear matter according to recent developments in hypernuclear physics @xcite . in the work of @xcite , the authors examined the properties of dense matter based on an extended rmf model including the full baryon octet . they presented several sets of eos including hyperons for simulations of core - collapse supernovae , which were connected with eos1 at low density in the simple procedure described in section 2.3 of their paper . the effect of @xmath25-wave pion condensation was also examined in @xcite , where the authors considered free thermal pions assuming the pion mass could not be affected by the interaction . in @xcite , the authors examined the possibility of @xmath25-wave pion condensation in dense matter by using the phenomenological optical potentials determined from the pionic atom or pion nucleus scattering data , and they concluded that @xmath25-wave pion condensation would hardly take place in neutron stars and especially have no chance if hyperons could participate in neutron star matter . the presence of boson condensation and deconfined quarks in neutron stars has been extensively discussed in many works @xcite . it has been suggested that the quark matter may exist in the core of massive neutron stars , and the hadron quark phase transition can proceed through a mixed phase of hadronic and quark matter @xcite . if deconfined quark matter does exist inside stars , it is likely to be in a color superconducting phase , and various color superconducting phases have been intensively investigated in recent years @xcite . in the works of @xcite and @xcite , the authors constructed the eos tables for simulations of core - collapse supernovae including the hadron quark phase transition at high density , which were connected with eos1 at low density . the hadron quark phase transition proceeded through a mixed phase obtained by the gibbs conditions for phase equilibrium , where the rmf model was used for the hadronic phase and the bag model was adopted for the quark phase . generally , the introduction of non - nucleonic degrees of freedom leads to a softening of the eos and thereby a corresponding reduction in the maximum mass of neutron stars . the recent measurement of the shapiro delay in the radio pulsar psr j1614 - 2230 yielded a mass of @xmath26 @xcite . such a high neutron star mass provides an important constraint on the eos at high density and rules out many predictions of non - nucleonic components in neutron star interiors . however , it is currently difficult to rule out all possible exotica with the @xmath27 observation , some theoretical calculations including hyperons and/or quarks could be compatible with the observation of psr j1614 - 2230 @xcite . among these exotic candidates , @xmath0 hyperons are the most likely to occur in dense matter . in addition , much more experimental information is now available for @xmath0 than other hyperons @xcite . from the experimental binding energies of single-@xmath0 hypernuclei , the potential depth of @xmath0 in nuclear matter is estimated to be around @xmath28 mev @xcite . several recent observations of double-@xmath0 hypernuclei ( see table 4 of @xcite ) indicate that the effective @xmath29 interaction should be considerably weaker than that deduced from the earlier measurement @xcite . in theoretical studies of @xmath0 hypernuclei , the rmf theory with nonlinear @xmath16 and @xmath17 terms is able to provide a reasonable description of single- and double-@xmath0 hypernuclei @xcite . the influence of @xmath0 hyperons on neutron star properties has been investigated within the rmf model @xcite . we would like to examine the effect of @xmath0 hyperons on the eos for simulations of core - collapse supernovae , while other hyperons such as @xmath24 and @xmath30 are ignored due to their relatively high threshold densities and lack of available experimental data . we construct the relativistic eos with the inclusion of @xmath0 hyperons , hereafter referred to as eos3 . for the contribution of @xmath0 hyperons in eos3 , we assume that the equilibrium condition @xmath31 is valid . in a supernova explosion , the dynamical timescale is of the order of milliseconds , which is long enough to establish equilibrium with respect to weak interactions that change strangeness on the timescale of microseconds or less @xcite . in fact , @xmath0 hyperons can have a noticeable effect on the properties of matter mainly at high density . it is known that weak equilibrium could be achieved at densities above @xmath32 where neutrinos begin to be trapped in the core @xcite . therefore , it is justified to use the equilibrium condition @xmath31 to determine the @xmath0 fraction of supernova matter at high density where @xmath0 hyperons have a noticeable contribution . in the work of @xcite , the authors assumed weak equilibrium among the full baryon octet when they determined the composition of supernova matter at a fixed density , temperature , and charge fraction . they examined the appearance of hyperons during the evolution of core collapse and bounce , and found that the effect of hyperons would be small in a spherical and adiabatic collapse of a @xmath33 star by the hydrodynamics without neutrino transfer @xcite . the hyperons are expected to appear and play an important role in the thermal evolution of protoneutron stars and the black hole formation from massive stars @xcite . we have two aims in this paper . the first is to improve the eos table according to the requirements of the users . in comparison with the earlier version ( eos1 ) , the following improvements are made in eos2 and eos3 ( see table 1 for details of the comparison ) . * the number of @xmath34 points is largely increased , the upper limit of @xmath34 is extended , and equal grid spacing for @xmath34 is used . * a linear @xmath19 grid is adopted instead of the logarithmic @xmath19 grid used in eos1 and the @xmath19 upper limit is extended . * the upper limit of @xmath35 is extended and equal grid spacing is used in the whole range of @xmath35 , whereas it is only approximately equal at high density in eos1 . the finer grids are favorable for better accuracy in the numerical simulations of core - collapse supernovae . the wide coverage of the conditions is necessary for numerical simulations of astrophysical phenomena including black hole formation , neutron star mergers , and nucleosynthesis . the second aim of this paper is to provide the eos table with the inclusion of @xmath0 hyperons for the use in simulations of core - collapse supernovae . the difference between eos2 and eos3 is that the contribution from @xmath0 hyperons is included in eos3 when the @xmath0 fraction is larger than @xmath36 . this paper is arranged as follows . in section [ sec:2 ] , we briefly describe the framework to calculate the eos table . we introduce the rmf model with the inclusion of @xmath0 hyperons and explain how to determine the rmf parameters . to make this paper self - contained , we give a brief description of the thomas fermi approximation for the non - uniform matter . in section [ sec:3 ] , we discuss our results without and with the inclusion of @xmath0 hyperons . we explore the properties of dense matter by examining the phase diagram , compositions , and thermodynamic quantities . section [ sec:4 ] is devoted to a summary . in appendix a , we give the definitions of the physical quantities tabulated in the eos . eos2 and eos3 are presented in electronic tables and can also be found on several web sites . in appendix b , we describe various checks made for the eos tables . we construct the eos within the rmf framework @xcite . for uniform matter without heavy nuclei formed , the rmf theory can be easily used to calculate the properties of matter . for non - uniform matter where heavy nuclei are formed in order to lower the free energy , we adopt the thomas fermi approximation based on the work by @xcite . the non - uniform matter can be modeled as a mixture of a single species of heavy nuclei , alpha particles , and free nucleons that exist outside of nuclei . the results of the rmf model are taken as input in the thomas fermi calculation , so the treatments of non - uniform matter and uniform matter in this eos are sufficient to obtain the table in a consistent manner . it would be preferable to treat the mixture of nuclei as recently done in @xcite , @xcite , and @xcite for the detailed treatment of electron captures on nuclei in supernova core . however , it is beyond the scope of the current update of the shen eos table , which has been routinely used in astrophysical simulations . we adopt the rmf theory with nonlinear @xmath16 and @xmath17 terms to describe homogeneous nuclear matter @xcite . we study the two cases without and with @xmath0 hyperons . in the rmf approach , baryons interact through the exchange of various effective mesons . the exchanged mesons considered in this work include isoscalar scalar and vector mesons ( @xmath16 and @xmath17 ) and an isovector vector meson ( @xmath37 ) . in some published studies @xcite , the two hidden - strangeness ( @xmath38 ) scalar and vector mesons ( @xmath39 and @xmath40 ) were included in a hyperon - rich system . it has been found that the attraction from @xmath39 exchange is almost canceled by the repulsion from @xmath40 exchange @xcite . therefore , we neglect the contribution from the exchange of these two hidden - strangeness mesons in this work . we start with the lagrangian of the rmf theory including @xmath0 hyperons , @xmath41\psi \nonumber\\ & & + \bar{\psi}_{\lambda}\left[i\gamma_{\mu}\partial^{\mu } -m_{\lambda } -g_{\sigma}^{\lambda}\sigma - g_{\omega}^{\lambda}\gamma_{\mu}\omega^{\mu } \right]\psi_{\lambda } \nonumber\\ & & + \frac{1}{2}\partial_{\mu}\sigma\partial^{\mu}\sigma -\frac{1}{2}m^2_{\sigma}\sigma^2-\frac{1}{3}g_{2}\sigma^{3 } -\frac{1}{4}g_{3}\sigma^{4 } \nonumber\\ & & -\frac{1}{4}w_{\mu\nu}w^{\mu\nu } + \frac{1}{2}m^2_{\omega}\omega_{\mu}\omega^{\mu } + \frac{1}{4}c_{3}\left(\omega_{\mu}\omega^{\mu}\right)^2 \nonumber\\ & & -\frac{1}{4}r^a_{\mu\nu}r^{a\mu\nu } + \frac{1}{2}m^2_{\rho}\rho^a_{\mu}\rho^{a\mu } , \end{aligned}\ ] ] where @xmath42 and @xmath43 denote the nucleon and @xmath0 hyperon fields , respectively . @xmath16 , @xmath44 , and @xmath45 are @xmath16 , @xmath17 , and @xmath37 meson fields with masses @xmath46 , @xmath47 , and @xmath48 . @xmath49 and @xmath50 are the antisymmetric field tensors for @xmath44 and @xmath45 , respectively . it is known that the inclusion of nonlinear @xmath16 terms is essential to reproduce the properties of nuclei quantitatively and provide a reasonable value for the incompressibility , while the nonlinear @xmath17 term is added in order to reproduce the density dependence of the vector part of the nucleon self - energy obtained in the relativistic brueckner hartree fock ( rbhf ) theory @xcite . we adopt the parameter set tm1 listed in table [ tab:2 ] , which was determined in @xcite by fitting some ground - state properties of nuclei including unstable ones . with the tm1 parameter set , the nuclear matter saturation density is @xmath51 @xmath52 , the binding energy per nucleon is @xmath53 mev , the symmetry energy is @xmath54 mev , and the incompressibility is @xmath55 mev . the rmf theory with the tm1 parameter set provides an excellent description of nuclear matter and finite nuclei @xcite , and it is also shown to agree satisfactorily with experimental data in studies of nuclei with deformed configurations @xcite . for the parameters of @xmath0 hyperons , we use the experimental mass value @xmath56 mev @xcite . concerning the coupling constants between mesons and @xmath0 hyperons , we take @xmath57 based on the naive quark model and @xmath58 determined by fitting experimental binding energies of single-@xmath0 hypernuclei @xcite , which produce an attractive potential of @xmath0 in nuclear matter at saturation density to be around @xmath28 mev . it is known that the inclusion of tensor coupling between @xmath17 and @xmath0 is important to produce small spin - orbit splittings of single-@xmath0 hypernuclei , but it does not contribute to homogeneous matter . the @xmath0 hyperon is a charge neutral and isoscalar particle , so that it does not couple to the @xmath37 meson . it is shown that these parameters can reproduce well the experimental data for both single- and double-@xmath0 hypernuclei @xcite . starting with the lagrangian ( [ eq : lrmf ] ) , we derive a set of euler lagrange equations . we employ the rmf approximation as described in @xcite , where the meson fields are treated as classical fields and the field operators are replaced by their expectation values . for homogeneous matter , the non - vanishing expectation values of meson fields are @xmath59 , @xmath60 , and @xmath61 . the equations of motion for the meson fields in homogeneous matter have the following form : @xmath62 the stationary dirac equations for nucleons and @xmath0 hyperons are given by @xmath63 where @xmath64 stands for the nucleons ( @xmath65 or @xmath66 ) . @xmath67 and @xmath68 are the effective nucleon mass and effective @xmath0 mass , respectively . @xmath25 denotes the index of eigenstates , while @xmath69 and @xmath70 are the single - particle energies . in homogeneous matter , baryons occupy single - particle states with the occupation probability @xmath71 ( @xmath72 , @xmath66 , or @xmath0 ) . at zero temperature , @xmath73 under the fermi surface , while @xmath74 above the fermi surface . for finite temperature , the occupation probability is given by the fermi dirac distribution , @xmath75 } = \frac{1}{1+\exp\left[\left(\sqrt{k^2+{m_{i}^{*}}^2}-\nu_{i}\right ) /t\right ] } , \\ f_{\bar{i}}^{s}=\frac{1}{1+\exp\left[\left(-\varepsilon_{\bar{i}}^{s}+\mu_{i } \right)/t\right ] } = \frac{1}{1+\exp\left[\left(\sqrt{k^2+{m_{i}^{*}}^2}+\nu_{i}\right)/t\right ] } , \end{aligned}\ ] ] where @xmath76 and @xmath77 denote the particle and antiparticle , respectively . @xmath78 and @xmath79 are the single - particle energies . the relation between the chemical potential @xmath80 and the kinetic part of the chemical potential @xmath81 is given by @xmath82 the chemical potential @xmath80 is related to the baryon number density @xmath83 as @xmath84 where the quantum number @xmath25 is replaced by the momentum @xmath85 when we do the integration in the momentum space instead of summing over the eigenstates . we denote @xmath86 as the total baryon number density and @xmath87 as the proton fraction in homogeneous matter . the coupled equations are solved at fixed @xmath88 and @xmath19 . at zero temperature , @xmath0 hyperons appear only at high density when the equilibrium condition @xmath31 could be satisfied . for finite temperature , a small number of @xmath0 hyperons may exist at low density with @xmath31 , and the @xmath0 fraction increases rapidly at high density . the thermodynamic quantities of homogeneous matter have been derived in @xcite , @xcite , and @xcite , so we simply write the expressions here . the energy density of nuclear matter including @xmath0 hyperons is given by @xmath89 the pressure is given by @xmath90 and the entropy density is given by @xmath91 .\end{aligned}\ ] ] in the low - temperature and low - density region , heavy nuclei may be formed in order to lower the free energy . we adopt the thomas fermi approximation in combination with assumed nucleon distribution functions and a free energy minimization to describe the non - uniform matter based on the work by @xcite . in this study , we take into account the contribution from @xmath0 hyperons only when the @xmath0 fraction @xmath92 is larger than @xmath36 . in the region where the heavy nuclei are formed , @xmath92 is quite small , therefore we neglect the contribution from @xmath0 hyperons in the thomas fermi calculation . the non - uniform matter is modeled as a mixture of a single species of heavy nuclei , alpha particles , and free nucleons that exist outside of nuclei , while the leptons can be treated as uniform non - interacting particles separately . for the system with a fixed proton fraction , the leptons play no role in the free energy minimization . hence we mainly pay attention to the baryon contribution in this study . we assume that each heavy spherical nucleus is located in the center of a charge - neutral cell consisting of a vapor of neutrons , protons , and alpha - particles . the nuclei form a body - centered - cubic ( bcc ) lattice to minimize the coulomb lattice energy . it is useful to introduce the wigner seitz cell to simplify the energy of a unit cell . the wigner seitz cell is a sphere whose volume is the same as the unit cell in the bcc lattice . the lattice constant @xmath93 is defined as the cube root of the cell volume , @xmath94 , where @xmath95 and @xmath96 are the baryon number per cell and the average baryon number density , respectively . we define the baryon mass density as @xmath97 with @xmath98 being the atomic mass unit @xcite . we calculate the coulomb energy using the wigner seitz approximation and adding an energy correction for the bcc lattice @xcite . this energy correction is negligible unless the nuclear size is comparable to the cell size . we assume the nucleon distribution function @xmath99 ( @xmath72 or @xmath66 ) in the wigner seitz cell as @xmath100 ^ 3 + n_i^{\rm{out } } , & 0 \leq r \leq r_i , \\ n_i^{\rm{out } } , & r_i \leq r \leq r_{\rm{cell } } , \\ \end{array } \right.\ ] ] where @xmath101 represents the distance from the center of the nucleus and @xmath102 is the radius of the wigner seitz cell defined by @xmath103 . the density parameters @xmath104 and @xmath105 are the densities at @xmath106 and @xmath107 , respectively . the parameters @xmath108 and @xmath109 determine the boundary and the relative surface thickness of the nucleus . for the distribution function of alpha - particle @xmath110 , which should decrease as @xmath101 approaches the center of the nucleus , we assume @xmath111 ^ 3 + n_{\alpha}^{\rm{out } } , & 0 \leq r \leq r_p , \\ n_{\alpha}^{\rm{out } } , & r_{p } \leq r \leq r_{\rm{cell } } , \\ \end{array } \right.\ ] ] which could give @xmath112 and @xmath113 . here we use the same parameters @xmath114 and @xmath115 for both proton and alpha - particle distribution functions in order to avoid the presence of too many parameters in the minimization procedure . the parameters @xmath116 and @xmath117 may be somewhat different from @xmath114 and @xmath115 due to the additional neutrons forming a neutron skin in the surface region . for a system with fixed temperature @xmath34 , proton fraction @xmath19 , and baryon mass density @xmath35 , there are eight independent parameters among the ten variables , @xmath93 , @xmath118 , @xmath119 , @xmath116 , @xmath117 , @xmath120 , @xmath121 , @xmath114 , @xmath115 , and @xmath122 . the thermodynamically favorable state is the one that minimizes the free energy density with respect to these eight independent parameters . in principle , the resulting nucleon distribution in the wigner seitz cell would depend on the form of the parameterization . it is also possible to determine the nucleon distribution by a self - consistent thomas fermi method without any form of parameterization @xcite . we have compared , in figures 1 and 2 of @xcite , the results obtained by the self - consistent thomas fermi method with those using the parameterization of equation ( [ eq : nitf ] ) , and found that they were in good agreement with each other for the cases considered . therefore , equation ( [ eq : nitf ] ) is considered to be a reasonable form of the nucleon distribution in the wigner seitz cell . in this model , the free energy density contributed from baryons is given by @xmath123 where the free energy per cell @xmath124 can be written as @xmath125 the bulk energy @xmath126 , entropy @xmath127 , and free energy @xmath128 are calculated by @xmath129 here @xmath130 , @xmath131 , and @xmath132 are the local energy density , entropy density , and free energy density at the radius @xmath101 , where the system can be considered as a mixed uniform matter of neutrons , protons , and alpha - particles . these local densities are the sum of the contributions from nucleons and alpha particles . we use the rmf theory described in section [ sec:2.1 ] to calculate the nucleon contribution , while the alpha - particles are treated as an ideal boltzmann gas . in general , the number density of alpha - particles is quite small , and therefore the ideal - gas approximation is considered to be a reasonable approximation for alpha - particles . we note that other treatments for alpha - particles have been developed and used in nuclear astrophysics @xcite . the free energy density of alpha - particles in the ideal - gas approximation is given by @xmath133 + n_{\alpha}\left(4m - b_{\alpha}\right),\ ] ] where @xmath134 is the number density of alpha - particles , and we have used the abbreviation @xmath135^{3/2}$ ] . the alpha - particle binding energy @xmath136 mev is taken from @xcite . we have to take into account the volume of alpha - particle , otherwise the alpha - particle fraction would become a large number at high density , where the alpha - particle should actually disappear . when we take into account the volume excluded by alpha - particles , the free energy densities of nucleons and alpha - particles are given by @xmath137 where @xmath138 is the fraction of space occupied by alpha - particles with the effective volume of alpha - particle @xmath139 taken from @xcite . we denote the effective number density of neutrons , protons , or alpha - particles as @xmath140 ( @xmath141 , @xmath142 , or @xmath143 ) . the inclusion of the volume excluded by alpha - particles has negligible effect in the low - density region , whereas it is necessary for the calculation at high density . as for the surface energy term @xmath144 due to the inhomogeneity of the nucleon distribution , we take the simple form @xmath145 the parameter @xmath146 is determined by performing the thomas fermi calculation for finite nuclei so as to reproduce the gross properties of nuclear masses , charge radii , and the beta stability line as described in the appendix of @xcite . the coulomb energy per cell @xmath147 is calculated using the wigner seitz approximation with an added correction term for the bcc lattice @xcite @xmath148\,\phi(r ) d^3r \,+\,\delta e_c,\ ] ] where @xmath149 represents the electrostatic potential calculated in the wigner seitz approximation , @xmath150 is the electron number density of a uniform electron gas ( @xmath151 ) , and @xmath152 is the correction term for the bcc lattice , which is approximated as @xmath153 here @xmath93 is the lattice constant , @xmath154 is taken from @xcite , and @xmath155 is the non - uniform part of the charge number per cell given by @xmath156 ^ 3 4\pi r^2 dr.\ ] ] because of the long - range nature of the coulomb interaction , @xmath147 is dependent on the lattice type . this dependence has been extensively discussed in @xcite . the system prefers the bcc lattice because it gives the lowest coulomb energy . in general , the coulomb energy is dependent on the temperature as discussed in @xcite and @xcite . the dependence of @xmath34 in equation ( [ eq : ec ] ) is contained in the particle distributions . we assume that equation ( [ eq : dec ] ) , which was derived at zero temperature in @xcite , remains valid at finite temperature . since we assume the lattice of nuclei with the wigner seitz approximation , we do not include a contribution from the translational energy of heavy nuclei . this contribution is small in general and does not affect the general behavior of the eos . it may be noticeable only within the limited region of temperature , where nuclei behave as gas without being dissociated into nucleons . we note that this is different from the treatment by @xcite , who included this term as a minor contribution . in this work , we construct the eos tables covering a wide range of temperature @xmath34 , proton fraction @xmath19 , and baryon mass density @xmath35 for the use of core - collapse supernova simulations . we present two sets of eos tables , namely eos2 and eos3 . eos2 takes into account only the nucleon degree of freedom , while eos3 includes additional contributions from @xmath0 hyperons . in comparison with the earlier version ( eos1 ) described in @xcite , several improvements are made in eos2 and eos3 according to the requirements of the users . we largely increase the number of @xmath34 points , and adopt a linear @xmath19 grid in eos2 and eos3 instead of the logarithmic @xmath19 grid used in eos1 . on the other hand , the numerical methods are improved to allow the calculation with equal grid spacing for @xmath35 , while it is only approximately equal at high density in eos1 . for a detailed comparison between the eos tables discussed in this paper , one can see table 1 in section [ sec:1 ] . in principle , the matter at extremely high temperatures and densities is beyond the applicability of the rmf theory , but we still include the results in these exotic regions since they are sometimes necessary in astrophysical simulations . in eos2 , we present results of matter without the inclusion of @xmath0 hyperons . for each @xmath34 , @xmath19 , and @xmath35 , we determine the thermodynamically favorable state that has the lowest free energy in the present model . we perform the free energy minimization for both non - uniform matter and uniform matter . here the phase of heavy nuclei formed together with free nucleons and alpha - particles is referred to as non - uniform matter , while the phase of nucleons mixed with alpha - particles without heavy nuclei is referred to as uniform matter . for non - uniform matter , the minimization procedure is realized by using the thomas fermi method which includes eight independent parameters as described in section [ sec:2.2 ] . for uniform matter , we perform the minimization with respect to converting two protons and two neutrons into an alpha - particle , in which there is only one independent parameter . by comparing the free energies of non - uniform matter and uniform matter , we determine the most favorable state and estimate the phase transition between non - uniform matter and uniform matter . we first discuss the phase diagram of nuclear matter at finite temperature . it is known that the density of the phase transition between uniform matter and non - uniform matter depends on both @xmath34 and @xmath19 . the non - uniform matter phase can exist only in the low - temperature region ( @xmath157 mev ) . in figure [ fig : trho ] , we show the phase diagram in the @xmath35@xmath34 plane for @xmath158 , @xmath159 , and @xmath21 . the shaded region corresponds to the non - uniform matter phase where heavy nuclei are formed . the dashed line is the boundary where the alpha - particle fraction @xmath160 changes between @xmath161 and @xmath162 . it is shown that heavy nuclei can exist in the medium - density and low - temperature region . the phase of nuclear matter at low density is a homogeneous nucleon gas with a small fraction of alpha - particles , the heavy nuclei are formed at some medium densities where the system can lower the free energy by forming heavy nuclei , and it becomes uniform matter as the density increases beyond @xmath163 . it is seen that the starting density of the non - uniform matter phase depends on @xmath34 strongly , while the ending density is nearly independent of @xmath34 . as the temperature increases , the density range of the non - uniform matter phase becomes narrower , and it disappears completely for @xmath164 mev . in figure [ fig : yprho ] , we show the phase diagram in the @xmath35@xmath19 plane for @xmath165 , @xmath166 , and @xmath12 mev . the shaded region corresponds to the non - uniform matter phase , while the dashed line is the boundary between @xmath161 and @xmath162 . it is shown that the dependence of the boundary on @xmath19 is relatively weak except at very small values of @xmath19 . in the case of @xmath167 mev ( middle panel ) , the non - uniform matter phase disappears at lower @xmath19 because it is difficult to form heavy nuclei with smaller values of @xmath19 . the alpha - particles still exist at @xmath168 mev , and @xmath160 reaches to be more than @xmath169 at some medium densities for @xmath170 as shown in the top panel of figure [ fig : yprho ] . we note that the phase diagram of nuclear matter at zero temperature has been discussed in @xcite . in the non - uniform matter , nucleon distributions are determined by minimizing the free energy density with respect to the independent parameters in the thomas fermi approximation . the heavy nuclei are assumed to form a bcc lattice in order to minimize the coulomb lattice energy . in figure [ fig : nm_t1 ] , we show the nucleon distributions along the straight line joining the centers of the nearest nuclei in the bcc lattice for the case of @xmath165 mev and @xmath171 . it is found that the nuclei become heavier and get closer with each other as the density increases . in principle , there exist free nucleons and alpha - particles outside heavy nuclei at finite temperature , but their densities are too small to be observed in figure [ fig : nm_t1 ] . we show in figure [ fig : nm_t10 ] the same quantities as in figure [ fig : nm_t1 ] but for the case of @xmath167 mev and @xmath171 . we note that the starting density of the non - uniform matter phase in this case is @xmath172 , so the heavy nucleus does not exist at @xmath173 for @xmath167 mev which is different from the case of @xmath165 mev as shown in the bottom panel of figure [ fig : nm_t1 ] . comparing with the top and middle panels of figure [ fig : nm_t1 ] , it is seen that there are much more free nucleons and alpha - particles outside nuclei , while the nucleons inside nuclei get less . this is because the role of entropy becomes more important at higher temperature and the free energy can be lowered if the nucleons are freed from nuclei . in figure [ fig : az ] , we plot the nuclear mass number @xmath174 and charge number @xmath175 as a function of the baryon mass density @xmath35 for @xmath171 at @xmath165 mev and @xmath167 mev . it is shown that @xmath174 and @xmath175 have relatively weak dependence on @xmath35 at lower density and they increase rapidly just before the phase transition at @xmath176 . for the same @xmath35 and @xmath19 , the values of @xmath174 and @xmath175 at @xmath167 mev are much smaller than those at @xmath165 mev . this is because more nucleons are freed from nuclei at higher temperature , and it is eventually impossible to form heavy nuclei for @xmath164 mev in the present model . in figure [ fig : xirho2 ] , we show the fraction of neutrons , protons , alpha - particles , and heavy nuclei as a function of the baryon mass density @xmath35 for @xmath171 at @xmath165 , @xmath166 , and @xmath177 mev . at low density , the matter is a uniform gas of neutrons and protons with a small fraction of alpha - particles . the alpha - particle fraction @xmath160 increases with increasing @xmath35 , but the formation of heavy nuclei at low temperatures causes a rapid drop of @xmath160 , @xmath178 , and @xmath179 , which is due to the fact that heavy nuclei use up most of the nucleons in non - uniform matter . when the density increases beyond @xmath163 , the heavy nuclei dissolve and the matter becomes uniform . the alpha - particles may exist up to @xmath180 where the volume excluded by alpha - particles plays an important role that it is unfavorable to have alpha - particles in the uniform matter at such high density . for @xmath164 mev , the heavy nuclei can not be formed , but there are finite values of @xmath160 , especially at medium densities as shown in the top panel of figure [ fig : xirho2 ] . the alpha - particle fraction decreases with increasing temperature , and we find that @xmath160 at @xmath168 mev is of the order of @xmath169 at some medium densities . the alpha - particle fraction may be significantly affected if the alpha - particle binding energy @xmath181 is dependent on the density and temperature as discussed in @xcite . for simplicity , we neglect this dependence and take @xmath136 mev in the present model . since we treat the uniform matter and non - uniform matter consistently using the same rmf theory , all the resulting thermodynamic quantities are consistent and smooth in the whole range . we will discuss the thermodynamic quantities in eos2 and compare with those in eos3 in the next section . in this section , we discuss results of matter with the inclusion of @xmath0 hyperons given in eos3 . it is found that the contribution from @xmath0 hyperons is negligible in the low - temperature and low - density region . in fact , at zero temperature @xmath0 hyperons appear only at high density when the equilibrium condition @xmath31 could be satisfied . at low temperature , such as @xmath167 mev , the @xmath0 fraction @xmath92 is smaller than @xmath36 at densities below normal nuclear matter density . for simplicity , we take into account the contribution from @xmath0 hyperons only when @xmath92 is larger than @xmath36 . in the non - uniform matter phase , @xmath92 is quite small , therefore we neglect the contribution from @xmath0 hyperons in the thomas fermi calculation . in figure [ fig : xirho3 ] , we show the fraction @xmath182 ( @xmath183 , @xmath66 , @xmath142 , @xmath143 , or heavy nuclei @xmath174 ) as a function of @xmath35 for @xmath171 at @xmath165 , @xmath166 , and @xmath177 mev . for lower temperatures , such as @xmath165 mev ( bottom panel ) and @xmath167 mev ( middle panel ) , there is a significant fraction of @xmath0 hyperons only at high density . it is shown that @xmath92 increases with increasing @xmath35 , which causes a decrease of @xmath179 . we note that @xmath178 , which is equal to @xmath19 in the uniform matter at high density , has been fixed to be @xmath159 in this figure . for @xmath184 mev ( top panel ) , @xmath92 is of the order of @xmath185 at low density , and increases rapidly at high density . comparing with the top panel of figure [ fig : xirho2 ] , it is seen that the inclusion of @xmath0 hyperons does not affect @xmath160 and @xmath178 significantly . we find the contribution from @xmath0 hyperons increases with increasing temperature . at @xmath168 mev , @xmath92 is about 12% at low density for the case of @xmath171 . on the other hand , @xmath92 decreases with increasing @xmath19 at fixed @xmath35 and @xmath34 , since there are less neutrons at higher @xmath19 to realize the equilibrium condition @xmath31 . we now discuss the thermodynamic quantities in eos3 and compare with those in eos2 . here we mainly discuss their properties at high density where there are noticeable differences between the results with and without @xmath0 hyperons . as for the behavior of thermodynamic quantities at low density , one can refer to our earlier work @xcite . in figure [ fig : f ] , we show the free energy per baryon @xmath186 as a function of the baryon mass density @xmath35 with @xmath158 and @xmath21 at @xmath165 , @xmath166 , and @xmath12 mev . the results with @xmath0 hyperons given in eos3 are shown by solid lines , while those without @xmath0 hyperons given in eos2 are displayed by dashed lines . it is found that the inclusion of @xmath0 hyperons can lower the free energy and this effect increases with increasing @xmath35 . however , this effect decreases with increasing @xmath19 as shown in figure [ fig : f ] , which is due to the fact that @xmath92 decreases with increasing @xmath19 . we show in figure [ fig : p ] the pressure @xmath142 as a function of @xmath35 with @xmath158 and @xmath21 at @xmath165 , @xmath166 , and @xmath12 mev . the pressure is calculated from the derivative of the free energy as described in appendix a. therefore , the effect of @xmath0 hyperons on the pressure is similar to the one observed in figure [ fig : f ] . it is obvious that the inclusion of @xmath0 hyperons tends to soften the eos at high density . in figure [ fig : s ] , we show the entropy per baryon @xmath187 as a function of @xmath35 with @xmath158 and @xmath21 at @xmath165 , @xmath166 , and @xmath12 mev . at @xmath165 mev ( bottom panel ) it is hard to see the difference between the results with and without @xmath0 hyperons , while at @xmath167 mev ( middle panel ) there are small differences at high density . for the case of @xmath168 mev ( top panel ) , the effect of @xmath0 hyperons can be seen in the whole range of density , since @xmath92 reaches @xmath188 ( @xmath189 ) at low density for @xmath158 ( @xmath21 ) at @xmath168 mev . in general , the inclusion of @xmath0 hyperons tends to increase the entropy . in @xcite , we have discussed the effect of the formation of nuclei on the entropy , which has a strong @xmath19 dependence . the behavior of @xmath187 at @xmath190 in the middle and bottom panels of figure [ fig : s ] is due to the formation of heavy nuclei in non - uniform matter . for neutron star matter at zero temperature , the inclusion of @xmath0 hyperons leads to a softening of the eos of neutron star matter and thereby a corresponding reduction in the maximum mass of neutron stars . we use eos3 to calculate the neutron star properties and find that the maximum mass of neutron stars is about 1.75 @xmath191 , whereas the value for eos1 and eos2 is 2.18 @xmath191 . according to the recent measurement of psr j1614 - 2230 ( @xmath27 ; @xcite ) , eos3 seems to be too soft due to the inclusion of @xmath0 hyperons , while eos1 and eos2 are compatible with the observation of psr j1614 - 2230 . this is a common difficulty for eoss with the inclusion of hyperons . so far , there are large uncertainties in the properties of matter at high density , which are crucial for determining the maximum mass of neutron stars . in this paper , we have presented two sets of the eos tables ( eos2 and eos3 ) covering a wide range of temperature @xmath34 , proton fraction @xmath19 , and baryon mass density @xmath35 for the use of core - collapse supernova simulations . the difference between eos2 and eos3 is that only the nucleon degree of freedom is taken into account in eos2 , while eos3 includes additional contributions from @xmath0 hyperons . in comparison with the earlier version ( eos1 ) described in @xcite , several improvements have been made in eos2 and eos3 according to the requirements of the users . we have largely increased the number of @xmath34 points , and adopted a linear @xmath19 grid in eos2 and eos3 instead of the logarithmic @xmath19 grid used in eos1 . in addition , we have performed the calculation with equal grid spacing in the whole range of density , while it is only approximately equal at high density in eos1 . we have presented a detailed comparison between these eoss in table 1 . we have employed the rmf theory with nonlinear @xmath16 and @xmath17 terms , which can provide an excellent description of nuclear matter and finite nuclei including unstable ones . the thomas fermi approximation in combination with assumed nucleon distribution functions and a free energy minimization has been adopted to describe the non - uniform matter that is modeled as a mixture of a single species of heavy nuclei , alpha particles , and free nucleons outside of nuclei . the rmf results have been taken as input in the thomas fermi calculation , so the treatments of non - uniform matter and uniform matter in this eos are sufficient to obtain the table in a consistent manner . it would be preferable to treat the mixture of nuclei as recently done in @xcite , @xcite , and @xcite for the detailed treatment of electron captures on nuclei in supernova core . however , it is beyond the scope of the current update of the shen eos table , which has been routinely used in astrophysical simulations . we have included the contribution from @xmath0 hyperons in eos3 , when the @xmath0 fraction is larger than @xmath36 . it is based on the consideration that the @xmath0 hyperon is most likely to occur in dense matter among all possible non - nucleonic degrees of freedom because it is the lightest hyperon with an attractive potential in nuclear matter . the potential depth of @xmath0 in nuclear matter is estimated to be around @xmath28 mev from the experimental binding energies of single-@xmath0 hypernuclei . several recent observations of double-@xmath0 hypernuclei suggest that the effective @xmath29 interaction should be weakly attractive . the rmf theory with nonlinear @xmath16 and @xmath17 terms has been extended to include hyperons , and it can provide a reasonable description of single- and double-@xmath0 hypernuclei @xcite . we have not included contributions from other hyperons , such as @xmath24 and @xmath30 , due to their relatively high threshold densities and lack of available experimental data . in eos3 , we have performed a consistent calculation including @xmath0 hyperons for the entire table , whereas the eos tables of high density matter were connected with eos1 at low density in the simple procedure described in section 2.3 of @xcite , so eos3 is considered to be constructed in a consistent manner for the inclusion of @xmath0 hyperons . we have examined the effect of @xmath0 hyperons on the properties of dense matter . the contribution from @xmath0 hyperons is negligible in the low - temperature and low - density region , whereas it tends to soften the eos at high density . in principle , the matter at extremely high temperatures and densities is beyond the applicability of the rmf theory , but we still include the results in these exotic regions since they are sometimes necessary in astrophysical simulations . it will continue to be a challenge for nuclear physics to provide realistic eos , which should be supported by microscopic theory and available experimental data , for use in astrophysical studies . we thank c. d. ott and t. fischer for helpful suggestions and fruitful discussions . k.s . expresses his thanks to k. nakazato , a. ohnishi , h. suzuki , and s. yamada for practical discussions on the development of the eos tables for simulations . h.s . and k.s thank the organizers of the micra2011 workshop hosted by the perimeter institute . this work was partially supported by the national natural science foundation of china ( grant nos . 10675064 and 11075082 ) , the grant - in - aid for scientific research on innovative areas ( nos . 20105004 and 20105005 ) , and the grant - in - aid for the scientific research ( no . 22540296 ) from the ministry of education , culture , sports , science and technology ( mext ) of japan . this work for numerical simulations of core - collapse supernovae is partially supported by the hpci strategic program of mext in japan , which assists k.s . to use the supercomputers at yukawa institute for theoretical physics ( yitp ) in kyoto university , research center for nuclear physics ( rcnp ) in osaka university , the university of tokyo , japan atomic energy agency ( jaea ) and high energy accelerator research organization ( kek ) . the eos tables are available on the web at + _ http://physics.nankai.edu.cn/grzy/shenhong/eos/index.html_ , + _ http://www.rcnp.osaka-u.ac.jp/@xmath192shen/_ , + _ http://user.numazu - ct.ac.jp/@xmath192sumi / eos / index.html_. + the table under the name @xmath193.tab is the main eos table , while @xmath193.t00and @xmath193.yp0 are those for @xmath4 and @xmath5 , respectively . eos2 and eos3 are presented in electronic tables , tables 38 , in this paper . we present the main eos table of eos2 ( eos3 ) in table 3 ( table 6 ) , which is 149 mb ( 167 mb ) in size . table 4 ( table 7 ) is the one of eos2 ( eos3 ) at @xmath4 which is 1.6 mb ( 1.8 mb ) , while table 5 ( table 8) is the one for @xmath5 which is 2.3 mb ( 2.6 mb ) . the tables are written in the order of increasing @xmath34 . in the tables on the web , the values of @xmath194 and @xmath34 ( in mev ) are given at the beginning of each block , and the blocks with different @xmath34 are divided by the string of characters cccccccccccc . in tables 38 of this paper , @xmath194 and @xmath34 are set in columns 1 and 2 . we note that the entry for @xmath194 is set to @xmath195 in the tables for the case of @xmath4 . for each @xmath34 , we present the results in the order of increasing @xmath19 and @xmath35 . the quantities in one line of tables 38 are defined as follows . logarithm of temperature : @xmath194 [ mev ] . * temperature : @xmath34 [ mev ] . logarithm of baryon mass density : @xmath196 [ @xmath197 . * \4 . baryon number density : @xmath88 [ @xmath198 . + the baryon number density is related to the baryon mass density as @xmath97 with @xmath199 mev being the atomic mass unit taken from @xcite . * proton fraction : @xmath19 . + the proton fraction @xmath19 of uniform matter is defined by @xmath200 where @xmath201 , @xmath202 , @xmath203 , and @xmath134 are the number density of protons , neutrons , @xmath0 hyperons , and alpha - particles , respectively . @xmath88 is the total baryon number density . for non - uniform matter case , @xmath19 is the average proton fraction defined by @xmath204 where @xmath205 and @xmath95 are the proton and baryon numbers per cell given by @xmath206 d^3r , \\ n_b & = & \int_{\rm{cell } } \left[\ , n_n\left(r\right ) + n_p\left(r\right ) + 4n_{\alpha}\left(r\right ) \,\right ] d^3r.\end{aligned}\ ] ] here , @xmath207 and @xmath208 are the proton and neutron distributions given by equation ( [ eq : nitf ] ) , and @xmath110 is the alpha - particle distribution given by equation ( [ eq : na ] ) . because the @xmath0 fraction is very small in the non - uniform matter phase , we neglect the contribution from @xmath0 hyperons in this case . free energy per baryon : @xmath186 [ mev ] . + the free energy per baryon is defined relative to the free nucleon mass @xmath209 mev in the tm1 parameter set as @xmath210 * \7 . internal energy per baryon : @xmath211 [ mev ] . + the internal energy per baryon is defined relative to the atomic mass unit @xmath199 mev as @xmath212 * \8 . entropy per baryon : @xmath187 [ @xmath213 . + the entropy per baryon is related to the entropy density via @xmath214 * \9 . mass number of the heavy nucleus : @xmath174 . + the mass number of the heavy nucleus is defined by @xmath215 4\pi r^2 dr,\ ] ] where @xmath216 is the maximum of @xmath114 and @xmath116 , which is considered as the boundary of the heavy nucleus . * charge number of the heavy nucleus : @xmath175 . + the charge number of the heavy nucleus is defined by @xmath217 * \11 . effective nucleon mass : @xmath218 [ mev ] . + the effective nucleon mass is obtained in the rmf theory for uniform matter . in the non - uniform matter phase , the effective nucleon mass is a function of space due to inhomogeneity of the nucleon distribution , so it is meaningless to list the effective nucleon mass for non - uniform matter . we replace the effective nucleon mass @xmath218 by the free nucleon mass @xmath219 in the non - uniform matter phase . free neutron fraction : @xmath179 . + the free neutron fraction is given by @xmath220 where @xmath221 is the cell volume , @xmath222 is the volume outside the nucleus , @xmath119 is the number density of free neutrons outside the nucleus , and @xmath88 is the average baryon number density . free proton fraction : @xmath178 . + the free proton fraction is given by @xmath223 where @xmath121 is the number density of free protons outside the nucleus . alpha - particle fraction : @xmath160 . + the alpha - particle fraction is defined by @xmath224 where @xmath225 is the alpha - particle number per cell obtained by @xmath226 and @xmath110 is the alpha - particle distribution given by equation ( [ eq : na ] ) . * heavy nucleus fraction : @xmath227 . + the heavy nucleus fraction is defined by @xmath228 where @xmath174 is the mass number of the heavy nucleus as defined in equation ( [ eq : an ] ) . pressure : @xmath142 [ @xmath229 . + the pressure is calculated through the thermodynamic relation @xmath230_{t , y_p}.\ ] ] * \17 . chemical potential of the neutron : @xmath231 [ mev ] . + the chemical potential of the neutron relative to the free nucleon mass @xmath219 is calculated through the thermodynamic relation @xmath232_{t , n_p},\ ] ] where @xmath233 is the sum of the neutron and @xmath0 densities . * chemical potential of the proton : @xmath234 [ mev ] . + the chemical potential of the proton relative to the free nucleon mass @xmath219 is calculated through the thermodynamic relation @xmath235_{t , n_{n+\lambda}}.\ ] ] * \19 . effective @xmath0 mass : @xmath236 [ mev ] . + the effective @xmath0 mass is obtained in the rmf theory for uniform matter . we replace the effective @xmath0 mass by the free @xmath0 mass @xmath56 mev when the @xmath0 hyperon is not taken into account . @xmath0 fraction : @xmath92 . + the @xmath0 fraction is given by @xmath237 where @xmath203 is the number density of @xmath0 hyperons in the uniform matter phase . we note that @xmath238 is adopted in the non - uniform matter phase . we have done the following checks on the eos tables . ccccc constituents & uniform matter & @xmath66 , @xmath142 , @xmath143 & @xmath66 , @xmath142 , @xmath143 & @xmath66 , @xmath142 , @xmath143 , @xmath0 + & non - uniform matter & @xmath66 , @xmath142 , @xmath143 , @xmath174 & @xmath66 , @xmath142 , @xmath143 , @xmath174 & @xmath66 , @xmath142 , @xmath143 , @xmath174 + @xmath34 & range & @xmath242 & @xmath243 & @xmath243 + ( mev ) & grid spacing & @xmath244 & @xmath245 & @xmath245 + & points & 32 ( including @xmath4 ) & 92 ( including @xmath4 ) & 92 ( including @xmath4 ) + & range & @xmath246 & @xmath247 & @xmath247 + @xmath19 & grid spacing & @xmath248 & @xmath249 & @xmath249 + & points & 72 ( including @xmath5 ) & 66 & 66 + @xmath35 & range & @xmath250 & @xmath251 & @xmath251 + @xmath252 & grid spacing & @xmath253 & @xmath254 & @xmath254 + & points & 104 & 110 & 110 + reference & & @xcite & this paper & this paper + cccccccccccccccccc -1.0 & 0.1 & 5.1 & 7.581e-11 & 0.01 & -1.524 & 6.408 & 14.27 & 91.37 & 29.40 & 938 & 0.9689 & 0 & 0 & 0.03107 & 7.344e-12 & -1.222 & -21.73 + -1.0 & 0.1 & 5.2 & 9.544e-11 & 0.01 & -1.502 & 6.408 & 14.04 & 91.39 & 29.32 & 938 & 0.9688 & 0 & 0 & 0.03116 & 9.245e-12 & -1.199 & -21.78 + -1.0 & 0.1 & 5.3 & 1.201e-10 & 0.01 & -1.480 & 6.408 & 13.82 & 91.56 & 29.33 & 938 & 0.9687 & 0 & 0 & 0.03121 & 1.164e-11 & -1.176 & -21.83 cccccccccccccccccc -100 & 0 & 5.1 & 7.581e-11 & 0.01 & -0.2446 & 6.261 & 0 & 96.11 & 28.40 & 938 & 0.9661 & 0 & 0 & 0.03383 & 5.848e-16 & 3.571e-05 & -24.46 + -100 & 0 & 5.2 & 9.544e-11 & 0.01 & -0.2446 & 6.261 & 0 & 96.11 & 28.40 & 938 & 0.9661 & 0 & 0 & 0.03383 & 8.889e-16 & 4.156e-05 & -24.46 + -100 & 0 & 5.3 & 1.201e-10 & 0.01 & -0.2446 & 6.261 & 0 & 96.12 & 28.41 & 938 & 0.9661 & 0 & 0 & 0.03383 & 1.328e-15 & 4.770e-05 & -24.46 cccccccccccccccccc -100 & 0 & 5.1 & 7.581e-11 & 0 & 2.281e-03 & 6.508 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -1.703e-13 & 3.553e-05 & -938 + -100 & 0 & 5.2 & 9.544e-11 & 0 & 2.240e-03 & 6.508 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -2.099e-13 & 4.142e-05 & -938 + -100 & 0 & 5.3 & 1.201e-10 & 0 & 1.151e-03 & 6.507 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -1.323e-13 & 4.828e-05 & -938 cccccccccccccccccccc -1.0 & 0.1 & 5.1 & 7.581e-11 & 0.01 & -1.524 & 6.408 & 14.27 & 91.37 & 29.40 & 938 & 0.9689 & 0 & 0 & 0.03107 & 7.344e-12 & -1.222 & -21.73 & 1115.7 & 0 + -1.0 & 0.1 & 5.2 & 9.544e-11 & 0.01 & -1.502 & 6.408 & 14.04 & 91.39 & 29.32 & 938 & 0.9688 & 0 & 0 & 0.03116 & 9.245e-12 & -1.199 & -21.78 & 1115.7 & 0 + -1.0 & 0.1 & 5.3 & 1.201e-10 & 0.01 & -1.480 & 6.408 & 13.82 & 91.56 & 29.33 & 938 & 0.9687 & 0 & 0 & 0.03121 & 1.164e-11 & -1.176 & -21.83 & 1115.7 & 0 cccccccccccccccccccc -100 & 0 & 5.1 & 7.581e-11 & 0.01 & -0.2446 & 6.261 & 0 & 96.11 & 28.40 & 938 & 0.9661 & 0 & 0 & 0.03383 & 5.848e-16 & 3.571e-05 & -24.46 & 1115.7 & 0 + -100 & 0 & 5.2 & 9.544e-11 & 0.01 & -0.2446 & 6.261 & 0 & 96.11 & 28.40 & 938 & 0.9661 & 0 & 0 & 0.03383 & 8.889e-16 & 4.156e-05 & -24.46 & 1115.7 & 0 + -100 & 0 & 5.3 & 1.201e-10 & 0.01 & -0.2446 & 6.261 & 0 & 96.12 & 28.41 & 938 & 0.9661 & 0 & 0 & 0.03383 & 1.328e-15 & 4.770e-05 & -24.46 & 1115.7 & 0 cccccccccccccccccccc -100 & 0 & 5.1 & 7.581e-11 & 0 & 2.281e-03 & 6.508 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -1.703e-13 & 3.553e-05 & -938 & 1115.7 & 0 + -100 & 0 & 5.2 & 9.544e-11 & 0 & 2.240e-03 & 6.508 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -2.099e-13 & 4.142e-05 & -938 & 1115.7 & 0 + -100 & 0 & 5.3 & 1.201e-10 & 0 & 1.151e-03 & 6.507 & 0 & 0 & 0 & 938 & 1.0 & 0 & 0 & 0 & -1.323e-13 & 4.828e-05 & -938 & 1115.7 & 0 , @xmath159 , and @xmath21 ( bottom to top ) in the @xmath35@xmath34 plane . the shaded region corresponds to the non - uniform matter phase where heavy nuclei are formed . the dashed line is the boundary where the alpha - particle fraction @xmath160 changes between @xmath161 and @xmath162 . ( a color version of this figure is available in the online journal.),width=8 ] , @xmath166 , and @xmath12 mev ( bottom to top ) in the @xmath35@xmath19 plane . the shaded region corresponds to the non - uniform matter phase where heavy nuclei are formed . the dashed line is the boundary where the alpha - particle fraction @xmath160 changes between @xmath161 and @xmath162 . the non - uniform matter phase does not exist at @xmath168 mev ( top ) . ( a color version of this figure is available in the online journal.),width=8 ] ( top ) and charge number @xmath175 ( bottom ) as a function of the baryon mass density @xmath35 for @xmath171 at @xmath165 mev ( solid ) and @xmath167 mev ( dashed ) . ( a color version of this figure is available in the online journal.),width=8 ] as a function of the baryon mass density @xmath35 with @xmath158 ( blue ) and @xmath21 ( red ) at @xmath165 , @xmath166 , and @xmath12 mev ( bottom to top ) . the results with @xmath0 hyperons given in eos3 are shown by solid lines , while those without @xmath0 hyperons given in eos2 are displayed by dashed lines . ( a color version of this figure is available in the online journal.),width=8 ]

the flux power spectrum of the lyman- forest in quasar ( qso ) absorption spectra is sensitive to a wide range of cosmological and astrophysical parameters and instrumental effects . modelling the flux power spectrum in this large parameter space to an accuracy comparable to the statistical uncertainty of large samples of qso spectra is very challenging . we use here a coarse grid of hydrodynamical simulations run with to obtain a `` best guess '' model around which we calculate a finer grid of flux power spectra using a taylor expansion of the flux power spectrum to first order . in this way , we investigate how the interplay between astrophysical and cosmological parameters affects their measurements using the recently published flux power spectrum obtained from 3035 sdss qsos ( mcdonald et al . 2004 ) . we find that the sdss flux power spectrum alone is able to constrain a wide range of parameters including the amplitude of the matter power spectrum , the matter density , the spectral index of primordial density fluctuations , the effective optical depth and its evolution . the thermal history of the intergalactic medium ( igm ) is , however , poorly constrained and the sdss data favour either an unplausibly large temperature or an unplausibly steep temperature - density relation . by enforcing a thermal history of the igm consistent with that inferred from high - resolution qso spectra , we find the following values for the best fitting model ( assuming a flat universe with a cosmological constant and zero neutrino mass ) : , , ( error bars ) . the values for and are consistent with those obtained by mcdonald et al . with different simulations for similar assumptions . we argue , however , that the major uncertainties in this measurement are still systematic rather than statistical . = 6.5truein # 1*mh : # 1 * [ 2]@series cosmology : intergalactic medium , cosmological parameters large - scale structure of universe quasars : absorption lines

astro-ph0508177

the last few years haven seen the establishment of the lyman-@xmath0 forest as one of the major observational tools to probe the matter power spectrum . measurements of the matter power from the lyman-@xmath0 forest data extend to smaller scales and probe a complementary redshift range than those using the cosmic microwave background ( cmb ) , galaxy surveys or weak gravitational lensing . the lyman-@xmath0 forest is thus ideally suited to constrain cosmological parameters which affect the power spectrum on small scales like the neutrino mass and the mass of dark matter ( dm ) particles ( seljak et al . 2004 ; viel et al . 2005 ) . in a combined analysis with cmb data the lyman-@xmath0 forest results have also been pivotal in tightening constraints on the power law index of the power spectrum of primordial density fluctuations @xmath3 . a consistent picture emerges suggesting that the fluctuation amplitude of the matter power spectrum @xmath1 is rather high , that the spectral index of primordial density fluctuations is consistent with the value @xmath9 and that there is no evidence for a running of the spectral index , significant neutrino mass or a deviation from a cold dark matter spectrum at small scales ( croft et al . 1998 , mcdonald et al . 2000 , hui et al . 2001 , croft et al . 2002 [ c02 ] , mcdonald 2003 , viel et al . 2003 , viel , haehnelt & springel 2004 [ vhs ] , desjacques & haehnelt 2004 , viel , weller & haehnelt 2004 , mcdonald et al . 2004 [ m04a ] , mcdonald et al . 2004b [ m04b ] , viel et al . 2005 , lidz et al . 2005 ) . major uncertainties are thereby the assumed effective optical depth and thermal history of the intergalactic medium ( igm ) and the numerical limitations in obtaining accurate theoretical predictions for the flux power spectrum for a large parameter space ( see vhs and m04a for a detailed discussion ) . vhs and viel , weller & haehnelt ( 2004 ) recovered the linear dark matter power spectrum amplitude and its slope from a new set of 27 high resolution high signal - to - noise qsos ( the luqas sample , kim et al . 2004 [ k04 ] ) and reanalysed the earlier results of c02 . viel et al . found @xmath10 , @xmath11 ( @xmath8 ) and no evidence for a ( large ) running spectral index when they combined the lyman-@xmath0 forest data with the wmap data . similar results have have been obtained subsequently by the sdss collaboration from a much larger sample of 3035 low - resolution low s / n spectra with significantly wider reshift coverage ( m04b ; seljak et al . these findings corrected earlier claims of spergel et al . ( 2003 ) who used the data of c02 combined with cmb and galaxy survey data to argue for a significant tilt away from the canonical harrison - zeldovich @xmath9 spectrum and possibly also for a running of the spectral index ( see seljak et al . ( 2003 ) for the first suggestion that this result may be due to the assumption of an unplausibly large effective optical depth ) . large samples of qso absorption spectra offer the opportunity to obtain the flux power spectrum for a wide range of redshifts with an accuracy at the percent level . despite the generally good agreement between different groups achieved in the last couple of years there are , however , still major open issues of what and how large the uncertainties in the cosmological and astrophysical parameters inferred from the flux power spectrum are ( see vhs and m04b for discussions with different views ) . it will be important to resolve these issues if further progress is to be made . the two data sets used by vhs and m04b and their theoretical modelling were very different . the sdss qso data set analysed by m04a consists of @xmath12 qso spectra at low resolution ( @xmath13 ) and low s / n ( @xmath14 per pixel ) spanning a wide range of redshifts , while the luqas and the c02 samples contain mainly high resolution ( @xmath15 ) , high signal - to - noise ( @xmath16 per pixel ) qso spectra with median redshifts of @xmath17 and @xmath18 , respectively . the analysis methods to infer cosmological and astrophysical parameters were also very different . m04b modelled the flux power spectrum using a large number of hpm simulations ( gnedin & hui 1998 ; viel , haehnelt & springel 2005 ) exploring a large multi - dimensional parameter space . viel et al . ( 2004 ) improved instead the effective bias method developed by c02 . they used a grid of full hydrodynamical simulations run with the tree - sph code ( springel , yoshida & white 2001 ; springel 2005 ) to invert the non - linear relation between flux and matter power spectrum . both methods have their own set of problems . m04b had to use the approximate hpm method which they could only calibrate on a small number of hydro simulations of small box size ( see viel , haehnelt & springel ( 2005 ) for a discussion of the accuracy of hpm simulations ) . m04b further had to compromise on box size and even time - stepping in order to explore a large parameter space . this lead to large number of uncertain corrections which added to another large set of corrections necessary because of the rather low resolution and s / n of their data . the method used by vhs on the other hand requires differentiating the 1d flux power spectrum to obtain the `` 3d '' flux power spectrum , which introduces rather large errors . this makes the assessment of statistical errors somewhat difficult . we refer to zaldarriaga , hui & tegmark ( 2001 ) ; gnedin & hamilton ( 2002 ) ; zaldarriaga , scoccimarro & hui ( 2003 ) and seljak , mcdonald & makarov ( 2003 ) for a further critical assessments of the use of the effective bias method to infer cosmological parameters . vhs concluded that the best they can do was to give an estimate of the many systematic uncertainties involved and to combine these in a conservative way in the final result . however , for attemtps to take advantage of the full redshift coverage and smaller errors of the observed sdds flux power spectrum the effective bias stops being useful . further motivated by the differences in the data and in the theoretical modelling performed by the two groups , we will analyse here the sdss flux power spectrum with high resolution , large box - size full hydrodynamical simulations . ideally , one would like to be able to repeat the analysis made by the sdss collaboration by using a very fine grid of full hydrodynamical simulations in order to be able to sample the multi - dimensional parameter space . however , at present , this approach is not feasible since hydrodynamical simulations are very time consuming . we therefore decided to concentrate most of the analysis on a small region of the parameter space , around the best - fit models obtained by the two groups . we thereby use a taylor expansion to approximate the flux power spectrum around a best - guess model . this approximation should be reasonably accurate for little displacements in parameter space . we have checked the validity of the approximation for a few models . however , we caution the reader that the final error bars on recovered astrophysical and cosmological parameters is likely to be somewhat underestimated due to unaccounted errors of this approximation . the plan of this paper is as follows . in section [ data ] we briefly describe the sdss data set , while the hydrodynamical simulations are discussed in section [ hydro ] . section [ results ] describes the technical details of our modelling of the flux power spectrum and the reader more interested in the results may go straight to section [ chifitting ] where we discuss our findings for the cosmological and astrophysical parameters . section [ conclu ] contains a summary and an outline of possible ways of improving these measurements . mcdonald et al . ( 2004a ) have presented the flux power spectrum of a large sample of 3035 absorption spectra of qso in the redshift range @xmath19 drawn from the dr1 and dr2 data releases of sdss . with a spectral resolution of @xmath13 the typical absorption features with a width of @xmath20 km / s are not resolved . the signal - to - noise of the individual spectra is rather low , s / n @xmath14 per pixel . the large number of spectra means , however , that the flux power spectrum on scales a factor of a few larger than the thermal cut - off can be measured with small statistical errors . m04a have re - analyzed the raw absorption spectra and have investigated the effect of noise , resolution of the spectrograph , sky subtraction , quasar continuum and associated metal absorption . m04a make corrections for these effects and give estimates of the errors for the most important of these corrections . the corrections are not small . the noise contribution to the flux power spectrum rises from 15 - 30 percent at the smallest wavenumbers to order unity at the largest wavenumbers and varies with redshift . the correction for uncorrelated metal absorption are generally a factor five to ten smaller than this . the correction for resolution varies from 1% at the smallest wavenumbers to a factor four at the largest wavenumbers . m04a also identified a correlated siiii feature in the absorption spectra for which they present an empirical fit to the effect on the flux power spectrum . as final result of their analysis m04a present their best estimate for the flux power spectrum @xmath21 at 12 wavenumbers in the range @xmath22(s / km)@xmath23 , equally spaced in @xmath24 for @xmath25 . at the time we started this work the two highest redshift bins were not publicly available so we will not use them in the following analysis . we will here use this flux power spectrum together with the recommended corrections to the data and the recommended treatment of the errors of these corrections . we will come back to this in more detail in section [ datacorr ] . note that we have also dropped the highest redshift bin at @xmath26 as we could not fit it well with our models for the flux power spectrum ( see section [ bestguess ] for more details ) . lccccc @xmath27 & @xmath28 & @xmath29 & @xmath30 & @xmath31 + b1 & 0.7 & 0.95 & 60 - 400 & 124.1 + b1.5 & 0.775 & 0.95 & 60 - 400 & 113.2 + b2 & 0.85 & 0.95 & 60 - 400 & 104.7 + b2.5 & 0.925 & 0.95 & 60 - 400 & 101.2 + b3 & 1 & 0.95 & 60 - 400 & 107.7 + b3.5 & 1.075 & 0.95 & 60 - 400&132.8 + c2 & 0.85 & 1 & 60 - 400 & 127.8 + c3 & 1 & 1 & 60 - 400 & 121.2 + b2@xmath32 & 0.85 & 0.95 & 30 - 200 & + b2@xmath33 & 0.85 & 0.95 & 30 - 200 & + b2@xmath34 & 0.85 & 0.95 & 30 - 200 & + b2@xmath35 & 0.85 & 0.95 & 30 - 200 & + b2@xmath36 & 0.85 & 0.95 & 30 - 200 & + b2@xmath37 & 0.85 & 0.95 & 30 - 200 & + b2@xmath38 & 0.85 & 0.95 & 30 - 200 & + b2@xmath39 & 0.85 & 0.95 & 30 - 200 & + b2@xmath40 & 0.85 & 0.95 & 30 - 400 & + b2@xmath41 & 0.85 & 0.95 & 120 - 400 & + @xmath42 is for 88 d.o.f . ( 96 data points , eight free parameters for the effective optical depth at the eight different redshifts ) errors for the noise and resolution correction are treated as suggested in m04a ; @xmath43 @xmath44 km / s / mpc ; @xmath45 @xmath46 ; @xmath47 simulation with non - equilibrium solver with reionization at @xmath48 ; @xmath49 non eq . colder version with reionization at @xmath50 ; @xmath51 non eq . version with late reionization at @xmath52 ; @xmath53 non eq . version with early reionization at @xmath54 ; @xmath55 same as @xmath56 ( equilibrium with reionization at @xmath57 ) but with colder equation of state to match the evolution of b2@xmath58 at @xmath59 . we have run a suite of full hydrodynamical simulations with ( springel , yoshida & white 2001 ; springel 2005 ) similar to those in vhs . we have varied the cosmological parameters , particle number , resolution , box - size and thermal history of the simulations . in table 1 we list the fluctuation amplitude @xmath60 , the spectral index @xmath3 as well as box size and number of particles of the different simulations ( note that few of these simulations are actually the same as those presented in vhs ) . the box - size and particle number are given in a form such that 60 - 400 corresponds to a box of length @xmath61 mpc with @xmath62 ( gas + dm ) particles . note that this is the box size of the simulations used for our final analysis and is a factor 6 larger than the largest of the hydrodynamical simulations used by m04b . was used in its treepm mode and we have used a simplified star formation criterion which speeds up the calculations considerably . the simulations were performed with periodic boundary conditions with an equal number of dark matter and gas particles and used the conservative ` entropy - formulation ' of sph proposed by springel & hernquist ( 2002 ) . the cosmological parameters are close to the values obtained by the wmap team in their combined analysis of cmb and other data ( spergel et al . all but two of the simulations in table 1 have the following parameters : @xmath63 , @xmath64 , @xmath65 and @xmath66 . the cdm transfer functions of all models have been taken from eisenstein & hu ( 1999 ) . for the remaining two simulations the hubble constant and matter density were varied to @xmath67 and @xmath68 , respectively . to facilitate a comparison with m04b we also give the amplitude of the linear power spectrum at the pivot wavenumber @xmath69 used by m04b , @xmath70 and the effective spectral index @xmath71 for the b2 simulation . most of the simulations were run with the _ equilibrium _ solver for the evolution of the ionization balance and temperature implemented in the public version of ( which assumes the gas to be in photo - ionization equilibrium ) with a uv background produced by quasars as given by haardt & madau ( 1996 ) , which leads to reionisation of the universe at @xmath72 . the assumption of photoionization equilibrium is valid for most of the evolution of the igm responsible for the lyman-@xmath0 forest . it is , however , a bad approximation during reionization , where it leads to an underestimate of the photo - heating rate and as a result to too low temperatures and generally too steep a temperature density relation ( theuns et al . 1998 ) . for the models run with the equilibrium solver we have thus as before increased the heating rates by a factor of 3.3 at @xmath73 in order to take into account the underestimate of the photo - heating due to the equilibrium solver and optical depth effects for the photo - heating of helium before helium is fully reionized at @xmath74 ( abel & haehnelt 1999 ) . this is necessary to obtain temperatures close to observed temperatures ( schaye et al . 2000 , ricotti et al . 2000 ) . some simulations were run with a _ non - equilibrium _ solver which has been implemented by james bolton into . for the simulations run with the non - equilibrium solver only the optical depth effect for the photoheating of helium have to be taken into account . the increase of the photo - heating rates at @xmath75 necessary to match observed temperatures was thus only a factor 1.8 . we have varied the thermal history of the simulations by re - mapping the redshift evolution of the uv background . for more details on the simulations and the non - equilibrium solver we refer to vhs , bolton et al . ( 2005a , 2005b ) and section [ thermal ] . hydro simulations are rather expensive in terms of cpu time . a typical simulation used in the analysis of vhs took 2 weeks of wall - clock time to reach @xmath76 on 32 processors on cosmos , an sgi altix 3700 . this made it impossible to fully sample a large multi - dimensional parameter space with full hydro - simulations . with the rather modest numerical resources available to us we have thus decided to take the following approach . we have first run a range of simulations with cosmological parameters close to those inferred by vhs and m04b which allowed us to explore a wider range of thermal histories of the igm to better understand the resulting uncertainties . we have then fitted the simulated flux power spectra to the sdss flux power spectrum as described in section [ chifitting ] . based on this and our previous studies in vhs we have chosen a `` best guess '' model which fits the data well . to improve our parameter choices further and to get at least approximate error estimates we have then explored a large multi - dimensional parameter space around our best - guess model . this also enabled us to explore the degeneracies between different parameters . in order to keep the required cpu time at a manageable level we have approximated the flux power spectra by a taylor expansion to first order around our best guess model . if @xmath77 is an arbitrary parameter vector close to the best guess model described by @xmath78 , we have assumed that : @xmath79 where @xmath80 are the n components of the vector @xmath77 . this requires the determination of only n derivatives which can be estimated by running one simulation each close to the best guess model . obviously this linear approximation will only hold as long as the changes in the flux power spectrum are small . within 1 to 2 @xmath81 of the best guess model this should , however , be the case . we have explicitly checked how these derivatives change around another fiducial model for some of the parameters and we will quantify this in section [ derivatives ] . once the derivatives are obtained we can then calculate an arbitrarily fine grid of flux power spectra around the best guess model with equation [ taylor ] . as discussed in detail by vhs and m04b there is a wide range of systematic uncertainties in the analysis of the flux power spectrum . the origins of these uncertainties fall broadly into five categories : * deficiencies of the data which have to be corrected ; * uncertainty of the effective optical depth ; * uncertainty of the thermal state of the igm ; * lack of ability to make accurate predictions of the flux power spectrum for a large parameter space ; * lack of ability to model other physical processes which potentially affect the flux power spectrum . we will discuss the resulting systematic uncertainties in turn . m04b have applied a number of corrections for noise , resolution and the correlated siiii feature identified by m04a . we have adopted the same corrections which are summarized below . for more details we refer to m04a and m04b . we allow for an error in the @xmath82dependent noise correction at each redshift by subtracting @xmath83 from @xmath21 and treat the @xmath84 as free parameter in the fit . m04b recommend to assume that the distribution of @xmath84 is gaussian with mean zero and width 0.05 which corresponds to a typical error in the noise correction of @xmath85 . m04a further recommend to allow for an overall error on the resolution correction , by multiplying @xmath21 with @xmath86 , where @xmath0 is treated as a free parameter with gaussian distribution with mean zero and width @xmath87 . to account for contamination by the correlated siiii feature we modify our model flux power spectra as suggested as follows , @xmath88 with @xmath89 $ ] , @xmath90 and @xmath91 km / s . the noise and resolution corrections and their errors depend strongly on the wavenumber @xmath92 as shown in figure [ f1 ] . note that m04a also subtracted a contribution of uncorrelated associated metal absorption which they estimated redward of the lyman-@xmath0 emission line . the effect of continuum fitting errors on the flux power spectrum is a further uncertainty which is discussed in k04 , m04a and tytler et al . continuum fitting errors are difficult to quantify and we again follow m04b in not attempting to model them . as discussed in vhs and seljak et al . ( 2003 ) the poorly known effective optical depth results in major uncertainties in any analysis of the lyman-@xmath0 forest flux power spectrum . the effective optical depth is to a large extent degenerate with the amplitude of the matter power spectrum . as shown in m04b the @xmath82dependence of changes in the flux power spectrum due to varying effective optical depth and amplitude of the matter power spectrum are , however , somewhat different . this allows at least in principle to break this degeneracy . we will come back to this point later . when modelling the effective optical depth we will investigate two cases . we will either let the effective optical depth vary independently in the different redshift bins or we will parameterize the evolution of the effective optical depth as a power law in redshift . in order to explore in more detail the uncertainties due to thermal effects on the flux power spectrum we have run simulations with a wide range of thermal histories ( table 1 ) . note that this is different from vhs where we have only performed a - posteriori rescalings of the temperature - density relation to investigate thermal effects . we have calculated the value @xmath93 as the median temperature of gas with logarithmic overdensity values between -0.1 and 0.1 . the value of @xmath94 we determined from the slope of the temperature - density relation for gas at mean density and 1.1 times the mean density . in the left panel of figure [ f3 ] we plot the different thermal histories ( temperature at mean density in the bottom panel and @xmath94 in the top panel ) . the linestyles denote the different models as follows . the continuous curve is our standard b2 model with the equilibrium solver . the dashed curves shows the same model with the non - equilibrium solver ( @xmath95 ) which results in a significantly smaller @xmath94 . the dotted and dot - dashed and curve shows the same model with the non - equilibrium solver for late reionization at @xmath96 ( @xmath97 ) and early reionization at @xmath54 ( @xmath98 ) . finally , the dashed - triple - dot curves shows a model with reionization at @xmath99 but smaller heating rate and thus a lower temperature , similar to that of @xmath100 . in the right panel we show the differences of the flux power spectrum relative to our fiducial model @xmath101 at three different redshifts . the differences are typically @xmath102 % . comparing @xmath101 and @xmath98 , which have a similar redshift evolution we find for example that a decrease of @xmath93 by 50% ( @xmath103 k ) results in a decrease in the flux power larger than @xmath104 . the effect of changing @xmath94 can be judged by comparing @xmath95 and @xmath97 at @xmath105 which have approximately the same temperature . decreasing @xmath94 by @xmath106 produces an increase in the flux power by 7.5 % ( 2.5 % ) , but we note that for @xmath107 s / km the differences increase strongly . we will later model the evolution of the thermal state of the igm with redshift as broken power laws for temperature and slope of the temperature - density relation . modelling the flux power spectrum accurately is numerically challenging and checks for convergence are important we have chosen to use here full hydro - simulations performed with described in section [ hydro ] . in vhs we found that simulations with a box size 60 @xmath108 mpc and with @xmath109 particles ( 60,400 ) are the best compromise for the analysis of the luqas sample . the lower resolution sdss power spectrum is not affected on large scale in the same way by continuum fitting errors as the high - resolution spectra which we used in vhs and for which continuum fitting errors appear to increase the flux power significantly at and above the scale corresponding to one echelle order ( kim et al . the usable range of wavenumbers is thus shifted to larger scales compared to flux power spectra obtained from high - resolution echelle data . ideally one would thus like a somewhat larger box size than that used in vhs . unfortunately this is currently not feasible with simulations that at least marginally resolve the jeans length . we have thus decided to use simulations with the same box size and resolution as in vhs . _ box size _ : for the final analysis we use simulations of 60@xmath110 mpc and with @xmath109 particles ( 60,400 ) . since the error bars for the observed 1d flux power spectrum are significantly smaller than those of the 3d flux power used by vhs we checked whether or not the flux power spectrum is affected by box size and/or limited resolution at the percent level . we have first checked how well the largest scales probed by sdss are sampled by our fiducial ( 60,400 ) simulations . in vhs this was not an issue since the first data point used in the analysis was at @xmath111 s / km , which was well sampled by the ( 60,400 ) simulations . our analysis here extends however to @xmath112 s / km and could be affected by cosmic variance . to test this , we have compared the flux power spectrum of our fiducial ( 60,400 ) simulation b2 with that of a ( 120,400 ) simulation of the same model ( b2@xmath41 ) . the power of the latter at @xmath113 s / km is typically larger by @xmath114% at @xmath115 , respectively . we correct for this by multiplying the flux power spectrum of the ( 60,400 ) simulation by corresponding factors . we will later use simulations with a factor two smaller box size but the same resolution as our fiducial simulation to explore the effects of the thermal history . we have thus also compared the flux power spectrum of our fiducial ( 60 - 400 ) simulation b2 with a ( 30 - 200 ) simulation of the same model ( b2@xmath32 ) and find that the latter underestimates the power by @xmath116 in the range @xmath117 ( s / km)@xmath118 at @xmath119 . our findings regarding the dependence of the flux power spectrum on the thermal history of the igm should thus be little affected by the fact that they were obtained from simulations of smaller box size . _ resolution _ : to quantify the effect of resolution , we compare the simulation b2@xmath32 with b2@xmath40 a simulation which has the same box sixe but eight times better mass resolution . the differences between b2@xmath120 and b2@xmath32 are less than 5 % for @xmath121 ( s / km ) , at any redshift , ( 3% in the range @xmath122(s / km ) @xmath123 ) . at @xmath124 s / km , the flux power spectrum of b2@xmath125 is larger by @xmath126 , at @xmath105 and @xmath127 , respectively . this is in agreement with the findings of m04b . we will correct for the limited resolution by multiplying the flux power spectrum of our fiducial ( 60,400 ) simulation by corresponding factors . both the resolution and box size errors are comparable or smaller than the statistical errors of the flux power spectrum . we explicitly checked that these corrections do not significantly affect the parameters of our best - fitting model . they do , however , affect the @xmath42 value of the best - fitting model . note that both the resolution and box size corrections could be model dependent . _ high column density systems / damped lyman-@xmath0 systems : _ we have recently pointed out that the absorption profiles of high column density systems / damped lyman-@xmath0 systems have a significant effect on the flux power spectrum over a wide range of scales ( viel et al . these absorption systems are caused by high - redshift galaxies and the gas in their immediate vicinity . numerical simulations still struggle to reproduce these systems correctly and they are thus a major factor of uncertainty in modelling the flux power spectrum . m04b have modelled the expected effect on the sdss flux power spectrum and we have included this correction using the k - dependence of fig . 11 in m04b . m04b recommend to assume that the distribution of the correction made @xmath128 is gaussian with mean 1 and width 0.3 . the correction is shown in figure [ f1 ] as the dotted curve . it is of order of @xmath129% at the largest scales and drops to 3% at the smallest scales . note that unlike m04b we have assumed that this correction does not vary with redshift . there are a range of physical processes some of which are difficult to model , that are described below . note that we have chosen not to try to include these effects in our analysis . _ spatial fluctuations of the ionization rate _ : the flux power spectrum is obviously sensitive to the neutral fraction of hydrogen which depends on the ionization rate . at low redshift @xmath130 the mean free path of ionizing photons is sufficiently large that spatial fluctuations of the rate should be too small to affect the flux power spectrum relevant for our investigation here ( meiksin & white 2004 ; croft 2004 ; mcdonald et al . 2005 ) . at higher redshift this is less obvious . quantitative modelling of the fluctuations requires however a detailed knowledge of the source distribution of ionizing photons which is currently not available . as we will discuss below there may be some tentative evidence that uv fluctuations do become important at the highest redshifts of the sdss flux power spectrum sample . _ galactic winds _ : there is undeniable observational evidence that the lyman-@xmath0 forest has been affected by galactic winds . associated metal absorption is found to rather low optical depth ( cowie et al . 1995 , schaye et al . searches for the effect of galactic winds on qso absorption spectra from lyman break galaxies close to the line - of - sight have also been successful . the volume filling factor of galactic winds and the material enriched with metals is , however , very uncertain ( e.g. pieri & haehnelt 2004 ; adelberger et al . 2005 ; rauch et al . 2005 ) . numerical simulations have generally shown the effect of galactic winds to be small ( theuns et al . 2002a ; kollmeier et al . 2003 ; desjacques et al . 2004 ; mcdonald et al . 2005 ; kollmeier et al . 2005 ) . _ temperature fluctuations _ : it is rather difficult to measure mean values of the temperature and little is known observationally about spatial fluctuations of the temperature ( theuns et al . 2002b ) . in the redshift range of interest the heating rate of the igm should be dominated by photo - heating of helium before helium is fully reionized . helium reionization should thus lead to spatial temperature fluctuations ( miralda - escud et al . 2000 ) , which will affect the neutral fraction of hydrogen through the temperature dependence of the recombination coefficient . temperature fluctuation may thus affect the flux power spectrum . quantitative modelling of their effect on the flux power spectrum requires numerical simulation of helium reionization including full radiative transfer and good knowledge of the spatial distribution of the source of ionizing photons . such modelling will also be uncertain . _ reionization history _ : the flux power spectrum is not only sensitive to the current thermal state of the igm but also to its past thermal history . this is because the spatial distribution of the gas is affected by pressure effects ( hui & gnedin 1997 ; theuns , schaye & haehnelt 2000 ; zaldarriaga , scoccimarro , hui 2001 ) . quantitative modelling of this effect is , however , again very uncertain as the heating rate at high redshift is expected to be dominated by photo - heating of helium . in order to get a feeling for the effect of a simple change of the redshift when reionization occurs we can compare @xmath131 and @xmath100 at @xmath105 . both simulations have approximately the same values for @xmath93 and @xmath94 at @xmath59 while reionization occurs at @xmath99 and @xmath132 , respectively . the differences are of the order of 3% , in agreement with the findings of m04a . in order to settle on a best guess model we have started with the coarse grid of hydro - simulations presented in vhs . we have complemented this with simulations of a wider range of @xmath60 and a wider range of thermal histories . we have fitted the flux power spectrum of all ( 60,400 ) simulations listed in table 1 to the sdss flux power spectrum allowing for errors in the correction to the data as described in section [ datacorr ] . we thereby leave the effective optical depth as a free parameter at all redshifts . the last column in table 1 gives the resulting @xmath133 values obtained using the full covariance matrix as given by m04a pmcdonal / lyaf / sdss.html ] . when performing the fits we noticed that the highest redshift bin at @xmath26 is generally fitted very poorly . omitting these 12 data points usually reduced the @xmath134 by about 26 for 11 degrees of freedom ( d.o.f . ) . this may either indicate some problem with the data or some insufficiency of the model . the latter would require , however , a very rapid change of the model with redshift as dropping any of the other redshift bins reduced the @xmath134 by the expected amount . uv fluctuations are probably the most plausible candidate for a rapid change toward the highest redshift bin . we have decided to drop the highest redshift bin for our analysis . this leaves 96 data points and 8 free ( unconstrained ) parameters corresponding to 88 degrees of freedom . the best fitting simulation appears to be b2.5 which has @xmath135 , when minimizing over the noise array @xmath84 , the resolution @xmath0 and the effective optical depth . the simulation b2 has a @xmath136 close to that of b2.5 , while the other models have typically @xmath137 . we regard these two models , with their thermal history , as reasonable good fits to the data set ( @xmath134 values higher than the quoted values have a probability of 16% and 11% to occur ) . we decided to choose b2 as our best guess model despite the fact b2 has a somewhat larger @xmath134 value than b2.5 for two reasons . first , the model is surprisingly close to the best fit model quoted by m04a ( which has @xmath138 , @xmath139 , @xmath140 and @xmath141 ) . secondly , we had already run more simulations exploring different thermal histories for b2 than for the other models . we have , however , checked as discussed in more detail in section [ chifitting ] that an expansion around the b2.5 model gives essentially the same results in terms of the inferred astrophysical and cosmological parameters . the key ingredient for our estimates of a fine grid of flux power spectra are the derivatives of the flux power spectrum ( see equation [ taylor ] ) close to the best guess model b2 . figure [ f2 ] shows the @xmath82dependence of the most relevant of these derivatives at three redshifts ( @xmath115 from left to right ) in the form of the change of the flux power spectrum for a finite change of the parameters . we compute these derivatives at the wavenumbers of the sdss flux power spectrum , although we can not directly cover the two smallest wavenumbers because we are using ( 30,200 ) simulations to estimate these derivatives . for these two points we have used an extrapolation with a 2nd order polynomial . this should be a reasonable approximation given the rather weak and smooth @xmath82dependence at these scales . in the top panel we show how the flux power spectrum depends on @xmath1 , @xmath3 , @xmath4 . a @xmath85 increase in the effective optical depth , translates into a 7% increase in the flux power spectrum with no dependence on redshift and with very little ( @xmath142 ) dependence on wavenumber . a change of the spectral index by @xmath143 leads to an increase of the flux power which rises steeply with wavenumber from @xmath144 to @xmath145 , while increasing @xmath1 by 9% increases the flux power by @xmath146 for @xmath147 ( s / km ) dropping to @xmath148 at the smallest scales considered here . the changes introduced by different values of @xmath2 and @xmath149 are shown in the middle panel . an increase of the hubble parameter by 8 km / s / mpc leads to an increase of order 3% , while the change due to an increase of @xmath2 shows a peaks at @xmath150 s / km of 3% at @xmath26 and depends less strongly on wavenumber at lower redshifts . the derivatives of the flux power spectrum with respect to changes in temperature and power - law index of the temperature density relations are shown in the bottom panel . the trends are the same as those shown in figure [ f3 ] . we note that the overall effect depends only weakly on wavenumber for @xmath121 s / km and steepens at smaller scales . note that the dependence of the derivatives on @xmath92 and @xmath151 is distinctively different for the different parameters . this is the reason why we will be able to obtain reasonably tight constraints for most of them . our results agree quite well with the results of vhs and m04b . we note , however , that a direct quantitative comparison with m04b is not obvious . our best guess model has still somewhat different parameters from their fiducial model and m04b have used hpm simulations with a range of corrections while we have used full hydro - simulations . one may ask how much these derivatives depend on our choice of the best - guess model . we have checked this for a few cases . the derivatives obtained by rescaling the effective optical depth around other models ( b1,b3,c3 ) are different by less then 0.5 % compared to those shown in figure [ f2 ] . the derivatives with respect to @xmath1 and @xmath3 differ by less than 2.5% and 1.5 % , respectively , compared to expanding around b2 . we have also directly compared the approximate flux power obtained with the taylor expansion with that extracted from hydrodynamical simulations for a few simulation with parameters that are about@xmath152 away from our best - guess model . for the @xmath153 and @xmath154 ( 60,400 ) the error of the aproximation is less than 3.5% which should be compared to the difference to the best - guess model which is @xmath14% . we have also run a few further ( 30,200 ) simulations with @xmath155 , @xmath156 and a hotter simulations with @xmath93=41000 k and @xmath157 . the approximations for the models with different @xmath2 and the spectral index @xmath3 are accurate to 1% . the error for the model with the different @xmath93 parameter is @xmath158% . by comparing with figure 3 we find that for @xmath152 displacements in the parameters @xmath3,@xmath1,@xmath159 and @xmath93 the error of the taylor series approximation is generally less than 30% of the difference between the models . this is not perfect but should be acceptable considering the expense of full - hydrodynamical simulations and the size of the parameter space . we have modified the code cosmomc ( lewis & bridle 2002 ) to run monte carlo markov chains for our set of parameters . after settling on a best guess model we have explored the parameter space close to the best guess model . for most of the analysis we have used 22 parameters for the @xmath134 minimization , some of them free some of them independently constrained . we briefly summarize the parameters here . we assume a flat cosmological model with a cosmological constant and neglect the possibility of a non - zero neutrino mass . there are thus four cosmological parameters that describe the matter distribution : the spectral index @xmath3 , the dark matter power spectrum amplitude @xmath1 , @xmath2 and @xmath149 . we use nine `` astrophysical '' parameters . two describe the effective evolution of the optical depth evolution , six describe the evolution of @xmath94 and @xmath93 and one describes the contribution of strong absorption systems . for the evolution of the optical depth we assume a power - law @xmath160^{\tau_{\rm eff}^s}$ ] . note that this is different from what we did when we fitted the suite of hydro - simulations where we let the effective optical depth vary independently at all redshift bins . we will come back to this in section [ optical ] . for the evolution of @xmath94 and @xmath93 we assume broken power - laws with a break at @xmath105 . the three parameters for the temperature are the amplitude at @xmath105 , @xmath161 and the two slopes @xmath162 and @xmath163 . @xmath94 is described in the same way by @xmath164 , @xmath165 and @xmath166 . the correction for the damped / high column density systems is modelled with the @xmath82dependence of figure 11 of m04b and an overall amplitude @xmath128 and no redshift evolution , as described in section [ moderrors ] . finally , we have a total of 9 parameters which model uncertainties in the correction to the data . eight parameters for the noise correction @xmath167 as described in section [ datacorr ] and one parameter @xmath0 describing the error of the resolution correction . note that all the parameters describing the errors of the corrections to the data and the correction for damped / high column density systems are constrained as suggested in m04b and described in section [ system ] . we will present here two sets of results . at first we will discuss the 1d and 2d likelihoods for the most significant parameters by assuming no priors on the cosmological and astrophysical parameters ( with the exception of the correction for damped systems ) . then we will show some results assuming priors on the hubble parameter and on the thermal history . the dashed curves in fig . [ f4 ] show the 1d ( marginalized ) likelihoods for our full set of parameters without imposing any priors on the cosmological and astrophysical parameters . the best fit model has now @xmath168 for 84 d.o.f .. a @xmath134 equal or larger than this should have a probability of 64% . our best - fitting model is thus an excellent fit , maybe with a hint that we are already marginally over - fitting the data . most parameters including @xmath1 , @xmath3 and @xmath2 are tightly constrained . the constraints on the thermal history are considerable weaker . this is not surprising as the rather low resolution of the sdss spectra means that the thermal cut - off at small scales is not actually resolved . the best fit value for @xmath169 is rather high , @xmath170 k , and may be in disagreement with the best fit values of m04b who find @xmath171 k without giving any error estimate . the dashed curves in figure [ f5 ] show 2d likelihood contours for various parameter combinations . the contours have been obtained by marginalizing over the rest of the parameters . we note that the best fit models prefer somewhat larger values of the spectral index and higher temperatures than those of our best guess model . the 2d contour plot of @xmath172 vs @xmath169 , suggests that in order to reconcile this measurement with a colder igm one would need an unplausibly steep temperature density relation with @xmath173 . the preferred values for the thermal state of the igm are actually outside the range suggested by a detailed analysis of the line width of the absorption feature using high - resolution spectra . this is shown in figure [ f6 ] where we compare the best - fitting evolution of the temperature @xmath93 ( hashed region ) to the measurements of the temperature by schaye et al . note again that the lower range of values corresponds to @xmath174 outside of the range of plausible values for a photoionized igm . the hubble constant is also only poorly constrained . this is again not surprising as the spectra are sensitive to the matter power spectrum in velocity space . the only dependence comes thus through the shape parameter @xmath175 , which is weak ( [ f2 ] ) . the best fitting values for our modelling without priors with errors obtained by marginalizing over all other parameters are listed in table 2 . as a further consistency check and to test how well the observed redshift evolution of the flux power spectrum probes the expected gravitational growth of structure we have also obtained constraints on @xmath1 and @xmath3 independently for each redshift bin . we get the following constraints for @xmath1 : @xmath176 , @xmath177 , @xmath178 , @xmath179 , @xmath180 , @xmath181 , @xmath182 , @xmath183 , @xmath184 , for @xmath185 , respectively . for @xmath3 we get : @xmath181 , @xmath186 , @xmath187 , @xmath188 , @xmath189 , @xmath190 , @xmath191 , @xmath192 , @xmath193 for @xmath185 , respectively . the total @xmath42 is @xmath194 . for 74 d.o.f . in this case . the constraints are weaker but perfectly consistent with our estimates from all redshift bins combined . the errors are significantly smaller than the expected growth of the amplitude between the lower and upper redshift end of the sample demonstrating that the flux power spectrum evolves as expected for gravitational growth . lcc parameter & without priors & with priors on thermal + & & history and @xmath149 + n & @xmath195 & @xmath196 + @xmath1 & @xmath197 & @xmath198 + @xmath2 & @xmath199 & @xmath200 + @xmath149 ( km / s / mpc ) & @xmath201 & @xmath202 + @xmath203 & @xmath204 & @xmath205 + @xmath206 & @xmath207 & @xmath208 + @xmath209 & @xmath210 & @xmath211 + @xmath212 & @xmath213 & @xmath214 + @xmath215 & @xmath216 & @xmath217 + @xmath218 k & @xmath219 & @xmath220 + @xmath221 & @xmath222 & @xmath223 + @xmath224 & @xmath225 & @xmath226 + @xmath128 & @xmath227 & @xmath228 + as discussed in the last section the sdss flux power spectrum of sdss published by m04a prefers unplausibly large temperatures for reasonable values of @xmath94 . we have thus repeated the analysis with ( gaussian ) priors on the thermal state of the igm , as follows , @xmath229 , @xmath230 and @xmath231 ( schaye et al . we have also added a prior on the hubble constant , @xmath232 km / s / mpc ( freedman et al . 2001 ) . the solid curves in figure [ f4 ] and figure [ f5 ] show the marginalized likelihoods for our analysis with priors . the filled ( coloured ) contours in figure [ f5 ] show the mean 2d likelihood . as expected the thermal parameters and the hubble constant are now more tightly constrained . the temperature evolution of the best - fitting model with prior is shown as the shaded region in figure [ f6 ] . it lies about 1@xmath81 above the imposed constraint suggesting that the observed flux power spectrum definitely prefers models with higher temperatures than those observed . rather than being an actual indication of higher temperature this is more likely indicative of a not understood systematic uncertainty in our modelling which makes the flux power spectrum mimic high temperatures . as discussed in section [ othereffects ] there are plenty of candidates for this . the best fit model with priors has @xmath233 for 88 degrees of freedom . a @xmath134 equal or larger than this should have a probability of 60% very similar to the case without priors . the best fitting values for our modelling without priors are also listed in table 2 ( right column ) . the most significant change caused by the introduction of the priors is a decrease of the spectral index by @xmath234 ( consistent with the @xmath8 errors ) . in figure [ f5 ] the strongest correlations are those in the @xmath235 , @xmath236 and @xmath237 planes . as expected a higher value of the @xmath238 requires a smaller @xmath1 quantitatively comparable to previous findings ( viel , weller & haehnelt 2004 ; vhs ; seljak et al . 2003 ) . the best - fitting temperature and slope of temperature density relation are again anti - correlated . a lower temperature corresponds to a steeper temperature density relation . we further note that @xmath1 and @xmath3 appear not to be correlated , in contrast to the corresponding parameters @xmath239 in m04b . if we take out the box size and resolution corrections we do not get significant changes in the final parameters but the @xmath42 increases by 1.3 , showing that the data prefer these corrections . the best fitting values for our modelling with priors with errors obtained by marginalizing over all other parameters are also listed in table 2 . as discussed before the effective optical depth is the largest uncertainty and it is very degenerate with the fluctuation amplitude . larger effective optical depths correspond to smaller @xmath60 values ( figure [ f5 ] ) . the shape of the derivative of the flux power spectrum with respect to @xmath4 and @xmath60 ( figure [ f2 ] ) is , however , sufficiently different to get still interesting constraints on @xmath4 . in the left panel of figure [ f7 ] we compare the best fitting @xmath240 for our set of ( 60,400 ) hydrodynamical simulations to measurements made from high and low resolution spectra . all values are shown relative to the fit to the observed evolution of the optical depth corrected for the contribution by associated metal absorption based on a set of high - resolution spectra by schaye et al . 2003 ( their figure 1 ) . the fit is given by @xmath241^{3.57}$ ] . the evolution obtained by kim et al . ( 2002 ) based on a different set of high - resolution spectra is shown as the dashed curve . the measurements at @xmath17 and @xmath242 made by vhs from the luqas sample ( kim et al . 2004 ) are shown as filled triangles . note that there is some overlap between the schaye et al . ( 2003 ) data set and the luqas sample of k04 . the dotted line shows the result of press , rybicki & schneider ( prs , 1993 ) from low - resolution spectra of rather poor quality . the shaded area shows the range adopted by vhs and viel , weller & haehnelt ( 2004 ) for their analysis . the different symbols show the results from our fits of the flux power spectrum of the hydro - simulations to the sdss flux power spectrum as described in section [ bestguess ] . the degeneracy between @xmath1 and @xmath4 is clearly visible and can be roughly described by @xmath243 similar to that found in vhs . in the right panel , the solid curve shows the evolution of @xmath4 for our best fitting model compared to the same observational estimates . as discussed we have modelled the evolution of @xmath240 as a power - law . we find @xmath244^{3.22\pm0.24}$ ] in agreement with schaye et al . ( 2003 ) and the values used by vhs . it also agrees well with the values inferred by lidz et al . ( 2005 ) from the flux probability distribution . the shaded area in the right panel shows the @xmath8 uncertainty of the best fitting model . we thus confirm the findings of m04b that the differences in the @xmath82dependence of the respective changes can break the degeneracy between @xmath1 and @xmath4 . the dot - dashed line and the triangles show the evolution of the optical depth if we let @xmath4 vary separately at all nine redshifts ( this increases the number of free parameters from 22 to 28 ) . letting the effective optical depth vary freely at high redshift increases the average inferred effective optical depth by about 5% and reduces the inferred errors . lcccc & @xmath1 & @xmath3 + spergel et al . ( 2003)@xmath245 & @xmath246 & @xmath247 + spergel et al . ( 2003)@xmath43 & & @xmath248 + viel , haehnelt & springel ( 2004)@xmath45 & @xmath249 & @xmath250 + viel , weller & haehnelt ( 2004)@xmath47 & @xmath180 & @xmath251 + desjacques & nusser ( 2004 ) & @xmath192 & + tytler et al . ( 2004 ) & 0.90 & + mcdonald et al . ( 2004b)@xmath49 & @xmath252 & @xmath253 + seljak et al . ( 2004)@xmath51 & @xmath254 & @xmath251 + this work ( priors case ) & @xmath198 & @xmath255 + luqas+croft et al . ( 2002 ) only ; @xmath47 luqas+croft et al . ( 2002 ) + wmap 1st year ; @xmath49 error bars extrapolated by the authors from the abstract of m04b ; @xmath51 sdss lyman-@xmath0 flux power spectrum + wmap . in table [ tab3 ] we list results for the most important parameters describing the matter distribution ( @xmath1 and @xmath3 ) obtained by recent studies using the lyman-@xmath0 forest by a variety of authors . the most direct comparison is again with m04b . at @xmath105 our results correspond to an fluctuation amplitude and effective slope at the pivot wavenumber @xmath256 s / km as defined by m04b of @xmath257 and @xmath258 . the values are consistent within @xmath259 with those of m04b . the agreement with otherauthors including our own work is at the same level . for comparison we also show the value for wmap alone . a consistent picture emerges for the lyman-@xmath0 forest data with a rather high fluctuation amplitude and no evidence for a significant deviation from @xmath9 or a running of the spectral index . this results holds for the lyman-@xmath0 forest data alone but is considerably strengthened if the lyman-@xmath0 forest data is combined with the cmb and other data . as discussed in section [ system ] there are many systematic uncertainties that affect the measurement of cosmological parameters with the lyman-@xmath0 flux power spectrum . in order to test the stability of the results we have removed the correction for high column density systems and the errors for the noise and resolution corrections in the case with priors . if we take out the correction due to strong absorption systems @xmath3 , decreases to @xmath260 , and the best fit value for @xmath261 becomes 0.381 , all the other parameters remain practically unchanged . the measured values do therefore practically not change but the value of @xmath42 increases by 0.5 . if we do not allow for the error to the noise and resolution corrections the inferred values do again not change but the value of @xmath42 increases by 1.8 . note that m04a have also subtracted an estimate of the contribution of associated metal absorption from the flux power spectrum and have applied a suite of other ( smaller ) corrections to the data ( see m04a for more details ) . many of these corrections would either not be necessary or substantially easier for spectra with somewhat better resolution ( @xmath262 ) and higher s / n . obtaining large samples of such spectra is observationally feasible and there is certainly room for improvement . note further that the sdss team will soon release a further measurement of the flux power spectrum performed by an independent group within the collaboration . we also note again that calculating the flux power spectrum by using a taylor expansion to first order around the best guess model is just an approximation . this was an important part of our analysis as it reduced the amount of cpu time necessary dramatically , but could be further improved by getting more accurate fits around the `` best guess '' model . we have also not attempted to correct for the possible effects of spatial fluctuations of the ionization rate , galactic winds , temperature fluctuation or the reionization history . this is because we are not convinced that it is already possible to model these effects with sufficient accuracy . this should hopefully change in future with improved observational constraints and numerical capabilities . one should further keep in mind that so far very little cross checking of hydrodynamical simulation run with different codes has been performed . in our study the rather poorly known thermal state of the igm is one of the major remaining uncertainties . further high - resolution spectroscopy and improved modelling will hopefully soon improve this . accurate modelling of the effect of high column density systems/ damped lyman-@xmath0 systems and improvement in the determination of the column density distribution of observed absorption systems in the poorly determined range around @xmath263 should also be a priority in further studies ( see the discussion in mcdonald et al . 2005 ) . it is certainly encouraging that the differences between our analysis and that of m04b are moderate despite considerable differences between their approximate hpm simulations calibrated with hydrodynamical simulations of rather small box size and our analysis with full hydrodynamical simulations of much larger box size here and in vhs . for the time being we would advice the conservative reader to double the formal errors quoted here . this will bring the error estimates to about the same size as the also conservative estimates for the errors of the fluctuation amplitude from the lyman-@xmath0 forest data alone in the analysis of vhs . the actual errors lie probably somewhere in between . we have compared the flux power spectra calculated from a suite of full large box - size high resolution ( @xmath264 mpc , @xmath109 particles ) hydrodynamical simulations with the sdss flux power spectrum as published by mcdonald et al . ( 2005 ) . we have identified a best - guess model which provides a good fit to the data . we have used a taylor expansion to first order to calculate flux power spectra in a multi - dimensional space of parameters describing the matter power spectrum and the thermal history of the igm with values close to those of our best - guess model . we have investigated the combined effect of cosmological and astrophysical parameters on the flux power spectrum with an adapted version of the markov - chain code cosmomc . our main results can be summarized as follows . * the flux power spectrum calculated directly from the simulation of a @xmath265cdm model ( @xmath63 , @xmath266 , @xmath65 and @xmath66,@xmath267 ) with a temperature density relation described by @xmath268 and @xmath269 gives an acceptable fit to the sdss flux power spectrum in the redshift range @xmath270 ( @xmath271 for 88 degrees of freedom ) . the fit can easily be further improved by small changes in the cosmological and astrophysical parameters . * at higher redshift the deviations from the observed flux power spectrum become significantly larger ( @xmath272 for 12 additional data points at @xmath26 ) suggesting either some problem with the data or a physical effect that changes rapidly with redshift . * we confirm the claim by mcdonald et al . ( 2004 ) that the degeneracy of the dependence of the flux power spectrum on the amplitude of the matter power spectrum and the effective optical depth can be broken for the published sdss flux power spectrum . it will be interesting to see if the same is true for the independent analysis of the sdss data to be released soon . * the sdss power spectrum alone can constrain the amplitude of the matter power spectrum , the matter density and the power - law index of primordial density fluctuation to within @xmath273 . the thermal state of the igm is however , poorly constrained and the sdss power spectrum formally prefers models with unplausible values of the parameters describing the thermal state . the dependence of the flux power spectrum on the assumed hubble constant is also very weak and the hubble constant is not well constrained . the exact values for the other cosmological parameters depend somewhat on the assumed prior for the thermal state and the details of the correction which have been applied to the data . * with a prior on the thermal history and the hubble constant motivated by the observations by schaye et al . ( 2000 ) and freedman et al . ( 2001 ) we obtain the following best - fitting values for the cosmological parameters @xmath5 , @xmath6 , @xmath7 ( @xmath8 error bars ) , and the effective optical depth is well described by the following power - law relation : @xmath244^{3.22\pm0.24}$ ] . the errors were obtained by marginalizing over a set of 22 parameters describing the matter distribution , thermal history of the universe , the effective optical depth and errors to various corrections to the data . the values for @xmath60 and @xmath3 are consistent with those found by m04a for the same data set with different simulations . they are also consistent with the results of other recent studies of lyman-@xmath0 forest data . the inferred optical depth is in good agreement with that measured directly from continuum - fitted high - quality absorption spectra . the simulations were run on the cosmos ( sgi altix 3700 ) supercomputer at the department of applied mathematics and theoretical physics in cambridge . cosmos is a uk - ccc facility which is supported by hefce and pparc . mv thanks pparc for financial support . we thank james bolton for providing us with the non - equilibrium solver for , antony lewis for useful suggestions and technical help and the anonymous referee for useful suggestions . abel t. , haehnelt m.g . , 1999 , apj , 520 , 13 adelberger et al . , 2005 , astro - ph/0505122 bolton j.s . , haehnelt m.g . , viel m. , springel v. , 2005a , mnras , 357 , 1178 bolton j.s . , haehnelt m.g . , viel m. , carswell r.f . , 2005b , arxiv : astro - ph/0508201 cowie l. , songaila a. , kim t .- s . , hu e.m . , 1995 , aj , 109 , 1522 croft r. a. c. , 2004 , apj , 610 , 642 croft r. a. c. , weinberg d. h. , katz n. , hernquist l. , 1998 , apj , 495 , 44 croft r. a. c. , weinberg d. h. , bolte m. , burles s.,hernquist l. , katz n.,kirkman d. , tytler d. , 2002 , apj , 581 , 20 [ c02 ] desjacques v. , nusser a. , 2005 , arxiv : astro - ph/0410618 desjacques v. , nusser a. , haehnelt m. g. , stoehr f. , 2004 , mnras , 350 , 879 eisenstein d. j. , hu w. , 1999 , apj , 511 , 5 freedman w.l . et al . , 2001 , apj , 553 , 47 gnedin n. y. , hui l. , 1998 , mnras , 296 , 44 gnedin n. y. , hamilton a. j. s. , 2002 , mnras , 334 , 107 haardt f. , madau p. , 1996 , apj , 461 , 20 hui l. , gnedin n. , 1997 , mnras , 292 , 27 hui l. , burles s. , seljak u. , rutledge r. e. , magnier e. , tytler d. , 2001 , apj , 552 , 15 lewis a. , bridle s. , 2002 , phys . d66 , 103511 ; http://www.cosmologist.info lidz a. , heitmann k. , hui l. , habib s. , rauch m. , sargent w.l.w . , 2005 , astro - ph/0505138 kim , t .- s . , carswell , r. f. , cristiani , s. , dodorico , s. , giallongo , e. 2002 , mnras , 335 , 555 kim , t .- s . , viel m. , haehnelt m.g . , carswell r.f . , cristiani s. , 2004 , mnras , 347 , 355 [ k04 ] kollmeier j.a . , weinberg d.h . , dav r. , katz n. , apj , 594 , 75 kollmeier j. , miralda - escude j. , cen r. , ostriker j.p . , 2005 , arxiv : astro - ph/0503674 mcdonald p. , 2003 , apj , 585 , 34 mcdonald p. , seljak u. , cen r. , bode p. , ostriker j.p . , 2005 , mnras , in press mcdonald p. , miralda - escud j. , rauch m. , sargent w.l . , barlow t.a . , cen r. , ostriker j.p . , 2000 , apj , 543 , 1 mcdonald p. et al . , 2004a , astro - ph/0407377 [ m04a ] mcdonald p. et al . , 2004b , astro - ph/0407378 [ m04b ] meiksin a. , white m. , 2004 , mnras , 350 , 1107 pieri m. , haehnelt m.g . , 2004 , 347 , 985 press w. h. , rybicki g. b. , schneider d. p. , 1993 , apj , 414 , 64 rauch m. et al . , 2005 , apj , in press ricotti m. , gnedin n. , shull m. , 2000 , apj , 534 , 41 schaye j. , theuns t. , rauch m. , efstathiou g. , sargent w. l. w. , 2000 , mnras , 318 , 817 schaye j. , aguirre a. , kim t .- s . , theuns t. , rauch m. , sargent w. l. w. , 2003 , apj , 596 , 768 seljak u. , mcdonald p. , makarov a. , 2003 , mnras , 342 , 79l seljak et al . , 2004 , astro - ph/0407372 spergel d. n. et al . 2003 , apjs , 148 , 175 springel v. , 2005 , astro - ph/0505010 springel v. , hernquist l. , 2002 , mnras , 333 , 649 springel v. , yoshida n. , white s.d.m . , 2001 , newa , 6 , 79 theuns t. , schaye j. , haehnelt m.g . , 2000 , 315 , 600 theuns t. , leonard a. , efstathiou g. , pearce f.r . , thomas p.a . , 1998 , mnras , 301 , 478 theuns t. , zaroubi s. , kim t .- s . , tzanavaris p. , carswell r.f . , 2002b , mnras , 332 , 367 theuns t. , viel m. , kay s. , schaye j. , carswell b. , tzanavaris p. , 2002a , apj , 578 , l5 tytler et al . , 2004 , apj , 617 , 1 verde l. , et al . 2003 , apjs , 148 , 195 viel m. , haehnelt m.g . , springel , 2004 , mnras , 354 , 684 [ vhs ] viel m. , weller j. , haehnelt m.g . , 2004 , mnras , 355 , 23l viel m. , haehnelt m.g . , springel , 2005 , astro - ph/0504641 viel m. , haehnelt m. g. , carswell r. f. , kim t .- , 2004b , mnras , 349 , l33 viel m. , matarrese s. , theuns t. , munshi d. , wang y. , 2003 , mnras , 340 , l47 viel m. , lesgourgues j. , haehnelt m.g . , matarrese s. , riotto a. , 2005 , phys.rev.d71 , 063534 zaldarriaga m. , hui l. , tegmark m. , 2001 , apj , 557 , 519 zaldarriaga m. , scoccimarro r. , hui l. , 2003 , apj , 590 , 1

we identify 17 possible 8.0 m - selected counterparts to the submillimeter galaxies in the cudss 14 field , derived from deep imaging carried out with the irac and mips instruments aboard the _ spitzer space telescope_. ten of the 17 counterparts are not the same as those previously identified at shorter wavelengths . we argue that 8.0 m selection offers a better means for identifying counterparts to submillimeter galaxies than near - infrared or optical selection . based on the panchromatic seds , most counterparts appear to be powered by ongoing star formation . power - law fits to the seds suggest that five objects in the 8.0 m - selected counterpart sample harbor dominant agns ; a sixth object is identified as a possible agn . the 3.6 - 8.0 m colors of the infrared - selected counterparts are significantly redder than the general irac galaxy population in the cudss 14 field .

astro-ph0603198

observations at 850 , initially with the submillimeter common user bolometer array ( scuba ) and subsequently with the max planck millimeter bolometer ( mambo ) , have revealed galaxies that are strong sources of dust emission but faint at visible wavelengths ( smail , ivison & blain 1997 , hughes et al . 1998 , barger et al . 1998 , eales etal 1999 , clements et al . 2004 ) . the scuba galaxy populations are important because they seem to represent the most luminous star - forming galaxies during the epoch when star formation rates were highest . the space density of submillimeter galaxies is orders of magnitude greater than that of @xmath2@xmath3 galaxies at low redshift ( blain et al . 2002 ) , implying that strong evolution has greatly altered the submillimeter galaxy populations since @xmath4 . moreover , some or all of the scuba population may be the progenitors of today s massive elliptical galaxies . the stars in nearby elliptical galaxies and in the bulges of disk galaxies are old and tend to be metal rich . at high redshifts , there ought to be a significant population of progenitor galaxies with star formation rates ( sfrs ) high enough to produce today s population of ellipticals and bulges ( lilly et al . no such population has been detected in deep optical surveys , but the scuba galaxies , or some of them , could represent the long - sought progenitors . understanding these so - called `` scuba galaxies '' and their place in cosmic evolution , however , requires complementary observations at other wavelengths and in particular some way to determine redshifts . observing scuba galaxies in visible light is difficult . the dust responsible for the submillimeter emission obscures the primary luminosity sources , making the scuba galaxies quite faint at visible wavelengths . even identifying the correct visible counterpart to a given scuba galaxy is difficult because of the large beamsize ( 14fwhm at 850 @xmath0 m ) of the james clerk maxwell telescope and other submillimeter facilities . the unavoidable result is that at visible and even near - infrared wavelengths , there are numerous potential counterparts in the scuba beam , and there is no easy way to decide which of the candidates is the true counterpart . one successful approach to finding counterparts has been to use radio interferometric observations to refine the scuba positions and thus select a specific visible object . chapman et al . ( 2003 , 2005 ) , for example , have used this approach and have shown that radio - identified scuba galaxies have a median spectroscopic redshift of @xmath5 . however , about 1/3 of scuba sources are not detectable with current radio observations ( chapman et al . 2005 ) and could represent a distinct population . it is therefore worthwhile seeking alternate ways to identify counterparts . this contribution describes a new program to recover submm - selected galaxies in the mid - infrared , so as to exploit the advantages this passband offers for better identifications of the counterparts to these important objects . our submm sample is comprised of 23 scuba - detected objects in the 14@xmath1 field of the canada - uk deep submillimetre survey ( cudss ; lilly et al . 1999 , eales et al . 2000 , webb et al . we base our infrared counterpart search on the very deep 8.0 @xmath0 m _ spitzer _ mosaic of the 14@xmath1 field . although the _ spitzer _ observations ( described below ) include coextensive observations at a number of passbands from 3.6 to 160 @xmath0 m , we use sources drawn specifically from the 8.0 @xmath0 m mosaic , for the following reasons . first , 8.0 @xmath0 m observations sample rest - frame _ k_-band emission at typical scuba galaxy redshifts @xmath6 . they are therefore much less affected than shorter - wavelength observations by extinction due to the dust known to be abundant in galaxies that are actively forming stars ( schmitt et al . 2006 ) . this rest - frame _ k_-band emission is also a good proxy for stellar mass ( cole et al . 2001 ) ; it is therefore more sensitive to total star formation history than shorter - wavelength observations that are relatively more affected by recent star formation . second , high - redshift galaxies like those selected at submm wavelengths typically have red infrared colors that are well - suited to detection at 8.0 @xmath0 m ( egami et al . 2004 , huang et al . 2004 , frayer et al . 2004b ) , or at least they have red optical and near - infrared colors ( pope et al . 2005 , frayer et al . 2004a , borys et al . 2004 , ivison et al . lockman hole galaxies were detected in 450 and 850 @xmath0 m scuba maps with stacking analyses at 5.8 and 8.0 @xmath0 m by serjeant et al . ( 2004 ) but not at 3.6 or 4.5 @xmath0 m . in addition , although the 3.6 and 4.5 @xmath0 m _ spitzer _ observations in the 14@xmath1 field are somewhat more sensitive than the 8.0 @xmath0 m mosaic , they tend to detect more of the blue , relatively numerous , low - redshift galaxies that are unrelated to the scuba - selected population . the resulting contamination problem is of course more severe at shorter wavelengths , e.g. , _ k_-band . at 8.0 @xmath0 m we avoid to a great degree the problems of false associations that affect these other bands , without missing sources : all potential 8.0 @xmath0 m counterparts we considered are also detected in the 3.6 @xmath0 m mosaic . third , the 8.0 @xmath0 m mosaics themselves are very deep down to a 5@xmath7 point source sensitivity of 5.8 @xmath0jy . we used two very simple models of star formation history to calculate that this depth is easily sufficient to detect typical scuba galaxies . both models ( one a constant star formation rate model , the other a burst ) give approximately equivalent results and suggest that at redshifts of @xmath8 the 8.0 @xmath0 m mosaics will detect star formation rates of @xmath9 and @xmath10@xmath11 yr@xmath12 , respectively . this is comfortably below the rates found for submm galaxies ( up to roughly @xmath13@xmath11 yr@xmath12 , e.g. , smail et al . 2004 , yun & carilli 2002 , lilly et al . 1999 ) , so we can expect _ a priori _ that their 8.0 @xmath0 m counterparts are detected in the _ spitzer _ data . this paper describes the _ spitzer _ observations , the 8.0 @xmath0 m counterpart identification procedure , and the inferences to be drawn from the panchromatic seds of infrared - selected scuba counterparts . a companion paper ( dye et al . 2006 ) describes the inverse process , whereby we determine the average submillimeter emission at the positions of the _ spitzer _ sources found in the field . the next section describes the _ spitzer_/irac and mips observations and presents infrared images at the known locations of the scuba galaxies . we then outline the scheme used to identify the most - likely infrared counterparts of the submillimeter galaxies in section [ sec : ids ] . the infrared properties of the likely counterparts are analyzed in section [ sec : discussion ] . the irac ( fazio et al . 2004 ) observations were carried out as part of spitzer guaranteed time observing program number 8 to deeply image the extended groth strip ( egs ) . the first visit to the field was in 2003 december . the field was observed again during a second epoch in 2004 june at a different position angle . the irac exposures ( each of which covered a 512@xmath1412 field of view with @xmath15 pixels @xmath16 in size ) totaled 52 dithered 200 s exposures at 3.6 , 4.5 , and 5.8 @xmath0 m , together with 208 dithered 50 s exposures taken concurrently at 8.0 @xmath0 m at all positions in a @xmath17 map . each dither cluster was separated from its neighbor by 290 , slightly less than the irac field of view . because the field was observed at position angles differing by roughly @xmath18 during the two epochs it was straightforward to remove most of the well - known instrumental artifacts ( multiplexer bleed , banding ) during mosaicing . the resulting point - source sensitivity was 24.0 , 24.0 , 21.9 , and 22.0 mag ( ab , 5@xmath7 ) , in the 3.6 , 4.5 , 5.8 , and 8.0 @xmath0 m bands , respectively . this corresponds to flux density levels of 0.9 , 0.9 , 6.3 , and 5.8 @xmath0jy in the four irac bands . all magnitudes given in this work are on the ab system . sources in the irac mosaics were identified and extracted using sextractor ( bertin & arnouts 1996 ) . the software was configured with a 2.5@xmath7 detection threshold and a minimum object area of at least 5 pixels . the result was a band - merged photometric catalog of objects for all four irac wavelengths ; photometry was performed using 3 diameter apertures centered at the positions computed at 3.6 . all sources here are pointlike in the irac bands . to permit comparison with total magnitudes in the other bands , appropriate corrections were applied to obtain total magnitudes from the aperture magnitudes . the density of 8.0 @xmath0 m sources brighter than 5.8 @xmath0jy is roughly 11 per square arcminute . the mips ( rieke et al . 2004 ) observations of the egs were carried out on 2004 june 19 and 20 . the mips 24 @xmath0 m channel ( @xmath19 m ; @xmath20 m ) used a 128@xmath21128 si : as array with a pixel scale of 255 pixel@xmath12 , providing a field of view of @xmath2254 . the 70 @xmath0 m channel ( @xmath23 m ; @xmath24 m ) uses a 32@xmath2132 ge : ga array with a pixel scale of 998 pixel@xmath12 , but one half of the array is not usable due to a high noise caused by a cabling problem . the effective 70 @xmath0 m array size is therefore 32 @xmath2116 pixels , providing a field of view of 52@xmath2126 . the area was observed in scan map mode at slow rate with scan legs 24 long . this results in an integration time of 100 sec pixel@xmath12 per scan pass ( 10 frames@xmath2110 s ) at both 24 and 70 @xmath0 m . in scan mode images at these two wavelengths are obtained simultaneously . pairs of scan map observations , with a cross - scan offset of one full array ( 296 ) , were executed 6 times , with an offset of 21 between successive pairs . the effective integration time is therefore @xmath251500 s ( 12 passes ) at 24 @xmath0 m for locations near the long centerline of the strip , decreasing to @xmath25700 s 5 from the centerline . at 70 @xmath0 m , the integration time is only half as much due to the loss of half of the detector array . the final scan maps cover an area 2.4@xmath2110 with an integration time @xmath26 sec pixel@xmath12 . a smaller 2@xmath216 strip in the center , encompassing most of the cudss field , was covered with an integration time of 13001600 seconds . the data were reduced and mosaiced with the mips data analysis tool ( gordon et al . the 24 @xmath0 m scan images were resampled and mosaiced at half the original instrument pixel scale ( 125 ) while at 70 @xmath0 m the original pixel scale was preserved . the point spread functions in the mosaics have fwhms of 6 and 18 at 24 and 70 @xmath0 m , respectively . the resulting 5@xmath7 point source sensitivities are respectively 70 @xmath0jy and 7 mjy at 24 and 70 @xmath0 m . we have drawn from several pre - existing photometric catalogs in order to construct seds spanning as wide a wavelength range as possible for our counterparts . for the u , b , v , and i band photometry , we used the canada - france deep field survey ( mccracken et al . 2001 , hereafter cfdf ) survey data . cfdf magnitudes are calculated within 2.5 diameter apertures . the 3@xmath7 cfdf limiting magnitudes are 27.71 , 26.23 , 25.98 , and 25.16 mag in u , b , v , and i , respectively . at r we use the very deep image acquired at the subaru telescope with _ suprime _ ; our r - band photometry has a @xmath27 limit of 26.6 mag ( miyazaki 2006 , private communication ) . for z - band we use the catalog constructed by brodwin et al . ( 2006 ) , converting vega to ab magnitudes by adding a constant 0.54 mag . the limiting magnitude of the z - band data is 25.0 ( @xmath28 ) , within 2.5 diameter apertures . the @xmath29-band photometry data are drawn from the observations described by webb et al . ( 2003 ) , converting to ab magnitudes using @xmath30 . @xmath29-band magnitudes were measured within 3 apertures . because the @xmath29-band data were acquired at two facilities , the depth is nonuniform but it generally extends down to @xmath31 mag ( @xmath32 ) . all visible and near - infrared sources considered here are small compared to the apertures used ; we therefore treat the aperture magnitudes as total magnitudes . two of the submillimeter galaxies ( 14.13 and 14.18 ) have _ infrared space observatory _ ( iso ) 15 @xmath0 m fluxes reported by flores et al . one of these ( 14.13 ) was observed using _ spitzer s _ _ infrared spectrograph _ ( irs ) in low - resolution mode by higdon ( 2004 ) , who also report a 16 @xmath0 m flux density from the irs peakup imager . finally , we have also made use of the chandra x - ray observatory ( cxo ) source catalog compiled by nandra et al . a montage showing our observations at r , 3.6 , 4.5 , 5.8 , 8.0 , and 24 @xmath0 m at the locations of each of the 23 scuba sources in the 14@xmath1 field is provided in figure [ fig : montage ] . to avoid spurious sources we restrict our investigation to portions of the 14@xmath1 field covered by at least ten 8 @xmath0 m irac exposures . with this constraint the irac coverage allows counterpart identifications for 20 of the 23 scuba galaxies in this field . we identify irac counterparts of the scuba sources using a two - part approach , consisting first of a proximity search of the irac 8.0 @xmath0 m source catalog , followed by an examination of the infrared seds of all possible matches . we first use a metric based on 8.0 @xmath0 m brightness and positional coincidence to select likely counterparts . we restrict the search to irac objects within 10 of each scuba source , for several reasons . the primary motivation is to adopt a ( conservative ) value slightly larger than the @xmath33 uncertainty in the scuba positions , so as not to miss any true counterparts that might lie at relatively large separations , while simultaneously avoiding such a large value that we include a large number of unassociated foreground objects in the search . we find that a 10 search radius often selected just one ( or at most three ) candidates , and was therefore well - matched to our irac source density . based on the source surface density down to the 5@xmath7 limit of our survey , on average one irac 8 @xmath0 m source should randomly reside within each 10 radius search area . borys et al . ( 2004 ) found 11 counterparts within 7 of 19 submillimeter objects in the hubble deep field north , implying a substantial fraction may lie beyond this radius . similarly , wang et al . ( 2004 ) identified five of their 20 x - ray counterparts and two of their six mid - infrared counterparts to submillimeter galaxies in the same field at separations greater than 7 . by contrast lilly et al . ( 1999 ) found at least 90% of the visible - wavelength counterparts to the submillimeter galaxies in the cudss 14@xmath1 field are present within 7 , but their analysis includes only six sources . our 10 criterion is also safely larger than the mostly - likely average 850 @xmath0m - infrared offset of 2 determined from the stacking analysis discussed by dye et al . ( 2006 ) . all potential counterparts are listed in table [ 8micids ] . in seven cases ( see below ) a single , obvious best counterpart was identified . in instances where the identifications were not clear - cut , we applied a second criterion based on infrared color to refine the identifications . specifically , we made use of @xmath29-band photometry from webb et al . ( 2003 ) as well as our own irac photometry to discriminate the most likely counterpart among multiple candidates . in doing so , we made one important assumption . we _ a priori _ assumed , based on spectroscopic studies ( chapman et al . 2003 , 2005 ) , that the true counterparts must reside at high redshift and that the seds must be consistent with this fact . we applied color criteria developed by huang et al . ( 2004 ; see their table 2 ) , who use the @xmath29-to - irac colors to constrain a galaxy s redshift into local , intermediate , and high - redshift regimes . those criteria are summarized as follows : a blue @xmath34 $ ] color corresponds to redshifts @xmath35 . red @xmath34 $ ] color is consistent with either an intermediate ( @xmath36 ) or high redshift ( @xmath371.3 ) , depending on whether the mid - infrared @xmath38 - [ 4.5]$ ] color is blue or red , respectively . in section [ ssec : notes ] we make frequent use of this technique to select the best ( highest - redshift ) candidate among multiple possibilities . details of the identifications for these 20 accessible sources are provided in the following section . seven sources have secure identifications : 14.1 , 14.3 , 14.13 , 14.17 , 14.18 , 14.19 , and 14.23 . of these , all but 14.17 and 14.23 have relatively small position uncertainties . we have possible identifications for ten others : 14.2 , 14.4 , 14.7 , 14.8 , 14.10 , 14.12 , 14.14 , 14.15 , 14.20 , and 14.22 . because their infrared colors and r - band morphologies are not those of dusty , high - redshift galaxies , we regard both potential 8.0 @xmath0 m identifications for 14.11 as dubious and do not include it in subsequent analysis . likewise 14.6 and 14.21 which have no significant 8.0 @xmath0 m counterpart within 10 of the scuba positions . objects 14.1 , 14.3 , 14.13 , and 14.18 have 1.4 ghz positions available from fomalont et al . a recalibration of the vla data by m. yun improved the effective beamwidth of this dataset to 4 fwhm and resulted in a radio identification for 14.19 ( webb et al . 2003 ) . _ cudss 14.1_. our single 8.0 @xmath0 m identification is coincident with both the webb et al . ( 2003 ) @xmath29-selected candidate and the i - selected candidate from lilly et al . ( 1999 ) within the positional uncertainties . there is also a mips 24 @xmath0 m source at this position . _ cudss 14.3_. our only 8.0 @xmath0 m candidate is coincident with the webb et al . ( 2003 ) candidate as well as the position observed by chapman et al . 2005 , who used keck to obtain a spectroscopic redshift of @xmath39 for this object . this object is also an obvious mips 24 @xmath0 m source . _ cudss 14.13_. for this object the best candidates at @xmath29 and 8.0 @xmath0 m agree within the positional uncertainties . this object is also a strong mips 24 @xmath0 m source and was detected by iso at 7 and 15 @xmath0 m at the same position . this is not surprising , because it is by far the brightest of all our irac candidates . this object was also observed with the infrared spectrograph on the _ spitzer space telescope _ ( higdon et al . 2004 ) and exhibits a continuum slope consistent with a seyfert 2 nucleus and no pah emission . our best position is coincident with that observed by chapman et al . ( 2005 ) , who used keck to obtain a spectroscopic redshift of @xmath40 for this object ; this places it in the lowest quartile of the submm galaxy redshift distribution . _ cudss 14.18_. for this object our 8.0 @xmath0 m search selects the same candidate as webb et al . found at @xmath29 . this is identical to the i - selected candidate from lilly et al . ( 1999 ) as well . within the positional uncertainties , all three positions ( irac+@xmath29-selected+i - selected ) are also coincident with the iso 7 and 15 @xmath0 m detections . the 8.0 @xmath0 m object is also a bright 24 @xmath0 m source . finally , the best irac position is coincident with that observed at keck by chapman et al . ( 2005 ) , who found a spectroscopic redshift of @xmath41 for this object . this is a surprisingly low redshift compared to most scuba sources . _ cudss 14.19_. for this object our 8.0 @xmath0 m search selects a different best candidate than at @xmath29 . the irac candidate position lies just 2 from the radio position and is coincident with significant 24 @xmath0 m emission . we identify what we believe to be secure infrared identifications for two objects that lack radio counterparts , 14.17 and 14.23 . _ cudss 14.17_. our 8.0 @xmath0 m search selects a single best candidate and disagrees with the @xmath29-selected candidate , which is not apparent in the mid - infrared mosaics . the 8.0 @xmath0 m candidate falls just 08 from the iso detection and is detected by mips at 24 @xmath0 m . thus despite the large offset ( 95 ) this identification appears secure . _ cudss 14.23_. our 8.0 @xmath0 m search selects two candidates . one ( b ) is fainter , blue in @xmath38-[4.5]$ ] , and offset 66 to the ne of the scuba position . the other , brighter source is red in @xmath38-[4.5]$ ] and @xmath42 $ ] and is a strong source of 24 @xmath0 m emission . it is offset only 2 from the scuba position . for these reasons we prefer candidate a. it is the same as the @xmath29-selected candidate from webb et al . we identify plausible infrared counterparts to ten sources , none of which has a radio identification ( but see discussion of 14.8 and 14.20 below ) . _ cudss 14.2_. although we detect the webb et al . ( 2003 ) candidate ( 14.2a in table [ 8micids ] ) in all irac bands , we also find a closer , fainter irac object ( b ) only 21 from the reported scuba position . however , candidate a is a source of mips 24 @xmath0 m emission and is much brighter at 8.0 @xmath0 m . for these reasons we adopt a as the more likely counterpart to 14.2 . the lilly et al . ( 1999 ) i - selected counterpart , which is distinct from both the @xmath29- and irac - selected positions , is not detected in the irac mosaic at 8.0 @xmath0 m . _ cudss 14.4_. at 8.0 @xmath0 m there are three possible counterparts . of these , c is approximately coincident with the webb et al . object c is also the faintest of the three candidates at 8.0 @xmath0 m and shows no significant 24 @xmath0 m emission . it is blue in @xmath34 $ ] , suggesting this object resides at @xmath43 . thus c seems unlikely to be the correct counterpart . object a presents a red @xmath38 - [ 4.5 ] > 0 $ ] color , but the @xmath34 $ ] upper limit is not sufficiently sensitive to constrain the redshift . object a is a mips 24 @xmath0 m source . object b has no counterpart in the 24 @xmath0 m catalog , but this is most likely because it is blended with an even brighter object to the nw . however object b presents a blue @xmath38 - [ 4.5]$ ] color , suggesting it lies at @xmath44 . thus candidate a seems the most likely scuba counterpart among the irac sources , despite its relatively large separation ( 87 ) from the scuba position . _ cudss 14.7_. there are two possible 8.0 @xmath0 m counterparts within the 10 search radius . one of them ( a , the closer and fainter of the two ) coincides with the webb et al . ( 2003 ) @xmath29-selected candidate . although both are detected in all irac bands , only the southern source ( b , the one more distant from the reported scuba position ) is unambiguously detected by mips . the southern source exhibits multiple components in our deep r - band image . it also has a redder @xmath45-[8.0]$ ] and @xmath46-[24.0]$ ] colors and is by far the brighter source at 8.0 @xmath0 m . for these reasons we prefer candidate b. however , the scuba source is extended toward both irac objects . in addition , both irac candidates have red @xmath29-to-4.5 @xmath0 m colors that suggest they both reside at @xmath47 . thus while we prefer b we can not rule out the possibility that both objects may contribute to the submillimeter flux . _ cudss 14.8_. our 8.0 @xmath0 m search selects two candidates ( a and b ) . there is emission at the location of the @xmath29-selected candidate ( b ) , but it is present at less than 5@xmath7 significance . candidate a is roughly 65 west of the best scuba position . there is 24 @xmath0 m emission at this position but not at the position of candidate b. interestingly , the @xmath29-selected candidate may be a source of radio emission ( chapman et al . 2005 ) , and a spectroscopic redshift @xmath48 has been obtained for it . the radio detection is regarded as somewhat marginal however ( @xmath49 , s. chapman 2005 , private communication ) , and so the association of candidate b with the scuba source is not established with certainty . object b is detected at 3.6 and 4.5 @xmath0 m with flux densities of 5.4 and 5.6 @xmath0jy , respectively , but it exhibits no significant emission at 5.8 , 8 , or 24 @xmath0 m . because b has a blue mid - infrared sed and only a is significantly detected 8.0 @xmath0 m , we prefer a as the counterpart . this object has red @xmath29-to-4.5 @xmath0 m colors that place it at @xmath47 , meaning it is possible that it could reside at the same redshift as candidate b as part of an interacting system . _ cudss 14.10_. our 8.0 @xmath0 m search selects a different best candidate ( a ) than webb et al . ( b ) , 98 sw of the best scuba position . candidate a shows @xmath42 = 1.5 $ ] and @xmath38-[4.5]=-0.04 $ ] , suggesting it lies in the range @xmath36 . candidate b , the @xmath29-selected object , though closer to the scuba position ( 44 away ) , is almost 1 magnitude fainter at k , is blue in @xmath38-[4.5]$ ] , and exhibits no significant emission at 5.8 , 8 , or 24 @xmath0 m . hence a is the more likely counterpart . _ cudss 14.12_. our 8.0 @xmath0 m diagnostic selects two candidates , neither of which is coincident with the @xmath29-selected candidate . the @xmath29-selected candidate is not detected in the 8.0 @xmath0 m mosaic . candidate a lies 55 w of the scuba coordinates and exhibits significant 24 @xmath0 m emission . neither a nor b is a source of significant 5.8 @xmath0 m emission . although b is brighter at 8.0 @xmath0 m than a , it exhibits no significant 24 @xmath0 m emission . hence a is the more likely candidate , although the identification is by no means secure . _ cudss 14.14_. at 8.0 @xmath0 m there are two candidates . neither is coincident with the @xmath29-selected candidate . significant mips 24 @xmath0 m emission is associated only with candidate b. likewise only this candidate has significant detections in all four irac bands . furthermore , only candidate b is red in @xmath38-[4.5]$ ] . candidate b is our preferred counterpart on the basis of its relative brightness , the irac colors , and the presence of 24 @xmath0 m emission at its position . _ cudss 14.15_. there are two possible 8.0 @xmath0 m counterparts . one of these ( a , the brightest ) is coincident with the best @xmath29-selected candidate and is a mips 24 @xmath0 m source . candidate b , however , exhibits comparable 24 @xmath0 m emission , has a red @xmath38-[4.5]$ ] color , and has an sed that peaks at 5.8 @xmath0 m , suggestive of a galaxy at @xmath50 . although a is brighter at 8.0 @xmath0 m , b is a more likely candidate because of these sed features . perhaps because the @xmath29-band coverage at this location is relatively shallow , only a @xmath29 upper limit can be determined for candidate b. the r - band source seen in figure [ fig : montage ] close to the apparent position of candidate b is not the visible - wavelength counterpart . only upper limits could be established in the bands blueward of i for this object . _ cudss 14.20_. the 8.0 @xmath0 m search selects a different best candidate than webb et al . . the irac candidate is not detected in our deep r - band image or the webb et al . i - band image but is detected with @xmath51 mag . it therefore has red @xmath29-to-4.5 @xmath0 m colors that place it at @xmath47 . there is no significant mips 24 @xmath0 m source within 10 of the scuba position . acquiring a visible - wavelength spectrum of this extremely optically - faint object would be very challenging . as is the case for 14.8 , we select a different position than preferred by chapman et al . ( 2005 ) , who obtained a spectroscopic redshift of @xmath48 for the @xmath29-selected candidate , where they detect 1.4 ghz emission . however , the radio identification , which is made at roughly 4@xmath7 significance ( s. chapman 2005 , private communication ) may be incorrect . _ cudss 14.22_. the 8.0 @xmath0 m search finds two candidates . one of these ( a ) is coincident with the the @xmath29-selected candidate to the nw . even though both are red in @xmath42 $ ] , the fact that they are both blue in @xmath38-[4.5]$ ] suggests that they lie in a relatively nearby redshift range @xmath36 and that we may have not detected the true counterpart of this object . however , only the southernmost candidate is a strong 24 @xmath0 m source , so we regard it as the more plausible irac counterpart . for the sake of completeness , we examined the relatively faint 24 @xmath0 m source apparent in figure [ fig : montage ] 103 nnw of the scuba position , just outside the search radius . it is blue throughout the irac bands and is a much weaker 24 @xmath0 m source ( 232 @xmath0jy ) than candidate b. it is therefore likely to be a low - redshift object and not associated with the scuba source . _ cudss 14.6_. despite the excellent irac coverage at the location of cudss 14.6 ( a total of 218 exposures ) , no sources in the irac 8.0@xmath0 m mosaic satisfy our selection criteria as a valid counterpart to cudss 14.6 . within 10 of the reported scuba position , and specifically at the position identified by webb et al . ( 2003 ) , there is no significant 5.8 , 8.0 , or 24 @xmath0 m emission . moreover , the _ k _ to 4.5 @xmath0 m colors of the _ k_-selected counterpart are blue , suggesting that object lies nearby . there is a faint source in the 8.0 @xmath0 m mosaic within the 10 search radius , 85 to the ne of the scuba position . however , its low flux density ( 5.5 @xmath0jy ) puts it below the 5@xmath7 sensitivity of the 8.0 @xmath0 m mosaic . like the _ k_-selected counterpart it also has a blue @xmath42 $ ] color , implying that it resides relatively nearby . the object appears in three of the irac mosaics ( it is not apparent at 5.8 @xmath0 m ) and is therefore unquestionably real , but it is too faint to satisfy our selection criteria . a significant 8.0 @xmath0 m detection ( 18 @xmath0jy ) is located just outside the search area , 101 to the ne of the scuba position . it is red in @xmath42 $ ] but exhibits a blue @xmath38-[4.5]$ ] color , suggesting it lies at an intermediate ( @xmath36 ) redshift and is therefore unlikely to be the true counterpart . both of the 8.0 @xmath0 m sources mentioned above are as likely as not to be due to chance superposition . in combination with the color information this strongly suggests that no plausible counterpart to the scuba source is present in the irac mosaics . _ cudss 14.11_. at 8.0 @xmath0 m there are two indistinguishable best candidates , one of which ( b ) is coincident with the @xmath29-selected candidate . candidate a is closer to the scuba position , only 26 away . neither shows any significant 24 @xmath0 m emission , and both exhibit the morphologies of elliptical galaxies in our deep r - band image . furthermore , the blue infrared colors ( @xmath42 = -0.53,-0.60 $ ] , and @xmath38-[4.5]=-0.48,-0.69 $ ] ) and _ r_-band morphologies are consistent with both objects being elliptical galaxies at @xmath35 . there is no significant 24 @xmath0 m emission within 10 of the scuba position . for all these reasons , we conclude that we have not detected the true counterpart to this scuba source . a speculative possibility is that the 850 @xmath0 m emission detected by scuba is lensed by the foreground , low - redshift objects . the closest 24 @xmath0 m source ( 176 @xmath0jy ) lies 157 ne of the scuba position and has blue colors throughout the irac bands . _ cudss 14.21_. there are no significant 8.0 @xmath0 m sources within 10 of the scuba position of this source . we therefore attempt no identification . there is significant 3.6 and 4.5 @xmath0 m emission at the location of the @xmath29-selected counterpart . this object is red in @xmath42 $ ] but blue in @xmath38-[4.5]$ ] , suggesting that it may be at @xmath36 . at 5.8 , 8.0 , and 24.0 @xmath0 m only upper limits are available . we regard cudss 14.21 as unidentified at 8 @xmath0 m . assuming that these sources are not highly dust - obscured galaxies at such extremely high redshift ( @xmath52 ) that their obscured rest - visible emission falls in the observed mid - infrared , we suggest three possible causes for the lack of plausible 8 @xmath0 m counterparts . in the case of cudss 14.11 , it may be that the two foreground elliptical galaxies are screening a high - redshift scuba source from view at visible - mid - infrared wavelengths . a second alternative is that the submillimeter flux densities may include contributions from multiple weak components that lie within the scuba beam but not precisely at the reported scuba position . the scuba maps are known to be highly confused below 2 mjy ( knudsen et al . 2006 , and references therein ) . weak submillimeter sources that combine to yield an apparent scuba detection might well be individually too faint to detect even in the deep irac mosaic . third , one of the scuba sources might be spurious , which would not be unreasonable given that the sample contains so many roughly 3@xmath7 sources . cudss 14.6 , 14.11 , and 14.21 are detected in the 850 @xmath0 m maps at 4.2 , 3.5 , and 3.0@xmath7 significance , respectively . three sources lie in a portion of the irac 8.0 @xmath0 m survey region for which fewer than 10 exposures were acquired . for these sources ( 14.5 , 14.9 , and 14.16 ) the depth of coverage was insufficient to reliably detect an infrared counterpart . excluding ` dubious ' associations , potential counterparts for 17 of 20 scuba galaxies having deep 8.0 @xmath0 m coverage in the 14@xmath1 field have been identified ( table 2 ) . of these , the seven secure identifications have a mean offset from the scuba positions of 4.4 `` and the ten possible identifications have a mean offset of 7.7 '' . if one uses the available 1.4 ghz positions in preference to the scuba positions only the mean offset for the set of secure identifications changes ( because no scuba sources with ` possible ' identifications have radio positions ) , to 24 . the mean offsets in right ascension and declination for the 17 counterparts are 25 and 09 ( scuba - 8.0 @xmath0 m ) , respectively . all these offsets are smaller than the scuba positional uncertainties . of course , a small number of the 8.0 @xmath0m - selected counterparts are chance associations and not physically associated with the corresponding scuba source . in the following we address this issue in two ways . first , we derive an estimate of the number of such chance associations using a statistical analysis . second , we compare the infrared - submm colors of our counterparts to those that have been reliably identified in a deep irac survey of the lockman hole region by egami et al . ( 2004 ) . we first employ a variant of the statistical formalism used by lilly et al . ( 1999 ) and others to assess the significance of possible matches . specifically , we estimate a statistic , usually denoted @xmath53 , that a candidate irac source lies within the search distance ( 10 ) of a scuba source to which it is unrelated . the statistic is @xmath54 where @xmath55 and @xmath56 is the surface density of irac sources at least as bright as the candidate . thus lower values of @xmath53 suggest a higher likelihood that the association of the irac and scuba sources is significant . unlike lilly et al . ( 1999 ) we treat all sources within the 10search radius equally with respect to their offset from the scuba position . in doing so we avoid unduly biasing our selection toward sources at small offsets from the uncertain scuba positions . we use table 1 of webb et al . ( 2003 ) for the scuba sources coordinates , except for cudss 14.1 , 14.3 , 14.13 , 14.18 , and 14.19 . the first four of these objects have vla positions ( eales et al . 2000 ) with lower uncertainties ( roughly 2 ) . similarly , we adopt the radio position reported for 14.19 by webb et al . ( 2003 ) , although it is offset from the nominal scuba position by 85 . we found no significant differences in our lists of most - likely matches when we repeated the analysis using the less - certain scuba positions . in order to interpret the @xmath53-values , we performed a monte carlo analysis in which randomly - generated catalogs of irac sources were searched for counterparts in a manner identical to that employed for the real sources . the random catalogs were constructed to present brightness distributions and source densities identical to those of the real catalog . the @xmath53-distributions for the random and real samples are compared in figure [ fig : pdist ] . we calculate the probability @xmath57 of finding a random or unassociated source as bright as the observed candidate as @xmath58 , where @xmath59 is the ratio of the number of random sources with smaller values of @xmath53 than the candidate to the total number of random sources generated in the monte carlo calculation . although the @xmath57 statistics are not meaningful for individual sources , the sum of the @xmath57 statistics tabulated in table [ 8micids ] is an estimate of the number of spurious identifications . we can not say which _ particular _ identifications are wrong , but the sums for the seven secure and the ten possible identifications are 0.61 and 2.61 , respectively , implying that the bulk of the errors apply to the latter subset , as one would expect . for the sample as a whole , the sum is 3.22 , suggesting that about three of our 17 identifications are incorrect . this estimate ( three false identifications ) is a pessimistic one , because the probability formalism from which it is derived does not account for our use of color information to winnow lists of multiple candidates down to single most - likely counterparts . three is in fact the maximum number of likely false sources , because we have taken into account the @xmath29 to mid - infrared colors as well as the presence of 24 @xmath0 m mips emission as identification criteria . when the estimate for the number of spurious sources is subtracted , our identification rate is ( conservatively ) 14/20 , or 70% . this is comparable to or better than rates described elsewhere . borys et al . ( 2004 ) recovered 11 of 19 ( 57% ) scuba source in the hdf - n . pope et al . ( 2005 ) reported a 72% recovery rate for scuba sources in the goods - n field , although this figure does nt account for possible erroneous associations . like pope et al . ( 2005 ) we recovered essentially all the scuba sources with secure radio positions ( but see the discussions of 14.8 and 14.20 above ) . more straightforward comparisons can be made with lilly et al . ( 1999 ) and webb et al . ( 2003 ) , because the @xmath57 statistics they provide permit one to estimate the number of incorrect associations in a manner identical to that applied to our own sample . lilly et al . ( 1999 ) identified counterparts for eight of the 12 submillimeter sources they observed at visible wavelengths . the sum of the resulting @xmath57 statistics was @xmath60 , suggesting a total identification rate of 7/12 or 58% . likewise the webb et al . ( 2003 ) @xmath29-band counterparts for which the @xmath57 statistics sum to @xmath61 , implying a total identification rate of 17/23 or 74% ( but see below ) . egami et al . ( 2004 ) presented mid - infrared through submillimeter flux densities measured by spitzer for scuba sources in a @xmath62 area of the lockman hole east . because these sources have radio counterparts from ivison et al . ( 2002 ) their positions are well - localized , and the mid - infrared identifications are secure . for this reason , the egami et al . ( 2004 ) submillimeter sample provides a useful basis with which to assess our own larger sample of infrared - selected counterparts . figure [ fig : irac_colors ] shows the distributions of submillimeter galaxies in the irac @xmath38-[4.5]$ ] versus @xmath45-[8.0]$ ] color - color space . the portion of this space occupied by the lockman hole sources and the secure identifications of the cudss 14@xmath1 field sources are approximately coextensive ; they define a locus in which other counterparts sources may be expected to fall . this is indeed the case for the ` possible ' sources taken as a group the distribution of these objects occupies a very similar region of this color - color space overall as do the secure sources . in general , when the counterpart is bright , the @xmath29- and 8.0 @xmath0m - selected counterparts are identical . however , the mid - infrared technique selects different counterparts than webb et al . ( 2003 ) in 10 cases ( cudss 14.4 , 14.7 , 14.8 , 14.10 , 14.12 , 14.14 , 14.15 , 14.19 , 14.20 , and 14.22 ) . the @xmath29- and 8.0 @xmath0m - selected techniques disagree for three additional cases ( cudss 14.6 , 14.11 , and 14.21 ) for which the mid - infrared technique identifies no plausible counterparts . cudss 14.11 is a nearby elliptical galaxy and is therefore unlikely to be the source of the submillimeter emission . cudss 14.6 and 14.21 have no plausible counterparts in the 8.0 @xmath0 m mosaics . the @xmath29- and 8.0 @xmath0m - selected counterparts therefore disagree in a total of 13 cases ( slightly more than half the total ) . with the exception of the counterpart to cudss 14.7a ( which as noted above may indeed be a contributor to the submillimeter emission ) , the distribution of the @xmath29-selected sources is clearly blue in both colors compared to the other objects . in particular it lies close to the region occupied by a passively evolving elliptical galaxy . only six @xmath29-selected counterparts are plotted because the remainder are undetected at both 5.8 and 8.0 @xmath0 m , however the @xmath38-[4.5]$ ] colors of all these other sources tend to be similarly blue in @xmath38-[4.5]$ ] and not like the distributions of the secure identifications ( table [ 8micids ] ) . we conclude that the 8.0 @xmath0m - selected counterparts are therefore more likely to be the correct counterparts in these discrepant cases . figure [ fig : mips_scuba ] shows the same populations but in @xmath63-[850]$ ] versus @xmath46-[24]$ ] color - color space . again , the distributions of the lockman hole counterparts and the secure sources from this work are similar . cudss 14.13 and 14.18 are understandable exceptions on account of their relatively low redshifts . and as before , the ` possible ' counterparts occupy a region of this color - color space that is reassuringly similar to that of the secure sources . that is not the case for the discrepant @xmath29-selected counterparts . these objects lie at the outskirts of the secure counterparts distribution . based on the lockman hole sources and the secure 8.0 @xmath0m - selected sources ( not including cudss 14.13 and 14.18 ) , one would expect the typical @xmath63-[850]$ ] color of a scuba galaxy to be between 3 and 4 magnitudes . not one of the discrepant @xmath29-selected counterparts lies in this range , and indeed only one ( 14.15a ) is even detected at 24 @xmath0 m . analogous trends are apparent in @xmath38-[8.0]$ ] versus @xmath46-[24]$ ] color - color space shown in figure [ fig : mips_irac ] . the secure 8.0 @xmath0m - selected sources occupy the same region of this color - color space as the lockman hole sources , and that region is spanned by the possible 8.0 @xmath0 m identifications . the low - redshift objects are not outliers in this distribution . as before , the discrepant @xmath29-selected counterparts tend to be blue in @xmath38-[8.0]$ ] and with the exception of 14.7a do not overlap with the securely identified counterparts ; half of these discrepant sources are so blue they are not even detected at 8 or 24 @xmath0 m . figure [ fig : sedplot ] shows the seds of our most - likely irac counterparts to the scuba sources in this field , showing the diversity of spectral shapes present within the sample . the galaxies divide roughly into two categories . the majority ( 12 objects ) show evidence in the irac bands for the 1.6 @xmath0 m bump arising from the h@xmath64 opacity minimum . in other words , these objects appear to be powered primarily by stars and not by a compact object . cudss 14.18 is among these sources , in agreement with chapman et al . ( 2005 ) . four objects in the sample ( sources 14.3 , 14.7 , 14.13 , and 14.19 ) show the monotonically rising seds indicative of agn - dominated emission . these sources fulfill all three of the irac - color agn selection criteria proposed by @xcite , @xcite , and @xcite . source 14.4 has an unusual sed , very faint in the three shortest - wavelength irac bands compared to the optical and 8 and 24 @xmath0 m . this source also has a monotonic rise through the irac bands , so we consider it a tentative agn candidate . simple color criteria developed from shallow irac surveys like those cited above are not adequate to separate agn from starburst galaxies at high redshift however . we therefore chose to perform a simple sed fitting procedure to determine whether a galaxy is agn - dominated , in exactly the same way as @xcite did for other sources in the extended groth strip . specifically , we fit power laws ( @xmath65 ) to irac flux densities as an empirical way of characterizing their seds . seven sources have acceptable power - law fits . source 14.14 has a blue power - law ( @xmath66 ) , source 14.18 has @xmath67 , and five others 14.3 , 14.7 , 14.12 , 14.13 , and 14.19 ) have good fits to red power - laws . source 14.4 does not fit a power - law well because of a very steep rise in the sed at 8.0 @xmath0 m . agns thus selected on the basis of the irac - mips seds show good agreement with agns discovered on the basis of their x - ray emission . four objects with secure mid - infrared power - law agn signatures are x - ray sources ( 14.3 , 14.7 , and 14.19 are nandra et al . 2005 objects c111 , c113 , and c128 respectively ; 14.13 is source c72 in the list of nandra et al . and source 23 in waskett et al . cudss 14.12 and 14.4 ( the tentative agn candidate ) , show no x - ray emission . our classification of 14.13 as agn - dominated agrees with higdon et al . ( 2004 ) and le floch et al . ( 2006 , in preparation ) , although chapman et al . ( 2005 ) call it a starburst . given the small numbers involved , our finding that between five and six members of the sample are agn - dominated is not inconsistent with the assertion by alexander et al . ( 2004 ) that at most 20% " of the luminosity of submillimeter galaxies arises from agn . they found that bright ( @xmath68 mjy ) scuba sources host agns at twice this rate ( 38% ) , although star formation tends to dominate the energy output . in summary , the number of scuba counterparts apparently dominated by an agn in the mid - infrared could be as few as five ( cudss 14.3 , 14.7 , 14.12 , 14.13 , and 14.19 ) or as many as six ( if 14.4 is included ) . there could be others not detectable using our approach : @xcite found that not all ( x ray selected ) agn show clear mid - infrared signatures . figure [ fig : fluxhist ] shows the 850 @xmath0 m flux density distribution of the 23 scuba galaxies . our 17 counterparts span the full range occupied by the sample , including all of the brightest scuba sources and all but one of the faintest . there is no evidence that our recovery rate is biased toward either the high- or low - brightness scuba sources , based on a kolmogorov - smirnov ( ks ) test comparing a sample consisting of the retrieved sources with another sample consisting of all 23 scuba galaxies . furthermore , agn - dominated objects occupy the middle of the distribution , without any obvious skew toward either the high- or low - brightness extreme of the distribution . the infrared color distributions of the irac counterparts to the scuba galaxies are shown in figure [ fig : histocolor ] , and the means and dispersions are given in table [ tab : colors ] . colors at longer wavelengths present greater diversity ( larger dispersions ) both for the scuba counterparts and for the full sample of all 8@xmath0m - detected galaxies in the field , but the color distributions are consistently tighter for the counterparts . in particular , the counterparts do not overlap the blue wings of the distributions of the field galaxies . the scuba counterparts are slightly redder in the irac bands relative to the full sample of all irac galaxies in the cudss 14@xmath1 field . although the difference in any one color is small , it amounts to nearly 1 magnitude in @xmath38-[8.0]$ ] . this is unlikely to be a selection effect . the seven counterparts ( 14.1 , 14.3 , 14.13 , 14.17 , 14.18 , 14.19 , and 14.20 ) identified without recourse to any color criteria show no significant color differences from the larger sample . given the dispersions , the differences in mean colors between scuba galaxies and field galaxies are only marginally significant . the ks test , however , is sensitive to the lack of blue galaxies in the counterpart sample . the ks - derived probabilities of drawing samples of scuba counterparts that are as different as observed from the sample consisting of all 8 @xmath0m - detected galaxies are @xmath69 , @xmath70 , and @xmath71 for the @xmath38-[4.5]$ ] , @xmath72-[5.8]$ ] , and @xmath45-[8.0]$ ] color distributions , respectively , if the underlying populations are the same . in other words , the color distributions are significantly different . figure [ fig : mipscolor ] shows the [ 8.0 ] - [ 24.0 ] and _ k _ - [ 3.6 ] color distributions for the irac - detected galaxies in this field . both distributions exhibit much larger dispersions than the irac - only colors plotted in figure [ fig : histocolor ] . despite the fact that in some cases the irac counterpart was selected on the basis of the presence and/or strength of mips 24 @xmath0 m emission , there are no significant differences in the mean [ 8.0 ] - [ 24.0 ] colors or dispersions of the full sample and the counterparts . the _ k _ - [ 3.6 ] color distributions are likewise indistinguishable ( given the wide range of colors exhibited by the sample ) . when we compare the color distributions using ks tests we find no evidence that the samples are different with respect to either _ k _ - [ 3.6 ] or [ 8.0 ] - [ 24.0 ] color . deep irac mosaics identify infrared counterparts for 17 of 20 scuba galaxies in the cudss 14@xmath1 field . the recovery rate of scuba galaxies at 8.0 @xmath0 m is comparable to or better than that found in other counterpart surveys , once the likelihood of mistaken identifications is estimated ( three or fewer for the current work ) . for more than half of the sample , the most - likely 8.0 @xmath0 m counterpart is different than the _ k_-selected counterpart . multiband irac - mips seds allow one to identify and avoid unlikely candidates in the scuba beam ( e.g. , elliptical galaxies ) . this suggests that the mid - infrared regime in general ( and the irac 8.0 @xmath0 m band in particular ) offers a more reliable means than ground - based optical or near - infrared surveys for identifying the true counterparts to the high - redshift , dusty starforming galaxies detected in submillimeter surveys . on the basis of power - law fits to the infrared irac - mips seds we infer that five or possibly six of the counterparts are agn - dominated . the 8.0 @xmath0 m counterparts have redder infrared colors than the general population of irac - detected galaxies in the field . five counterparts have irac - mips seds that are fit well by red power laws and are likely to be agn - dominated . one more is fit well by a slightly blue power - law and may also harbor an active nucleus . four of these six objects are detected in x - ray observations , lending further support to the agn hypothesis . the remaining counterparts are likely to be dominated by starburst emission . this work is based on observations made with the _ spitzer space telescope _ , which is operated by the jet propulsion laboratory , california institute of technology under nasa contract 1407 . support for this work was provided by nasa through contract number 1256790 issued by jpl / caltech . we are grateful to the anonymous referee , whose comments and suggestions greatly improved the clarity and presentation of this work . iraf is distributed by the national optical astronomy observatories , which are operated by the association of universities for research in astronomy , inc . , under cooperative agreement with the national science foundation . facilities : . lcrrrrrccll @xmath7314.1 & & 40.7 & 44.8 & 32.5 & 41.1 & 191 & 14 17 40.22 & 52 29 06.5 & 1.07(0.28 ) & 0.11 + @xmath7314.2 & a & 38.5 & 45.0 & 41.5 & 27.0 & 150 & 14 17 51.33 & 52 30 24.9 & 6.55 & 0.18 + 14.2 & b & 2.0 & 3.3 & 5.5 & 8.6 & @xmath74 & 14 17 51.93 & 52 30 30.8 & 2.12 & 0.45 + @xmath7314.3 & & 35.0 & 41.0 & 48.6 & 48.4 & 408 & 14 18 00.58 & 52 28 21.3 & 0.87(2.32 ) & 0.078 + @xmath7314.4 & a & 2.7 & 3.6 & 5.2 & 27.2 & 239 & 14 17 43.43 & 52 28 05.8 & 8.71 & 0.18 + 14.4 & b & 52.8 & 49.2 & 40.7 & 40.9 & & 14 17 42.41 & 52 28 11.6 & 9.07 & 0.11 + 14.4 & c & 7.4 & 8.7 & 9.5 & 6.2 & @xmath74 & 14 17 43.80 & 52 28 16.8 & 4.71 & 0.52 + 14.6 & & 1.8 & 1.8 & @xmath75 & 5.5 & @xmath74 & 14 17 57.02 & 52 29 14.7 & 8.50 & 0.51 + 14.7 & a & 8.4 & 10.6 & 11.7 & 12.9 & @xmath74 & 14 18 00.95 & 52 29 50.0 & 1.68 & 0.35 + @xmath7314.7 & b & 5.3 & 10.6 & 20.2 & 51.3 & 150 & 14 18 01.15 & 52 29 42.0 & 7.02 & 0.071 + @xmath7314.8 & a & 21.4 & 24.5 & 24.7 & 15.3 & 275 & 14 18 02.01 & 52 30 16.3 & 6.43 & 0.31 + 14.8 & b & 5.4 & 5.6 & 4.9 & 2.9 & @xmath74 & 14 18 02.70 & 52 30 10.4 & 5.25 & 0.76 + @xmath7314.10 & a & 50.1 & 48.4 & 34.8 & 38.3 & 307 & 14 18 03.00 & 52 29 33.2 & 9.81 & 0.12 + 14.10 & b & 4.3 & 3.2 & @xmath75 & @xmath76 & @xmath74 & 14 18 03.98 & 52 29 40.6 & 2.20 & 0.86 + 14.11 & a & 153.6 & 99.5 & 66.7 & 36.0 & @xmath74 & 14 17 46.89 & 52 32 36.1 & 2.69 & 0.13 + 14.11 & b & 225.3 & 140.8 & 92.7 & 50.3 & @xmath74 & 14 17 47.32 & 52 32 42.1 & 4.60 & 0.075 + @xmath7314.12 & a & 3.7 & 5.7 & @xmath75 & 8.0 & 80 & 14 18 04.70 & 52 28 56.0 & 5.48 & 0.47 + 14.12 & b & 3.1 & 4.2 & @xmath75 & 18.3 & @xmath74 & 14 18 05.69 & 52 29 03.3 & 8.60 & 0.33 + @xmath7314.13 & & 580.1 & 981.7 & 1448.5 & 2225.5 & 5646 & 14 17 41.88 & 52 28 23.5 & 0.67(6.46 ) & 0.0013 + 14.14 & a & 9.9 & 8.3 & 3.4 & 6.2 & @xmath74 & 14 18 09.06 & 52 31 00.5 & 4.78 & 0.52 + @xmath7314.14 & b & 9.4 & 10.7 & 7.8 & 6.9 & 120 & 14 18 08.21 & 52 31 08.1 & 6.16 & 0.50 + 14.15 & a & 30.8 & 26.7 & 19.7 & 17.1 & 103 & 14 17 29.87 & 52 28 21.5 & 5.83 & 0.28 + @xmath7314.15 & b & 7.4 & 10.8 & 15.3 & 12.1 & 93 & 14 17 28.44 & 52 28 19.8 & 7.90 & 0.37 + @xmath7314.17 & & 47.1 & 47.0 & 34.6 & 77.0 & 266 & 14 17 24.44 & 52 30 46.1 & 9.48 & 0.049 + @xmath7314.18 & & 159.7 & 138.1 & 143.0 & 164.7 & 1078 & 14 17 42.11 & 52 30 25.7 & 0.61(1.51 ) & 0.029 + @xmath7314.19 & & 15.3 & 20.6 & 21.1 & 25.2 & 84.0 & 14 18 11.26 & 52 30 12.3 & 2.33(8.59 ) & 0.20 + @xmath7314.20 & & 18.5 & 21.2 & 18.6 & 13.9 & @xmath74 & 14 17 50.51 & 52 30 55.2 & 8.87 & 0.34 + 14.22 & a & 29.2 & 28.1 & 19.6 & 19.1 & @xmath74 & 14 17 55.08 & 52 32 08.4 & 7.91 & 0.25 + @xmath7314.22 & b & 72.1 & 66.7 & 49.1 & 50.3 & 500 & 14 17 56.80 & 52 31 57.8 & 9.94 & 0.075 + @xmath7314.23 & a & 31.7 & 43.4 & 47.1 & 34.0 & 529 & 14 17 46.21 & 52 33 22.2 & 1.97 & 0.14 + 14.23 & b & 21.4 & 17.5 & 20.1 & 20.1 & @xmath74 & 14 17 46.68 & 52 33 29.6 & 6.60 & 0.24 lcccccccccc 14.1 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.2a & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & [email protected] & + 14.3 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.4a & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & @xmath78 & @xmath79 & + 14.7b & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & & [email protected] & + 14.8a & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.10a & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.12a & @xmath80 & @xmath81 & [email protected] & [email protected] & [email protected] & @xmath78 & @xmath82 & + 14.13 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 1.150 + 14.14b & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.15b & @xmath80 & @xmath81 & @xmath83 & @xmath84 & [email protected] & & @xmath85 & + 14.17 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.18 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 0.661 + 14.19 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & + 14.20 & @xmath80 & @xmath81 & @xmath83 & 27.7@xmath86 & [email protected] & [email protected] & [email protected] & + 14.22b & @xmath80 & @xmath81 & @xmath83 & [email protected] & [email protected] & & [email protected] & + 14.23a & @xmath80 & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & lrrrrrrr 8 @xmath0 m galaxies & 726 & @xmath87 & @xmath88 & @xmath89 & @xmath90 & @xmath91 + 8 @xmath0 m counterparts & 17 & @xmath92 & @xmath93 & @xmath94 & @xmath95 & @xmath96 + counterparts subsample & 7 & @xmath97 & @xmath93 & @xmath94 & @xmath98 & @xmath99

we extend our earlier multidimensional , magnetohydrodynamic simulations of coronal rain occurring in magnetic arcades with higher resolution , grid - adaptive computations covering a much longer ( hour ) timespan . we quantify how in - situ forming blob - like condensations grow along and across field lines and show that rain showers can occur in limit cycles , here demonstrated for the first time in 2.5d setups . we discuss dynamical , multi - dimensional aspects of the rebound shocks generated by the siphon inflows and quantify the thermodynamics of a prominence - corona - transition - region like structure surrounding the blobs . we point out the correlation between condensation rates and the cross - sectional size of loop systems where catastrophic cooling takes place . we also study the variations of the typical number density , kinetic energy and temperature while blobs descend , impact and sink into the transition region . in addition , we explain the mechanisms leading to concurrent upflows while the blobs descend . as a result , there are plenty of shear flows generated with relative velocity difference around 80 km s in our simulations . these shear flows are siphon flows set up by multiple blob dynamics and they in turn affect the deformation of the falling blobs . in particular , we show how shear flows can break apart blobs into smaller fragments , within minutes .

1507.00882

observations show recurrent formation and reshuffling of cool and dense material in coronal loops . the small scale ( @xmath2 km ) coronal rain is observed as cold , dense elongated blob - like features condensing in a much hotter loop and descending along one of its legs . the rain is guided by the loop magnetic field @xcite , dropping from heights of tens of mm into the chromosphere @xcite . similar phenomena have been observed by analysing absorption profiles in euv spectral lines @xcite . seen to propagate from the top of the loop towards its footpoints @xcite , systematic intensity variations in euv spectral lines are confirmed as downflows of cool plasma , rather than representing slow magneto - acoustic waves @xcite . besides downflows towards footpoints , @xcite also found upflows towards the loop apex . @xcite observed and tracked coronal rain in 30 active region loops and found more than one hundred descending condensations within 71 minutes . tracing the cool material towards loop footpoints , @xcite observed propagating patterns suggesting a hot upflow following the downflows , supplying hot plasma into the loops . @xcite suggested that coronal rain consists of plenty of small blobs , with sizes around 300 km in width and 700 km in length on average and they also estimated the occurrence rate of coronal rain in active region loops to be once every two days . since the solar corona is swamped with magnetic loops , this suggests a scenario where coronal rain is rather common . considering the very small sizes involved , one of the most attractive features of coronal rain is that it can be used to probe the local magnetic field structure , or that it can expose valuable properties of the local thermodynamic conditions inside coronal loops @xcite . indeed , the magnetic field structure has a much longer lifetime than the timescale for condensations to form and fall @xcite . additionally , due to the low temperature ( of order @xmath3 k ) of these condensations , coronal rain is normally observed in cold chromospheric lines @xcite . coronal rain results from thermal instability , with its non - linear counterpart and evolution also known as thermal non - equilibrium or catastrophic cooling . the linear thermal instability takes places whenever radiative losses locally overcome the heating input and is governed by well - known stability criteria for uniform radiative plasma conditions @xcite . these can be met in the coronal temperature range , as one encounters locally negative slopes in the radiative loss function @xmath4(t ) as function of temperature . when thermal conduction is insufficient in transporting enough energy to cooling ( and condensing ) material , the temperature reduces over time . as a consequence of thermal instability , temperature and gas pressure drop dramatically in the perturbed region , resulting in matter sucked in from the surroundings to the perturbed region , forming an increasingly larger and cooler condensation . this runaway effect will continually increase the density and decrease the temperature of condensations until heating and cooling achieve a balance again at lower temperatures and higher densities.numerical simulations have contributed to our understanding of thermal instability over the last 40 years @xcite . early numerical work shows that in the million degrees solar corona , small temperature contrasts could be enhanced by line and recombination - driven radiative losses within several minutes @xcite . catastrophic cooling drives recombination of elements in the cool condensations , making them partially ionized and visible in cool chromospheric lines . @xcite concluded that the rate of condensation is determined by hydrodynamical processes mainly . an important progress in modeling was obtained in @xcite , by using a spatially dependent heating increase that is localized nearer to the chromospheric footpoints than to the loop midpoint . with this localized heating at the footpoints , @xcite pointed out that another key requirement to generate a stable , prominence - like condensation is a dipped geometry in the loop . with an adaptive grid code , @xcite showed , in a 1d model , that the complete growth of a condensation reached a quasi - steady state after @xmath5 5000 s. in a similar 1d setup , @xcite calculated the linear instability criterion from numerical results and proved that the onset of coronal condensation indeed satisfies the linear isochoric instability criterion @xcite . in the solar corona , the fairly high densities required for the instability onset are thought to be obtained by evaporating material with heating located near the footpoints of coronal loops in the chromosphere or by direct mass injection into the corona @xcite , resulting e.g. from nano - heating events . influenced by magnetohydrodynamic forces ( gravity , lorentz force and gas pressure gradients ) , condensations , once formed , either fall from the corona down to the chromosphere as coronal rain or they collect into larger structures and remain suspended in the corona over long time periods as prominences , supported by the magnetic field . many numerical works addressed formation and dynamics of coronal rain , but adopted simplifying one - dimensional ( 1d ) approximations reducing the problem to gas dynamic , thermodynamic evolutions along individual field lines @xcite . since coronal rain occurs in many active region loops , the heating input is generally thought to be concentrated at the loop footpoints @xcite , which evaporates chromospheric plasma into the loops and increases the density . with a persistent heating , the anisotropic thermal conduction and optically thin radiation lead these coronal hot loops to reach thermally unstable regimes with a higher density in a timescale of hours @xcite . then catastrophic cooling sets in locally , resulting in the fast formation of cool condensations , as demonstrated in 1d models @xcite . numerical simulations by @xcite emphasized that a loss of equilibrium at the loop apex and the process of catastrophic cooling is caused by constant heating concentrated at the footpoints of the loop rather than a drastic decrease of the total loop heating which was used in earlier models . @xcite compared observations from an eit shutterless campaign with simulations of coronal loops and confirmed that observed localized brightenings and fast flows are consistent with this model . an important conclusion from @xcite was that the structure and dynamics of the coronal rain blobs are more sensitive to the pressure variations arising from catastrophic cooling than to gravity itself . this is in agreement with @xcite , who suggested that the internal pressure evolution of the loops , rather than gravity , determines the condensation speeds . furthermore , @xcite indicated that if a loop is predominantly heated by alfvn waves , coronal rain is inhibited since they tend to heat the loop uniformly . hence , coronal rain may not only point to the spatial distribution of the heating in coronal loops but also to the agent of the heating itself . they thus propose coronal rain as a marker for coronal heating mechanisms . @xcite pointed out that steady heating is not necessary to sustain the condensation . once the condensation is formed , it keeps growing even after localized heating ceases . @xcite simulated a three - dimensional sheared double arcade with a large ensemble of 1d independent flux tubes and observed the formation of both prominence threads and coronal rain . recently , @xcite presented the first multidimensional , magnetohydrodynamic simulations which captured the initial formation and the long - term sustainment of the coronal rain phenomenon . there we found that coronal rain in arcades is always accompanied by fast counter - streaming siphon flows in neighbouring flux bundles and we statistically analysed 80 minutes of virtual coronal rain in terms of sizes , mass , and velocity patterns . our 2.5d simulations showed how blobs deform into v - shaped patterns , and had blobs that levitate , evaporate in - situ , or fall into the transition region at speeds below free - fall . iris data recently revealed also many coronal rain impact events , with up to supersonic speeds above sunspots @xcite . we therefore revisited our mhd setup from @xcite , at even further increased numerical resolution and for much longer time , going up to 6 hours in total . we now analyse blob formation and blob impact into the transition region in more detail , focusing on multi - dimensional aspects not probed by 1d setups . furthermore , the high - resolution coronal imager ( hi - c ) in july 2013 provided a much more detailed look at the fine structure and dynamics in the solar corona . with data from hi - c , @xcite reported that anti - parallel flows have been directly imaged along fundamental filament threads within the million degree corona . they measured relative flow velocities of similar magnitude as in our previous simulations , namely 70 - 80 km s@xmath1 . both observations and our simulations hence suggest that such counter - streaming flows are likely commonplace . we observed that siphon flows establish naturally in a raining arcade , with velocity differences on adjacent field lines up to 80 km s@xmath1 . we thus also extended our simulations to further argue how our setup in a low field ( order 12 g ) magnetic arcade relates to the observed clumps of falling coronal rain @xcite and to unresolved fine - scale structure in solar coronal loop - tops @xcite . the paper is then organized as follows : in 2 we describe the numerical setup , in 3.1 we describe the multidimensional aspects of the condensations , focusing on rebound shocks and their prominence - corona - transion region ( pctr ) structure , 3.2 discusses the condensation rates and the long term coronal rain limit cycle obtained , 3.3 quantifies blob impact on the transition region and concurrent upflows , in 3.4 we investigate the counter - streaming flows , and in 3.5 we describe the shear flow effects . conclusions are drawn in 4 . our numerical setup follows our previous 2.5d thermodynamic mhd simulation from @xcite , which includes gravity , field - aligned heat conduction and radiative cooling and parametrized heating terms , on a rectangular plane with horizontal extension -40 mm @xmath6 mm and vertical extension 0 @xmath7 mm . the governing equations are as follows : @xmath8 @xmath9 @xmath10 @xmath11 where @xmath12 , @xmath13 , and @xmath14 are respectively temperature , density , magnetic field , velocity , and unit tensor , with the ratio of specific heats @xmath15=5/3 , and a total energy density as @xmath16 ; @xmath17 is the total pressure , consisting of magnetic pressure and thermal pressure @xmath18 ; * * g**@xmath19*@xmath20 * is the gravitational acceleration with the solar surface gravitational acceleration @xmath21 and the solar radius @xmath22 ; @xmath23 and @xmath24 are respectively the heating and radiative loss rates ; @xmath25 is the thermal conductivity tensor . assuming a 10:1 abundance of hydrogen and helium of completely ionized plasma , we obtain @xmath26 , where @xmath27 is the proton mass and @xmath28 is the number density of hydrogen . we use the ideal gas law @xmath29 , where @xmath30 is the boltzmann constant . we also adopt @xmath31 as the radiative cooling term , where @xmath32 is the radiative loss function for optically thin emission , quantified by @xcite using a recommended set of quiet - region element abundances , as used in our previous work @xcite . in the calculations , @xcite used a complete and self - consistent atomic data set and an accurate atomic collisional rate over a wide temperature range . below 10,000 k , we set @xmath32 to vanish because the plasma there is optically thick and no longer fully ionised . we use the exact integration method as introduced by @xcite to evaluate the radiative loss term . the use of explicit , ( semi-)implicit , and exact integration methods in grid - adaptive simulations has been compared in @xcite . the term containing @xmath33 quantifies the anisotropic thermal conduction along the magnetic field lines , composed by the unit vector @xmath34 along the magnetic field and the spitzer conductivity @xmath35 as @xmath36 @xmath37 erg cm@xmath1 s@xmath1 k@xmath39 . we employ a linear force - free magnetic field for the initial magnetic configuration , which is characterised by a constant angle @xmath40 as follows : @xmath41 @xmath42 @xmath43 with @xmath44 , the arcade makes a @xmath45 angle with the neutral line ( @xmath46 ) . @xmath47 mm is the horizontal size of our domain from -40 mm to 40 mm , and when adopting @xmath48 g , our magnetic arcade has a total box averaged field strength of 2.9 g. for the initial thermal structure , we set a uniform temperature of 10000 k below a height of 2.7 mm and choose a temperature profile with height ensuring a constant vertical thermal conduction flux ( i.e. , @xmath49 = 2 @xmath50 10@xmath51 erg @xmath52 s@xmath1 ) above this height , as also exploited by other authors @xcite . the initial density is then derived by assuming hydrostatic equilibrium with the number density of 1.2 @xmath50 10@xmath53 @xmath54 at the bottom and the initial velocity field of all plasma is static . since the corona needs to achieve a self - consistent thermal structure , we employ a background heating rate decaying exponentially with height into the whole system all the time , @xmath55 where @xmath56 erg @xmath54 s@xmath1 and @xmath57 mm . this heating is meant to balance the radiative losses and heat conduction related losses of the corona in its steady state . the slight difference in heating scale height between 50 mm in equation ( [ heat0 ] ) above and 36 mm in equation ( 2 ) in @xcite improves numerical stability at the top boundary and prevents it from cooling down during the longer timescale run performed here . with the above initial setup the whole system now is out of thermal equilibrium . we integrate the governing equations in time with heating @xmath58 active until the system achieves a quasi - equilibrium state . after 72 minutes , the above configuration reaches a quasi - equilibrium state shown in fig . [ 3d ] , which represents a 3d impression of the numerical box quantifying the temperature and number density profile and selected magnetic field lines . the @xmath59 in fig . [ 3d ] means that after reaching the quasi - equilibrium state , we reset the time of the system back to zero for the next stage of simulation . as seen in this fig . [ 3d ] , the numerical relaxation phase leads to some thermodynamic structuring in the final arcade . some chromospheric plasma is quickly evaporated into coronal loops at the beginning of the relaxation , but this material gradually loses its kinetic energy . as a result , the final relaxed state of the system is identified as the time when the maximal residual velocity in the simulation is less than 5 km s@xmath1 . in this end state , [ 3d ] shows a relatively thin transition region located at heights between 3 mm and 5 mm , which connects the chromosphere to corona . this transition region is higher above the neutral line , due to less downward thermal flux there because of the strong horizontal magnetic field . the plasma beta is 0.06 at 20 mm height above the neutral line while the temperature and number density there are around 1.7 mk and @xmath60 . the total mass ( per unit length in the ignored dimension ) of hot plasma in the corona is around @xmath61 g cm@xmath1 . following this equilibrated system , we turn on a relatively strong heating @xmath62 . this extra heating is localized near footpoints in the chromosphere with formula as @xcite : @xmath63 @xmath64 @xmath65 where @xmath66 erg @xmath54 s@xmath1 , @xmath67 mm , @xmath68 mm , @xmath69 mm , @xmath70 mm@xmath71 and @xmath72 mm@xmath71 . this choice of strong base heating contrast ( @xmath73 ) between @xmath62 and @xmath74 can mimic flare related chromospheric evaporation . however , this @xmath62 heating decreases with height to very small values and reaches one - tenth of @xmath74 heating at 10 mm . as a result , the @xmath62 heating dominates heating in the chromosphere and the transition region , while the @xmath74 heating plays a more important role in the heating in the corona . the parameter @xmath67 mm represents the height of the transition region in the quasi - equilibrium system . @xmath75 is the magnetic potential depending on the location and decaying exponentially with height into the whole system . because the magnetic potential along a single magnetic field line is constant , we add extra heating at both feet of all magnetic field lines identified by @xmath75 in the range of @xmath76 . since catastrophic cooling is very sensitive to the heating decay scale and the length of magnetic field lines @xcite , the heating decay scale @xmath77 is set to larger values for longer field lines by the above formulae . we use the mpi - parallelized adaptive mesh refinement ( amr ) versatile advection code _ mpi - amrvac _ @xcite to run the simulation . an effective resolution of @xmath78 or an equivalent spatial resolution of 20 km in both directions is obtained through six amr levels . this represents an effective fourfold improvement in resolution with respect to our earlier model @xcite . our numerical strategy to advance the governing partial differential equations uses a three - step runge - kutta type scheme . for flux computations , a third - order - accurate limited reconstruction @xcite is introduced to calculate the variable evaluation from cell center to cell edge . we adopt a suitably mixed prescription between a diffusive total variation diminishing lax - friedrichs and contact - resolving harten - lax - van leer with contact restored ( hllc ) scheme @xcite . for boundary treatment , we employ 2 grid layers exterior to the domain as ghost cells to prescribe cell center values . considering the left and right physical boundary , density , energy , @xmath79 and @xmath80 momentum components , @xmath81 and @xmath82 are set symmetrically , while @xmath83 and @xmath84 are adopted antisymmetrically to ensure zero face values . in bottom boundary ghost cells , we use the primitive variables ( @xmath85 ) to set all velocity components antisymmetrically to enforce both no - flow - through ( vertical ) and no - slip ( horizontal ) , while the @xmath86 are fixed to the initial analytic expressions of equation ( [ bfield ] ) , and the stratification of density is kept at pre - determined values from the initial condition , as well as the pressure . we always resolve the bottom region up to @xmath87 mm at the maximal resolution . while for the top conditions , we set all velocity components as antisymmetric , and adopt a discrete pressure - density extrapolation from the top layer pressure with a maximal temperature @xmath88 k. for the magnetic field , we use a two - cell zero - gradient extrapolation to determine @xmath89 in the ghost cells and improve @xmath81 from a second order one - sided centered difference evaluation of @xmath90 . in our previous work @xcite , we already described the formation process for the first condensation and emphasized how it perturbed the overall force balance in a 2d fashion . in this work , we discuss more of the multi - dimensional details for the forming condensations , and compare the results of our 2.5d simulations with previous 1d simulation works @xcite and insights from observations @xcite , in particular paying attention to the cross - field effects . the forming process of the first condensation in our 2.5d simulation is shown by fig . [ first ] which presents the temporal evolution of number density ( left columns ) , temperature ( middle columns ) and gas pressure ( right columns ) at @xmath91 101.2 , 101.5 and 102.2 minutes . when we compare these results with the corresponding fig . 5 and fig . 7 of 1d hydrodynamic simulations in @xcite , we conclude that all three parameters behave similarly in the forming process , as the number density increases rapidly from @xmath92 @xmath54 to @xmath93 @xmath54 , while the temperature decreases down to 0.01 mk . along each arched field line , this is analogous to the sudden thermal instability onset in 1d runs . this similarity confirms the applicability of restricted 1d model efforts which assume a rigid 1d loop under the prevailing plasma @xmath94 conditions , which takes on a local value of around 0.06 . the middle panel in the right column of fig . [ first ] also shows a significantly increased gas pressure inside the condensation and a layer of low gas pressure surrounding it after its formation . in the bottom panels of fig . [ first ] , we notice that density , temperature and gas pressure all reveal a front propagating as expanding wings on both sides of the condensation . this phenomenon is because fast siphon inflows are driven into the forming condensation by a strong pressure gradient between the lower gas pressure around the condensation and relatively higher gas pressure away from the condensation , as seen in the middle panel in the right column of fig . [ first ] . these two siphon flows meet up with the blobs , and dynamically impact on the blob to generate two rebound shocks . hence , while thermal instability and runaway cooling triggers a growing condensation , one also forms two rebound shock fronts that propagate away from the blob . the slightly different formation time at different parts of the condensation on adjacent magnetic field lines @xcite , which are due to gradual variations in length and chromospheric footpoint conditions , is the reason that these two expansion shock fronts display a fan - shaped structure , forming earliest in the blob center and spreading away from the blob . this fan - shaped structure of the rebound shocks is also clearly observed in @xcite . however , not every condensation realizes this nearly left - right symmetric situation as seen near the loop apex for this first condensation from fig . [ first ] . due to slightly asymmetric conditions already prevailing after the numerical relaxation process and due to perturbations from existing condensations , most of the following condensations initiate in loop limbs ( also shown in the online movies of @xcite ) . the field - projected gravity force on the limbs leads to asymmetric plasma distributions , as seen e.g. in the number density map in panel a of fig . [ rs ] at @xmath95 minutes , the moment when local catastrophic cooling begins there ( about 10 minutes after the first condensation ) . the higher central gas pressure indicates the initial forming location of this condensation in panel b of fig . [ rs ] . due to its limb - loop location , the number density map points out that the right of the condensation holds a relatively denser ( @xmath96 @xmath54 ) and wider plasma distribution than the left part ( @xmath97 @xmath54 ) . still , strong pressure gradients drive siphon flows from both sides towards this condensation . after a short time at @xmath98 , the denser and heavier plasma at the right of this condensation realizes a ( left - directed ) siphon flow with a slower speed of 23 km / s , compared to the left siphon flow ( which is right - directed ) at a speed of 42 km / s , shown by the velocity magnitude map in panel c of fig . [ rs ] . as discussed above , the impact of siphon flows on the condensation naturally generates rebound shocks , whose speeds are determined by the original speeds of the siphon flows and the mass contrast between the condensation and the siphon flows . the slower and heavier siphon flow on the right of the blob here leads to a much slower rebound ( right - directed ) shock seen to separate at 7 km / s , while the left one ( left - directed ) travels at 21 km / s . these two rebound shocks are identifiable in the gas pressure map in panel e of fig . [ rs ] at @xmath99 minutes . the condensation itself has a velocity of 5 km / s , meaning that basically the right rebound shock barely can sweep up and heat little siphon flow plasma . because the central condensation keeps sucking in plasma from nearby and the rebound shock at the right of the blob is too slow to sweep and heat up plasma , the gas pressure there does not rise to a higher value and keeps a strong pressure gradient at the right of the blob , as shown in panel e of fig . about 3 minutes later at @xmath100 minutes , this persistent pressure gradient at the right of the blob accelerates the left - directed siphon flow to a higher speed of 52 km / s ( shown in panel i of fig . [ rs ] ) , therefore the corresponding right - directed rebound shock finally speeds up to 28 km / s and begins to sweep and shock - heat the plasma on its way . in short , initial asymmetric situations on the condensation can lead to a complicated thermal and dynamical evolution and result in a delay of rebound shocks spreading at one side of the condensation . additionally , we also find another special case , namely blob a in fig . [ os ] , which has only one rebound shock on its left side . [ os ] shows the gas pressure map ( a ) and ( b ) at @xmath91 134.8 and 137.6 minutes , with a dotted isocontour of the number density at @xmath101 @xmath54 overplotted . this density contour at @xmath101 @xmath54 is one of the criteria which identifies whether a cell contains cool plasma belonging to coronal rain , as used further on . panel a in fig . [ os ] indicates a similar situation for blob a as in the second row of fig . [ rs ] in which only the left rebound shock spreads out . in contrast to what happens in the third row of fig . [ rs ] , for blob a we do not find a right rebound shock in fig . [ os ] until the collision and merging of blob a with blob b. the reason is that when the thermal instability triggers the condensation labeled there as blob a , another existing condensation labeled as blob b in the same coronal loop already depleted the plasma between these two blobs . therefore the small pressure gradient in the emptied loop between the two blobs can not drive a fast siphon flow to create a strong rebound shock for blob a , even though the gas pressure on the right of blob a is low enough ( panel a in fig . [ os ] ) , afterwards , when blob a catches up and merges with blob b because of the strong pressure gradient outside these two blobs , the rebound shock at the right side of blob a is still not fast enough to show clear separation and propagation . we also observe the details of a gas pressure substructure within these shock - bounded regions around the condensation in the simulations . these reveal the establishment of a prominence - corona - transition - region ( pctr ) like structure around all blobs . the gas pressure substructure around the first condensation consists of three components shown in panel a of fig . [ pctr ] and panel i of fig . [ first ] , namely a high gas pressure outside of the condensation , a low gas pressure at the boundary of the condensation , and a higher gas pressure in the center of the condensation . actually not only this first condensation in fig . [ first ] has this kind of gas pressure substructure , but also all the blobs which establish a dynamic equilibrium around themselves have it , e.g. all the blobs in fig . [ rs ] and fig . [ os ] . to better quantify this , we identify a field line crossing the center of the blob shown in panel a of fig . [ pctr ] and plot gas pressure , temperature and radiative loss along this field line in panel c of fig . the temperature declines from a coronal temperature of 0.35 mk to a cool plasma temperature of 0.01 mk in 200 km and density increases from @xmath102 @xmath54 to @xmath103 @xmath54 , therefore basically this 200 km area could be considered as a pctr . within this area , we find that two highly radiative loss peaks exist , introduced by a temperature around 0.02 mk . this corresponds to the two dips of gas pressure at the boundary of the blob . these two strong radiation areas also indicate the location in which catastrophic cooling takes place that ensure that the condensation keeps growing . indeed , the two dips in gas pressure always relocate with the boundary of the blobs , coincident with the strong emissive loss . although the temperature of 0.01 mk inside the blob is lower than in the surrounding coronal plasma , a much higher density at the center of the condensation ( @xmath104 @xmath54 ) leads to a little higher gas pressure there . the high gas pressure outside of the condensation reflects the post shock conditions prevailing there after the rebound shocks run against the condensation inflows . note that our resolution is such that we have about 7 grid points along the field line through the pctr at each side of the blob in fig . [ pctr ] , clearly resolving the pctr around the blob in our simulation . the gas pressure difference between inside and outside the condensation is found to persist throughout the lifetime of the blobs and plays a role in the movement of the blobs . especially when the blobs fall along the field lines toward footpoints , the gas pressure and temperature ahead of the descending blob increase as shown in panel b of fig . [ pctr ] , due to the blob compressing the plasma ahead of it in the loop and the strong evaporation at the loop footpoints . we also identify a field line crossing the center of the blob shown in panel b of fig . [ pctr ] and plot gas pressure , temperature and radiation loss along this field line in panel d of fig . [ pctr ] , which shows also an obvious pctr . due to the gravity variation and the strong gas pressure gradient between the two sides of the blob , the lower part of this blob has a higher density distribution , which naturally leads to a higher radiative loss . this strong gas pressure gradient slows down the acceleration of the blob in its descent . this was also pointed out in @xcite , where we stated that sometimes , it can even lift lighter blobs to cross the loop apex . panel a of fig . [ tm ] shows the temporal evolution of total mass of cool ( solid ) and hot ( dashed ) plasma in the corona , and panel b of fig . [ tm ] represents the number of blobs for the entire time interval of our 2.5d simulation . the criteria to identify whether a cell contains cool plasma belonging to coronal rain are that ( i ) the number density is higher than @xmath105 @xmath54 , ( ii ) the temperature is lower than @xmath106 k , and ( iii ) the location is above the chromosphere - corona - transition - region . we dynamically locate the height of the transition region at each @xmath107-position as @xmath108 by searching the vertical position of the ( first ) maximum gradient value of temperature from the bottom boundary . each blob is defined as a collection of neighbouring cells which hold cool plasma . however , if the number of grid cells in one blob is smaller than 10 at our highest resolution , we remove this blob from the blob list to avoid counting spurious transient features that do not collect into a clearly resolvable blob , and also to mimic the observational resolution . as stated before , we adopt a 4 times higher resolution than in @xcite , but also extend the simulation to a two times longer time of around 370 minutes ( previously 190 mins ) . by running our 2.5d simulation for these much longer times , we find that the whole coronal rain process shows limit cycles , which has been discussed in earlier 1d simulations @xcite , as well as in observational work @xcite . this is the first time that we can report limit cycles of coronal rain in a multidimensional simulation , which confirms that constant heating conditions which provide enough energy , can form secondary ( or even more ) coronal rain cycles in a single arcade . from panel a and b of fig . [ tm ] , we find the time interval between the first and secondary cycle to be around 175 minutes , when measured between successive maxima in cool mass matter . panel a of fig . [ tm ] shows the temporal evolution of total mass of hot coronal plasma which is the mass in the corona , excluding the cool plasma identified by the above criteria . we find that at @xmath109 minutes , the total mass of cool plasma reaches its peak in the first cycle in panel a of fig . [ tm ] , while at @xmath110 minutes the catastrophic cooling process has cooled down most of the hot plasma in the corona shown in panel a of fig . from about @xmath111 minutes , blobs begin to fall into the transition region , then the evaporation of plasma in the chromosphere driven by the extra heating @xmath62 fills the evacuated loops left by blobs which already sank into the chromosphere . from this moment , until the onset of the secondary cycle of our coronal rain shower at @xmath112 minutes , it takes about 120 minutes , which is of similar duration as the time for the first cycle to reach its onset ( about 100 minutes ) . so although we infer from the total mass evolution of cool plasma in panel a that there is only about 50 minutes between the ending of the first and the beginning of the second cycle , actually the continued heating at the chromosphere spontaneously begins to fill the empty corona already 70 minutes before . we also see that the total mass of hot plasma before the onset of the secondary cycle is higher than in the first cycle ( panel a ) , which leads to a longer lasting secondary cycle with more mass in condensations . panel a of fig . [ tm ] indicates that at @xmath113 minutes ( before the first blob falls into the chromosphere ) , there is at least @xmath114 g cm@xmath1 cool plasma in the corona , which originally was hot plasma . meanwhile panel a of fig . [ tm ] also suggests that compared with the corona before the onset of catastrophic cooling at @xmath115 minutes , the decrement in the same time of total mass of hot plasma at @xmath109 minutes is only @xmath116 g cm@xmath1 . the difference between the increase in cool plasma and the decrease in hot , indicates that during these 30 minutes since onset at @xmath115 minutes , the evaporation in the chromosphere evaporates @xmath117 g cm@xmath1 into the corona , i.e. at an evaporation rate of 2.2 g cm@xmath1 s@xmath1 . we can similarly estimate an evaporation rate of 2.3 g cm@xmath1 s@xmath1 between the onset of the secondary cycle and the moment its first blob falls into the transition region . till the onset of the first cycle at @xmath115 minutes , the increment of total hot plasma from turning on the extra heating @xmath62 is about @xmath118 g cm@xmath1 in total , further confirming this evaporation rate of 2.2 g cm@xmath1 s@xmath1 . based on these estimates , we infer that anywhere in both simulated cycles , the constant extra heating @xmath62 leads to a nearly constant evaporation rate . we can thus extrapolate to even more cycles expected further on , and interpret these limit cycles as a chronological sequence of mass recycling between chromosphere and corona : heating in the chromosphere brings plasma to the corona by evaporation , where it ultimately triggers catastrophic cooling , the cooling process manages itself into a coronal rain where plasma drains back to the chromosphere , and persistent heating causes the chromospheric material to evaporate again towards the corona . although the duration and peak value of the total mass in both computed cycles are similar , their initial condensation rates ( in contrast to the previously discussed evaporation rate ) computed from the temporal variation of their total mass curve work out to be 6.7 and 4.5 g cm@xmath1 s@xmath1 , respectively and thus are different . it is known from linear thermal instability theory @xcite and 1d simulation results in @xcite , that this initial condensation rate in catastrophic cooling depends on parameters controlling the energy input from heating . one notices that the condensation rate ( the local derivative of the solid curve in panel a of fig . [ tm ] ) varies dramatically even within one cycle , despite a constant heating energy input in our multidimensional simulation . we now will interpret the reason for the changes seen in the condensation rate , by surveying especially the process of growth for the first condensation which forms under a relatively simple and almost symmetric condition . the solid line in panel a of fig . [ first_t ] shows the temporal evolution of the mass accumulation for this first condensation ( the one from fig . [ first ] ) from @xmath115 to 110 minutes . its near linear behavior quantifies that the condensation rate remains almost constant in this time interval at a value of about 2.3 g cm@xmath1 s@xmath1 . we deliberately do not discuss what happens to the first condensation after @xmath119 minutes , since afterwards it breaks into two smaller blobs . in the same figure panel a , the dashed line displays the growth of the total mass of cool plasma as seen on a single field line through the center of the first condensation , i.e. in a 1d fashion . to show this , we identify the group of grid points which are passed by the field line . the total mass of cool plasma determined on the single field line keeps growing in time , but its growth rate is much smaller than that for the whole 2d condensation . panel b of fig . [ first_t ] quantifies the temporal evolution of typical lengths for the first condensation , where we quantify both the length parallel to the magnetic field lines and the length perpendicular to the magnetic field lines . this indicates that blob growth in the perpendicular direction is much faster than in the parallel one , which can be seen visually as well in all columns in fig . [ first ] and fig . as discussed in @xcite , the low pressure region surrounding the first condensation onset leads to magnetic restoring forces on adjacent loops . these inturn influence in which location the catastrophic cooling will take place on the adjacent loops , which are all close to the thermal instability onset . the different growth rates found for blob sizes in these two directions then relate to the fairly fast ` growth ' along the perpendicular direction due to sympathetic runaway cooling onset , versus the slower growth seen in the parallel one , which is the only one found in 1d setups . the average density of each cell of the condensation is quantified in panel c of fig . [ first_t ] , and this density stays basically the same in the forming process , meaning that the total mass of the condensation is just proportional to the increasing number of neighbouring grid cells that contain cool plasma . while the number of cells in the condensation increases in both directions , the larger condensation rate of the whole blob in panel a of fig . [ first_t ] again directly reflects the faster growth in size in the perpendicular direction . we conclude that the growth of total mass of individual blobs in our simulation is mainly determined by the onset of catastrophic cooling in neighbouring loops rather than the growth along the loops in which catastrophic cooling gets triggered . we can indeed verify this 2d growth aspect by further showing a correlation between the total mass of cool plasma and two other measures , which holds up even for a longer time than the first 10 minutes , i.e. when several blobs have started to form . this is shown in panel d in fig . [ first_t ] where we plot the temporal evolution of total mass of cool plasma , the size of the onset transition region , and the total blob region width . the total blob region width indicates the total width of all magnetic loops where catastrophic cooling takes place on . the size of onset transition region means the corresponding width as found at the transition region height , of all the loops undergoing catastrophic cooling . because the magnetic arcade configuration adopted , these size measures for the affected loops give higher values for higher locations , i.e. the total blob region width always exceeds the ( field aligned remapped ) onset transition region size . the latter size of the onset transition region shows a nice correlation with the total mass of cool plasma evolution . in the simulation , we observe plenty of blobs hitting the transition region and disappearing into the lower chromosphere , as also known to occur in observations @xcite . @xcite observed high - speed downflows and concurrent upflows in coronal loops close to the footpoints and argue in favor of upflows in coronal loops at higher temperatures . @xcite confirmed that the high - speed downflows represent the cool plasma , which is corresponding to the falling blobs in our simulation ( see once more also the online movie of @xcite ) . meanwhile , our 2.5d simulation also shows the possibility of triggering concurrent upflows as observed by @xcite and @xcite . panel a in fig . [ hit1 ] shows the number density map at @xmath120 at a moment when falling blobs sink into the transition region and compress the plasma on its way ( at about @xmath121 mm ) . panel b in fig . [ hit1 ] shows the vertical velocity map at the same location and instant , which clearly displays the concurrent upflows rising at the tails of the declining blobs in the same field line bundles . hence , this answers the question in @xcite whether the upflows can flow along the same field lines as the downflows . these upflows in our simulation are actually rebound shocks from the impact of the blobs on the transition region ( tr ) . they arise immediately when the blobs impact on the tr , and spread from one footpoint to another footpoint in around 5 minutes with a velocity of around 50 km s@xmath1 . from panel a we can see the enhanced density left after passage of these rebound shocks . however , panel c in fig . [ hit1 ] which quantifies temperature indicates that the temperature in the loop already increases before the rebound shocks have reached far into the loop , since the parametrized background heating @xmath122 heats the low density loops left by falling blobs very efficiently . panel d shows also the temperature , but now on a larger domain and at a later time . it shows that afterwards the rebound shocks heat the low density loops to an even higher temperature of 2.0 mk . after the rebound shocks reach the other footpoint , the loops are at high temperature of about 2 mk but with a low number density of @xmath102 @xmath54 . we distinguish this from further upflows coming from evaporation due to the extra strong heating @xmath62 located in the chromosphere . this enhances the density to @xmath123 @xmath54 again and the temperature to 2.3 mk . however , these upflows from evaporation rise with a much slower velocity of around 15 km s@xmath1 . to quantify even further the detailed fate of a blob when it hits and descends into the tr , fig . [ hit ] shows the temporal evolution of the mass , density , velocity , kinetic energy , momentum , and temperature of the first coronal rain blob to hit the transition region from the corona and to sink down into the chromosphere . the vertical dashed line in each panel of fig . [ hit ] points at @xmath124 minutes when this blob hits the transition region . because the density and temperature of plasma in the transition region is comparable with those of the blobs , we can not use only the density and temperature as a criterium to distinguish blobs when they are near or partially below the transition region anymore . in order to identify plasma belonging to the blob as it hits and descends in the transition region after @xmath124 minutes , we change our criteria to require the local velocity to be larger than 3 km s@xmath1 and the location are below the transition region line @xmath108 after @xmath124 minutes . since the velocity of plasma in the transition region is almost zero , this velocity - based criterion captures the location of sinking blobs . in panel a of fig . [ hit ] , we find that the mass identified as blob material by the above criteria begins to increase at t@xmath5 132.7 minutes . this is because the mass detected not only includes the blob itself , but also counts mass compressed and accelerated by the blob impact . at @xmath125 minutes , the total mass affected reaches its peak at six times the original blob mass . after @xmath125 minutes , due to the combined influence of reflection - transmission processes at the transition region , and the higher gas pressure from the impact , the velocities in much of the blob impacted area decrease to values smaller than the criterion 3 km s@xmath1 . this is then seen as a mass decrease in our panel a. in panel b of fig . [ hit ] , the density versus time profile keeps rising while the blob hits the transition region . as we know , this blob impact compresses the transition region plasma swept up by the blob and transfers momentum from the sinking blob to the impacted plasma , and therefore in panel c of fig . [ hit ] we find that the average velocity of the region identified keeps decreasing during the whole process , as well as the kinetic energy shown in panel d. panel e of fig . [ hit ] shows the total momentum of the mass identified . due to the gravitational acceleration , the blob momentum keeps increasing until it reaches its maximum value at @xmath91 136 minutes , then it reduces quickly . this is a combination of the mass evolution in panel a and the velocity info from panel c. the momentum and velocity decreasing after the impact relate to momentum transfer to the surrounding transition region and upper chromosphere plasma , until the regions selected by the velocity - based criteria vanishes : the local conditions settle to static chromosphere conditions . panel f of fig . [ hit ] shows the average temperature evolution during the blob impact . the temperature increases before hitting the tr due to the compressional heating when the blob descends through the higher gas pressure region just above the transition region . after the impact , since also more cooler material gets identified as impacted matter , one settles back to upper chromospheric temperature values . the impact speed of blobs in fig . [ hit ] is around 30 km s@xmath1 , and the highest impact speed of all blobs in our simulation is around 60 km s@xmath1 and number densities range from 4 to 6@xmath126 @xmath54 . our maximum impact speed is much lower than the falling speeds reported in @xcite which went up to 200 km s@xmath1 . they report that these coronal rain events with high impact speeds are correlated with local brightenings which probably indicate an increase of density and temperature in the transition region . panel b of fig . [ hit ] and panel a of fig . [ hit1 ] confirm the expected increase of the number density of impacts in our simulation . we also find another interesting phenomenon in our numerical simulation , namely the self - consistent establishment of counter - streaming flows . such anti - parallel flows are very commonly found in solar observations , especially also in prominences @xcite . panels a , b and c in fig . [ sf ] respectively show the signed velocity magnitude map ( with the sign taken from the horizontal velocity component ) , the gas pressure map and the number density map at @xmath127 minutes . panel d shows the signed velocity magnitude map as in panel a , but at a later time , namely at @xmath128 minutes . these four panels in fig . [ sf ] display many cases of counter - streaming flows established on neighbouring field line bundles in our simulation and allow to explain the origins of counter - streaming flows . after thermal instability inducing a runaway catastrophic cooling and initial growth in an almost static state , the condensations lose their delicate force balance and begin to slide towards one footpoint along magnetic field lines . whether a particular condensation segment slides to the left or right is influenced by its initial location and local total force balance ( gravity , gas pressure gradient and magnetic field force ) . once in motion , they are accelerated by the field - projected gravitational force , meanwhile catastrophic cooling keeps taking place around the condensations . as discussed in section [ s - rebound ] , the initial catastrophic cooling process depletes the local plasma and sucks in fast inflows , then the spontaneous rebound shocks heat the plasma and increase the gas pressure . afterwards , no stronger inflows are driven again due to the increased gas pressure . however , there can be several blob pairs lying in the same or neighbouring field line bundles , as shown in panel c of fig . [ lp ] shows gas pressure maps with magnetic field lines at @xmath129 and 113.0 minutes . the black contour relates to the temperature distribution and is an isocontour at 0.1 mk . both the gas pressure and temperature in panel a in fig . [ lp ] indicate the clear pctr around the blob as previously discussed in fig . [ pctr ] . after 3 minutes , the panel b of fig . [ lp ] shows two white ( low ) pressure sections after the blob breaks into three segments . these low pressure sections slant through the field lines and they are the elongated pctr cross sections from the original parts of the whole blob in panel a of fig . [ lp ] . because the strong radiation in the pctr , the temperatures inside these elongated regions stay low during their deformation . as a result , we could consider these cross sections to undergo an isothermal expansion . because the condensed mass in these narrow regions grows much slower than their areal growth due to elongation , the densities inside these elongated cross sections decrease faster as well as the gas pressure . this leads to blob sequences with low pressure sections in between them . this is also seen in panel b of fig . [ sf ] where a sequence of blobs show up with white ( low ) pressure sections in between them . the depleted areas trigger siphon inflows to refill these regions . then this pair of siphoned fast inflows establish the counter - streaming flows between the pair of neighbouring blobs . panels a and b in fig . [ sf ] also show that falling condensations with larger velocities induce larger density depletions and lower gas pressure areas on their way down , which leads to faster inflows than those found for more static condensations . panel d in fig . [ sf ] indicates another different origin of counter - streaming flows at @xmath128 minutes . as we discuss in section [ s - impact ] , we observe that after blobs decline into the transition region , concurrent upflows rise up towards the loop apex . upflows labeled as a in panel d in fig . [ sf ] are the concurrent upflows shown in panel b of fig . [ hit1 ] , but about 3 minutes later ( concurrent with the later temperature panel d of fig . [ hit1 ] ) . upflows arising from blob impacts also have the chance to establish a counter - streaming flow if there is an opposite flow pattern in the neighbouring loops . the difference between these two different origins for counter - streaming flows is that the one based on depleted sections between a pair of blobs lying on neighbouring field line bundles can last through the whole falling process of blobs , or on time scales of about 10 minutes , while the other ones will vanish after the upflows refill the loops , typically in a shorter time scale of about 5 minutes . the sheared flows that are established by the detailed blob dynamics could also in return influence the further evolution of the condensations . an example is shown in panel a of fig . [ se ] , where we show a signed velocity map , with overlaid contours of the density distribution of condensations at levels of 7 , 25 and 50 @xmath130 @xmath54 at @xmath131 minutes . concentrating on the density feature labeled with a , after its initial formation , sheared flows already begin to take shape . about 10 minutes later , this segment a is seen as segment a1 and a2 in panel b of fig . [ se ] and the condensation has broken into two distinct segments with increasing separation between them . segment a2 is also going to break into two segments a little later . at the @xmath91 123.7 minutes in panel a of fig . [ se ] , this segment a feature is more like one whole elongated condensation . however , by @xmath132 in panel b in fig . [ se ] , several condensations behave totally separate to each other . another example is the one of segment b in panel a and panel b. this breaks up into segment b1 and segment b2 in panel c at @xmath91 136.2 minutes . then segment b1 further breaks into segment b1 and b3 in panel d at @xmath91 139.8 minutes . this gradual change from one elongated dense blob or strand breaking into several segments , surrounded by fast sheared flows , hints at the influence of kelvin - helmholtz instabilities ( khi ) . however , there is no clear vorticity pattern emerging in our simulation that would clearly identify khi development , which may not have enough time to develop . we speculate that other khi related substructure may well arise under different parameter settings ( field strength , heating scale height ) , but already establish that sheared flows contribute to the breaking up of elongated condensations into smaller fragments . we extended our earlier multidimensional , magnetohydrodynamic simulations of coronal rain occurring in magnetic arcades hosting chromospheric , transition region , and coronal plasma . the main new results can be summarized as follows . 1 . we find that after the initial formation stage of condensations , expansion rebound shock fronts introduced by fast siphon inflows display a fan - shaped structure , typically . the local conditions of where condensations form influences the detailed dynamics and expansion of these rebound shocks , and can lead to asymmetric expansion fronts or only one - sided expansion shock fronts . we discussed the process of establishing a structured prominence - coronal - transition - region ( pctr ) around coronal rain condensations . the strong radiation loss at the boundary of blobs results in local dips in the gas pressure structure at the blob boundary where the temperature sharply rises from 0.01 mk to a coronal temperature of 0.5 mk . by extending our 2.5d simulation to a longer time of 6 hours , we obtain a secondary cycle of coronal rain in the simulation . this secondary cycle confirms the deductions from previous 1d simulations and observations that by providing consistent and enough energy , coronal rain can form a secondary cycle or even more . 4 . we study the condensation rate in our 2.5d simulation and find the growth of cool mass in the corona to show a good correlation especially with the faster growth rate in the length of condensations in the direction perpendicular to the field lines . this indicates that the growth of cool mass is dominated by the onset of runaway cooling in neighbouring loops . this significantly exceeds the rates obtained in studies of this growth rate in 1d models , purely along field lines , as we also need to understand the expansion speed of onset of runaway cooling in neighbouring loops . by performing detailed quantitative analysis , we also find that no matter what happens in the corona , a constant heating in the chromosphere keeps on evaporating a certain amount of hot plasma into the corona , establishing a mass cycle . we look into the impact of declining blobs on the transition region , and find that their rebound shocks can spread as upflows from one footpoint to another footpoint . following the rebound shocks , evaporation driven upflows with a slower velocity refill the loops and heat them to 2.3 mk again . 6 . plenty of counter - streaming flows are found in our simulation , and we demonstrated several reasons for forming these flows . one is that the extremely low gas pressure area between two neighbouring coronal rain blobs drives strong siphon flows towards it . these shear flows accompany the blobs until they fall into the transition region . the counter - streaming flows also in return influence the deformation of the blobs , which can break into several segments , starting from an elongated one . the research has been sponsored by an odysseus grant of the fwo vlaanderen . the results were obtained in the ku leuven goa project goa/2015 - 014 and by the interuniversity attraction poles programme initiated by the belgian science policy office ( iap p7/08 charm ) . part of the simulations used the infrastructure of the vsc - flemish supercomputer center , funded by the hercules foundation and the flemish government - department ewi . we acknowledge fruitful discussions with p. antolin , and a very helpful referee report . alexander , c. e. , walsh , r. w. , rgnier , s. , et al . 2013 , , 775 , l32 antiochos , s. k. , & klimchuk , j. a. 1991 , , 378 , 372 antiochos , s. k. , macneice , p. j. , spicer , d. s. , & klimchuk , j. a. 1999 , , 512 , 985 antiochos , s. k. , macneice , p. j. , & spicer , d. s. 2000 , , 536 , 494 antolin , p. , shibata , k. , & vissers , g. 2010 , , 716 , 154 antolin , p. , & verwichte , e. 2011 , , 736 , 121 antolin , p. , & rouppe van der voort , l. 2012 , , 745 , 152 beckers , j. m. 1962 , australian journal of physics , 15 , 327 , m. & torrilhon , m. 2009 , journal of computational physics , 228 , 4118 chae , j. , wang , h. , qiu , j. , et al . 2001 , , 560 , 476 colgan , j. , abdallah , j. , jr . , sherrill , m. e. , et al . 2008 , , 689 , 585 dahlburg , r. b. , antiochos , s. k. , & klimchuk , j. a. 1998 , , 495 , 485 de groof , a. , bastiaensen , c. , mller , d. a. n. , berghmans , d. , & poedts , s. 2005 , , 443 , 319 de groof , a. , berghmans , d. , van driel - gesztelyi , l. , & poedts , s. 2004 , , 415 , 1141 fang , x. , xia , c. , & keppens , r. 2013 , , 771 , l29 farrugia , c. j. , gratton , f. t. , bender , l. , et al . 1998 , , 103 , 6703 field , g. b. 1965 , , 142 , 531 fontenla , j. m. , avrett , e. h. , & loeser , r. 1991 , , 377 , 712 goldsmith , d. w. 1971 , , 19 , 86 hildner , e. 1974 , , 35 , 123 kamio , s. , peter , h. , curdt , w. , & solanki , s. k. 2011 , , 532 , a96 karpen , j. t. , antiochos , s. k. , hohensee , m. , klimchuk , j. a. , & macneice , p. j. 2001 , , 553 , l85 karpen , j. t. , antiochos , s. k. , & klimchuk , j. a. 2006 , , 637 , 531 karpen , j. t. , tanner , s. e. m. , antiochos , s. k. , & devore , c. r. 2005 , , 635 , 1319 karpen , j. t. , antiochos , s. k. , klimchuk , j. a. , & macneice , p. j. 2003 , , 593 , 1187 karpen , j. t. , & antiochos , s. k. 2008 , , 676 , 658 kawaguchi , i. 1970 , , 22 , 405 keppens , r. , meliani , z. , van marle , a. j. , delmont , p. , vlasis , a. , & van der holst , b. 2012 , jcp , 231 , 718 keppens , r. , & xia , c. 2014 , , 789 , 22 , r. & porth , o. 2014 , journal of computational and applied mathematics , 266 , 87 kleint , l. , antolin , p. , tian , h. , et al . 2014 , , 789 , l42 levine , r. h. , & withbroe , g. l. 1977 , , 51 , 83 leroy , j .- 1972 , , 25 , 413 luna , m. , karpen , j. t. , & devore , c. r. 2012 , , 746 , 30 , z. , keppens , r. , & giacomazzo , b. 2008 , , 491 , 321 mendoza - briceo , c. a. , sigalotti , l. d. g. , & erdlyi , r. 2005 , , 624 , 1080 mok , y. , drake , j. f. , schnack , d. d. , & van hoven , g. 1990 , , 359 , 228 mok , y. , miki , z. , lionello , r. , & linker , j. a. 2005 , , 621 , 1098 mok , y. , miki , z. , lionello , r. , & linker , j. a. 2008 , , 679 , l161 mller , d. a. n. , de groof , a. , hansteen , v. h. , & peter , h. 2005 , , 436 , 1067 mller , d. a. n. , hansteen , v. h. , & peter , h. 2003 , , 411 , 605 mller , d. a. n. , peter , h. , & hansteen , v. h. 2004 , , 424 , 289 oshea , e. , banerjee , d. , & doyle , j. g. 2007 , , 475 , l25 parker , e. n. 1953 , , 117 , 431 porth , o. , xia , c. , hendrix , t. , moschou , s. p. , & keppens , r. 2014 , , 214 , 4 schrijver , c. j. 2001 , , 198 , 325 scullion , e. , rouppe van der voort , l. , wedemeyer , s. , & antolin , p. 2014 , arxiv:1409.1920 townsend , r. h. d. 2009 , , 181 , 391 tripathi , d. , mason , h. e. , dwivedi , b. n. , del zanna , g. , & young , p. r. 2009 , , 694 , 1256 van marle , a. j. , meliani , z. , keppens , r. , & decin , l. 2011 , , 734 , l26 xia , c. , chen , p. f. , keppens , r. , & van marle , a. j. 2011 , , 737 , 27 xia , c. , chen , p. f. , & keppens , r. 2012 , , 748 , l26 xia , c. , keppens , r. , & guo , y. 2014 , , 780 , 130 wang , y .- m . 1999 , , 520 , l71 minutes after relaxation , we show a 3d view on the quasi - equilibrium stage of our simulation , which serves as initial condition when extra localized heating is turned on . ( the @xmath59 minutes means the resetting of time to zero from now on . ) the randomly selected field lines are colored by temperature , the back cross - section shows the temperature while the front @xmath133 cross - section shows the number density map . ] and 102.2 minutes ( top to bottom rows ) , we show the number density ( left column ) , temperature ( middle column ) and gas pressure ( right column ) maps in a zoomed ( about @xmath134 ) area . this shows the formation process of the first condensation . ] and 117.6 minutes ( top to bottom rows ) , we show evolutions of two - sided rebound shocks in number density ( left column ) , gas pressure ( middle column ) , and plasma velocity magnitude ( right column ) maps . the panel c shows the plasma velocity magnitude map at @xmath135 minutes . this blob shows clear left - right asymmetric behaviour in its rebound shock pair pattern . ] 134.8 and 137.6 minutes , we show the gas pressure map ( a ) and ( b ) at times indicated , with a dotted isocontour of the number density at @xmath101 @xmath54 . the thin grey lines are magnetic field lines . there are two blobs a and b in the same loop , with consequences for the way siphon flows can induce or prevent rebound shock patterns.,scaledwidth=70.0% ] ( left ) and @xmath136 minutes ( right ) , we show in the top row panels ( a ) and ( b ) the gas pressure maps . panels ( c ) and ( d ) plot gas pressure , temperature , and radiation loss along the selected field line crossing the blob center . , scaledwidth=90.0% ] min ; ( b ) the horizontal velocity component map ; ( c ) the temperature map . these three panels show the same local area with chromosphere and transition region variations , while a larger area view is shown in panel ( d ) , giving a later temperature map at @xmath128 min.,width=566 ] minutes . panel ( b ) shows the gas pressure map and panel ( c ) shows the number density map at the same time . in panel ( d ) , the signed velocity magnitude map is shown later at @xmath128 min , where the label a points to the upflows resulting from the rebound event shown in detail in fig . [ hit1].,width=566 ] ( a ) , 132.6 ( b ) , 136.2 ( c ) and 139.8 ( d ) minutes . the black contours relate to the number density distribution with levels at 7 , 25 and 50 @xmath130 @xmath54 . this clearly shows how shear flow effects induce blob fragmentation and evolution.,width=566 ]

this paper experimentally and theoretically investigates the fundamental bounds on radio localization precision of far - field received signal strength ( rss ) measurements . rss measurements are proportional to power - flow measurements time - averaged over periods long compared to the coherence time of the radiation . our experiments are performed in a novel localization setup using 2.4ghz quasi - monochromatic radiation , which corresponds to a mean wavelength of 12.5 cm . we experimentally and theoretically show that rss measurements are cross - correlated over a minimum distance that approaches the diffraction limit , which equals half the mean wavelength of the radiation . our experiments show that measuring rss beyond a sampling density of one sample per half the mean wavelength does not increase localization precision , as the root - mean - squared - error ( rmse ) converges asymptotically to roughly half the mean wavelength . this adds to the evidence that the diffraction limit determines ( 1 ) the lower bound on localization precision and ( 2 ) the sampling density that provides optimal localization precision . we experimentally validate the theoretical relations between fisher information , cramr - rao lower bound ( crlb ) and uncertainty , where uncertainty is lower bounded by diffraction as derived from coherence and speckle theory . when we reconcile fisher information with diffraction , the crlb matches the experimental results with an accuracy of 97 - 98% . radio localization , cramr - rao bounds , fisher information , bienaym s theorem , sampling theorem , speckles , uncertainty principle .

1507.07385

localization involves the process of obtaining physical locations using radio signals . radio signals are exchanged between radios with known and unknown positions . radios at known positions are called reference radios . radios at unknown positions are called blind radios . localization of blind radios reduces to fitting these measured radio signals to appropriate propagation models . propagation models express the distance between two radios as a function of the measured radio signals . these measured radio signals are often modeled as deterministic radio signals with noise using a large variety of empirical statistical models @xcite . radio localization usually involves non - linear numerical optimization techniques that fit parameters in the propagation model given the joint probability distribution of the ensemble of measured radio signals . localization precision is usually expressed as the root mean squared error ( rmse ) . in the field of wireless sensor networks , the estimation bounds on localization precision are often calculated by the cramr - rao lower bound ( crlb ) from empirical signal models with independent noise @xcite . in general , localization precision depends on whether the measured radio signals contain phase information . phase information can only be retrieved from measurements that are instantaneous on a time scale that is short relative to the oscillation period of the signals . the smallest measurable position difference depends on how well phase can be resolved . this is usually limited by the speed and noise of the electronics of the system . less complex and less expensive localization systems are based on measurements of time - averaged power flows or received signal strengths ( rss ) . time - averaging is usually performed over timespans that are large compared to the coherence time of the radios , so that phase information is lost . rss localization is an example of such less - complex systems . hence , when determining the bounds on localization precision , it makes sense to distinguish between the time scales of the signal measurements , i.e. between instantaneous and time - averaged signal and noise processing . speckle theory @xcite describes the phenomenon of power - flow fluctuations due to random roughness from emitting surfaces . this theory shows that independent sources generate power - flow fluctuations in the far field that are always cross - correlated over a so - called spatial coherence region . the linear dimension of this region equals the correlation length . this correlation length has a lower bound in far - field radiation . in all practical cases of interest , the lower bound on the correlation length of this far - field coherence region is of the order of half the mean wavelength of the radiation . this lower bound is called the diffraction limit or rayleigh criterion @xcite . it is not obvious that this lower bound on correlation length holds in our wireless sensor network setup with its characteristic small cylindrical antennas . therefore , we derive this lower bound on the correlation length with the corresponding cross - correlation function from the maxwell equations using the ieee formalism described by @xcite . our derivation of this lower bound on the correlation length and corresponding cross - correlation function leads to the well - known van cittert - zernike theorem , known in other areas of signal processing like radar @xcite , sonar @xcite and optics @xcite . our novel experimental setup validates that power - flow fluctuations are cross - correlated over a correlation length of half a wavelength by increasing the density of power - flow measurements from one to 25 power - flow measurements per wavelength . in the field of wireless sensor networks , the estimation bounds on localization precision are usually determined by empirical signal models with independent noise over [ space , time ] @xcite . hence , the correlation length of the radiation is assumed to be infinitely small . this representation renders practical relevance as long as the inverse of the sampling rate is large compared to the correlation length , so that correlations between measurements are negligibly small . this paper experimentally and theoretically investigates radio localization in the far field when sufficient measurements are available to reveal the bounds on localization precision . we use bienayme s theorem @xcite to show that the ensemble of correlated power - flow measurements has an upper bound on the finite number of independent measurements over a finite measuring range . we show that this finite number of independent measurements depends on the correlation length . when we account for the finite number of independent measurements in the fisher information , the crlb on localization precision deviates 2 - 3% from our experimental results . we show that this approach provides practically identical results as the crlb for signals with correlated gaussian noise @xcite , given the lower bound on correlation length . hence , the lower bound on correlation length determines the upper bound on localization precision . there are a few papers that assume spatially cross - correlated gaussian noise on power - flow measurements @xcite . these papers consider cross - correlations caused by shadowing that extend over many wavelengths . @xcite states that their cross - correlation functions do not satisfy the propagation equations . such cross - correlations have no relation to wavelength of the carrier waves as the diffraction limit is not embedded in the propagation model like we derive . correlation , coherence and speckle properties of far - field radiation are governed by the spreads of fourier pairs of wave variables that show up as variables in the classical propagation equations of electromagnetic waves . the products of these spreads of fourier pairs express uncertainty relations in quantum mechanics , which are formulated as bandwidth relations in classical mechanics @xcite and signal processing @xcite . @xcite mathematically establishes a relationship between fisher information , the crlb and uncertainty . as uncertainty is lower bounded by diffraction , fisher information is upper bounded and crlb is lower bounded . our experimental work reveals quantitative evidence for the validation of this theoretical work . this paper is organized as follows . section [ sec : experimental_setup ] describes the experimental setup . section [ sec : theory ] derives the propagation and noise models from first principles for our experimental setup for each individual member of the ensemble of measurements . section [ sec : model ] describes the signal model , maximum likelihood estimator ( mle ) and crlb analysis for the ensemble of measurements using the results of section [ sec : theory ] . section [ sec : experimental ] presents the experimental results in terms of spatial correlations and upper bounds on localization precision . finally , section [ sec : discussion ] provides a discussion and section [ sec : conclusions ] summarizes the conclusions . [ fig : measurement_setup ] and [ fig : photo ] show the two - dimensional experimental setup . [ fig : measurement_setup ] shows a square of @xmath0m@xmath1 , which represents the localization surface . we distinguish between two types of radios in our measurement setup : one reference radio and one blind radio . the reference radio is successively placed at known positions and is used for estimating the position of the blind radio . the crosses represent the @xmath2 manual positions of the reference radio ( @xmath3 ) and are uniformly distributed along the circumference of the square ( one position every half centimeter ) . rather than placing a two - dimensional array of reference radios inside the @xmath0m@xmath1 square , it may suffice to place a significantly smaller number on the circumference of the rectangle and get similar localization performance . measuring field amplitudes on a circumference rather than measuring across a surface is theoretically expressed by green s theorem as is shown in section [ sec : theory ] . whether sampling power flows on circumferences instead of sampling across two - dimensional surfaces suffices has yet to be verified by experiment . this paper aims to show the practical feasibility of this novel technique , which was first proposed by @xcite . m@xmath1 ) to measure rss to the blind radio . the blind radio is located at the origin . for illustrative purposes , this figure only shows @xmath4 of the @xmath2 measurement positions.,scaledwidth=48.0% ] [ t ] cm apart . the inset on the right shows a close - up picture of the radio.,scaledwidth=48.0% ] the red circle represents the position of the blind radio . we place the blind radio at an unsymmetrical position , namely @xmath5 . we only use one blind radio and one reference radio to minimize influence of hardware differences between radios . the blind and reference radios are both main powered to minimize voltage fluctuations . the blind radio has a power amplifier and broadcasts messages with maximum power allowed by etsi to maximize snr . both blind and reference radios have an external dipole antenna . the antennas have the same vertical orientation and are in line - of - sight ( los ) for best reception . this implies that the polarization is vertically oriented perpendicular to the two - dimensional localization plane . the length of the antenna is half a wavelength . its diameter is roughly one twentieth of a wavelength . we keep the relative direction of the printed circuit boards on which the antennas are mounted constant by realigning them every @xmath6 cm . in order to minimize interference from ground reflections , we place the radios directly on the ground , so that their antennas are within one wavelength height . the ground floor consists of a reinforced concrete floor covered by industrial vinyl . we minimize interference from ceiling reflections by placing a @xmath7cm@xmath1 aluminum plate one centimeter above the blind radio antenna . the ground and the aluminum plate minimize the influence from signals in the z - direction , so that we only have to consider signals in two dimensions . all reference radio positions are in the far field of the blind radio . a photo of our setup is shown in fig . [ fig : photo ] . at each of the @xmath2 reference radio positions , the reference and blind radio perform a measurement round . a measurement round consists of @xmath9 repetitive multiplexed rss measurements to investigate and quantify the measurement noise and apply crlb analysis to this measurement noise . each measurement round consists of @xmath10 measurement sets that consist of @xmath11 rss measurements on an unmodulated carrier transmitted by the blind radio . between each measurement set of @xmath11 rss measurements , the reference and blind radios automatically turn on and off ( recalibrate radios ) . although we did not expect to find any difference from these two different forms of multiplexing rss signals , our experiments should verify this , which they did . hence , a measurement round consists of @xmath10 measurement sets , each set consisting of @xmath11 repetitive multiplexed rss measurements . measurement rounds and measurement sets are represented by @xmath12 . index @xmath13 identifies the position of the reference radio and thus the measurement round , index @xmath14 identifies the @xmath10 measurement sets , and index @xmath15 identifies the @xmath11 individual rss measurements in each measurement set . @xmath16 represents the measured power in dbm and is time - averaged over @xmath17 according to the radio chip specification @xcite . the coherence time is @xmath6 , which implies a coherence length of @xmath18 km . the averaging time of @xmath17 is a factor of five larger than the specified coherence time of the carrier wave of a typical 802.15.4 radio @xcite . practically , this means that these power measurements lose all phase information @xcite . in theory , this means that we measure the time - averaged power flow or poynting vector as the cross - sections of the antennas are the same . the blind radio transmits in ieee 802.15.4 channel @xmath19 in order to minimize interference with wi - fi channels @xmath20 and @xmath21 . the reference radio performs power - flow measurements in the same channel and sends the raw data to a laptop over usb , which logs the data . we use matlab to analyze the logged data . between each measurement round , we change the position of the reference radio by @xmath22 cm and push a button to start a new measurement round . note that @xmath22 cm is well within the @xmath23 diffraction limit of half the mean wavelength . in summary , each measurement set takes one second ; each measurement round takes roughly one minute ( @xmath24 seconds ) . the experiment consists of @xmath2 measurement rounds and takes in total @xmath25 hours ( @xmath2 minutes @xmath26 @xmath25 hours ) . in practice , we spend almost three weeks in throughput time to perform this complete data collection . this section formulates the propagation and noise model of radio localization systems operating in the far field . this model holds for each individual member of the ensemble of @xmath2 power - flow measurements over space described in section [ sec : experimental_setup ] . our propagation model is based upon em inverse source theory , where we follow the ieee formalism given by @xcite . this formalism is needed to derive the lower bound on correlation length of wireless networks using power - flow measurements on a time scale long compared to the coherence length of the hardware . the measurement configuration of practical interest is composed of the wave - carrying medium , i.e. the atmosphere , a number of source domains and one receiver domain . our measuring chamber has a variety of contrasting obstacles like a ceiling , walls , a ground floor , pillars , and a small aluminum table hiding the ceiling from the transmitter . most of these secondary sources are so far away from the receiver that they are neglected in our theoretical representation . we account for the following domains : * the wave - carrying medium denoted by @xmath27 , with electric permittivity @xmath28 , magnetic permeability @xmath29 and em wave speed @xmath30 . it is locally reacting , spatially invariant , time invariant and lossless . to locate positions in space , the observer employs the ordered sequence of cartesian coordinates @xmath31 , or @xmath32 , with respect to a given origin @xmath33 , while distances are measured through the euclidean norm @xmath34 . the fourth dimension is time and is denoted by @xmath35 . * a transmitter with bounded support @xmath36 with @xmath37 . here , @xmath38 denotes the spatial support carrying electric currents with volume density @xmath39 , and @xmath40 denotes the spatial support carrying magnetic currents with volume density @xmath41^{-}$].the transmitter transmits unmodulated em currents at a temporal frequency of @xmath42ghz with a temporal frequency bandwidth of @xmath43khz @xcite . the polarization of the em currents is assumed to be perpendicular to the ground plane of localization space , which is assumed to be the @xmath44 plane . the electromagnetic volume current densities are usually represented by equivalent surface current densities , which physically relate to the bounded charges and currents in terms of polarization and magnetization @xcite . * volume scatterer(s ) or volume noise with bounded spatial support @xmath45 with @xmath46 . here , @xmath47 denotes the spatial support carrying electric currents with volume density @xmath48 , and @xmath49 denotes the spatial support carrying magnetic currents with volume density @xmath50^{-}$ ] . we assume this noise to be negligible in our setup as we identify it with the thermal noise in all contrasting media in the measuring chamber . * surface scatterer(s ) or surface noise with bounded spatial support @xmath51 with @xmath52 . here , @xmath53 denotes the surface boundary carrying electric currents with surface density @xmath54 , and @xmath55 denotes the surface boundary carrying magnetic currents with surface density @xmath56^{-}$ ] . the surface noise is assumed to be caused by surface roughness of the ground floor as described by @xcite . * a receiver with bounded spatial support @xmath57 in which the electric field strength @xmath58 and the magnetic field strength @xmath59^{-}$ ] are accessible to measurement through the received time - averaged power flow @xmath60^{- } e_{q } dt \text{.}\ ] ] the receiver measures the power flow of an unmodulated carrier with an observation time of @xmath61 , long compared to the coherence time of @xmath6 @xcite . the mapping sources @xmath62 field is unique if the physical condition of causality is invoked . for the received signal at the point , @xmath63 , of observation , the relation between the em vector and scalar potentials , @xmath64^{- } \right\}$ ] , of the received power flow and the real em currents on the surfaces of the bounded sources is given by @xcite @xmath65^{- } \right\}(\boldsymbol{x } , t ) = g(\boldsymbol{x},t ) \stackrel{(\boldsymbol{x } ) } { * } \stackrel{(t ) } { * } \left\ { j_{k } , \left [ k_{i , j } \right]^{- } \right\}(\boldsymbol{x } , t ) \text{,}\ ] ] where the spatial and temporal convolutions extend over the surfaces of the spatial - temporal supports of the pertaining bounded primary and induced sources located at @xmath66 in the measurement configuration . according to section [ sec : mc ] , the spatial supports are the surfaces of the transmitter and of the ground floor . in two dimensions , the surfaces become circumferences of their cylindrical cross sections . equation links the extended sources to the well - known retarded potentials . in our measurement configuration , where the polarization of transmitted and received signals are directed perpendicular to the @xmath44 plane , the array of green s functions that couples the sources to their respective potentials takes on the scalar form @xmath67 in , @xmath68 is the ( 3 + 1)-space - time dirac distribution operative at @xmath69 and @xmath70 @xmath71 d \frac{\omega_{0}}{2 \pi } \text{.}\ ] ] the bandwidth , @xmath72 , of temporal frequency spectrum of the hardware is usually mathematically represented by its inverse , the coherence time , @xmath73 , as described by @xcite @xmath74 where we assume a narrow gaussian line shape relative to the carrier frequency as is the case for quasi - monochromatic radiation @xmath75 time and position are connected by the constant speed of light , @xmath76 , as are angular temporal frequency and wavelength @xmath77 @xmath78 to simplify the notation , we represent this time dependence of the source currents in the dirac distribution by limiting the integration interval in to the frequency bandwidth , @xmath72 , @xmath79 \\ \text{sinc } \left ( \delta \omega_{0 } \left(t - \frac{\left| \boldsymbol{x } \right|}{c_{0}}\right ) \right ) \frac{\delta \omega_{0}}{2 \pi}\end{gathered}\ ] ] for @xmath80 , the slowly varying envelope of the sinc - function approaches unity , and we end up with a plane - wave representation leading to a spherical wave in . the sinc - function can be replaced by a normalized gaussian or lorentzian as is applied in @xcite . for our work , the only time dependence that is important is that our observation time is long compared to the coherence time so that all phase information is lost . the convolution integral over the extended spatial supports in assumes its far - field approximation when the point of observation , @xmath63 , is far away from any part of the extended spatial supports , the coordinates of which are denoted by @xmath66 , so that @xmath81 where @xmath82 denotes the unit vector in the direction of observation , @xmath63 . in the far - field approximation , the potentials in are replaced by the far - field electromagnetic field strengths @xcite with the given polarization along the z - axis @xmath83 where the tilde , @xmath84 , denotes the sum of a macroscopic average of all primary and secondary fields including a small noise term due to the surface roughness of the ground floor @xmath85 in , the far - field green s function is given by substituting in and in . the transverse far - field magnetic field strength , @xmath86^{-1}$ ] , is directed perpendicular to the z - axis and @xmath82 and equals the far - field electric field strength multiplied by @xmath87 . in the far - field approximation , the time - averaged power flow of becomes inversely proportional to @xmath88 with the help of , so that without any noise , reduces to @xmath89 in , @xmath90 denotes the absolute value of the time - averaged poynting vector given by , and @xmath91 denotes the reference power flow at reference far - field distance of , say , @xmath92 m . equation can be read as the non - logarithmic gain model . in the inverse - source problem formulation pertaining to our measurement configuration , information is extracted from an ensemble of measured , time - averaged power - flow signals . this information reveals the nature and location of the scattering volume and surface sources in . the mapping of the @xmath93 is known to be non - unique . a detailed analysis in this respect can be found in @xcite . in radio localization , an algorithm is used that is expected to lead to results with a reasonable degree of confidence . such an algorithm is based on the iterative minimization of the norm of the mismatch in the response between an assumed propagation model for each individual measurement and an ensemble of observed power - flow signals . the influence of noise can be accounted for by using an input power - flow signal perturbed by an additive noise signal as is usually done in the log - normal shadowing model ( lnsm ) . for independent , uncorrelated noise , reduces to the non - logarithmic gain model @xcite @xmath94^{1/2}/10 \ll 1$}.\end{gathered}\ ] ] in , @xmath95 denotes the gaussian variable of the independent zero - mean noise with standard deviation @xmath96^{1/2}$ ] . under the condition that the perturbation is small , we obtain @xmath97^{1/2}/10 \ll 1$}.\ ] ] the linearization of is established by recursive expansion . with this linearization of the perturbed , time - averaged power flow with added noise , the nature of the correlations can be derived from , , and and then be combined with and . we apply the rationale used by van cittert and zernike . according to this rationale , each surface element on the contrasting surfaces acts as a spatially incoherent point source represented by the green s function of . the contrasting surfaces observed by the receiver in the far - field in our localization setup are the surfaces of the primary and secondary sources . the primary source is the blind transmitter , and the main secondary source in our measuring chamber is the ground floor . the wave vector of the primary source is fixed in size , i.e. @xmath98 , as the primary source is not extended and can be considered as a point source . the wave vectors @xmath98 of the plane wave amplitudes originating from secondary stochastic point sources induced under grazing incidence all over the ground floor are the same in size . all these scattered plane waves that are directed towards the vertical antenna of the receiver are again directed parallel to the ground floor . as the vertical receiving antenna has a circular cross - section , the projections of all these wave vectors on the planes tangential to this cross section give rise to a spatial - frequency bandwidth of @xmath99 . the expectation value , @xmath100 , of the superposition of these plane waves in the far field at a slightly displaced position , @xmath101 , all of the same amplitude but with random phases follows from , and @xmath102 = \mathbb{e } \left [ g^{\infty}(\boldsymbol{x } + \delta \boldsymbol{x } ) \stackrel{(\boldsymbol{x } + \delta \boldsymbol{x } ) } { * } \delta j_{z}^{s}(\boldsymbol{x } ) \right ] \nonumber \\ & = \mathbb{e } \left [ \exp\left [ i k_{0 } \widehat{\boldsymbol{x } } \cdot \delta \boldsymbol{x } \right ] \delta e_{z}^{\infty}(\boldsymbol{x } ) \right ] \nonumber \\ & = \mathbb{e } \left [ \exp\left [ i k_{0 } \widehat{\boldsymbol{x } } \cdot \delta \boldsymbol{x } \right ] \right ] \mathbb{e } \left [ \delta e_{z}^{\infty}(\boldsymbol{x})\right ] \text { , with } \frac{|\delta \boldsymbol{x}|}{|\boldsymbol{x}| } \ll 1 \text{,}\end{aligned}\ ] ] where we have taken out the time dependence to simplify the notation . in , the third equal sign results from the spatial invariance of the propagating medium and from the fact that the variance in vertical surface roughness is statistically independent of horizontal correlations in the far - field . the ensemble average of all individual measurements over space of @xmath103 $ ] can be replaced by an average over the fourier conjugated space because of the assumed wide - sense stationary ( wss ) random noise process @xcite . in one dimension , the expectation value of @xmath103 $ ] equals the sinc - function @xmath104 \right ] = \frac{1}{2 k_{0 } } \int_{-k_{0}}^{k_{0 } } \exp\left [ i k \widehat{\boldsymbol{x } } \cdot \delta \boldsymbol{x } \right ] dk \\ = \text{sinc}(k_{0 } \delta x ) \text{,}\end{gathered}\ ] ] and in two dimensions the jinc - function @xmath105 , with @xmath106 . the first - order spatial correlation function of the perturbed electric field strengths follows from by multiplying its deterministic form for @xmath107 by @xmath108 and by neglecting the second - order terms of the perturbed field strength . we then take the expectation value @xmath109 } { \mathbb{e } \left [ \left| e_{z}^{\infty}(\boldsymbol{x } ) \right|^{2 } \right ] } \cong\\ 1 + \frac{\mathbb{e } \left [ \delta \left| e_{z}^{\infty}(\boldsymbol{x } ) \right|^{2 } \right]}{\mathbb{e } \left [ \left| e_{z}^{\infty}(\boldsymbol{x } ) \right|^{2 } \right ] } \text{sinc}(k_{0 } \delta x ) \text{,}\ ] ] where the superscript star , @xmath110 , denotes complex conjugate and the ratio in the right member denotes the normalized variance of the perturbed field . the sinc- and jinc - functions have their first zeros at @xmath111 and @xmath112 , which are called the diffraction limit or rayleigh criterion in one and two dimensions . they can be considered as the far - field correlation functions for each individual measurement with a spatial correlation length that equals the diffraction limit . equation is equivalent with the van cittert - zernike theorem @xcite . this theorem links far - field spatial correlations of noise to diffraction . the diffraction limit acts as a fundamental lower bound on the spatial correlation or coherence region for time scales @xmath113 , and hence , as an upper bound on the localization precision as we will see in section [ sec : model ] . equations - generally allow for extracting the phases from the em field measurements as long as the timescales are short relative to an oscillation period @xmath114 . under those time scales , the van cittert - zernike theorem is equivalent with the wiener - khintchine theorem . using and , we extend the perturbed input signals of and from independent to second - order spatially correlated noise to @xmath115 } { \mathbb{e } \left [ p^{2}(\boldsymbol{x } ) \right ] } \cong 1+\frac{\mathbb{e}\left [ \delta \left| e_{z}^{\infty}(\boldsymbol{x } ) \right|^{4 } \right ] } { \mathbb{e}\left [ \left| e_{z}^{\infty}(\boldsymbol{x } ) \right|^{4 } \right ] } \text{sinc}^{2}(k_{0 } \delta x ) \nonumber \\ \cong 10^{\frac{\mathbb{e}\left[x^2\right ] \text{sinc}^{2}(k_{0 } \delta x)}{10 } } \text { , for } \mathbb{e}\left[x^2\right]^{0.5}/10 \ll 1 \text { , } t_{m } \gg \tau_{c } \text{.}\end{aligned}\ ] ] the normalized scattering cross - sections in and resulting from surface roughness equal the measurement variance @xmath96 $ ] at each individual reference - radio position . as we shall see in the next subsection , this measurement variance has a fundamental lower bound within the time scale of our measurements ( @xmath113 ) . although the correlation function derived only holds for correlations in the close neighborhood of each point of observation , can be extended to larger distances when the path losses are accounted for as is usually done in lnsm . the correlation function adopted in the literature is an exponentially decaying function @xcite . reference @xcite notes that the exponentially decaying correlation function does not satisfy the propagation equations of em theory . our correlation functions hold for each individual measurement position of the reference radio . the extension to an ensemble of measurement positions with a derivation of the measurement variance of this ensemble is given in section [ sec : model ] . section [ sec : model ] investigates two signal models for this extension and compares the computed measurement variances for the two correlation functions of power flows as a function of the sampling rate . equation contain two products of the fourier pairs [ angular frequency @xmath116 , time @xmath35 ] and [ spatial frequency @xmath118 , position @xmath119 . these two sets of conjugate wave variables naturally lead to an uncertainty in physics called the heisenberg uncertainty principle . hence , uncertainty is a basic property of fourier transform pairs as described by @xcite @xmath120 and @xmath121 equations and are the well - known uncertainty relations for classical wave variables of the fourier pairs [ energy , time ] and [ momentum , position ] . in these fourier pairs , energy and momentum are proportional to angular frequency and wave number , the constant of proportionality being the reduced planck constant . although these uncertainty relations originally stem from quantum mechanics , they fully hold in classical mechanics as bandwidth relations . in information theory , they are usually given in the time domain and are called bandwidth measurement relations as described by @xcite . in fourier optics , they are usually given in the space domain as described by @xcite . heuristically , one would expect the lower bounds on uncertainty in time , @xmath122 , and position , @xmath123 , to correspond to half an oscillation cycle , as half a cycle is the lower bound on the period of energy exchange between free - space radiation and a receiving or transmitting antenna giving a stable time - averaged poynting vector . when one defines localization performance or precision as the inverse of the rmse , one would expect the localization precision of the ensemble not to become infinitely large by continuously adding measurements . when the density of measurements crosses the spatial domain of spatial coherence as computed from , the power - flow measurements become mutually dependent . the link between uncertainties and fisher information was first derived by @xcite . as we show in section [ sec : experimental ] , our experiments reveal convincing evidence for this link . the first five subsections describe the signal model , mle and crlb usually applied in the field of rss - based radio localization for independent noise @xcite and cross - correlated noise @xcite . section [ sec : bienayme ] reconciles diffraction and the nyquist sampling theorem with the crlb using cross - correlated noise and bienaym s theorem . in addition , it introduces the notion of the maximum number of independent rss measurements and relates this to fisher information . in section [ sec : sim ] , simulations verify that the estimator is unbiased and efficient in the setup as described in section [ sec : experimental_setup ] . we extend the notations introduced in section [ sec : experimental_setup ] and [ sec : theory ] that describe our measurement configuration with the bold - faced signal and estimator vectors usually employed in signal processing . the experimental setup consists of a reference radio that measures power at @xmath2 positions @xmath124 of an unmodulated carrier transmitted by the blind radio . at each position , the reference radio performs @xmath125 repetitive multiplexed power measurements @xmath12 . these measurements are used to estimate the blind radio position , which is located at the origin @xmath126 . we adopt the empirical lnsm for modeling the power over distance decay of our rss measurements . as the cross - sections of the blind transmitter and reference receiver are given and equal , the power as well as the power - flow measurements are assumed to satisfy the empirical lnsm @xcite @xmath127 where @xmath128 in and , @xmath129 identifies the power - flow measurement . @xmath130 denotes the ensemble mean of power - flow measurements at far - field distance @xmath131 in dbm . @xmath132 represents the power flow at reference distance @xmath133 in dbm . @xmath134 represents the path - loss exponent . @xmath135 represents the noise of the model in db due to fading effects . @xmath135 follows a zero - mean gaussian distribution with variance @xmath136 and is invariant with distance . equations and are equivalent with the @xmath137 of , where the path - loss exponent equals two . one usually assumes spatially independent gaussian noise @xcite . in @xcite , the cross - correlation between power - flow measurements is independent of wavelength and is modeled with an exponentially decaying function over space by @xmath138 \text{,}\ ] ] where @xmath139 denotes the correlation length . section [ sec : noise ] shows that the lower bound on correlation length depends on the wavelength and equals half the mean wavelength , @xmath140 . in section [ sec : noise ] , shows that the cross - correlation function of power - flow measurements takes on the form of the diffraction pattern @xcite @xmath141 in section [ sec : crlb ] , we show the implications of using the cross - correlation function of and the cross - correlation function of that satisfies the propagation model derived from first principles . the non - linear least squares problem assuming the lnsm with correlated noise is denoted by @xmath142 where @xmath143 denotes the vector of power - flow measurements @xmath144 \text{,}\ ] ] where @xmath145 denotes the vector of power - flow measurements calculated by the lnsm as expressed by @xmath146 \text{,}\ ] ] where @xmath147 denotes the vector of distances between the reference radio and the blind radio @xmath148 \text{.}\ ] ] here , @xmath130 denotes power flow at far - field distance @xmath149 as expressed by . in , @xmath150 denotes the set of unknown parameters , where @xmath151 $ ] . @xmath152 denotes the covariance matrix . usually power - flow measurements are assumed to be independent so that @xmath153 , where @xmath154 is the identity matrix @xcite . in case of correlated noise @xcite , the elements in the covariance matrix are defined by @xmath155 , where @xmath156 where @xmath157 denotes the correlation distance between reference radio positions @xmath129 and @xmath158 . equation denotes the correlation coefficients as expressed by and . usually , the three parameters @xmath159 , @xmath134 , @xmath160 $ ] of the lnsm are calibrated for a given localization setup @xcite . the blind radio position is assumed to be known when the lnsm is calibrated . the lnsm assumes that @xmath161 is equal for each rss measurement , so that @xmath132 and @xmath134 can be estimated independent of @xmath161 @xmath162 = \arg \min_{\boldsymbol{\theta } = [ p_{r_{0 } } , \eta ] } v(\boldsymbol{\theta } ) \text{.}\ ] ] in , @xmath163 is defined as the standard deviation between the measurements and the fitted lnsm using @xmath164 and @xmath165 . equation gives for all our power - flow measurements the best fit when the lnsm parameters are calibrated at @xmath166 we use these calibrated lnsm parameter values to calculate the bias and efficiency of our estimator from simulations . in addition , we use these calibrated lnsm parameter values to calculate the cross - correlations in the far - field and crlb . we use the mle as proposed by @xcite , with the physical condition of causality invoked on @xmath134 and with boundary conditions on @xmath167 , to estimate the blind radio position @xmath168 = \arg \min_{\boldsymbol{\theta } = [ x , y , p_{r_{0 } } , \eta ] } v(\boldsymbol{\theta } ) \nonumber \\ & \text{subject to } \nonumber \\ & \eta>0 \nonumber \\ & -1 \leq x \leq 2 \nonumber \\ & -2 \leq y \leq 1 \text{,}\end{aligned}\ ] ] our estimator processes the noise as independent , so that @xmath169 . section [ sec : sim ] shows that this estimator is unbiased and efficient in our simulation environments with uncorrelated and correlated noise . crlb analysis provides a lower bound on the spreads of unbiased estimators of unknown deterministic parameters . this lower bound implies that the covariance of any unbiased estimator , @xmath170 , is bounded from below by the inverse of the fisher information matrix ( fim ) @xmath171 as given by @xcite @xmath172 in , @xmath150 represents the set of unknown parameters , and @xmath173 represents the estimator of these parameters . in case of multivariate gaussian distributions , the elements of the fim are given by @xcite @xmath174_{a , b } = \left [ \frac{\partial \mathbf{\bar{p}}}{\partial \theta_{a } } \right]^{t } c^{-1 } \left [ \frac{\partial \mathbf{\bar{p}}}{\partial \theta_{b } } \right ] + \\ \frac{1}{2 } \left [ c^{-1 } \frac{\partial c}{\partial \theta_{a } } c^{-1 } \frac{\partial c}{\partial \theta_{b}}\right]\text{.}\end{gathered}\ ] ] in our case , the covariance matrix is not a function of unknown parameter set @xmath150 so that the second term equals zero . the elements of the @xmath175 fim associated with the estimator defined by , @xmath176 $ ] , are given in the independent measurement case by @xcite and in the correlated measurement case without nuisance parameters by @xcite . the elements of @xmath171 consists of all permutations in set @xmath150 . after some algebra , @xmath177 reduces to @xmath178 where @xmath179 we use to calculate the crlb of the mle expressed by using the calibrated lnsm parameter values given by : @xmath180 $ ] . the crlb on rmse for unbiased estimators is computed from @xmath181 here @xmath182 represents the trace of the matrix . we use the calibrated lnsm parameters as given by .,title="fig:",scaledwidth=50.0% ] fig . [ fig : crlb ] shows the lower bound on the rmse calculated by and as a function of the number of rss measurements . when we assume independent rss measurements , @xmath183 , the rmse decreases to zero with an ever increasing number of independent rss measurements . this is in accordance with the theoretical bound analysis presented by @xcite . hence , there is no bound on localization precision with increasing sampling rate . when we account for the spatial cross - correlations between rss measurements , @xmath184 , the bound on radio localization precision reveals itself when sufficient rss measurements are available . we evaluate the crlb with the cross - correlation function expressed by the diffraction pattern in and the exponentially decaying cross - correlation function as expressed by . they basically converge to the same bound on localization precision . apparently , the bound on localization precision depends on the bound on the correlation length rather than on the form of the cross - correlation functions . the determinant of the covariance matrix starts to decrease when correlations start to increase , until the fisher information and thus localization precision stabilizes on a certain value . when the cross - correlations are equal to the diffraction pattern as expressed by , the covariance matrix becomes singular when the sampling rate goes above a certain threshold , which equals the inverse of the diffraction limit or the nyquist sampling rate . hence , the multivariate gaussian distribution becomes degenerate and the crlb can not be computed without regularization . this section connects diffraction to the sampling theorem , covariance and fisher information using bienaym s theorem @xcite . bienaym s theorem offers a statistical technique to estimate measurement variance of correlated signals . it states that when an estimator is a linear combination of @xmath15 measurements @xmath185 @xmath186 the covariance of this linear combination of measurements equals @xmath187 and is equivalent to the measurement variance . in , @xmath188 is a weighting factor , @xmath189 represents the standard deviation of measurement @xmath129 , and @xmath190 represents the correlation coefficient between measurement @xmath135 and @xmath191 . when all measurements have equal variance @xmath192 and equal weights @xmath193 , reduces to @xmath194 in , @xmath195 is the spatially averaged correlation of @xmath196 measurement pairs and is given by @xmath197 and @xmath198 represents the number of measurements that effectively decreases estimator covariance and is given by @xmath199 @xcite shows in the time domain that the cross - correlation of bandwidth limited fluctuations on signals with a periodic wave character is proportional to the point - spread - function ( psf ) of a measurement point . in the space domain , the cross - correlation of far - field power - flow measurements takes on the form expressed by the diffraction pattern in . in addition , @xcite shows that the radius , @xmath200 , of the first zero @xmath201 equals the inverse of the minimum ( nyquist ) sampling rate . the sampling theorem determines the sampling rate that captures all information of a signal with finite bandwidth . for signals with a periodic wave character measured in the far field , noise is cross - correlated over the far - field spatial coherence region of the signal . this far - field coherence region assumes the far - field diffraction pattern and is equivalent to the psf of power - flow measurements @xcite . the lower bound on the far - field region of spatially correlated signals and noise is equal to the diffraction limit @xcite and to the inverse of the upper bound on spatial frequencies divided by @xmath202 , i.e. @xmath203 . the finite spatial - frequency bandwidth of the far field relates diffraction to the sampling theorem @xcite and whittaker - shannon interpolation formula or cardinal series @xcite . hence , this coherence region determines the maximum number of independent measurements over a finite measuring range . this maximum number of independent measurements is also known as the degrees of freedom @xcite , and is given by @xmath198 in . we decompose into two factors @xmath204 where @xmath205 represents the ratio of the total number of measurements and effective number of measurements , and @xmath206 represents the covariance of the estimator assuming that all measurements are independent . the global spatial average of correlation coefficients , @xmath195 , as given by can be computed using the correlation function for any localization setup by assuming an ever increasing sampling rate . for our localization setup of fig . [ fig : measurement_setup ] , we compute @xmath207 processing all @xmath2 rss measurements over space . substituting this value in gives @xmath208 . to account for the effects of spatial correlations on the covariance of non - linear unbiased estimators , we heuristically rewrite into the following postulate @xmath209 where @xmath171 is equal to the fisher information assuming independent measurements . our postulate rigorously holds when the variance and thus the fisher information of each wave variable at each measurement point is equal , assuming that the crlb holds as defined in . our postulate of is in agreement with the concept that the degrees of freedom of signals with a periodic wave character is independent of the estimator . postulate is also in line with that crlb efficiency is approximately maintained over nonlinear transformations if the data record is large enough @xcite . hence , is a good performance indicator when the correlation length is small relative to the linear dimensions of localization space . [ fig : crlb ] shows the lower bound on the rmse calculated by and using the two cross - correlation functions expressed by and . the solid curves represent the results of . our approach expressed in differs less than 1 mm from the crlb for signals with correlated gaussian noise @xcite over the entire range shown by fig . [ fig : crlb ] . the influence of correlated measurements on the crlb is apparent for @xmath210 as converges asymptotically to roughly half the mean wavelength for both cross - correlation functions . hence , the correlation length and thus the finite number of independent rss measurements determines the bound on localization precision . note that the effective number of measurements converges to practically identical values for both cross - correlation functions , @xmath208 . the sinc - squared cross - correlation function expressed by is bandwidth limited , so that the covariance matrix becomes degenerative at roughly the nyquist sampling rate . the exponentially decaying cross - correlation function expressed by is not bandwidth limited , so that the covariance matrix stays full rank . in this case , the fisher information decreases per individual rss measurement with increasing sampling density until the localization precision stabilizes . when @xmath211 , provides practically the same results as the crlb for independent measurements ( @xmath2122 mm difference in all cases ) . hence , one can assume independent measurements as long as the inverse of the sampling rate is large compared to the correlation length . the bound on estimator covariance becomes of interest when the density of measurements surpasses the bound imposed by the spatial coherence region of far - field radiation of the transmitters . when the number of measurements goes to infinity , reduces to @xmath213 which does not approach zero . hence , bienaym s theorem reveals the link of measurement variance to diffraction , the sampling theorem , and fisher information . we performed @xmath214 simulation runs using the measurement setup described in fig . [ fig : measurement_setup ] to quantify ( 1 ) the bias and ( 2 ) the efficiency of our estimator in . we performed two sets of simulations , one with independent noise and one with correlated noise expressed by . we did not perform simulations with cross - correlation function , because the covariance matrix is singular above the nyquist sampling rate and thus for our measurement setup . note that these simulations do not provide an experimental validation of the model used . an estimator is unbiased if @xcite @xmath215 we are interested in the bias of the estimated blind radio position @xmath216 , so we define the bias as @xmath217 simulations show that the bias is of the order of @xmath218 mm for independent and correlated noise . estimator efficiency is defined as the difference between the covariance of the estimator and crlb . we quantify this difference by @xmath219 simulations show that the estimator and crlb differs at most by @xmath220 mm for both independent and correlated noise . hence , our estimator in has a negligible bias and a high efficiency in our simulation environments with independent and correlated noise . this section presents the experimental results of the two - dimensional measurement setup described in section [ sec : experimental_setup ] . the first subsection presents the far - field spatial cross - correlation function of the rss signals and estimate the spread of these spatial correlations . we then determine the spatial fourier transform of these cross - correlations to look for an upper bound at a spatial frequency of @xmath221 , in line with the spatial resolution being bounded by the diffraction limit . finally , we determine the global rmse of and show its asymptotic behavior to the diffraction limit with the increase of the density of sampling points . we compare this experimentally determined rmse with the rmse determined by the crlb for independent and correlated noise . we show that the crlb for independent noise underestimates the rmse computed from our measurements . [ fig : coherence ] shows the spatial cross - covariance function between power flows as a function of distance in wavelengths . the spatial cross - covariance function is calculated using the deviations from the lnsm expressed by and calibrated propagation parameters expressed by . hence , spatial cross - covariance and thus cross - correlations are distance independent by cross correlating the deviations from the calibrated lnsm . fig . [ fig : coherence ] shows that the spatial cross - correlations go to a minimum over a distance of roughly half the mean wavelength , which corresponds to the diffraction limit . the small difference between the black and red curve indicates that noise resulting from repetitive multiplexed measurements over time is negligible compared to the cross - correlated noise in the far field . [ t ] million rss measurements . the red curve shows the measured cross - correlations using @xmath20 rss measurement per reference radio position.,title="fig:",scaledwidth=48.0% ] [ t ] . theoretically , it is represented by tri(k ) , as it is the fourier transform of . the vertical line represents the spatial frequency that corresponds to the diffraction limit , which forms an upper bound in the spectrum.,title="fig:",scaledwidth=48.0% ] in the case of fisher information and thus crlb analysis , information is additive when measurements are independent @xcite . one usually assumes signal models with independent noise@xcite . hence , localization precision increases with an ever increasing amount of independent measurements over a finite measuring range . however , when measurements are correlated , information and localization precision gain decrease with increasing correlations . [ fig : coherence ] shows that when space - measurement intervals become as small as the diffraction limit , measurements become spatially correlated and mutually dependent . our measurements show that rss signal measurements are spatially correlated over a single - sided region of roughly half the mean wavelength . it corresponds to the diffraction pattern expressed by derived in section [ sec : noise ] . therefore , increasing reference radio density beyond one per half a wavelength has a negligible influence on fisher information gain and thus on localization precision . our experimental results in the next subsection confirm this . in case of independent measurements , fig . [ fig : coherence ] would show an infinitely sharp pulse ( dirac delta function ) . [ fig : fft ] shows the measured power spectrum of cross - correlated noise in rss signals , @xmath222 , i.e. the spatial fourier transform of the cross - covariance of fig . [ fig : coherence ] . theoretically , it is represented by the spatial fourier transform of . this figure shows that the energy is mainly located in lower spatial frequencies and it diminishes over a single - sided interval of @xmath223 . this upper bound corresponds to the diffraction limit . the nyquist sampling rate provides an estimate of the minimum sampling rate to fully reconstruct the power - flow signal over space without loss of information , which equals the single - sided bandwidth of our power spectrum . the vertical black line represents the spatial frequency associated with the nyquist sampling rate , which equals @xmath224 . our experimental results in the next subsection confirm this by showing that the localization precision does not increase by sampling beyond this sampling rate . in case of independent measurements , fig . [ fig : fft ] would show a uniform distribution . [ fig : localization ] shows the experimental results of our measurement setup described in section [ sec : experimental_setup ] . localization precision is given as the inverse of rmse , which is computed from . [ t ] independent rss measurement per reference radio position ( green curve).,title="fig:",scaledwidth=48.0% ] the red curve in fig . [ fig : localization ] shows the rmse as a function of the number of measurements per wavelength . the rmse decreases with increasing number of rss measurements over space until sufficient measurements are available . at a critical measurement density , the bound on localization precision becomes of interest . the rmse ( red and black curves ) converges asymptotically to roughly half the mean wavelength as one would expect from the diffraction limit . the rmse represented by the red curve is based on processing all @xmath225 rss signal measurements per reference radio position . the rmse represented by the black curve is based on processing one rss signal measurement per reference radio position . the negligible difference between the red and black curves shows that the number of repeated rss signal measurements per reference radio has a negligible influence on the rmse . the crlb for independent measurements starts deviating from the rmse ( red and black curves ) when the sampling rate is increased beyond one rss signal per half the mean wavelength ( see black dotted curve ) , as one would expect from the diffraction limit and the nyquist sampling rate over space @xcite . spatial correlations between rss signals increase rapidly with increasing rss measurement density beyond one sample per half the mean wavelength ( fig . [ fig : coherence ] ) . correlated rss signals can not be considered as independent . @xcite has shown that fisher information is upper bounded by uncertainty in line with . as @xmath226 is upper bounded in spatial frequencies by @xmath227 , the spread or uncertainty in position , @xmath228 , is lower bounded . hence its inverse is upper bounded , as is fisher information . coherence and speckle theory have shown that uncertainty is lower bounded by the diffraction limit . hence , the crlb at a sampling density of one sample per half the mean wavelength ( vertical dotted black curve ) should equal the measured bound on rmse . we define the measured bound on rmse as the rmse processing all 60 million measurements . our measurements show that the difference between this crlb and the measured bound on rmse is @xmath229-@xmath230% ( 1 mm ) . hence , our experiments validate the theoretical concepts introduced by @xcite . our experiments reveal evidence that the crlb can not be further decreased by increasing the number of measurements . on the other hand , at @xmath6 rss signal measurements per wavelength , the rmse is a factor of four higher than the one calculated by the crlb for independent noise . this difference can not be explained by the difference between the covariance of the estimator and the crlb ( see section [ sec : model ] ) . the difference can , however , be explained by calculating the degrees of freedom of our localization setup using bienaym s theorem and the lower bound on correlation length . when one substitutes this lower bound in the and substitutes in , one obtains the asymptotic behavior of the crlb with correlated noise as is shown in fig . [ fig : crlb ] . the measured asymptote in fig . [ fig : localization ] deviates @xmath229-@xmath230% from bienaym s theorem and from the crlb for correlated noise . all @xmath4 million power measurements and matlab files are arranged in a database at linkping university @xcite . our novel localization setup of using reference radios on the circumference of a two - dimensional localization area instead of setting them up in a two - dimensional array worked well . this implies that one does not have to know the phases and amplitudes on the closed surfaces around extended sources to reconstruct the positions of these extended sources . our experiments show that it suffices to measure time - averaged power flows when a localization precision of about half a wavelength is required ( in our case @xmath21 cm ) . time averaging on a time scale that is long compared to the temporal coherence time is usually employed in rss localization , so that phase information is lost . we expect such a setup to work well when all radio positions are in los . for a practical implementation in nlos environments , we refer to @xcite . the propagation equations of section [ sec : theory ] are than applied to design an efficient setup and algorithm to locate the blind radios . in section [ sec : theory ] , we distinguished between primary and secondary cross - correlated noise in the far field . in our experimental setup , where signal levels are large compared to noise levels , it is reasonable to neglect thermal and quantum noise . scattering from the surface roughness of the primary source can not provide the the spatial frequency bandwidth of fig . [ fig : fft ] because of the small size of the transmitting antenna . we expect that the cross - correlated noise originates from the surface roughness of the large area between the transmitter and receiver . hence , our @xmath2 time - averaged rss measurements are correlated over space in line with the derived correlation model of as was verified by our experiments . noise can originate from a variety of sources . in our setup , the noise level of the antenna plus electronics on a typical @xmath231 radio is about @xmath232dbm below signal level as specified in @xcite . the reproducible ripples on our rss signals originated in part from small interference effects from undue reflections from mostly hidden metallic obstacles in the measuring chamber . such obstacles like reinforced concrete pillars could not be removed . in a real indoor office environment , these interference effects usually dominate rss signals . in these environments the cross - correlated noise in rss signals is still determined by the diffraction limit , which in turn , determines the bound on localization precision . multi - path effects and fading do not change the stochastic properties of the secondary extended sources where the noise is generated . as explained in the literature @xcite , such models are usually clarified by ray - tracing . ray - tracing implies the geometrical - optics approximation , so that the wave lenghth is set to zero and diffraction effects are not considered . when we set the correlation length of the exponentially decaying correlations as expressed by @xcite equal to the diffraction limit , the crlb converges to practically the same bound on localization precision . it is remarkable that this bound is revealed by our simple experiment using various noise and signal models . our experiments reveal evidence that the diffraction limit determines the ( 1 ) bound on localization precision , ( 2 ) sampling rate that provides optimal signal reconstruction , ( 3 ) size of the coherence region and uncertainty , ( 4 ) upper bound on spatial frequencies and ( 5 ) upper bound on fisher information and lower bound on the crlb . with the mathematical work of @xcite , a rigorous link between fisher information , crlb and uncertainty was established . as uncertainty has a lower bound according to coherence and speckle theory , so must fisher information have an upper bound , so that the crlb has a lower bound , all when the noise processes are assumed to be gaussian distributed . our experiments were able to validate those theoretical concepts . this is further confirmed by applying bienaym s theorem . an interesting observation is that despite the fact that power flows have twice the spatial bandwidth than radiation flows determined by amplitudes , their spatial coherence regions are the same in size , as are their degrees of freedom . our paper describes a novel experimental and theoretical framework for estimating the lower bound on uncertainty for localization setups based on classical wave theory without any other prior knowledge . bienaym s theorem and existing crlb for signals with correlated gaussian noise reveal our postulate , so that the lower bound on uncertainty corresponds to the upper bound on fisher information . our experimental results can not be explained by existing propagation models with independent noise . it took almost three weeks in throughput time to perform the @xmath4 million measurements , generating @xmath233 repetitive multiplexed measurements . we tried to minimize multi - path effects by avoiding the interfering influence of ground and ceiling reflections . making sure to minimize reflections of other metallic obstacles in the measuring chamber were challenging but could be overcome . this allowed us to reveal a performance bound in theory and experiment for measurements without phase information . our novel two - dimensional localization setup where we positioned the reference radios on the circumference of the localization area rather than spreading them out in a two - dimensional array over this area worked well in our los setup . our measurements show that localization performance does not increase indefinitely with an ever increasing number of multiplexed rss signals ( space and time ) , but is limited by the spatial correlations of the far - field rss signals . when sufficient measurements are available to minimize the influence of measurement noise on localization performance , the bound on localization precision is revealed as the region of spatially correlated far - field radiation noise . the determination of this region of spatial correlations is straightforward and can be directly calculated from rss signals . within this region of correlated rss signals , assumptions of independent measurements are invalidated , so that the crlb for independent noise underestimates the bounds on radio localization precision . this underestimation is removed by accounting for the correlations in the noise . the bound on the correlations is given by the fundamental bound on the correlation length that we derived from first principles from the propagation equations . the crlb is linked to the uncertainty principle as measurement variance is directly related to this principle as we showed . the bounds on the precision of rss- and tof - based localization are expected to be equal and of the order of half a wavelength of the radiation as can be concluded from our experiments and underlying theoretical modeling . sampling beyond the diffraction limit or the nyquist sampling rate does not further resolve the oscillation period unless near - instantaneous measurements are performed with the a priori knowledge that signal processing is assumed to be based on non - linear mixing . future research is aimed at the inclusion of strong interference effects such as show up in practically any indoor environment . the authors would like to thank adrianus t. de hoop , h. a. lorentz chair emeritus professor , faculty of electrical engineering , mathematics and computer sciences , delft university of technology , delft , the netherlands , for his comments on applying green s theorem as a propagation model in electromagnetic theory . secondly , they would like to thank carlos r. stroud , jr . of the institute of optics , rochester , ny , for sharing his insight on fourier pairs and uncertainties . finally , they would like to thank gustaf hendeby and carsten fritsche of linkping university , sweden , for their comments . 35 g. toraldo di francia , `` resolving power and information , '' j. optical soc . 45 , no . 7 , 1955 . a. j. stam , `` some inequalities satisfied by the quantities of information of fisher and shannon , '' inform . and control , vol . 2 , pp . 101112 , 1959 . a. papoulis , `` error analysis in sampling theory , '' proc . ieee , vol . 54 , no . 7 , pp . 947 - 955 , july 1966 . m. gudmundson , `` correlation model for shadow fading in mobile radio systems , '' ieee electronics letters , 21452146 , 7 nov . h. hashemi , `` the indoor radio propagation channel , '' proc . ieee , vol . 81 , no . 7 , pp . 943 - 968 , july 1993 . s. kay , `` fundamentals of statistical signal processing : estimation theory , '' prentice hall , 1993 . j. s. lee , i. jurkevich , p. dewaele , p. wambacq , a. costerlinck , `` speckle filtering of synthetic aperture radar images : a review '' , remote sensing reviews , vol . 313 - 340 , 1994 . l. cohen , `` time - frequency analysis , '' prentice hall , 1994 . e. s. keeping , `` introduction to statistical inference , '' dover publications , 1995 . m. v. de hoop and a. t. de hoop , `` wavefield reciprocity and optimization in remote sensing , '' proceedings of the royal society of london , pp . 641682 , 2000 . m. a. pinsky , `` introduction to fourier analysis and wavelets , '' amer . soc . , 2002 . n. patwari , a. o. h. iii , m. perkins , n. s. correal , and r. j. odea , `` relative location estimation in wireless sensor networks , '' ieee trans . signal process . 51 , no . 8 , pp . 21372148 , aug . 2003 . wireless medium access control ( mac ) and physical layer ( phy ) specifications for low rate wireless personal area networks ( wpan ) , 802.15.4 - 2003 , 2003 . j. w. goodman , `` introduction to fourier optics , '' roberts and company publishers , 2004 . j. w. goodman , `` speckle phenomena in optics : theory and applications , '' roberts and company publishers , 2006 . n. patwari and a. o. hero iii , `` signal strength localization bounds in ad hoc and sensor networks when transmit powers are random , '' in ieee workshop on sensor array and multichannel processing , july 2006 . r. a. malaney , `` nuisance parameters and location accuracy in log - normal fading models , '' ieee trans . wireless commun . , vol . 6 , pp . 937 - 947 , march 2007 . n. patwari and p. agrawal , `` effects of correlated shadowing : connectivity , localization , and rf tomography , '' ieee / acm int . conf . on inf . processing in sensor networks , april 2008 . f. gustafsson , f. gunnarsson , `` localization based on observations linear in log range , '' proc . 17th ifac world congress , seoul , 2008 . y. shen , and m. z. win , `` fundamental limits of wideband localization part i : a general framework , '' ieee trans . theory , vol . 4956 4980 , oct . 2010 . b. j. dil and p. j. m. havinga , `` rss - based self - adaptive localization in dynamic environments , '' int . internet of things , pp . 5562 , oct . l. mandel and e. wolf , `` optical coherence and quantum optics , '' cambridge university press , 2013 . b. j. dil , `` lost in space , '' ph.d . dissertation , dept . , twente univ . , enschede , netherlands , 2013 . a.t . de hoop , `` electromagnetic field theory in ( n+1)-spacetime : a modern time - domain tensor / array introduction , '' proceedings of the ieee 101 , pp . 434450 , 2013 . a. herczyski , `` bound charges and currents , '' american journal of physics , vol . 3 , 202205 , 2013 . g. brooker , `` modern classical optics , '' oxford university press , 2014 . t. szabo , `` diagnostic ultrasound imaging : inside out , '' 2014 . j. w. goodman , `` statistical optics , '' wiley , 2015 . p. lpez - dekker , m. rodriguez - cassola , f. de zan , g. krieger , and a. moreira , `` correlating synthetic aperture radar ( cosar ) , '' ieee trans . geoscience and remote sensing , no . 99 , pp.1 - 17 , 2015 . http://www.ti.com/lit/ds/symlink/cc2420.pdf , p. 55 , march 2015 . b. j. dil , `` database of fundamental bound measurements and matlab files , '' linkping university , sweden , 2015 .

recent direct imaging discoveries suggest a new class of massive , distant planets around a stars . these widely separated giants have been interpreted as signs of planet formation driven by gravitational instability , but the viability of this mechanism is not clear cut . in this paper , we first discuss the local requirements for fragmentation and the initial fragment mass scales . we then consider whether the fragment s subsequent growth can be terminated within the planetary mass regime . finally , we place disks in the larger context of star formation and disk evolution models . we find that in order for gravitational instability to produce planets , disks must be atypically cold in order to reduce the initial fragment mass . in addition , fragmentation must occur during a narrow window of disk evolution , after infall has mostly ceased , but while the disk is still sufficiently massive to undergo gravitational instability . under more typical conditions , disk - born objects will likely grow well above the deuterium burning planetary mass limit . we conclude that if planets are formed by gravitational instability , they must be the low mass tail of the distribution of disk - born companions . to validate this theory , on - going direct imaging surveys must find a greater abundance of brown dwarf and m - star companions to a - stars . their absence would suggest planet formation by a different mechanism such as core accretion , which is consistent with the debris disks detected in these systems .

0909.2644

motivated by the recent discovery of massive planets on wide orbits , we explore the requirements for making gas giants at large separations from their host star via gravitational instability , hereafter , gi . in particular , we consider the formation mechanism for the system hr 8799 which contains three @xmath0@xmath1 objects orbiting at distances between @xmath0 30 and 70 au @xcite . the standard core accretion model for planet formation , already strained in the outer solar system , has difficulty explaining the presence of these objects . while gi is an unlikely formation mechanism for close in planets @xcite , for more widely separated planets , or sub - stellar companions , the viability gi - driven fragmentation deserves further investigation . in the inner regions of a protoplanetary disk , gas can not cool quickly enough to allow a gravitationally unstable disk to fragment into planets @xcite . for this reason , core accretion in which solid planetesimals collide and grow into a massive core which then accretes a gaseous envelope has emerged as the preferred mechanism for forming planets at stellar separations @xmath2 au . planets at wider separations have only recently been discovered by direct imaging around the a - stars hr 8799 @xcite , fomalhaut @xcite , and possibly beta pic . searches at large radii surrounding solar - type stars have yet to turn up similar companions @xcite . standard core accretion models can not form these planets , though further investigation is warranted . in favor of this possibility , all three systems show some evidence of processes related to core accretion : all have infrared excess due to massive debris disks at large radii . this is at least partially a selection effect as these systems were targeted due to the disks presence . nevertheless , these debris disks are composed of reprocessed grains from collisions of planetesimals . the disks long lifetimes prohibit a primordial origin for small grains they are removed quickly by radiation pressure and poynting - roberston drag @xcite and so must be regenerated from collisions between larger bodies that formed through the coagulation of solids at early times . therefore , planetesimal formation , a necessary ingredient in core accretion models , has taken place ( e.g. , * ? ? ? * ; * ? ? ? in addition , other a - stars host planets at 12 au @xcite which , although they have a distinct semi - major axis distribution from planets orbiting g and m stars , are likely formed by core accretion . if future surveys demonstrate that this distribution extends smoothly to wide separation planets , then simplicity would argue against a distinct formation mechanism for the wide giants . yet the standard core accretion model faces a serious problem at large distances . the observed lifetimes of gas disks are short , at most a few myr @xcite . in contrast , typical core accretion times increase with radius and exceed 10 myr beyond 20 au . whether or not this theoretical difficulty can be overcome will require careful modeling of the interactions between planetesimals and the young gas disk . could wide orbit planets have formed at smaller radii , and migrated outwards ? @xcite , have investigated the possibility of forming the hr 8799 system via scattering in the absence of dynamically important gas , but find that putting three massive planets into such closely spaced yet wide orbits is unlikely . @xcite have suggested that under favorable circumstances outward migration in resonances might be feasible . alternatively , the core of a giant planet could be scattered outward by a planet , or migrate outward before accreting its gas envelope , either by interactions with the gas disk ( type iii migration ; e.g. @xcite ) or with planetesimals embedded in the gas ( capobianco , duncan , & levison , in prep ) . neither mechanism has yet been shown to move a core to such large distances , though this possibility has not been ruled out . given these difficulties , it is natural to search for other formation mechanisms , and gi @xcite stands out as a promising alternative . if any planets form by gi , the recently discovered directly imaged planets are the most likely candidates @xcite . in this paper , we examine this possibility in more detail , considering the expected mass scale of fragments and the effect of global disk evolution on the formation process . the inferred masses for the hr 8799 planets are close to the deuterium burning limit of @xmath3 @xcite . for simplicity , we take this as a the dividing line between planets and brown dwarfs and we refer to the hr 8799 objects as planets throughout . however , there is no reason for a given formation mechanism to function only above or below this threshold , and in fact , we will argue that if the hr 8799 planets formed by gi , their histories are more akin to those of higher - mass brown dwarfs than to lower - mass planets . to constrain gi as a mechanism for wide giant planet formation , we set the stage by describing the hr 8799 system in [ sec - hr8799 ] . we review the standard requirements for fragmentation in [ idealgi ] and discuss the initial mass scale of fragments in [ sec - masses ] . in [ sec - giaccrete ] we show that under typical disk conditions , fragments will continue to accrete to higher masses . we then discuss important global constraints on planet formation provided by star formation models , disk evolution timescales , and migration mechanisms in [ sec - starform ] , and [ sec - migration ] . we compare predictions of our analysis with the known wide substellar companions and exoplanets in [ sec - observe ] , suggesting that future observations will provide a definitive answer to the formation mechanism for hr 8799 . in the appendices we re - examine the heating and cooling properties of disks that are passively and actively heated , with special attention to the implications for irradiated disks , which become increasingly relevant for more massive stars . the planets around hr 8799 probe a previously unexplored region of parameter space @xcite because they are more distant from their host star . the three companions to hr 8799 are observed at separations from their host star of 24 , 38 , and 68 au . their masses , estimated using the observed luminosities of the planets in conjunction with cooling models , have nominal values of 10 , 10 , and 7 @xmath4 , respectively . a range in total mass of 1937 @xmath4 is derived from uncertainties in the age of the host star @xcite . interpretation of the cooling models generates substantial additional systematic uncertainty recent measurements suggest that these models may overpredict the masses of brown dwarfs by @xmath025% @xcite . @xcite have demonstrated that for planetary masses in the stated range , orbital stability over the age of the system requires that the planets occupy at least one mean - motion resonance , and that for doubly - resonant orbital configurations total masses of up to at least @xmath5 can be stable . hr 8799 has been called a scaled - up solar system " in terms of the stellar flux incident on its giant planets @xcite . however for understanding the formation of this system it is more useful to consider dynamical times , and disk mass requirements . because the dynamical time at fixed radius scales only as @xmath6 , the dynamical times are larger at the locations of the hr 8799 planets that at the solar system giants . since the total mass in planets greatly exceeds the @xmath7 in the solar system , we can infer that ( as with some other extrasolar systems ) the primordial disk around hr 8799 was more massive than the solar nebula and/or there was greater efficiency of planet formation , especially in the retention of gas . compared to solar system giants , longer dynamical times make core accretion more difficult and larger disk masses make gi more plausible . we first determine where , and under what local disk conditions , fragmentation by gi is possible . following @xcite , @xcite and @xcite , we argue that for a disk with surface density @xmath8 and temperature @xmath9 to fragment , it must satisfy two criteria . first , it must have enough self - gravity to counteract the stabilizing forces of gas pressure and rotational shear , as quantified by toomre s q : @xmath10 @xcite , where @xmath11 is the isothermal sound speed of the gas with mean particle weight @xmath12 appropriate for a molecular gas , @xmath13 is the gravitational constant , @xmath14 is the boltzmann constant , and @xmath15 is the orbital frequency . equation ( [ eq - q ] ) specifies the onset of axisymmetric instabilities in linear theory that can give rise to bound clumps @xcite . the second criterion that must be satisfied for fragmentation to proceed is the so - called cooling time criterion . the heat generated by the release of gravitational binding energy during the contraction of the fragment must be radiated away on the orbital timescale so that increased gas pressure does not stall further collapse @xcite . this implies @xmath16 here @xmath17 is a constant of order unity , @xmath18 is the adiabatic index of the gas , and @xmath19 is the stefan - boltzmann constant . we take @xmath20 @xcite for disk vertical optical depth @xmath21 and gas opacity @xmath22 . numerical models of collapse in barotropic disks measure the critical value @xmath17 through the inclusion of a loss term @xmath23 in the equation for the internal energy , @xmath24 . estimates of @xmath17 range from @xmath25 , depending on @xmath18 @xcite , the numerical implementation of cooling @xcite , and the vertical stratification in the disk . we assume @xmath26 , appropriate for molecular hydrogen , and we adopt @xmath27 here . although @xmath17 was measured in disks whose temperature is controlled by viscous heating , we show in appendix [ sec - irradcool ] that the same expression ( modulo slightly different coefficients ) should apply when irradiation sets the disk temperature , as will be the case in disks prone to fragmentation ( see appendix [ sec - visc ] ) . a disk satisfying toomre s criterion for instability ( equation [ eq - q ] ) but not the cooling time criterion ( equation [ eq - tcool ] ) experiences gi - driven angular momentum transport which regulates the surface density of the disk so that @xmath28 and @xmath29 does not reach substantially smaller values ( c.f . appendix [ sec - irradcool ] ) . we can therefore use toomre s criterion to define a relationship between @xmath8 and @xmath9 at fragmentation , as a function of period : @xmath30 where for convenience we define @xmath31 . we shall hereafter set @xmath32 . given equation([eq - sigmaq ] ) , we can rewrite equation([eq - tcool ] ) to generate a single criterion for fragmentation which depends on temperature and location : @xmath33 where @xmath34 . somewhat counterintuitively , the critical cooling constraint requires that a disk be sufficiently hot to fragment . the value of @xmath35 depends on both @xmath15 and @xmath9 . evaluating this criterion relies on the disk opacity , which we return to in [ sec - opacity ] . we take the initial mass of a fragment to be the mass enclosed within the radius of the most unstable wavelength , @xmath36 in a @xmath37 disk , or : @xmath38 @xcite , where @xmath39 is the disk scaleheight . @xcite have shown that even when the gi is non - axisymmetric , the most unstable axisymmetric wavelength @xmath40 is one of the dominant growing modes , suggesting that this is a reasonable estimate . while more numerical follow up will be necessary to pin down the true distribution of fragments born through gi , at present simulations show that this estimate may well be a lower limit , but is the correct order of magnitude @xcite . using equation([eq - sigmaq ] ) , we rewrite the fragment mass explicitly as a function of temperature and location : @xmath41 where @xmath42 . equation ( [ eq - mfragt ] ) demonstrates that at a given disk location , fragment masses depend only on temperature , with lower temperatures generating smaller fragments , subject to the minimum temperature required for fragmentation by equation([eq - tcoolb ] ) . @xcite pointed out that there exists an absolute minimum fragment mass at any disk location , when the disk satisfies the equalities in equation([eq - q ] ) and equation([eq - tcool ] ) and has @xmath43 . the minimum temperature required for fragmentation scales as @xmath44 ( equation [ eq - tcoolb ] ) , so the critical temperature and fragment mass are minimized at @xmath45 , the optical depth for which cooling is most efficient . the corresponding minimum mass as a function of location is : @xmath46 which occurs for disk temperatures : @xmath47 equation ( [ eq - minmass ] ) corresponds to @xmath37 , @xmath48 , and @xmath45 . this minimum mass is only achieved for temperatures consistent with @xmath49 . once an opacity law is specified which relates @xmath9 and @xmath50 , the problem becomes overconstrained these three criteria can only be satisfied at a single disk location , and equations [ eq - minmass]-[eq - mintemp ] are valid at only one radius in the disk . we now proceed to evaluate the critical disk temperatures and fragment masses for realistic opacity laws , demonstrating that planetary - mass fragments can only form at large separations from their host star . at low temperatures , when @xmath51 k , ice grains are the dominant source of opacity @xcite . above @xmath0@xmath52 k , ices begin to sublimate . in the cold regime applicable to the outer regions of protoplanetary disks , we consider two realistic opacity laws , one which is characteristic of grains in the interstellar medium ( ism ) , and one which is characteristic of grains that have grown to larger sizes due to processes within the disk . we assume a rosseland mean opacity which scales as @xmath53 where the both the exponent , @xmath54 , and the constant @xmath55 are not well constrained in protoplanetary disks . they depend on the number - size distribution and composition of the dust grains , and on the dust - to - gas ratio . for ism like grains , the rosseland mean opacity may be approximated by an opacity law with @xmath56 , or @xmath57 as long as the ice grains dominating the opacity are smaller than a few tens of microns . this opacity law is observationally confirmed in the ism ( * ? ? ? * and references therein ) . for our fiducial model we use @xmath58 for @xmath9 in kelvin , a fit to the standard dust model by . throughout , we quote @xmath22 per gram of gas for a dust - to - gas ratio of @xmath59 . as grains grow larger , they eventually exceed the wavelength of the incident radiation , and so the opacity is determined by the geometric optics limit . in this case the rosseland mean opacity is independent of temperature , so the exponent @xmath60 . in this limit : @xmath61 for a fixed dust mass in grains of size @xmath62 , @xmath63 . for our fiducial model we choose @xmath64 which is valid for @xmath65 k and typical grain sizes of order 300@xmath66 ( see figure 6 of * ? ? ? observations of emission from optically thin protoplanetary disks show evidence of grain growth at millimeter wavelengths . specifically the measured opacities @xmath67 with @xmath68@xmath69 in the rayleigh - jeans tail imply particle growth toward the millimeter wavelengths of the observations @xcite . although most observed disks have had more time for grain growth to proceed , class 0 sources also show evidence of grain growth @xcite . alternatively , these objects could be optically thick at millimeter wavelengths , which would mimic the effects of grain growth . using these opacity laws we see that cooling proceeds with a different functional form in the optically thick and optically thin limits . in the optically thick regime , @xmath70 , which when combined with equation([eq - tcoolb ] ) indicates that to fragment , the disk must have temperature @xmath71 for @xmath72 . in the special case @xmath73 , the fragmentation constraint is temperature - independent , as discussed in [ sec - ism ] . in the optically thin regime , @xmath74 . the cooling time is independent of @xmath8 , so we can rewrite equation([eq - tcoolb ] ) as : @xmath75 fig . [ fig - tcool ] shows the dependence of the cooling time on disk temperature for each opacity law at two different radii . since fragmentation can only occur when @xmath76 , the minimum temperatures at which fragmentation is allowed are specified by the intersection of the cooling curves with the @xmath77 boundary . for an optically thin disk with @xmath73 , equation([eq - tcoolcritthin ] ) implies that in order to fragment , the disk must have temperatures in excess of : @xmath78 colder disks , even when optically thin , can not cool quickly enough to fragment . for an optically thick disk , @xmath56 turns out to be a special case : the temperature dependence drops out of equation([eq - tcoolcritthick ] ) , giving instead a critical radius beyond which fragmentation can occur , independent of the disk temperature : @xmath79 @xcite first pointed out the existence of a minimum critical radius for fragmentation . in fig . [ fig - a100 ] we illustrate how the two fragmentation criteria create a radius rather than temperature cutoff . at radii larger than the critical radius defined above , any @xmath80 combination which gives @xmath81 will fragment ( so long as the opacity law remains valid ) . at smaller radii , no combination of @xmath8 and @xmath9 which gives @xmath81 will fragment because the disk can not simultaneously satisfy the cooling time criterion . as shown in fig . [ fig - tcool ] , for the large grain opacity , @xmath82 for all relevant temperatures . in this case equation([eq - tcoolcritthick ] ) requires : @xmath83 the corresponding minimum mass for these temperatures is : @xmath84 we have scaled equations ( [ eq - tminz ] ) and ( [ eq - mminz ] ) to an effective critical radius for this opacity law . although fragmentation can occur inside 43 au at sufficiently high temperatures , the fragments exceed @xmath3 , making it irrelevant for planet formation . smaller values of @xmath85 move this boundary inward , allowing for fragmentation into lower mass objects at smaller radii , although the scaling with opacity is shallow . grain growth to larger sizes could plausibly reduce @xmath85 . if , for example , grains dominating the disk opacity have grown up to 1 mm without altering the dust - to - gas ratio , equation([eq - ind_kap ] ) implies that in the geometric optics limit @xmath86 . in this case , the minimum radius is pushed inward to 26 au ( see also * ? ? ? thus far we have determined the minimum masses allowed as a function of radius with the temperature as a free parameter . we now calculate actual disk temperatures , which at large radii are typically higher than the minima . in this case we must evaluate fragment masses using equation([eq - mfragt ] ) . to consider the case favorable to gi planet formation , we consider the lowest plausible disk temperatures in order to minimize the fragment masses . we estimate the disk temperature using the passive flared disk models of @xcite . this model likely underestimates the temperatures in actively accreting systems because it ignores significant backheating " from the infall envelope and surrounding cloud @xcite . although viscous heating will also contribute to the temperature , we ignore its modest contribution to obtain the lowest reasonable temperatures and fragment masses . disk irradiation dominates over viscous heating in this regime ( see appendix b ) . we consider the inner region where the disk is optically thick to blackbody radiation . in this regime , the temperature of a flared disk in radiative and hydrostatic equilibrium is : @xmath87 where @xmath88 measures the grazing angle at which starlight hits the disk ; @xmath88 is dependent on the degree of disk flaring measured at the height of the photosphere @xcite . grain settling may reduce the height of the photosphere ( set here to 4 times the scaleheight ) and thus @xmath88 . for the standard radiative equilibrium model , the disk flaring scales approximately as @xmath89 . we shall find when we calculate the disk temperature that a gravitationally unstable disk remains optically thick , justifying the use of this formula . to estimate the the stellar luminosity we use the stellar evolution models of @xcite , which include both nuclear burning and accretion energy . the accretion luminosity depends on both the current accretion rate and the accretion history ( in so far as it effects the stellar radius ) , so we obtain the lowest luminosity estimates by allowing the star to accrete at a constant , low accretion rate . we use the stellar luminosity after accreting to 90% of its current mass ( or @xmath90 assuming roughly 10% is still in the disk ) . we choose an accretion rate of @xmath91 as a lower bound because a star accretingat a lower accretion rate throughout its history has a formation timescale that is too long . accretion rates an order of magnitude larger give comparable luminosity ( when the star has only reached @xmath92 ) to the present day luminosity of @xmath93 @xcite . lowering the accretion rate below this value does not significantly lower the stellar luminosity because the accretion energy contribution is small . the luminosity calculated for an accretion rate of @xmath91 translates to temperatures of : @xmath94 which we shall use as our fiducial temperature profile . in the outer regions of the disk where fragmentation is allowed , the disk temperatures are of order @xmath95k . these temperatures exceed the minimum threshold for fragmentation , and so the mass of fragments will be set by equation ( [ eq - mfragt ] ) . other analytic and numerical models of stellar irradiation predict temperatures in agreement with or higher than our estimate . similarly , models of disks in ophiuchus have similar temperatures for 1 myr old stars of lower mass ( and thus luminosity ) , implying that our model temperatures are low , though not unrealistic @xcite . in fig . [ fig - cartoon ] we illustrate the constraints on fragment masses from this irradiation model , calculated using equation([eq - mfragt ] ) . we show the fiducial disk temperatures of equation([eq - mint ] ) , along with temperatures consistent with luminosities up to the present - day luminosity . for our fiducial opacity laws , the expected fragment masses are only marginally consistent with gi planet formation fragments form near the upper mass limit of @xmath3 . lower opacities produced by grain growth might allow fragmentation into smaller objects at closer radii . whether grain - growth has proceeded to this extent in such young disks is unclear . for realistic disk temperatures , it is conceivable that fragments may be born at several @xmath4 . we now consider their subsequent growth , which may increase their expected mass by an order of magnitude or more . the final mass of a planet depends sensitively on numerous disk properties ( effective viscosity , column density , scaleheight ) and the mass of the embedded object . in order to constrain the mass to which a fragment will grow , we can compare it to two relevant mass scales : the disk isolation mass and the gap opening mass . halting the growth of planetary mass objects is a relevant problem independent of the formation mechanism . however the gi hypothesis requires that the disk is ( or was recently ) sufficiently massive to have @xmath96 , implying that the disk is actively accreting . the core accretion scenario does not face this restriction . we estimate an upper mass limit for fragments by assuming that they accrete all of the matter within several hill radii : @xmath97 here @xmath98 is a numerical constant representing from how many hill radii , @xmath99 , the planet can accrete @xcite . it is instructive to compare the ratio of the isolation mass to the stellar mass : @xmath100 we find that large isolation masses are always expected in gravitationally unstable disks . [ allmasses ] illustrates the scaling of equation ( [ eq - miso - m ] ) with @xmath29 and the disk aspect ratio , @xmath101 . for our fiducial temperature profile , @xmath102 at 70 au . for low values of @xmath29 and comparable @xmath101 , the isolation mass exceeds @xmath103 of the stellar mass . for the ideal disk values cited above ( equation [ eq - rminthick ] ) , the isolation mass is : @xmath104 this mass is nearly two orders of magnitude larger than the minimum mass . growth beyond the isolation mass is possible either through mergers or introduction of fresh material to accrete by planet migration or disk spreading . objects which grow to isolation mass can not be planets , and so we turn to mechanisms that truncate fragment growth below the isolation mass . massive objects open gaps in their disks when gravitational torques are sufficiently strong to clear out nearby gas before viscous torques can replenish the region with new material . @xcite . the gap width is set by the balance between the two torques : @xmath105 where @xmath106 is the gap width , @xmath107 is a geometric factor derived in @xcite , @xmath108 is the planet to star mass ratio , and @xmath109 is the @xcite effective viscosity . this can be used to derive the standard minimum gap opening mass by requiring that @xmath110 : @xmath111 gap opening requires @xmath112 and @xmath113 . the latter requirement is automatically satisfied for fragments formed by gi . while the effects of gi are often parameterized through an @xmath109 viscosity , @xcite have pointed out that @xmath109 , a purely local quantity , may not adequately describe gi driven transport , which is inherently non - local . @xcite have shown that for sufficiently thin disks , the approximation is reasonable : in order to form objects of planetary mass , the disk must be relatively thin and at least marginally within this limit . however , even in this thin - disk limit , it is not clear that gi driven torques will exactly mimic viscous ones at gap - opening scales . equation ( [ eq - gapmass ] ) implies that the gap opening mass is less than or equal to the fragment mass for effective viscosities consistent with gi . we use @xmath114 , as this is consistent with active gi @xcite . if the local disk viscosity is lower , fragments will always form above the gap - opening mass . gap - opening slows accretion onto the planet , but does not starve it of material completely . accretion rates through gaps remain uncertain for standard core accretion models , and numerical models are not available for accretion onto the distended objects formed through gi fragmentation . nevertheless simulations of accretion through gaps in low viscosity disks @xcite demonstrate that accretion is slower through larger gaps , and this qualitative conclusion likely remains valid as long as gaps form . analogous to the isolation mass , we consider a gap starvation mass " that is related to the ratio of the gap width to planet hill radius . rewriting equation([eq - gapwidth ] ) we find the ratio of gap width to hill radius is : @xmath115 note that @xmath116 recovers the canonical gap opening estimate appropriate for jupiter : @xmath117 , modulo order unity coefficients ( cf . if we make the simplifying assumption that gap accretion terminates when the gap width reaches a fixed number of hill radii , @xmath118 , we can calculate a gap starvation mass . we expect that for a gap to truncate accretion , @xmath119 , the width in hill radii used to calculate the isolation mass ( [ sec - isolation ] ) . terminating accretion is an unsolved problem for jupiter in our own solar system , and so to provide a further constraint on @xmath118 , we refer to the numerical simulations of @xcite ( see also @xcite for a detailed explanation of the numerical work ) . their runs 2l and 2lj exhibit asymptotic mass growth after @xmath02.5 myr for a constant - mass , low viscosity ( @xmath120 ) disk under conditions appropriate to the formation of jupiter . using equation([eq - gaptohill ] ) , we solve for the width of the gap generating this fall - off in accretion rate and find @xmath121 . the need for an extremely large and well - cleared gap reflects the integrated effects of low - level accretion through the gap and onto the planet over the disk lifetime of a few myr . even a slow trickle of material onto the planet can contribute to significant growth . using @xmath122 , the gap - opening starvation mass for hr 8799 scaled to both the simulated solar - system viscosity and to the expected gi viscosity is : @xmath123 in order to limit growth to planetary masses , the effective viscosity must be two orders of magnitude below that expected in gi unstable disks , roughly @xmath124 . more restrictively this requires that other local transport mechanisms such as the mri be weaker than currently predicted by simulations they produce @xmath125 at least in disks with a net magnetic flux . [ jupiterscale ] illustrates the scaling of gap starvation mass with radius for several values of @xmath109 . it appears that active disks face severe obstacles in producing planetary mass objects unless the disk disappears promptly after their formation . although we expect fragmentation to make the disk more stable by lowering the local column density , there is little reason to expect a recently massive disk to be so quiescent . , and indicate the planetary mass regime , and the region in which disk fragmentation is likely . we use @xmath126 , scaled to simulation 2lj of jupiter formation in @xcite ( labeled l09 in the figure ) . the radial scaling is derived assuming @xmath89 . for the low viscosity case , we normalize the scale height to jupiter at @xmath127 au in a 115 k disk for comparison with l09 . for the higher viscosities , we normalize the disk scale height to the lowest expected temperatures ( equation [ eq - mint ] ) . for comparison we show the hr 8799 planets as black circles . ] although it is unlikely that the disk will fragment into a sufficiently large number of planetary mass objects to completely deplete the disk of mass @xcite , the formation of multiple fragments simultaneously may limit accretion through competition for disk material by opening overlapping gaps . the current separation between the planets is such that gaps larger than roughly three hill radii overlap , so depending on their migration history , this could limit growth ( see also [ sec - migration ] ) . from equation([eq - low_tgap ] ) we see that if gaps are forced to be smaller by a factor of two due to competition with another planet , the expected masses are decreased by a factor of 8 . this effect would imply that fragments in multiple systems should be lower in mass . note that when simulated disks fragment into multiple objects simultaneously , they generally have orbital configurations like hierarchical multiples rather than planetary systems @xcite . whether the same conditions required to limit fragment growth reduced disk mass and/or viscosity after fragmentation can allow the retention of a planetary system of fragments remains to be simulated . mechanisms for gas dispersal such as photoevaporation may be necessary to stunt planetary growth , even for models of jupiter in our own solar system @xcite . dissipation timescales for a - star gas disks are thought to be short , less than 2 - 3 myr @xcite , which could halt growth before the gap - opening starvation mass is achieved . radiative transfer models such as @xcite have calculated that photoevaporation by the central star will become important at radii of @xmath128 au around one myr for an a star ( see also * ? ? ? this timescale coincides with the expected fragmentation epoch , and may aid in shutting off accretion both onto the disk , and onto the planets . we now consider how the disk can reach the fragmentation conditions described in [ idealgi ] in the context of a model for star formation . due to the effects of infall onto the disk , we find that planet formation via gi can only occur when the fragmentation epoch is concurrent with the end of the main accretion phase , as the protostar transitions from a class i to a class ii object @xcite . fragmentation at earlier times leads to the formation of more massive companions , while fragmentation at later times is unlikely because disks are too low in mass @xcite . because the cooling time constraint is easily satisfied for expected disk temperatures , disks are likely driven to fragmentation by lowering @xmath29 . even when @xmath29 is above the threshold for fragmentation , torques generated by self - gravity ( e.g. spiral arms ) can drive accretion . when the infall rate onto the disk is low , a self - gravitating disk regulates its surface density and hence @xmath29 so that the torques are just large enough to transport the supplied mass down to the star , thereby avoiding fragmentation . however , gi can not process matter arbitrarily quickly because the torques saturate . thus the disk will be driven toward fragmentation if the infall rate becomes too high . this critical accretion rate is a function of disk temperature : @xmath129 @xcite . numerical simulations show that gi saturates at @xmath130 @xcite . if the infall rate onto the disk exceeds @xmath131 the disk can no longer regulate the surface density to keep @xmath29 just above unity , and fragmentation will occur . because the conversion of accretion energy to thermal energy at large radii is inefficient , disks can not restabilize through heating to arbitrarily high accretion rates @xcite . if fragmentation occurs due to rapid accretion , as described above , it is difficult to limit subsequent fragment growth . as demonstrated in [ sec - gapstarve ] , gap opening does not limit accretion efficiently when the effective viscosity , in this case , @xmath132 , is high . a more general barrier to making small fragments during rapid infall is the large reservoir of material passing by the fragment as star formation proceeds @xcite . the specific angular momentum , @xmath133 , of accreting disk material typically increases with time ( modulo small random fluctuations in a turbulent core ) , landing at a circularization radius , @xmath134 , which is larger than the fragment s orbit . since the fragment s hill radius is roughly 10% of the disk radius , newly accreted material undergoes many orbits in the fragment s sphere of influence as it tries to accrete onto the central star , and some fraction will accrete onto the fragment itself . this process is less efficient if global gi modes drive fragments to smaller radii , because their hill radii shrink . however , growth to stellar ( or sub - stellar ) masses may still occur as long as migration timescales are not faster than accretion timescales . the latter scenario implies that the disk can not fragment at early times and still reproduce a single - star system like hr 8799 . the trend toward continued growth and even mass equalization following disk fragmentation is observed in numerous simulations with ongoing accretion @xcite . simulations of planet formation by gi with ongoing accretion also illustrate this behavior @xcite . consequently , if the hr 8799 planets formed by gi , they must have fragmented at the tail end of accretion from the protostellar core onto the disk . most likely , this requires that the protostellar core have properties such that its infall rate reaches the critical value in equation ( [ eqn - critmdot ] ) just as the core is drained of material . if this coincidence in timing does not occur , fragmentation produces a substellar , rather than a planetary , companion . whether fragmentation at this epoch can produce non - heirarchical orbits like hr 8799 remains to be explored . driving the disk unstable with external accretion corresponds to excessive growth of fragments . it is therefore tempting to consider mechanisms to lower @xmath29 through disk cooling , while holding the column density fixed . because the disks are dominated by irradiation , changes in the viscous dissipation due to accretion are unlikely to affect a significant temperature change , and so lowering @xmath29 requires order of magnitude changes in the stellar luminosity due to the weak scaling of @xmath135 . fu orionis type outbursts can cause rapid changes in luminosity . to result in planetary mass fragments , the luminosity drop following an fu ori outburst would need to reach at least the minimum luminosity used in equation ( [ eq - mint ] ) on timescales shorter than an outer disk dynamical time . the accretion of gi formed embryos onto the protostar is a proposed source of the outbursts @xcite . if this occurs , perhaps a final generation of gravitational fragments , the so - called last of the mohicans " @xcite , would remain as detectable companions . although the lack of infall in this scenario might ease the gap opening constraints , fragments face all of the other difficulties discussed above in remaining low in mass . a final consideration for making wide orbit planets through gi is their subsequent migration history . if formed via gi , the hr 8799 planets had to migrate inward to their current locations . there is an independent reason to believe that migration did in fact take place in this system . as discussed by @xcite , the long term stability of hr 8799 requires a resonant orbital configuration , most plausibly a 4:2:1 mean motion resonance , which likely resulted from convergent migration in the protoplanetary gas disk . we demonstrate below that while this history is plausible , it requires special disk conditions . inward type ii migration ( appropriate for gap opening planets ) is expected at formation , because the fragmenting region is within the part of the disk accreting onto the star . type ii migration of a single planet occurs on the disk viscous timescale : @xmath136 where we have used @xmath137 . this timescale is short enough to allow substantial migration during the lifetime of the gas disk . if fragments migrated inward independently , they could never become captured in resonance , as the innermost planet would migrate farther and farther from its neighbor . however , as shown in [ sec - giaccrete ] , planetary gaps may overlap if they are within several hill radii of each other , comparable to the current separations between the hr 8799 planets . this overlap alters the torques felt by , and thus the migration of , the planets . as shown by @xcite , if multiple planetary gaps interact , convergent migration is possible . gap interaction allows an outer planet to shield an inner planet from the material , and thus torques , of the outer disk , slowing or halting its inward migration and allowing the outer planet to catch up . this mechanism is invoked by @xcite to generate convergent migration and resonance capture in the planets orbiting gj 876 . under the assumption that gap overlap allows convergent migration and resonance capture , we now ask : what is the overall direction of the subsequent migration ? we note that if the planets migrated a substantial distance after resonance capture , eccentricity damping by the gas disk was likely necessary ( c.f . we do not consider eccentricity damping further here . once two planets are caught in mean - motion resonance , the torque on an individual planet from the gas disk can cause both planets to migrate , with angular momentum transfer mediated by the resonance . @xcite ( see also * ? ? ? * ) have argued that the torque imbalance on a pair of gap - opening resonant planets can even reverse the direction of migration , although this relies on a significant difference between planet masses . nevertheless , understanding the planets overall migration requires understanding how overlapping gaps alter the torque balance on the group of planets . guided by our interest in clean gaps that limit the growth of planets ( [ sec - giaccrete][sec - starform ] ) , we consider the following simplified problem . we imagine that the three planets have cleared , and are embedded in , a single large gap which is sufficiently clean that any disk gas passing through is dynamically unimportant . because the system is locked in a double mean - motion resonance , we assume that an imbalance in the torques acting on the two edges of the gap can cause all three planets to migrate . we can then ask : is there a sufficient flux of angular momentum from the outer disk to cause the planets to migrate inward with the viscous accretion of the disk ? when in the 4:2:1 resonance , the total angular momentum of the planets is roughly @xmath138 , where we have assumed that the planets are roughly equal in mass , @xmath139 , while @xmath140 and @xmath141 are the separation and keplerian angular velocity of the outermost planet . the angular momentum flux from the outer disk is large enough to move the planets on a viscous time @xmath142 when : @xmath143 where @xmath144 is the mass flux through radius @xmath140 . using the above inequality , we can calculate a critical disk surface density at the current location of the outer planet such that the disk can push the planets inward : @xmath145 for @xmath146 and our fiducial disk temperatures ( equation [ eq - mint ] ) , this constraint is always satisfied when @xmath37 . the disk is unable to cause inward migration when @xmath147 at 70 au , which is equivalent to @xmath148 . if the planets do share a clean common gap , a large fraction of the disk would be effectively cleared of gas while a massive outer disk is still present . a similar mechanism has been invoked to explain transitional disks , which contain holes at radii of a few tens of au and smaller @xcite . transport of disk gas through a less well - cleared gap could substantially alter this picture . in summary , it is possible to envision a scenario in which the hr 8799 planets migrate inward to their current locations in such a way that their orbits converge , allowing resonance capture . this scenario is consistent with other constraints on gi planet formation : shortly after formation , the disk must have low accretion rates and decline in mass in order to ( a ) limit the growth of fragments , ( b ) allow for large , overlapping gaps . more stringent constraints will require future work on migration in gravitationally unstable disks , particularly in the presence of multiple planets massive enough to clear large , overlapping gaps . while there is a regime of parameter space in which planet formation is possible , typical conditions produce more massive ( @xmath149 ) companions . if gi fragmentation ever forms planets , then fragmentation should typically form more massive objects . consequently , these planets would constitute the low mass tail of a distribution of disk - born companions . if the mass distribution is continuous there should be more sub - stellar companions than planets at comparable distances of 50 - 150 au . observing this population is a strong constraint on the formation mechanism , but current data are insufficient to draw conclusions . have compiled the known sub - stellar companions in this range of radii to date . this range of separations falls beyond the well - established inner brown dwarf desert " @xcite , and has not been well probed due to observational difficulties at these low mass ratios . note that the overall dearth of brown dwarf companions to solar mass stars is not a selection effect . we illustrate the observational constraints by plotting the companions from along with the known exoplanets compiled by the exoplanet encyclopedia as a function of mass ratio and projected separation in fig . [ fig - bdplot]a , and as a function of minimum fragment masses and fragmentation radii in fig . [ fig - bdplot]b . we compare these with the hr 8799 planets , fomalhaut b , and the solar system giants . we distinguish between stars of different masses because disk fragmentation becomes more likely for higher mass stars @xcite . at present , neither the population of substellar companions nor the population of exoplanets is continuous out to hr 8799 . many selection biases are reflected in fig . [ fig - bdplot ] , and these gaps in particular may be due to selection effects ; resolution and sensitivity make it difficult to detect both wide orbit planets , and close - in low mass brown dwarfs . we note that while there are actually fewer planets at distances less than 1 au , the cutoff above 5 au is unphysical . there is not yet any indication of an outer cut off radius in the exoplanets : if they continued out to larger separation , the distribution would easily encompass the hr 8799 and fomalhaut systems . data from ongoing surveys like that which found hr 8799 are necessary to verify the true companion distribution as a function of mass and radius . if these planets are formed through gi , we would expect observations to fill in the gap between hr 8799 and higher mass ratio objects to show a continuous distribution . if these planets are formed via core accretion , than observations may fill in the plot on the opposite side of hr 8799 , occupying a region of parameter space for which neither core accretion nor gi is currently a successful formation mechanism . we have demonstrated that while gi - driven fragmentation is possible at wide distances from a stars , fragment masses typically exceed the deuterium burning planet " limit . in contrast , the formation of sub - stellar and stellar companions is more likely because moderate disk temperatures and active accretion onto and through the disk drive disk - born objects to higher masses . if the hr 8799 planets did form by gi , the following criteria had to be met : 1 . fragments should form beyond @xmath150 au : inside of this location the disk will not fragment into planetary mass objects even if @xmath151 . grain growth is required for fragmentation at the lower end of this range . 2 . temperatures must be colder than those of typical disks to limit the initial fragment masses . the disk must be driven unstable at a special time : infall onto the disk must be low , but the disk must remain massive ( e.g. the end of the class i phase ) . the disk must only become unstable to fragmentation at this point because earlier episodes of instability should lead to sub - stellar or stellar companion formation . the subsequent growth of fragments must be limited through efficient gap clearing necessitating low disk viscosity or early gap overlap . disk dispersal via photoevaporation may also be necessary . the three fragments must form at the same epoch separated by several hill radii , implying that the entire outer disk becomes unstable simultaneously . migration must be convergent . this likely requires the gaps of the planets to overlap so as to starve the inner most planet of disk material , thereby preventing runaway inward migration . if these conditions are met , then the planets in hr 8799 could comprise the low - mass tail of the disk - born binary distribution , the runts of the litter . in this case one would expect to find a larger number of brown dwarfs or even m stars in the same regime of parameter space surrounding a - stars at distances of @xmath152 au . ongoing direct imaging surveys of a and f stars will provide a strong constraint on the formation mechanism for this system : if hr 8799 is the most massive of a new distribution of widely separated planets , our analysis suggests that formation by gi is unlikely . on the contrary , the discovery of a population of brown dwarf and m - star companions to a - stars would corroborate formation via disk fragmentation . * acknowledgements : * the authors thank eric ford and jonathan tan for organizing the astrowin workshop at the university of florida , which inspired this research , chris matzner , for invaluable comments on this manuscript , and yoram lithwick , roman rafikov , ken rice , pawel artymowicz , eugene chiang , sally dodson - robinson , aaron boley , and stella offner for helpful discussions . the authors also thank an anonymous referee for helpful comments . kmk is supported in part by an ontario graduate scholarship . rmc is funded by an institute for theory and computation fellowship at harvard university . 85 natexlab#1#1 , p. , & montmerle , t. 1994 , , 420 , 837 , s. m. , wilner , d. j. , hughes , a. m. , qi , c. , & dullemond , c. p. 2009 , , 700 , 1502 , h. h. , beichman , c. a. , gillett , f. c. , de jong , t. , houck , j. r. , low , f. j. , neugebauer , g. , walker , r. g. , & wesselius , p. r. 1984 , , 278 , l23 , s. a. , & papaloizou , j. c. b. 1999 , , 521 , 650 , m. r. 2000 , , 314 , 33 , s. v. w. , henning , t. , & nakagawa , y. 2000 , protostars and planets iv , 533 , k. r. , & lin , d. n. c. 1994 , , 427 , 987 , a. c. 2009 , , 695 , l53 , i. a. , & bate , m. r. 1994 , , 269 , l45 , a. p. 1997 , science , 276 , 1836 , g. , chen , x. , lin , d. n. c. , nelson , r. p. , & papaloizou , j. c. b. 1999 , , 514 , 344 , n. , dalessio , p. , hartmann , l. , wilner , d. , walsh , a. , & sitko , m. 2002 , , 568 , 1008 , j. m. , mamajek , e. e. , hillenbrand , l. a. , & meyer , m. r. 2006 , , 651 , l49 , g. , & baraffe , i. 2000 , , 38 , 337 , e. , kite , e. , kalas , p. , graham , j. r. , & clampin , m. 2009 , , 693 , 734 , e. , & youdin , a. 2009 , arxiv e - prints , e. i. , & goldreich , p. 1997 , , 490 , 368 , k. m. , & cassen , p. 1997 , , 477 , 398 , c. j. , harper - clark , e. , & lodato , g. 2007 , , 381 , 1543 , p. , lodato , g. , & clarke , c. j. 2009 , , 393 , 1157 , a. , masset , f. , & morbidelli , a. 2009 , , 705 , l148 , a. , morbidelli , a. , & masset , f. 2006 , icarus , 181 , 587 , p. , calvet , n. , & hartmann , l. 2001 , , 553 , 321 , g. , kley , w. , & henning , t. 2003 , , 586 , 540 , s. e. , veras , d. , ford , e. b. , & beichman , c. a. 2009 , , 707 , 79 , t. j. , liu , m. c. , & ireland , m. j. 2009 , , 692 , 729 , b. , clarke , c. j. , & drake , j. j. 2009 , , 699 , 1639 , d. c. , & murray - clay , r. a. 2008 , arxiv e - prints , t. p. , stone , j. m. , & hawley , j. f. 2000 , , 530 , 464 , s. , papaloizou , j. , lesur , g. , & heinemann , t. 2007 , , 476 , 1123 , c. f. 2001 , , 553 , 174 , p. , lithwick , y. , & sari , r. 2004 , , 42 , 549 , p. , & lynden - bell , d. 1965 , , 130 , 125 , g. 1997 , , 285 , 403 , u. , & hollenbach , d. 2009 , , 690 , 1539 , l. , & kenyon , s. j. 1996 , , 34 , 207 , l. a. , strom , s. e. , vrba , f. j. , & keene , j. 1992 , , 397 , 613 , r. , coffey , j. , scholz , a. , brandeker , a. , & van kerkwijk , m. h. 2006 , , 648 , 1206 , a. , youdin , a. , & klahr , h. 2009 , , 697 , 1269 , j. a. , fischer , d. a. , marcy , g. w. , wright , j. t. , driscoll , p. , butler , r. p. , hekker , s. , reffert , s. , & vogt , s. s. 2007 , , 665 , 785 , p. , graham , j. r. , chiang , e. , fitzgerald , m. p. , clampin , m. , kite , e. s. , stapelfeldt , k. , marois , c. , & krist , j. 2008 , science , 322 , 1345 , w. 2000 , , 313 , l47 , k. m. , & matzner , c. d. 2006 , , 373 , 1563 , k. m. , matzner , c. d. , & krumholz , m. r. 2008 , , 681 , 375 , k. m. , matzner , c. d. , krumholz , m. r. , & klein , r. i. 2010 , , 708 , 1585 , m. r. , klein , r. i. , & mckee , c. f. 2007 , , 656 , 959 , m. r. , klein , r. i. , mckee , c. f. , offner , s. s. r. , & cunningham , a. j. 2009 , science , 323 , 754 , m. r. , & thompson , t. a. 2007 , , 661 , 1034 , d. , marois , c. , doyon , r. , & barman , t. 2009 , , 694 , l148 , a .- gratadour , d. , chauvin , g. , fusco , t. , ehrenreich , d. , mouillet , d. , rousset , g. , rouan , d. , allard , f. , gendron , . , charton , j. , mugnier , l. , rabou , p. , montri , j. , & lacombe , f. 2009 , , 493 , l21 , m. h. , & peale , s. j. 2002 , , 567 , 596 , y. 2007 , , 374 , 515 , h. f. , & stewart , g. r. 2001 , icarus , 153 , 224 , d. n. c. , & papaloizou , j. 1986 , , 309 , 846 , d. n. c. , & papaloizou , j. c. b. 1993 , in protostars and planets iii , ed . e. h. levy & j. i. lunine , 749835 , j. j. 1987 , icarus , 69 , 249 , j. j. , hubickyj , o. , dangelo , g. , & bodenheimer , p. 2009 , icarus , 199 , 338 , g. , & rice , w. k. m. 2005 , , 358 , 1489 , s. h. , seibert , m. , & artymowicz , p. 1999 , , 526 , 1001 , c. , macintosh , b. , barman , t. , zuckerman , b. , song , i. , patience , j. , lafrenire , d. , & doyon , r. 2008 , science , 322 , 1348 , f. , & snellgrove , m. 2001 , , 320 , l55 , f. s. , & papaloizou , j. c. b. 2003 , , 588 , 494 , t. , & hanawa , t. 2003 , , 595 , 913 , c. d. , & levin , y. 2005 , , 628 , 817 , c. , & zuckerman , b. 2004 , , 127 , 2871 , s. a. , & hillenbrand , l. a. 2009 , , 181 , 62 , d. , & bjorkman , j. e. 2009 , , 702 , l163 , e. l. , & close , l. m. 2009 , arxiv e - prints , s. s. r. , klein , r. i. , mckee , c. f. , & krumholz , m. r. 2009 , , 703 , 131 , j. b. , hubickyj , o. , bodenheimer , p. , lissauer , j. j. , podolak , m. , & greenzweig , y. 1996 , icarus , 124 , 62 , j. b. , mckay , c. p. , & christofferson , b. m. 1985 , icarus , 64 , 471 , r. r. 2005 , , 621 , l69 . 2007 , , 662 , 642 . 2009 , , 704 , 281 , r. r. , & de colle , f. 2006 , , 646 , 275 , w. k. m. , lodato , g. , & armitage , p. j. 2005 , , 364 , l56 , v. s. 1960 , annales dastrophysique , 23 , 979 , d. , henning , t. , helling , c. , ilgner , m. , & sedlmayr , e. 2003 , , 410 , 611 , n. i. , & sunyaev , r. a. 1973 , , 24 , 337 , d. , & whitworth , a. p. 2009 , , 392 , 413 , a. 1964 , , 139 , 1217 , e. i. , & basu , s. 2006 , , 650 , 956 , s. , burkert , a. , whitworth , a. , naab , t. , & gritschneder , m. 2009 , , 400 , 13 , a. n. , & shu , f. h. 2002 , , 580 , 494 , b. , & song , i. 2009 , , 493 , 1149 @xcite has suggested that the cooling time might be altered in an irradiated disk . here we consider the cooling time for thermal perturbations to a disk , and show that a simple formula [ equation([eq - t_lum ] ) ] gives the cooling time for arbitrary levels of irradiation , at least in radiative optically thick disks . consider ambient radiation striking an optically thick disk with a normal flux @xmath153 ( where the numerical factor in the last definition is purely for later convenience ) . note that @xmath154 depends both on the irradiation field ( from the host star and/or light emitted and reflected from a surrounding envelope ) and also on the disk s surface geometry , e.g. flaring angle . by incorporating these variables into @xmath154 we attempt a general calculation . at the photosphere , where the optical depth to the disk s self - emission @xmath155 , energy balance gives : @xmath156 here , @xmath157 is the luminous flux from any internal sources of energy dissipation , e.g. viscous accretion , shocks , or the gravitational binding energy released by a collapsing fragment . the factor of two reflects that half of the radiation is emitted from each surface of the disk . since optical light is absorbed above the ir photosphere , about half the irradiation free streams out before it heats the disk @xcite . we can absorb this reduction into the definition of @xmath154 . we assume heat is transferred by radiative diffusion as : @xmath158 since convection is suppressed by irradiation and may be a negligible correction in any event @xcite . integration from the midplane at @xmath159 and @xmath160 to the photosphere gives : @xmath161 we now drop the subscripts from @xmath162 and the midplane @xmath163 , which we will soon take as the characteristic temperature ( ignoring order unity corrections from height averaging ) . furthermore we apply equation([eq - eb ] ) and take the @xmath164 limit ( meaning that we do nt need to know the precise location of the photosphere ) to express @xmath165 which shows that the midplane temperature is controlled by the larger of @xmath166 and @xmath167 . the cooling timescale to radiate away thermal fluctuations ( generated e.g. by gi ) is @xmath168 where a temperature perturbation @xmath169 has an excess heat @xmath170 , and @xmath171 is the specific heat for a mean molecular weight @xmath172 and adiabatic index @xmath18 . strongly compressive motions , which are not at constant pressure , will introduce order unity corrections that we ignore . the excess luminous flux , using equation([eq - f ] ) , is @xmath173 \nonumber \\ & = & { 32 \sigma t^3 \delta t \over 3 \tau } \times \left\ { \begin{array}{cl } ( 1 - \beta/4 ) & ~{\rm if}~ t \gg t_o , \beta \neq 4 \\ 1 & ~{\rm if } ~t \simeq t_o \end{array } \right.\ , , \end{aligned}\ ] ] where @xmath174 . the point is that the escaping flux varies by only an order unity factor between the strongly ( @xmath175 ) and weakly ( @xmath176 ) irradiated regimes . typical grain opacities , @xmath177 , ensure the correction is order unity ( and also ensure that we can ignore the catastrophic heating that would occur if @xmath178 ) . combining equations ( [ eq - t_lumdef ] ) and ( [ eq - deltaf ] ) with the definition of heat capacity we find that the cooling time is simply @xmath179 where the isothermal sound speed @xmath180 . from this derivation , we see that the cooling time obeys the simple form of equation([eq - t_lum ] ) for all levels of irradiation which only introduces an order unity @xmath54 correction . the cooling time depends on @xmath54 for weakly irradiated disks because changes to the opacity alter the amount of flux that escapes from the midplane . in highly irradiated disks , opacity changes have little effect because the small difference between the midplane and surface temperatures drives a weak flux . while increasing the irradiative flux incident on a disk decreases the cooling time by raising @xmath9 , it will not trigger fragmentation in a @xmath96 disk , since it increases @xmath29 . we will not explore optically thin or convective disks at this time . we assume that @xmath181 is the fragmentation criterion independent of irradiation . when the cooling time is longer , the disk is presumed to enter a state of gravito - turbulence @xcite . we take this term to mean a quasi - steady state of gravitationally driven turbulence , on scales @xmath182 wherein viscous dissipation of gi turbulence regulates @xmath96 . thus the cooling time can be translated to a the critical value of @xmath109 at which gravito - turbulent accretion disks will fragment . while @xcite find that an @xmath109-threshold is more robust than one for @xmath183 when the adiabatic index varies , they did not include irradiation , which we contend would reveal that cooling is ultimately the more physical criterion , but the issue is best settled by simulation . the emitted flux with an @xmath109-viscosity and @xmath96 gives @xmath184 which combined with equation([eq - f ] ) and equation([eq - t_lum ] ) without the @xmath54 correction gives @xmath185 when irradiation is weak enough that @xmath186 , we recover the standard @xmath187 criterion for fragmentation ( ignoring the accumulated order unity coefficients ) . however for stronger irradiation with @xmath188 , fragmentation occurs for @xmath189 , a lower threshold . gravito - turbulent models require modification when @xmath190 , i.e. an even lower level of irradiation than needed to affect the @xmath109 fragmentation threshold . in this case the disk shows some similarities to isothermal disks , and should have lower amplitude density perturbations , because there is insufficient viscous dissipation to support order unity thermal perturbations ( see the related discussion in @xcite ) . we note that simulations of isothermal disks do develop gi , exhibit gi - driven transport and fragment @xcite , but they do not appear particularly turbulent . we now show that viscous heating is relatively unimportant in the outer reaches of irradiated a - star disks ( see also section 3 of @xcite ) . for an optically thick disk with an ism opacity law , balancing viscous heating and emitted radiation gives @xmath191 the solution for the midplane temperature is @xmath192 lower than the irradiation temperatures shown in fig . [ fig - cartoon ] by more than a factor of four . note that the surface density falloff @xmath193 is also quite steep for a constant @xmath194 , viscous disk with the ism opacity law . if the disk is optically thin , then the balance between heating and cooling gives : @xmath195 this temperature profile is shallow , but still colder than the irradiation temperature at large radii where the disk becomes optically thin .

we present calculations of the exchange interactions and curie temperatures in cr - based pnictides and chalcogenides of the form crx with x = as , sb , s , se and te , and the mixed alloys crasx with x = sb , s , se , and te . the calculations are performed for zinc blende ( zb ) structure for 12 values of the lattice parameter between 5.44 and 6.62 , appropriate for some typical ii - vi and iii - v semiconducting substrates . electronic structure is calculated via the linear muffin - tin - orbitals ( lmto ) method in the atomic sphere approximation ( asa ) , using empty spheres to optimize asa - related errors . whenever necessary , the results have been verified using the full - potential version of the method , fp - lmto . the disorder effect in the as - sublattice for crasx ( x = sb , s , se , te ) alloys is taken into account via the coherent potential approximation ( cpa ) . exchange interactions are calculated using the linear response method for the ferromagnetic ( fm ) reference states of the alloys , as well as the disordered local moments ( dlm ) states . these results are then used to estimate the curie temperature from the low and high temperature side of the ferromagnetic / paramagnetic transition . estimates of the curie temperature are provided , based on the mean field and the more accurate random phase approximations . dominant antiferromagnetic exchange interactions for some low values of the lattice parameter for the fm reference states in crs , crse and crte prompted us to look for antiferromagnetic ( afm ) configurations for these systems with energies lower than the corresponding fm and dlm values . results for a limited number of such afm calculations are discussed , identifying the afm[111 ] state as a likely candidate for the ground state for these cases .

0912.1760

half - metallic ferromagnets with high ( room temperature and above ) curie temperatures @xmath1 are ideal for spintronics applications , and as such , much experimental and theoretical@xcite effort has been devoted in recent years to the designing of and search for such materials . among these , cr - doped dilute magnetic semiconductors ( dms ) @xcite or cr - based alloys and in particular cras and crsb @xcite in zinc blende(zb ) structure have attracted particular attention , not only because of the possibility of complete spin polarization of the carriers at the fermi level , but also for their possible high @xmath1 . akinaga @xcite were able to grow zb thin films of cras on gaas ( 001 ) substrates by molecular beam epitaxy , which showed ferromagnetic behavior at temperatures in excess of 400 k and magnetic moments of 3@xmath2 per cras unit . theoretical calculations by akinaga @xcite and several other theoretical calculations since then @xcite have verified the half - metallic character of cras . the high value of @xmath1 has also been supported by some of these studies@xcite . thin films of crsb grown by solid - source molecular beam epitaxy on gaas , ( al , ga)sb , and gasb have been found to be of zb structure and ferromagnetic with @xmath1 higher than 400 k@xcite . galanakis and mavropoulos@xcite , motivated by the successful fabrication of zb cras , crsb and mnas@xcite , have examined the possibility of half - metallic behavior in ordered zb compounds of transition metals v , cr and mn with the @xmath3 elements n , p , as , sb , s , se and te . their theoretical study shows that the half - metallic ferromagnetic character of these compounds is preserved over a wide range of lattice parameters . they also found that the half - metallic character is maintained for the transition element terminated ( 001 ) surfaces of these systems . yamana @xcite have studied the effects of tetragonal distortion on zb cras and crsb and found the half - metallicity to survive large tetragonal distortions . of course , the ground states of many of these compounds in the bulk are known to be different from the zb structure , the most common structure being the hexagonal nias - type . zhao and zunger@xcite have argued that zb mnas , cras , crsb , and crte are epitaxially unstable against the nias structure , and zb crse is epitaxially stable only for lattice constants higher than 6.2 , remaining half - metallic at such volumes . they also find that even though the ground state of crs is zb , it is antiferromagnetic at equilibrium lattice parameter , and thus not half - metallic . these results reveal the challenge experimentalists face in synthesizing these compounds in zb structure . however , the possibility remains open that such difficulties will be overcome with progress in techniques of film - growth and materials preparation in general . recently , deng @xcite were successful in increasing the thickness of zb - crsb films to @xmath4 3 nm by molecular beam epitaxy using ( in , ga)as buffer layers , and li @xcite were able to grow @xmath4 4 nm thick zb - crsb films on nacl ( 100 ) substrates . in view of the above situation regarding the state of experimental fabrication of these compounds and available theoretical results , it would be appropriate to study the variation of magnetic properties , particularly exchange interactions and the curie temperature , of cr - based pnictides and chalcogenides as a function of the lattice parameter . towards this goal , we have carried out such calculations for the compounds crx ( x = as , sb , s , se and te ) and the mixed alloys cras@xmath0x@xmath0 with x = sb , s , se and te . essentially we study the effect of anion doping by choosing elements of similar atomic sizes ( neighboring elements in the periodic table ) , one of which , namely sb , is isoelectronic to as , while the others ( s , se , te ) bring one more valence electron to the system . the mixed pnictide - chalcogenide systems offer further opportunity to study the effects of anion doping . the alloying with other 3d transition metals ( both magnetic , e.g. fe or mn , or non - magnetic , e.g. v ) on cation sublattice would also change the carrier concentration and bring about strong d - disorder which can additionally modify the shape of the fermi surface . this , however , is not the subject of the present paper . almost all theoretical studies on these alloys so far address aspects of electronic structure and stability of these alloys only . although a few theoretical estimates of exchange interactions and the curie temperature for cras at equilibrium lattice parameter have appeared in the literature , a detailed study of the volume dependence of these quantities is missing . for the other alloys , crsb , crs , crse and crte , no theoretical results for the exchange interaction , curie temperature and their volume dependence exist at present . the mixed pnictide - chalcogenide systems offer the possibility of not only creating these alloys over a larger range of the lattice parameter , but also with a larger variation in the exchange interactions . this is because at low values of the lattice parameter the dominant cr - cr exchange interactions in the chalcogenides can be antiferromagnetic , while for the pnictides they are ferromagnetic . the pnictide - chalcogenide alloying is important from the experimental viewpoint of stabilizing the zb structure on a given substrate , via the matching of the lattice parameter of the film with that of the latter . although the present study is confined to the zb structure only , we hope that it will provide some guidance to the experimentalists in their search and growth of materials suitable for spintronic devices . electronic and magnetic properties of crx ( x = as , sb , s , se and te ) and cras@xmath0x@xmath0 ( x= sb , s , se and te ) were calculated for lattice parameters varying between 5.45 and 6.6 , appropriate for some typical ii - vi and iii - v semiconducting substrates . calculations were performed using the tb - lmto - cpa method@xcite and the exchange - correlation potential given by vosko , wilk and nusair@xcite . in our lmto calculation we optimize the asa ( atomic sphere approximation ) errors by including empty spheres in the unit cell . we use the fcc unit cell , with cr and x ( as , sb , s , se and te ) atoms located at ( 0,0,0 ) and ( 0.25,0.25,0.25 ) , respectively , and empty spheres at locations ( 0.5,0.5,0.5 ) and ( -0.25,-0.25,-0.25 ) . for several cases , we have checked the accuracy of the lmto - asa electronic structures against the full - potential lmto results@xcite and found them to be satisfactory . for the mixed alloys cras@xmath0x@xmath0 ( x= sb , s , se and te ) , the as - sublattice of the zb cras structure is assumed to be randomly occupied by equal concentration of as and x atoms . the disorder in this sublattice is treated under the coherent potential approximation ( cpa)@xcite . our spin - polarized calculations assume a collinear magnetic model . in the following we will present results referred to as fm and dlm . the fm results follow from the usual spin - polarized calculations , where self - consistency of charge- and spin - density yields a nonzero magnetization per unit cell . although we call this the fm result , our procedure does not guarantee that the true ground state of the system is ferromagnetic , with the magnetic moments of all the unit cells perfectly aligned . this is because we have not explored non - collinear magnetic states , nor all antiferromagnetic ( afm ) states attainable within the collinear model . indeed , our results for the exchange interactions in some cases do suggest the ground states being of afm or complex magnetic nature . for lack of a suitable label , we refer to all spin - polarized calculations giving a nonzero local moment as fm state calculations . within the stoner model , a nonmagnetic state above the curie temperature @xmath1 would be characterized by the vanishing of the local moments in magnitude . it is well - known and universally accepted that the neglect of the transversal spin fluctuations in the stoner model leads to an unphysical picture of the nonmagnetic state and a gross overestimate of @xmath1 . an alternate description of the nonmagnetic state is provided by the disordered local moment ( dlm ) model , where the local moments remain nonzero in magnitude above @xmath1 , but disorder in magnitude as well as their direction above @xmath1 causes the global magnetic moment to vanish . combining aspects of the stoner model and an itinerant heisenberg - like model , heine and co - workers@xcite have developed a suitable criterion for a dlm state to be a more appropriate description of the nonmagnetic state than what is given by the stoner model . within the collinear magnetic model , where all local axes of spin - quantization point in the same direction , dlm can be treated as a binary alloy problem and thus described using the coherent potential approximation ( cpa)@xcite . we have carried out such dlm calculations , assuming the cr - sublattice to be randomly occupied by an equal number of cr atoms with oppositely directed magnetic moments . the object for performing the dlm calculations is two - fold . if the total energy in a dlm calculation is lower than the corresponding fm calculation , we can safely assume that the ground state ( for the given lattice parameter and structure ) is not fm , albeit of unknown magnetic structure . the exchange interactions in the dlm state can also be used to compute estimates of @xmath1 , and such estimates of @xmath1 may be considered as estimates from above the magnetic - nonmagnetic transition . @xmath1 computed from exchange interactions in the fm state are estimates from below the transition . of course , if the ground state is known to be ferromagnetic , then estimates of @xmath1 based on exchange interactions in the fm reference state are the appropriate ones to consider . in some cases where the fm results point to the possibility of the ground state magnetic structure being afm or of complex nature , we have carried out a limited number of afm calculations to provide some insight into this problem ( see section [ subsec : stabilityjq ] ) . we have computed the spin - resolved densities of states ( dos ) for all the alloys for varying lattice parameters , and for both the fm and dlm configurations . the fm calculations show half - metallic character , due to the formation of bonding and antibonding states involving the @xmath5 orbitals of the cr - atoms and the @xmath3 orbitals of the neighboring pnictogen ( as , sb ) or chalcogen ( s , se , te ) . the hybridization gap is different and takes place in different energy regions in the two spin channels . the critical values of the lattice parameters above which the fm calculations show half - metallic character agree well with those reported by galanakis and mavropoulos@xcite . the dos for the alloys of the type crx ( x = sb , s , se , te ) have been presented by several other authors@xcite and thus will not be shown here . in figs.[fig1 ] and [ fig2 ] we show the dos for the mixed alloys cras@xmath0sb@xmath0 and cras@xmath0se@xmath0 , for lattice parameters above and below the critical values for the half - metallic character . according to galanakis and mavropoulos@xcite , half - metallicity in zb cras appears between the lattice parameters of 5.45 and 5.65 . the latter corresponds to the lattice parameter of the gaas substrate . for crsb half - metallicity appears at a lattice parameter between 5.65 and 5.87 . the mixed alloy cras@xmath0sb@xmath0 , as shown in fig.[fig1 ] , is not quite half - metallic at the lattice parameter of 5.65 , and fully half - metallic at the lattice parameter of 5.76 . replacing sb with se in the above alloy , i.e. for cras@xmath0se@xmath0 , brings the critical lattice parameter down slightly . as shown in fig.[fig2 ] , at a lattice parameter of 5.65 , cras@xmath0se@xmath0 is half - metallic , although barely so . in our calculation crs and crse are half - metallic at a lattice parameter of 5.65 , and not so at a lattice parameter of 5.55 . crte is not half - metallic at a lattice parameter of 5.76 , but at a lattice parameter of 5.87 . for both crs and crse the critical value should be close to 5.65 , and for crte it should be close to 5.87 . note that in general the half - metallic gap is larger in the chalcogenides than in the pnictides . this is due to larger cr - moment ( see section [ sec : mag.mom . ] ) for the chalcogenides , which results in larger exchange splitting . this explains the difference in the half - metallic gaps in figs . [ fig1 ] and [ fig2 ] for similar lattice parameters . fig . [ fig3 ] compares the total dos of cras for the fm and dlm calculations for the equilibrium lattice parameter 5.65 . higher dos at the fermi level for the dlm calculation , compared with the fm calculation , is an indication that the band energy is lower in the fm state . indeed , as indicated in table [ table1 ] , compared with the dlm state the total energy for zb cras is lower in the fm state for the lattice parameters from 5.44 to 5.98 . in fact , this holds for lattice parameters up to 6.62 , showing the robustness of ferromagnetism in cras over a wide range of the lattice parameter . this is also true for crsb . sb@xmath0 for lattice parameters ( a ) 5.55 , ( b ) 5.65 , ( c ) 5.76 @xmath6 and ( d ) 5.87 , respectively . , width=360 ] se@xmath0 for lattice parameters ( a ) 5.55 , ( b ) 5.65 , ( c ) 5.76 @xmath6 and ( d ) 5.87 , respectively . , width=360 ] for the dlm and fm states.,width=360 ] in table [ table1 ] we show the variation of total energies per atom in ry with the lattice parameter for crx ( x = as , sb , s , se and te ) in the dlm and fm states . the purpose of tabulating these energies is not to determine the bulk equilibrium lattice parameters in the zb structure , as this has already been done by several authors@xcite . our results for equilibrium zb phase lattice parameters agree with those found by galanakis and mavropoulos@xcite . the important point is that for crs and crse at low values of lattice parameters the dlm energies are lower than the fm energies , showing clearly that the fm configuration is unstable . the result for crs is in line with the observation by zhao and zunger@xcite , who find zb crs to be antiferromagnetic with an equilibrium lattice parameter of 5.37 . as shown later ( section [ sec : exchange ] ) , the exchange coupling constants for the cr atoms in the fm calculations are negative , indicating the instability of the ferromagnetic spin alignment . the tendency to antiferromagnetism in crse at compressed lattice parameters is also revealed in a study by sasaio@xmath7lu @xcite for crte at lower lattice parameters the fm energy is lower than the dlm energy , but the exchange constants for the cr - atoms in the fm calculations are still negative ( see discussion in section [ sec : exchange ] ) , signaling the possibility of the ground states in crte at low values of the lattice parameter being neither dlm nor fm . note that in our discussion ground state implies the lowest energy state in zb structure . for crs , crse and crte the ground states at low lattice parameters can be of an antiferromagnetic ( afm ) or complex magnetic structure . a lower total energy may also mean a lower band energy , and in some cases , the latter may be reflected in a lower density of states at the fermi level . this is shown in fig.[fig4 ] , where for crs at the lowest lattice parameter of 5.44 @xmath6 the dos at the fermi level is lower in the dlm state than in the fm state . the deviation from ferromagnetism at low values of the lattice parameter for crs , crse and crte is also revealed by our study of the lattice fourier transform of the exchange interaction between the cr atoms in the fm state ( section [ sec : exchange ] ) . the search for an antiferromagnetic state with lower energy is possible within our collinear magnetic model by enlarging the unit cell in various ways . we have pursued this issue to a limited extent , by considering 001 , 111 afm configurations for crs , crse and crte at low values of the lattice parameter ( see discussion in section[sec : exchange ] ) . a satisfactory resolution of such issues is possible only by going beyond the collinear model . lcccccc lattice parameter ( ) & 5.44 & 5.55 & 5.65 & 5.76 & 5.87 & 5.98 + * cras * + dlm energy & -1653.4039 & -1653.4021 & -1653.3995 & -1653.3960 & -1653.3920 & -1653.3876 + fm energy & -1653.4068 & -1653.4055 & -1653.4034 & -1653.40026 & -1653.3966 & -1653.3922 + * crsb * + dlm energy & -3762.4738 & -3762.4781 & -3762.4807 & -3762.4821 & -3762.4822 & -3762.4814 + fm energy & -3762.4770 & -3762.4813 & -3762.4841 & -3762.4857 & -3762.4862 & -3762.4856 + + + * crs * + dlm energy & -723.4504 & -723.4468 & -723.4426 & -723.4381 & -723.4332 & -723.4280 + fm energy & -723.4491 & -723.4464 & -723.4434 & -723.4395 & -723.4351 & -723.4302 + * crse * + dlm energy & -1737.9146 & -1737.9139 & -1737.9123 & -1737.9100 & -1737.9071 & -1737.9036 + fm energy & -1737.9139 & -1737.9132 & -1737.9124 & -1737.9110 & -1737.9087 & -1737.9057 + * crte * + dlm energy & -3918.9074 & -3918.9124 & -3918.9158 & -3918.9179 & -3918.9189 & -3918.9188 + fm energy & -3918.9078 & -3918.9128 & -3918.9161 & -3918.9181 & -3918.9194 & -3918.9201 + our spin - polarized calculations for the fm reference states lead to local moments not only on the cr atoms , but also on the other atoms ( as , sb , s , se , and te ) as well as the empty spheres . sandratskii @xcite have discussed the problem associated with such induced moments in case of the heusler alloy nimnsb and the hexagonal phase of mnas . usually such systems can be divided into sublattices with robust magnetic moments and sublattices where moment is induced under the influence of the former . these authors argue that the treatment of the induced moments as independent variables in a heisenberg hamiltonian may lead to artificial features in the spin - wave spectra , but these artificial features do not drastically affect the calculated curie temperatures of the two alloys , nimnsb and hexagonal mnas . clearly , in our case the sublattice with the robust magnetic moment is the cr - sublattice . among the three other sublattices , the magnitudes of the induced moments decrease in the following order for the two robust ferromagnets cras and crsb : x - sublattice ( x = as , sb ) , sublattice es-1 ( the sublattice of empty spheres that is at the same distance with respect to the cr - sublattice as the x - sublattice ) , sublattice es-2 ( sublattice of empty spheres further away from the cr - sublattice ) . this trend is particularly valid for low values of the lattice parameter . the induced moments originate from the tails of the orbitals ( primarily @xmath8 ) on the nearby cr - atoms . this is particularly true for the moments induced on the empty spheres . the magnitudes of the induced moments on the two empty sphere sublattices decrease as the lattice parameter increases , and so do the differences in their magnitudes . the signs of the moments on es-1 and es-2 may be the same for small lattice parameters , but are opposite for large lattice parameters . the sign of the moment on the x - sublattice is opposite to that on the cr - sublattice and the magnitudes of the moments on the two sublattices increase with increasing lattice parameters , due to decreased hybridization between cr-@xmath8 and x-@xmath3 orbitals . above a critical value of the lattice parameter , the moment per formula unit ( f.u . ) saturates at a value of 3.0 @xmath2 , as the half - metallic state is achieved , while the local moments on the cr- and x - sublattices increase in magnitude , remaining opposite in sign . the maximum ratio between the induced moment on x ( x = as , sb ) and the moment on cr is 0.18 for cras and 0.15 for crsb , occurring at the highest lattice parameter of 6.62 @xmath6 studied . the maximum ratio between the induced moment on es-1 and that on cr is 0.06 , occurring at the lowest lattice parameter of 5.44 @xmath6 studied . magnetic moments of cras and crsb per formula unit ( f.u . ) as well as the local moment at the cr site are shown in fig . [ fig5 ] , where we compare the two compounds with each other for their magnetic moments in the fm and dlm states . the same results are presented in fig . [ fig6 ] , comparing the moment per f.u.in the fm state with the cr local moment in the fm and dlm states separately for each compound . it is to be noted that there are no induced moments for the dlm reference states , i.e. the moments on the non - cr sublattices are several orders of magnitude smaller than the robust moment on the cr atoms . the total moment per formula unit in the dlm state is zero by construction . the local moment on the cr atom for the dlm reference state is usually less than the corresponding fm value for smaller lattice parameters , and larger for larger lattice parameters ( fig.[fig6 ] ) . similar trends in the variation of the local moment on cr and the induced moments on the other sublattices for the fm reference states as a function of lattice parameter are revealed for crx ( x = s , se , te ) , except that the moments on es-1 are always an order of magnitude larger than those on es-2 . i addition , the induced moments on es-2 are @xmath9 times larger than those on x sublattice for smaller values of the lattice parameter , with the two becoming comparable in magnitude for larger lattice parameters . the induced moments on es-1 and x - sublattices are never larger than @xmath4 5% of the moment on the cr atoms . the induced moments for the dlm reference states are several orders of magnitude smaller than the cr - moments , and can be safely assumed to be zero . results for zb crs , crse and crte are presented in figs.[fig7 ] and [ fig8 ] . the moment per f.u . reaches the saturation values of 4@xmath2 for crs , crse , and crte in the half - metallic state , as discussed in detail by galanakis and mavropoulos@xcite . the saturation values of the moments for all these alloys ( crx , x = as , sb , s , se , and te ) satisfy the so - called `` rule of 8 '' : @xmath10 , where @xmath11 is the total number of valence electrons in the unit cell . the number 8 accounts for the fact that in the half - metallic state the bonding @xmath12 bands are full , accommodating 6 electrons and so is the low - lying band formed of the @xmath13 electrons from the @xmath3 atom , accommodating 2 electrons . the magnetic moment then comes from the remaining electrons filling the @xmath8 states , first the @xmath14 states and then the @xmath5 . the saturation value of 3@xmath2/f.u . , or the half - metallic state , appears for a larger critical lattice constant in crsb than in cras . similarly , the critical lattice constants for the saturation magnetic moment of 4@xmath2/f.u . are in increasing order for crs , crse and crte . the local moment on the cr atom can be less / more than the saturation value , depending on the moment induced on the non - cr atoms and empty spheres . -[fig8 ] ) , fm calculations produce induced moments on non - cr spheres representing the x - atoms ( x = as , sb , s , se , te ) , and one set of empty spheres . the dlm calculations produce no such induced moments , i.e. , the moments reside on the cr - atoms only . see text for discussion.,width=360 ] fig.[fig9 ] shows the variation of the magnetic moment with the lattice parameter for the random alloys cras@xmath0x@xmath0 ( x = sb , s , se , te ) , where 50% of the as - sublattice is randomly occupied by x - atoms . the saturation moment per f.u . for cras@xmath0sb@xmath0 in the half - metallic state is 3@xmath2 , with the results falling between those for cras and crsb shown in fig.[fig6 ] . for cras@xmath0x@xmath0 ( x = s , se , te ) , the saturation moment per f.u . is 3.5@xmath2 . the local cr - moment deviates from the saturation value in the half - metallic state , being higher than the saturation value for all lattice parameters above 6.1 . from figs.[fig5]-[fig9 ] it is clear that the magnetic moment per formula unit is closer to the magnetic moment of the cr atoms in the fm calculations than in the dlm calculations . local cr - moments in the dlm calculations are suppressed w.r.t . the fm results for low lattice parameters and enhanced for larger lattice parameters . as shown in table [ table1 ] the total energy of the fm state is lower than that of the corresponding dlm state in almost all cases , except for some compressed lattice parameters for crs and crse . however , the consideration of the dlm state does provide an advantage in that there are no associated induced moments , i.e. , the dlm calculations produce moments that reside on the robust magnetic sublattice only . mapping of the total energy on to a heisenberg hamiltonian , therefore , does not result in exchange interactions involving atoms / spheres with induced moments and all associated artificial / non - physical features referred to by sandratskii @xcite currently , most _ first - principles _ studies of the thermodynamic properties of itinerant magnetic systems proceed via mapping @xcite the system energy onto a classical heisenberg model : @xmath15 where @xmath16 are site indices , @xmath17 is the unit vector pointing along the direction of the local magnetic moment at site @xmath18 , and @xmath19 is the exchange interaction between the moments at sites @xmath18 and @xmath20 . the validity of this procedure is justified on the basis of the adiabatic hypothesis- the assumption that the magnetic moment directions are slow variables on all the characteristic electronic time scales relevant to the problem , and thus can be treated as classical parameters . the energy of the system for a given set of magnetic moment directions is usually calculated via methods based on density functional theory ( dft ) . one of the most widely used mapping procedures is due to liechtenstein @xcite it involves writing the change in the energy due to the deviation of a single spin from a reference state in an analytic form using the multiple scattering formalism and by appealing to the magnetic variant of the andersen force theorem@xcite . the force theorem , derived originally for the change of total energy due to a deformation in a solid , dictates that the differences in the energies of various magnetic configurations can be approximated by the differences in the band energies alone@xcite . the energy of a magnetic excitation related to the rotation of a local spin - quantization direction can be calculated from the spinor rotation of the ground state potential . no self - consistent calculation for the excited state is necessary . a second approach is based on the total energy calculations for a set of collinear magnetic structures , and extracting the exchange parameters by mapping the total energies to those coming from the heisenberg model given by eq.([e1 ] ) . such calculations can be done using any of the standard dft methods . however , unlike the magnetic force theorem method , where the exchange interactions can be calculated directly for a given structure and between any two sites , several hypothetical magnetic configurations and sometimes large supercells need to be considered to obtain the values of a modest number of exchange interactions . in addition , some aspects of environment - dependence of exchange interactions are often simply ignored . the difference between these two approaches is , in essence , the same as that between the generalized perturbation method ( gpm)@xcite and the connolly - williams method@xcite in determining the effective pair interactions in ordered and disordered alloys . a third approach is a variant of the second approach , where the energies of the system in various magnetic configurations corresponding to spin - waves of different wave - vectors are calculated by employing the generalized bloch theorem for spin - spirals@xcite . the inter - atomic exchange interactions can be calculated by equating these energies to the fourier transforms of the classical heisenberg - model energies . this approach , known as the frozen magnon approach , is similar to the frozen phonon approach for the study of lattice vibrations in solids . in this work , we have used the method of liechtenstein , which was later implemented for random magnetic systems by turek , using cpa and the tb - lmto method@xcite . the exchange integral in eq.([e1 ] ) is given by @xmath21 dz \ ; , \ ] ] where @xmath22 represents the complex energy variable , @xmath23 , and @xmath24 , representing the difference in the potential functions for the up and down spin electrons at site @xmath18. in the present work @xmath25 represents the matrix elements of the green s function of the medium for the up and down spin electrons . for sublattices with disorder , this is a configurationally averaged green s function , obtained via using the prescription of cpa . the integral in this work is performed in the complex energy plane , where the contour includes the fermi energy @xmath26 . the quantity @xmath19 given by eq . ( [ eq - jij ] ) includes direct- , indirect- , double - exchange and superexchange interactions , which are often treated separately in model calculations . the negative sign in eq.([e1 ] ) implies that positive and negative values of @xmath19 are to be interpreted as representing ferromagnetic and antiferromagnetic interactions , respectively . a problem with the mapping of the total energy to a classical heisenberg hamiltonian following the approach of liechtenstein @xcite is that it generates exchange interactions between sites , where one or both may carry induced moment(s ) . of course this problematic scenario appears only for the fm reference states , as the dlm reference states do not generate induced moments . in the present work the liechtenstein mapping procedure , applied to fm reference sates , generates exchange interactions between the cr atoms , between cr and other atoms x ( x = as , sb , s , se , te ) , and also between cr atoms and the nearest empty spheres es-1 . depending on the lattice parameter , this latter interaction is either stronger than or at least comparable to that for the cr - x pairs . the exchange interactions between cr atoms and the furthest empty spheres es-2 are always about one or two orders of magnitude smaller than the cr - es1 interactions and can be neglected . in cras , the ratio of the nearest neighbor cr - es1 to cr - cr interaction varies from 0.2 - 0.25 at low lattice parameters to 0.06 - 0.07 at high values of the lattice parameter . in crsb , these ratios are smaller , varying between 0.14 and 0.05 . the cr - es1 exchange interactions are also relatively strong in magnitude in crs , crse and crte . one important point is that while these interactions are positive for nearest neighbors for all lattice parameters , cr - cr nearest neighbor interaction is negative for low values of lattice parameters in crs and crse . in crte , this interaction changes sign from positive to negative and then back , as the lattice parameter is varied in the range 5.44 - 6.62 . as mentioned earlier , the calculation for the dlm reference states do not produce induced moments , and thus no exchange interactions other than those between the cr atoms . sandratskii@xcite have discussed the case when , in addition to the interaction between the strong moments , there is one secondary , but much weaker , interaction between the strong and one induced moment . in this case , the curie temperature , calculated under the mean - field approximation ( mfa ) , seems to be enhanced due to this secondary interaction , irrespective of the sign of the secondary interaction . in other words , the curie temperature would be somewhat higher than that calculated by considering only the interaction between the strong moments . the corresponding results under the random phase approximation ( rpa ) have to be obtained by solving two equations simultaneously . one can assume that the rpa results for the curie temperature follow the trends represented by the mfa results , being only somewhat smaller , as observed in the absence of induced moments . in our case , since there are at least two secondary interactions ( cr - x and cr - es1 ) to consider in addition to the main cr - cr interaction , the influence of these secondary interactions is definitely more complex . in view of the above - described situation involving secondary interactions between cr- and the induced moments for the fm reference states , we have adopted the following strategy . since no induced moments appear in calculations for the dlm reference states , the curie temperature @xmath1 for these can be calculated as usual from the exchange interaction between the cr - atoms , i.e. the strong moments . for these cases the calculation of @xmath1 can proceed in a straightforward manner by making use of the mean - field approximation ( mfa ) or the more accurate random - phase approximation ( rpa)@xcite . one can obtain the mfa estimate of the curie temperature from @xmath27 where the sum extends over all the neighboring shells . an improved description of finite - temperature magnetism is provided by the rpa , with @xmath28 given by @xmath29^{-1 } \ , .\ ] ] here @xmath30 denotes the order of the translational group applied and @xmath31 is the lattice fourier transform of the real - space exchange integrals @xmath32 . it can be shown that @xmath33 is always smaller than @xmath34 @xcite . it has been shown that the rpa curie temperatures are usually close to those obtained from monte - carlo simulations @xcite . as shown by sandratskii @xcite the calculation of @xmath1 using rpa is considerably more involved even for the case where only one secondary interaction needs to be considered , in addition to the principal interaction between the strong moments . the complexity of the problem increases even for mfa , if more than one secondary interaction is to be considered . the same comment applies to stability analysis using the lattice fourier transform of the exchange interactions . the deviation of the nature of the ground state from a collinear and parallel alignment of the cr moments in the fm reference states could be studied by examining the lattice fourier transform of the exchange interaction between the cr atoms : @xmath35 , if all the secondary interactions could be ignored . this is definitely not possible for many of our fm results , where several pairs of interaction need to be considered , and @xmath36 is a matrix bearing a complicated relationship to the energy as a function of the wave - vector * q*. thus , in the following the results for @xmath1 will be presented mostly for the dlm reference states . for comparison , in a small number of cases we will present @xmath1 calculated for the fm reference states using only the cr - cr exchange interactions as the input . of course , this will be done with caution only for cases where we have reason to believe that the results are at least qualitatively correct . some fm results will also be included towards the stability analysis based on @xmath36 derived from cr - cr interactions only . again , this will be done with caution , only if the corresponding results can be shown to be meaningful via additional calculations . the cr - cr exchange interactions for all the alloys studied and for both fm and dlm reference states become negligible as the inter - atomic distance reaches about three lattice parameters or , equivalently , thirty neighbor shells . the same applies to the cr - x and cr - es interactions for the fm cases , these interactions in general being somewhat smaller . the cr - cr interactions for the dlm reference states are more damped compared with the corresponding fm results , showing less fluctuations in both sign and magnitude . the distance dependence of the exchange interactions between the cr atoms in cras is shown in fig.[fig10 ] for several lattice parameters . although the nearest neighbor interaction is always positive ( i.e , of ferromagnetic nature ) , the interactions with more distant neighbors are sometimes antiferromagnetic . such antiferromagnetic interactions are more common in cras for lower lattice parameters . with increasing lattice parameter , interactions become predominantly ferromagnetic , and by the time the equilibrium lattice parameter of 5.52 @xmath6 is reached , antiferromagnetic interactions mostly disappear . we have calculated such interactions up to the 405th neighbor shell , which amounts to a distance of roughly 8 lattice parameters . although the interactions themselves are negligible around and after the 30@xmath37 neighbor shell , their influence on the lattice sums continues up to about 100 neighbor shells . by about the neighbor 110@xmath37 shell ( a distance of @xmath4 5 lattice parameters ) the interactions fall to values small enough so as not to have any significant effect on the calculated lattice fourier transform of the exchange interaction and the curie temperature ( see below ) . it is clear from fig.[fig10 ] that ferromagnetism in cras is robust and exists over a wide range of lattice parameters . the distance dependence of the cr - cr exchange interactions in crsb is very similar to that in cras for both fm and dlm reference states . for crs , crse , and crte the situation is somewhat different . for crs and crse , the fm reference states for some low lattice parameters yield cr - cr interactions that are antiferromagnetic even at the nearest neighbor separation . for crte , at the lowest lattice parameter studied ( 5.44 ) the nearest neighbor interaction for the fm reference state is ferromagnetic , but becomes antiferromagnetic with increasing lattice parameter , changing back to ferromagnetic at higher lattice parameters . for all three compounds , the interactions are predominantly ferromagnetic at higher lattice parameters . figs.[fig11 ] and [ fig12 ] show the distance dependence of the exchange interactions calculated for the fm reference states in crse and crte , respectively , for several lattice parameters . predominant nearest neighbor antiferromagnetic interactions between the cr atoms result in negative values of the curie temperature , when calculated via eqs . ( [ e2 ] ) or ( [ e3 ] ) . these results for the curie temperature for the fm reference states can be discarded as being unphysical on two grounds : because of the neglect of the interactions involving the induced moments and also because they point to the possibility that the ground state is most probably antiferromagnetic or of complex magnetic structure . the antiferromagnetic cr - cr interactions mostly disappear , when calculated for the dlm reference states . this could be interpreted as being an indication that the actual magnetic structure of the ground states for these low lattice parameters in case of crs , crse and crte is closer to a dlm state than to an fm state . in fig.[fig13 ] we show the cr - cr exchange interactions for the dlm reference states in case of crse for the same lattice parameters as those considered for fig.[fig11 ] . a comparison of the two figures shows that all interactions have moved towards becoming more ferromagnetic for the dlm reference states , the nearest neighbor interaction for the lowest lattice parameter staying marginally antiferromagnetic . between the cr atoms in cras for various lattice parameters @xmath38 , calculated for the fm and dlm reference states . the distance between the cr atoms is given in units of the lattice parameter @xmath38 ( the same applies to figs.[fig11]-[fig13 ] ) . the main plot in fig.[fig10 ] shows the distance dependence up to 2.25@xmath38 , while the inset shows the values between 2.25@xmath38 and 5@xmath38 . although the individual values of @xmath19 are small beyond about 2.25@xmath38 , their cumulative effects on the total exchange constant and the curie temerature can not be neglected ( see text for details ) . comparison of the insets for the fm and dlm cases shows that the interactions are more damped for the dlm case , being at least an order of magnitude smaller for distances beyond @xmath42.25 - 2.5@xmath38 or 15 - 20 neighbor shells . similar comments apply to the interactions presented in figs.[fig11]-[fig13].,width=379 ] . , width=336 ] . , width=336 ] . , width=336 ] the deviation of the nature of the ground state from the reference state can be studied by examining the lattice fourier transform of the corresponding exchange interactions between the cr atoms : @xmath35 . as pointed out earlier , for the fm reference states this procedure suffers from the drawback of neglecting the effects of all other interactions involving the induced moments . for the dlm reference states there are no induced moments , so the relationship between the energy and @xmath36 is simpler , but a physical picture of the spin arrangement corresponding to a particular wave - vector @xmath39 is harder to visualize . for the fm reference states , if there were no moments other than those on the cr atoms , a maximum in @xmath36 at @xmath40 would imply that the ground state is ferromagnetic with collinear and parallel cr magnetic moments in all the unit cells . a maximum at symmetry points other than the @xmath41-point would imply the ground state being antiferromagnetic or a spin - spiral state . a maximum at a wave - vector @xmath39 that is not a symmetry point of the bz would imply the ground state being an incommensurate spin spiral . the presence of induced moments and the consequent interactions involving non - cr atoms and empty spheres spoil such interpretations based on @xmath36 derived from cr - cr interactions alone . however , the tendencies they reveal might still be useful . it is for this reason that we study the fourier transform @xmath36 , defined above , for both fm and dlm reference states . in fig.[fig14 ] we have plotted this quantity for cras . the results for crsb are quite similar . the maximum in @xmath36 at the @xmath41-point for all lattice parameters and for both fm and dlm reference states can be taken as an indication that the ground state magnetic structure is ferromagnetic for cras for all the lattice parameters studied . the same comment applies to crsb . the apparent lack of smoothness in @xmath36 shown for the fm reference states is a consequence of the fact that there are other additional bands ( involving induced moments ) , which are supposed to cross the band shown , but have not been computed . for crs , crse , and crte ( see figs.[fig15]-[fig16 ] ) , the deviation of the ground state for low lattice parameters from the parallel arrangement of cr moments is reflected in the result that the maximum moves away from the @xmath41-point for the fm reference states . at high values of the lattice parameter the maximum returns to the @xmath41-point . the curves for crse are similar to those for crs and have therefore not been shown . the fact that the maximum for the dlm reference sates lies at the @xmath41-point in most cases is again an indication that the ground state magnetic structure is closer to the dlm state than to the fm state . the conclusions based on the fm reference state results in figs.[fig15 ] and[fig16 ] may be suspect on ground of neglecting the interactions involving the induced moments . however , to explore whether they do carry any relevant information we have carried out additional calculations for the three compounds crs , crse and crte for two commonly occurring antiferromagnetic configurations : afm[001 ] , afm[111 ] . note that another commonly occurring afm configuration afm[110 ] is not unique , i.e. there are several configurations that could be seen as an afm[110 ] arrangement ( see fig3 . of ref . [ ] , table 2 of ref . the simplest among these is actually equivalent to afm[100 ] . the results for the total energy for the two afm calculations are shown in table [ table2 ] and compared with the corresponding fm and dlm total energies . for crs , the lowest energy state for lattice parameters 5.44 and 5.55 @xmath6 is afm[111 ] , exactly as suggested by the maximum in @xmath36 appearing at the l - point in fig.[fig15 ] for the fm reference state and for these two lattice parameters . as the lattice parameter increases beyond 5.55 , antiferromagnetic interactions diminish . for the next higher lattice parameter 5.66 @xmath6 in table [ table2 ] , the lowest energy state is dlm . this may suggest that the ground state has a complex magnetic structure , which remains to be explored . for higher lattice parameters the fm state has the lowest energy . for crse , afm[111 ] state has the lowest energy up to the lattice parameter 5.66 , as is also supported by the maximum of @xmath36 at l - point . the @xmath36 curves for crse are similar to those of crs , and have not been shown . for crte , at the lowest lattice parameter of 5.44 @xmath6 the lowest energy state is fm , as is also indicated by the maximum of @xmath36 at the @xmath41-point . for higher lattice parameters 5.65 and 5.76 , even though the @xmath36 curves point to the possibility of an afm[111 ] ground state , the fm state energy turns out to be the lowest among the configurations studied . it could be concluded that in this case a proper relationship between the energy and @xmath36 , obtained without the neglect of the induced moments , would point to the ground state being fm . for these three chalcogenides , for lattice parameters above 5.65 - 5.7 @xmath6the ground state should be fm . [ cols="<,^,^,^,^,^,^ " , ] we determine the curie temperature using eqs . ( [ e2 ] ) and ( [ e3 ] ) . for the dlm reference states , these produce estimates of @xmath1 from above the ferromagnetic@xmath42paramagnetic transition , and are free from errors due to induced moments . however , these estimates are high compared with properly derived values of @xmath1 from below the transition . the latter estimates would require the use of fm reference states ( where the ground states are known to be fm ) and thus a proper treatment of the induced moments . for cras and crsb , the magnetic state is ferromagnetic for all the lattice parameters considered . hence , for the sake of comparison we have calculated the @xmath1 for the fm reference states using eqs . ( [ e2 ] ) and ( [ e3 ] ) as well . according to the results of sandratskii @xcite the correctly calculated @xmath1 values , in the presence of interactions involving all the induced moments , would be higher . thus , the correct estimates of @xmath1 should lie somewhere between the dlm results and the fm results obtained with the neglect of the induced moments . in fig . [ fig17 ] we show these results for cras , crsb . we have used up to 111 shells in the evaluation of eq . ( [ e2 ] ) and for the lattice fourier transform of @xmath43 in eq . ( [ e3 ] ) , after having tested the convergence with respect to the number of shells included . the estimated computational error corresponding to the chosen number of shells used in these calculations is below @xmath44 . for comparison we also include the results for the mixed alloy cras@xmath0sb@xmath0 , for which the calculated @xmath1 values fall , as expected , in between those of cras and crsb . since rpa values are more accurate than mfa values , our best estimates of @xmath1 for cras range from somewhat higher than 500 k at low values of the lattice parameter , increasing to 1000 - 1100 k around the mid lattice parameter range ( 5.75 - 5.9 ) and then decreasing to around 600 k for higher lattice parameters ( 6.5 @xmath6 and above ) . for crsb these estimates are consistently higher than those for cras : 1100 k , 1500 k and 1200 k , respectively . the estimates for cras are similar to those provided by sasaiolglu @xcite for crs , crse , the results obtained with the fm reference states would clearly be wrong , in particular , for the low values of the lattice parameters , for which we have shown the ground state to be antiferromagnetic within our limited search . there is a possibility that the ground state for certain lattice parameters might have a complex magnetic structure . for crte , even though the ground state appears to be ferromagnetic , there are considerable antiferromagnetic spin fluctuations , making the fm estimates unreliable . in fig . [ fig18 ] we show the @xmath1 values for crs , crse and crte for the dlm reference states . the values for lattice parameters for which the ground state has been shown to be antiferromagnetic in the preceding section should be discarded as being inapplicable . similar results for the alloys cras@xmath0x@xmath0 ( x = s , se and te ) are shown in fig . [ fig19 ] for dlm reference states . for these , the ground state is ferromagnetic for all lattice parameters . however , because of the neglect of the induced moments related effects , our results for the curie temperatures for the fm reference states are lower than the properly calculated values . thus in fig.[fig19 ] we show the dlm results only , which are devoid of the induced moment effects and provide us with estimates of @xmath1 from above the transition . these are expected to be somewhat higher than the properly computed values for fm reference states . thus , for these alloys the trend revealed in fig . [ fig19 ] for the variation of @xmath1 with lattice parameter is correct . the estimates themselves are qualitatively correct , albeit somewhat higher than the correct values . only the rpa values are plotted in fig . [ fig19 ] , which are more reliable than the mfa values . for comparison , we also show the results for the pnictides cras , crsb , and cras@xmath0sb@xmath0 , which are isoelectronic among themselves , but have half an electron per unit cell less than the mixed alloys cras@xmath0x@xmath0 ( x = s , se and te ) sb@xmath0 are also shown . , width=360 ] x@xmath0 alloys with x = s , se and te . for comparison , the results for cras , crsb and cras@xmath0sb@xmath0 are also shown . all results shown are for dlm reference states , and as such , should be considered as upper limits for @xmath1 . , width=336 ] the differences between the results for the pnictides , chalcogenides and the mixed pnictide - chalcogenides can be summarized as follows . the pnictides , cras , crsb , and cras@xmath0sb@xmath0 , are strong ferromagnets at all the lattice parameters studied ( 5.44 - 6.62 ) . in the dlm description , their @xmath1 stays more or less constant ( apart from a minor increase ) as the lattice parameter increases from 5.4 /aa to 6.1 , and then decreases beyond ( figs . [ fig17 ] and [ fig19 ] . the chalcogenides are antiferromagnetic or have complex magnetic structure for low lattice parameters . in the dlm description , their @xmath1 in the ferromagnetic state increases and then becomes more or less constant as the lattice parameter increases ( fig . [ fig18 ] ) . the mixed alloys cras@xmath0x@xmath0 ( x = s , se , te ) are ferromagnetic at all the lattice parameters studied . in the dlm description , their @xmath1 rises and then falls as the lattice parameter is increased from 5.44 to 6.62 . a comparison of the results presented in figs . [ fig17]-[fig19 ] shows that large changes in @xmath1 take place by changing the number of carriers . changes due to isoelectronic doping are small compared with changes brought about by changing carrier concentration . our _ ab initio _ studies of the electronic structure , magnetic moments , exchange interactions and curie temperatures in zb crx ( x = as , sb , s , se and te ) and cras@xmath0x@xmath0 ( x = sb , s , se and te ) reveal that half - metallicity in these alloys is maintained over a wide range of lattice parameters . the results for the exchange interaction and the curie temperature show that these alloys have relatively high curie temperatures , i.e. room temperature and above . the exceptions occur for the alloys involving s , se and te at some low values of lattice parameters , where significant inter - atomic antiferromagnetic exchange interactions indicate ground states to be either antiferromagnetic or of complex magnetic nature . a comparison of total energies for the fm , dlm , and two zb antiferromagnetic configurations ( afm[001 ] and afm[111 ] ) show the lowest energy configuration to be afm[111 ] for crs and crse for compressed lattice parameters ( table[table2 ] ) . the possibility of afm ground states for compressed lattice parameters for crs was noted by zhao and zunger@xcite and for crse by sasioglu @xcite . our search for the antiferromagnetic ground states is more thorough than what was reported in these two studies . an extensive study of several antiferromagnetic configurations as well as ferrimagnetic and more complex magnetic structures for crs , crse and crte is currently underway . the mixed pnictide - chalcogenide alloys cras@xmath0x@xmath0 ( x= s , se , te ) do not show any tendency to antiferromagnetic spin fluctuations for the entire range of the lattice parameter studied . presumably the pnictogens suppress antiferromagnetic tendencies . such alloys may play an important role in fabricating stable zb half - metallic materials , as the concentration of the pnictogens and the chalcogens may be varied to achieve lattice - matching with a given substrate . as long as the concentration of as or sb is higher than the chalcogen concentration , half - metallic ferromagnetic state can be achieved . there is a large variation in the curie temperature of these alloys ( fig . [ fig19 ] ) as the lattice parameter varies from the low ( @xmath4 5.4 ) to the mid ( @xmath4 6.1 ) range of the lattice parameters studied . this variation is much smaller for the isoelectronic alloys cras , crsb and cras@xmath0sb@xmath0 ( fig . [ fig17 ] ) over this range of lattice parameters . note that most ii - vi and iii - v zb semiconductors have lattice parameters in this range . large changes in @xmath1 can be brought about by changing the carrier concentrations . the pnictides in general have a higher @xmath1 than the chalcogenides . our results for the curie temperature , the lattice fourier transform of the exchange interactions , and the resulting stability analysis are based on the exchange interactions between the cr atoms only . for the fm reference states this causes some errors due to the neglect of the effects of the induced moments . the dlm results are free from such errors . it is expected that the present study will provide both qualitative and quantitative guidance to experimentalists in the field . 99 k. sato and h. katayama - yoshida , semicond . * 17 * , 367 ( 2002 ) . see k. sato , t. fukushima and h. katayama - yoshida , j. phys . : matter * 19 * , 365212 ( 2007 ) , and references therein . see b. belhadji , l. bergqvist , r. zeller , p.h . dederichs , k. sato and h. katayama - yoshida , j. phys . : condens . matter * 19 * , 436227 ( 2007 ) , and references therein . h. saito , v. zayets , s. yamagata , and k. ando , phys . lett * 90 * , 207202 - 1 ( 2003 ) . k. sato and h. katayama - yoshida , jpn . . phys . * 40 * , l651 ( 2001 ) . h. akinaga , t. manago , and m. shirai , jpn . j. appl . phys.*39 * , l1118 ( 2000 ) . s. li , j - g duh , f. bao , k - x liu , c - l kuo , x. wu , liya l , z. huang , and y du , j. phys . phys . * 41 * 175004 ( 2008 ) . m. shirai , j. appl . phys . * 93 * , 6844 ( 2003 ) . h. akinaga , m. mizuguchi , k. nagao , y. miura , and m. shirai in _ springer lecture notes in physics _ * 676 * , 293 - 311 ( springer - verlag , berlin 2005 ) . k. yamana , m. geshi , h. tsukamoto , i. uchida , m. shirai , k. kusakabe , and n. suzuki , j. phys . : condens . matter * 16 * , s5815 ( 2004 ) . l. kahal , a. zaoul , m. ferhat , j. appl . phys . * 101 * , 093912 ( 2007 ) . i. galanakis and p. mavropoulos , , 104417 ( 2003 ) ; see also i. galanakis , , 012406 ( 2002 ) . pask , l.h . yang , c.y . fong , w.e . pickett , and s. dag , , 224420 ( 2003 ) . t. ito , h. ido , and k. motizuki , j. mag . mag . mat . * 310 * , e558 ( 2007 ) . l - j shi and b - g liu , j. phys . : condens . matter * 17 * , 1209 ( 2005 ) . m. zhang , j. phys . : matter * 15 * , 5017 ( 2003 ) . j. kbler , , 220403(r ) ( 2003 ) . b. sanyal , l. bergqvist , and o. eriksson , , 054417 ( 2003 ) . xie , y - q . xu , b - g . liu , and d.g . pettifor , 037204 ( 2003 ) . zhao , f. matsukura , k. takamura , e. abe , d. chiba , and h. ohno , appl . lett . * 79 * , 2776 ( 2001 ) . k. ono , j. okabayashi , m. mizuguchi , m. oshima , a. fujimori , and h. akinaga , j. appl . phys . * 91 * , 8088 ( 2001 ) . zhao and a. zunger , , 132403 ( 2005 ) . deng , j.h . zhao , j.f . niu , f.h . yang , x.g . wu , and h.z . zheng , j. appl . phys . * 99 * , 093902 ( 2006 ) . j. kudrnovsk and v. drchal , , 7515 ( 1990 ) . i. turek , v. drchal , j. kudrnovsk , m. ob , and p. weinberger , _ electronic structure of disordered alloys , surfaces and interfaces _ ( kluwer , boston - london - dordrecht , 1997 ) . vosko , l. wilk , and m. nusair , can . j. phys . * 58 * , 1200 ( 1980 ) . savrasov , and d.yu . savrasov , , 12181 ( 1992 ) . v. heine , j.h . samson , and c.m.m . nex , j. phys . phys . * 11 * , 2645 ( 1981 ) . v. heine and j.h . samson , j. phys . * 13 * , 2155 ( 1983 ) . h. hasegawa , j. phys . jpn . * 46 * , 1504 ( 1979 ) . pettifor , j. magn . mater * 15 - 18 * , 847 ( 1980 ) . staunton , b.l . gyorffy , a.j . pindor , g.m . stocks , and h. winter , j. phys . f * 15 * , 1387 ( 1985 ) . pindor , j. staunton , g.m . stocks , h. winter , j. phys . f * 13 * , 979 ( 1983 ) . e. sasaio@xmath7lu , i. galanakis , l.m . sandratskii , and p. bruno , j. phys : condens . matter * 17 * 3915 ( 2005 ) . sandratskii , r. singer , and e. sasiolu , , 184406 ( 2007 ) . m. pajda , j. kudrnovsk , i. turek , v. drchal , and p. bruno , phys . b * 64 * , 174402 ( 2001 ) . liechtenstein , m.i . katsnelson and v.a . gubanov , j. phys.f : met.phys . * 14 * , l125 ( 1984 ) . a. i. liechtenstein , m. i. katsnelson , v. p. antropov , v. a. gubanov , j. magn . magn . mater . * 67 * , 65 ( 1987 ) . liechtenstein , m.i . katsnelson and v.a . gubanov , solid.state.commun . * 51 * , 1232 ( 1984 ) . a.i . liechtenstein , m.i . katsnelson , v.p . antropov and v.a . gubanov , j.magn.magn.mater . * 21 * , 35 ( 1988 ) . gubanov , a.i . liechtenstein , a.v . postnikov _ magnetism and the electronic structure of crystals _ , edited by m. cardona , p. fulde , k. von klitzing , h .- j . queisser ( springer , berlin , 1992 ) . see , e.g. , o.k . andersen , o. jepsen , and d. gltzel , in _ highlights of condensed matter theory _ , edited by f. bassani ( north - holland , amsterdam , 1985 ) , p.59 . v. heine , _ solid state physics _ * 35 * ( academic press , new york ) , 1 ( 1980 ) . a. oswald , j. phys . f * 15 * , 193 ( 1985 ) . f. ducastelle , `` order and phase stability in alloys '' ( north - holland , amsterdam , 1991 ) . m. sluiter , and p.e . a. turchi , phys . b * 40 * , 11215 ( 1989 ) . connolly and a.r . williams , phys . b * 27 * , 5169 ( 1983 ) . z.w . lu , s .- h . wei , a. zunger , s. frota - pessoa , and l.g . ferreira , phys . b * 44 * , 512 ( 1991 ) . l.m . sandratski , j. phys . : condens . matter 3 , 8565 ( 1991 ) . i. turek , j. kudrnovsk , v. drchal , and p. bruno , philos . mag . * 86 * , 1713 ( 2006 ) . wang , r.e . prange , and v. korenman , phys . b * 25 * , 5766 ( 1982 ) . j. rusz , l. bergqvist , j. kudrnovsk , and i. turek , phys . b * 73 * , 214412 ( 2006 ) . l. jiang , q. feng , y. yang , z. chen , and z. huang , sol . . comm . * 139 * , 40 ( 2006 ) . s. curtarolo , d. morgan and g. ceder , comp . phase diagrams and thermochemistry * 29 * , 163 ( 2005 ) .

we report significant anisotropies in the projected two - dimensional ( 2d ) spatial distributions of globular clusters ( gcs ) of the giant virgo elliptical galaxy ngc4649 ( m60 ) . similar features are found in the 2d distribution of low - mass x - ray binaries ( lmxbs ) , both associated with gcs and in the stellar field . deviations from azimuthal symmetry suggest an arc - like excess of gcs extending north at 4 - 15 kpc galactocentric radii in the eastern side of major axis of ngc4649 . this feature is more prominent for red gcs , but still persists in the 2d distribution of blue gcs . high and low luminosity gcs also show some segregation along this arc , with high - luminosity gcs preferentially located in the southern end and low - luminosity gcs in the northern section of the arc . gc - lmxbs follow the anisotropy of red - gcs , where most of them reside ; however , a significant overdensity of ( high - luminosity ) field lmxbs is present to the south of the gc arc . these results suggest that ngc4649 has experienced mergers and/or multiple accretions of less massive satellite galaxies during its evolution , of which the gcs in the arc may be the fossil remnant . we speculate that the observed anisotropy in the field lmxb spatial distribution indicates that these x - ray binaries may be the remnants of a star formation event connected with the merger , or maybe be ejected from the parent red gcs , if the bulk motion of these clusters is significantly affected by dynamical friction . we also detect a luminosity enhancement in the x - ray source population of the companion spiral galaxy ngc4647 . we suggest that these may be younger high mass x - ray binaries formed as a result of the tidal interaction of this galaxy with ngc4649 .

1401.2994

recent work is bringing forth a picture of complex and diverse globular cluster ( gc ) populations in elliptical galaxies , consistent with several merging episodes of which the different gc populations are the fossil remnant ( @xcite , for m87 ; @xcite , for ngc4365 ) . our work on the elliptical galaxy ngc4261 has shown large - scale spiral - like features in the two - dimensional ( 2d ) gc distribution @xcite , which would have been unreported in the study of radial distributions . this continuing merging evolution would be expected in the framework of simulations based on the @xmath0cdm model of hierarchical galaxy formation @xcite . the @xmath0cdm model of hierarchical galaxy formation @xcite has been successful in reproducing several results of both deep and large sky surveys ( e.g. , @xcite , @xcite , @xcite ) . however , direct observational validation of continuous galaxy evolution via satellite merging from observations of galaxies in the local universe is only recently appearing in the literature , and has been mostly limited to the milky way and local group . large - scale surveys of the stellar population of the milky way have shown an increasing complex dynamical environment of satellite dwarf galaxies interacting gravitationally with and being accreted by our galaxy ( e.g. the sloan digital sky survey results of @xcite ) . the sagittarius stream @xcite is a well - studied example of this phenomenon . a wide - area survey of the m31-m33 region confirms this picture , detecting streams and dwarf galaxies in the halo of m31 and in the region between m31 and m33 @xcite . although these subtle details may not be perceivable beyond our local group , gcs may provide a marker of these interactions in farther away systems . observationally , for example , the sagittarius dwarf / stream is associated with several gcs ( e.g. , @xcite , @xcite ) . although the parent dwarf galaxy may be disrupted , and therefore hard to be detected against the background of the more luminous dominant galaxy , gc may retain the information of the encounter . this is suggested by simulations - albeit limited to the tidal evolution of gcs in dwarf spheroidal dark matter halos - which show that gcs with combined high mass and high density will survive the encounter @xcite . the giant elliptical galaxy ngc4649 ( m60 ) in the virgo cluster was recently surveyed with several chandra and hst exposures , completely covering the area within the d25 ellipse @xcite and extending with variable coverage to larger radii . the resulting catalogs of gcs @xcite and low mass x - ray binaries ( lmxbs ) @xcite give us a unique opportunity to study the properties of these populations and to explore further the gc - lmxb connection ( see review , * ? ? ? * ) , with full spatial coverage . ngc4649 is a relatively isolated galaxy , except for the close neighbor spiral ngc4647 , with which ngc4649 may be tidally interacting @xcite . in the first follow - up paper based on these chandra and hst data @xcite , we studied the radial distributions of the gcs and lmxbs populations and compared them with that of the optical surface brightness , which is a good proxy of the stellar mass for the old stellar population of ngc4649 . the radial distributions of red and blue gcs differ , as generally observed in other galaxies ( see review , * ? ? ? * ; * ? ? ? the blue gc radial profile is definitely wider than that of the stellar light . the red gcs profile is steeper , close to that of the stellar light , with a noticeable `` dip '' in the denser centermost region ( r@xmath1 3 kpc at the ngc4649 distance of 16.5 mpc @xcite ) , where tidal disruption of gcs may be more efficient . the lmxbs associated with gcs follow the same radial distributions as their parent gc populations . roughly three times more of these lmxbs are found in red rather than in blue gcs , as generally observed in early - type galaxies ( @xcite ; @xcite ; @xcite ; see @xcite for a theoretical explanation ) . the field lmxbs - those without a gc counterpart - follow the stellar light , i.e. , the mass distribution of old stars , within the @xmath2 ellipse @xcite . @xcite also report a departure of the lmxb radial distribution from that of the stellar mass at larger galactocentric radii , which could also be associated with more luminous sources . as discussed there , this over - luminosity may point to a younger field lmxb population : evolutionary models of native field lmxb populations have shown that lmxbs become fainter with the increasing age of the parent stellar population @xcite . here we complete our study of the spatial distribution of the gc populations in ngc4649 , by examining their 2d observed distributions , to explore the presence of irregularities that may point to mergers or satellite accretion by ngc4649 . given the association and possible evolutionary relation between gcs and lmxbs ( see * ? ? ? * and refs . therein ) , we also extend this study to the lmxb population of ngc4649 . in particular we seek to explore if the regularity observed in the radial distributions of these sources still persists in the azimuthal dimension . in section 2 we describe the catalogs of gcs and lmxbs used in this paper . these data were analyzed with the method described in @xcite . this method is based on the k - nearest neighbor ( knn ) density estimator @xcite , augmented by monte carlo simulations , to estimate the statistical significance of any spatial feature ( section [ sec : method ] ) . the results are presented in section [ sec : densityresidual ] and discussed in section [ sec : discussion ] . we summarize our finding in section [ sec : conclusions ] . in this paper we adopt a distance of 16.5 mpc to ngc4649 @xcite . at this distance , an angular separation of 1@xmath3 corresponds to a linear distance of @xmath44.8 kpc . the gc and lmxb samples used in this work are the results of the joint chandra - hst large - area mapping of ngc4649 ( pi fabbiano ) . the gc sample is based on the gc catalog extracted by @xcite from six hst acs pointings of ngc4649 . it contains 1603 gcs with photometry in @xmath5 and @xmath6 filters , selected by requiring @xmath7 and @xmath8 and discarding all sources with size consistent with the psf ( presumed to be foreground stars ) . this catalog includes 841 red gcs with @xmath9 , and 762 blue gcs with @xmath10 . the color threshold was derived from @xcite after verifying that significant color substructures are visible in the whole interval of galactocentric distances explored by the catalog of gcs . while the shapes and positions of the red and blue features in the color distribution of gcs ( see figure 6 in @xcite ) show a slight dependence on the galactocentric distance within ngc4649 , the @xmath11 value allows a clear separation of the two gcs color classes over the whole range of distances . for our analysis we have also considered a split of the sample in luminosity , resulting in 467 high - luminosity gcs with @xmath12 , and 1136 low - luminosity gcs with @xmath13 . as discussed later in section [ sec : method ] , small variations of both color and luminosity boundaries do not alter the results of our analysis . in this particular case , the magnitude threshold used to define low- and high - luminosity gcs was chosen only to provide luminosity classes of similar sizes ( see table [ tab : summary ] ) . figure [ fig : positions ] ( left ) shows the spatial distributions of the gcs in the plane of the sky . the hst pointings overlap marginally with the @xmath2 elliptical isophote of the neighboring spiral galaxy ngc4647 . we have excluded all the gcs within the ngc4647 @xmath2 ellipse from the sample used to evaluate the observed density profile of gcs in ngc4649 ( see section [ sec : method ] ) , but we have kept them in the density maps . the lmxb sample is from the catalog of @xcite , where identifications with gcs are based on @xcite . this catalog was derived from six _ chandra _ acis - s3 pointings of ngc4649 , reaching a total exposure of 299 ks and yielding a total of 501 x - ray sources with 0.3 - 8.0 kev luminosities ranging from @xmath14 to @xmath15 ergs s@xmath16 . out of the total 427 lmxbs within the hst combined footprint , 161 are gc - lmxbs and 266 are found in the stellar field of ngc4649 ( field lmxbs ) . moreover , 74 lmxbs were detected in the region external to the hst footprint ( see @xcite ) . of the x - ray sources detected in the area not overlapped by the hst observations used to extract the catalog of gcs used in this paper , twelve of them can be identified with gcs in the ground - based catalog of @xcite ; the other x - ray sources in this area are likely to be mostly background active galactic nuclei ( agns ) , based on the contamination estimates of @xcite . to avoid the large uncertainties on the sample of x - ray sources detected at larger radii , in this paper we will derive the density reconstruction of only the gcs and field lmxbs located within the hst footprint . only @xmath17 background agns are expected within the @xmath2 elliptical isophote of ngc4649 ; these are likely to contaminate the field lmxb sample , because the gc identification excludes this contamination in the sample of gc - lmxbs . we have also explored luminosity classes based on a threshold of @xmath18 erg s@xmath16 for the lmxbs . the two luminosity classes comprise 339 and 162 sources respectively . the higher luminosity class may contain a larger proportion of black hole binaries or of younger lmxbs ( @xcite ) . the spatial distribution of the lmxbs is shown in figure [ fig : positions ] ( right ) . the number of members of each class of sources investigated in this paper is reported in table [ tab : summary ] . lccccc & @xmath19 & @xmath20 & @xmath21 & @xmath22 & @xmath23 + gcs & 1603 & 841[642/198 ] & 762[493/269 ] & 467 & 1136 + & @xmath19 & @xmath24 & @xmath25 & @xmath22 & @xmath23 + lmxbs & 501 & 161[28/133 ] & 266(74 ) & 162 & 339 + + [ tab : summary ] we determined the 2d spatial distributions of gcs and lmxbs by applying a method based on the knn density estimator @xcite . we used simulations to evaluate the significance of the results . a detailed explanation of our application of the knn method can be found in @xcite . the assumption behind the knn method is that the density is locally constant . the uncertainty on the knn density scales with the square root of @xmath26 , where @xmath26 is the index of the nearest neighbor used to calculate the density and the fractional accuracy of the method increases with increasing k at the expense of the spatial resolution . for each sample discussed in this paper , we calculated the 2d surface densities for @xmath26 ranging from 2 to 9 , and a regular grid with spacing @xmath27 @xmath28 ( @xmath29 ) and @xmath30 @xmath31 ( @xmath32 ) for the gcs , and @xmath27 @xmath33 ( @xmath34 ) and @xmath30 @xmath33 ( @xmath34 ) for the lmxbs . these grids define 2d cells of `` pixels '' . this choice of grids is justified by the different areas covered by the hst and _ chandra _ observations and produces similar average number of source per pixel for both gcs and lmxbs . this choice permits to compare meaningfully the statistical reliability of the features determined in density and residual maps for both classes of sources . the density in the boundary pixels was weighted according to the fraction of the pixel within the observed region . to assess the statistical significance of features suggested by the knn density maps ( hereafter `` observed '' density maps ) , we employed monte - carlo simulations , using the observed radial distributions of each of the classes of gcs and lmxbs investigated , as seeds . the azimuthal distributions were independently extracted from a uniformly random distribution between 0@xmath35 and @xmath36 . a geometrical correction on the expected number of gcs per elliptical annulus was applied to take into account the eccentricity of the galaxy . we performed @xmath37 simulations for each of the samples described in table [ tab : summary ] , and analyzed the results as done with the observed data ( see above ) . figure [ fig : radialprofiles ] compares observed and simulated mean radial density profiles for the gc and lmxb samples , respectively . to characterize the deviations from the average 2d trends of the gc and lmxb distributions , we calculated maps of residuals by subtracting pixel - by - pixel the mean simulated density map from the corresponding observed density map , and normalizing to the value of the density in the average simulated map . the pixel - by - pixel distributions of the simulated knn densities are well approximated by gaussians , simplifying the calculation of the statistical significance of the residuals . to evaluate the latter , for each set of simulated density maps we calculated the fraction of pixels with values above the 90-th percentile of the densities in the observed maps ( the `` extreme '' pixels ) . since these extreme pixels in both observed density and residual maps tend to be spatially correlated , we also evaluated the fraction of simulations with at least one group of contiguous extreme pixels ( with area equal to 18 pixels ) as large as the observed . since we did not impose any specific geometry to the groups of simulated contiguous extreme pixels , these fractions are upper limits to the fraction expected for a given spatial distribution of residuals . the results are given in section [ sec : densityresidual ] . as discussed above , the radial distributions of the simulated distributions of sources follow the observed radial distributions , while the azimuthal distribution is uniform in the @xmath38 $ ] interval . if the incompleteness of the observed distribution of sources is a function only of the radial distance from the center without azimuthal dependencies , it will affect in the same way both observed and simulated density maps . accordingly , in this paper we have not included completeness corrections for either gc and lmxb samples because both populations are affected by incompleteness which only depends on the radial dependence from the center of the ngc4649 galaxy ( see @xcite for gcs and lmxbs incompleteness maps respectively ) . lccccccc density & & & & & & + & @xmath39 & @xmath40 & @xmath41 & @xmath42 & @xmath43 & @xmath44 + all gcs ( red@xmath45blue ) & 99.5%(1.8% ) & 49.5%(1.2% ) & 6%(0.3% ) & 0%(0% ) & 0%(0% ) & 0%(0% ) + red gcs & 96%(3.1% ) & 25.9%(2.5% ) & 2.3%(1.4% ) & 0%(0.3% ) & 0%(0% ) & 0%(0% ) + blue gcs & 100%(4.7% ) & 60.5%(1.2% ) & 4%(0% ) & 1.2%(0% ) & 0%(0% ) & 0%(0% ) + high - l gcs & 100%(3.4% ) & 74%(0.9% ) & 15%(0.2% ) & 2.8%(0% ) & 0%(0% ) & 0%(0% ) + low - l gcs & 99.5%(1.3% ) & 49%(0.3% ) & 5.7%(0% ) & 0%(0% ) & 0%(0% ) & 0%(0% ) + all lmxbs & 97%(73% ) & 86%(56.4% ) & 68%(34.8% ) & 45.5%(12.9% ) & 33.5%(8% ) & 21.5%(0.5% ) + gc - lmxbs & 85.5%(54.6% ) & 64.5%(33.1%)&45.5%(12.2% ) & 33.5%(3.5% ) & 25.5%(1.4% ) & 20.5%(0.3% ) + field lmxbs & 85%(34% ) & 64%(28.8% ) & 52%(17.5% ) & 43%(11.8% ) & 12%(5.4% ) & 20%(0% ) + high - l lmxbs & 99.5%(87.4% ) & 94%(76.9% ) & 88%(65.3% ) & 71%(43.9% ) & 60%(19.3% ) & 53.5%(0.8% ) + low - l lmxbs & 61%(17.9% ) & 28%(12.8% ) & 15.5%(3.7% ) & 7.5%(0.5% ) & 4%(0% ) & 3%(0% ) + residuals & & & & & & + & @xmath39 & @xmath40 & @xmath41 & @xmath42 & @xmath43 & @xmath44 + all gcs ( red@xmath45blue ) & 65.2%(10.5% ) & 8.2%(4.8% ) & 1.1%(0.7% ) & 0%(0% ) & 0%(0% ) & 0%(0% ) + red gcs & 54.8%(8.9% ) & 4.2%(1.1% ) & 0.8%(0.5% ) & 0%(0.1% ) & 0%(0% ) & 0%(0% ) + blue gcs & 71.8%(0.2% ) & 9.1%(0% ) & 1.7%(0% ) & 0%(0.4% ) & 0%(0% ) & 0%(0% ) + high - l gcs & 89.5%(12.2% ) & 16%(1.9% ) & 4.5%(0.6% ) & 0.5%(0% ) & 0%(0% ) & 0%(0% ) + low - l gcs & 88.6%(7.2% ) & 5.2%(5.4% ) & 0%(1.1% ) & 0%(0.4% ) & 0%(0% ) & 0%(0% ) + all lmxbs & 78.2%(10.5% ) & 51.7%(4.8% ) & 22.1%(2.7% ) & 13.5%(0% ) & 2.3%(0% ) & 0%(0% ) + gc - lmxbs & 64%(8.9% ) & 41.7%(3.1% ) & 30.4%(1.5% ) & 14.2%(0% ) & 2.5%(0% ) & 0.1%(0% ) + field lmxbs & 81%(12.2% ) & 59.4%(9.3% ) & 43.9%(4.7% ) & 28.7%(2.4% ) & 6.4%(1.3% ) & 0.5%(0% ) + high - l lmxbs & 85.9%(17.2% ) & 78.4%(12.9%)&52.1%(9.6% ) & 34.7%(6.5% ) & 21%(5% ) & 9.7%(0.8% ) + low - l lmxbs & 57.4%(7.2% ) & 43.8%(5.4% ) & 28.5%(3.2% ) & 13.9%(1.6% ) & 3.2%(0% ) & 0.1%(0% ) + [ tab : statistics ] for each gc and lmxb sample , table [ tab : statistics ] summarizes the percentages of simulated maps with a number of extreme pixels exceeding that in the observed density and residual maps for values of @xmath26 ranging from 4 to 9 . given the size of the samples of gcs and lmxbs used in this paper , larger values of @xmath26 , while enhancing possible low - contrast large - scale features , would degrade the spatial resolution of the density and residual maps . the fraction of extreme pixels in the simulated density and residual maps decreases for increasingly larger @xmath26 values : for gcs , it is zero for @xmath43 ; for lmxbs , very small and zero values are obtained for @xmath44 in the case of residual maps , while significantly positive values occur even for @xmath44 for the density maps . however , once the spatial clustering of the extreme pixels is taken into account , the lmxbs density maps have negligible fractions of extreme simulated pixels when @xmath44 , making the asymmetries at these scales unlikely to be the results of random statistical fluctuations . in the following we will show and discuss the results obtained for @xmath43 for all gcs classes and @xmath44 for lmxbs . table [ tab : numover ] gives the number of sources of each class located within pixels associated to over - density with significance greater than 1 , 2 , and 3@xmath46 for the gc ( @xmath43 ) and lmxb ( @xmath44 ) samples . in table [ tab : numover ] , the fraction of sources located in the @xmath47 , @xmath48 and @xmath49 over - density pixels of the map and the number of sources in excess to the expected number ( based on the radial profiles shown in figure [ fig : radialprofiles ] ) are shown in parentheses and square brackets , respectively . figure [ fig:2dmapsngc4649_gc ] ( left ) shows the 2d knn density map of the spatial distribution of the entire gc sample for @xmath43 . gcs are concentrated in the center of the galaxy , with an over - density elongated along the major axis of ngc4649 . figure [ fig:2dmapsngc4649_gc ] ( center ) shows the same density map where pixels are color - coded according to the average color of the gcs located in each pixel . as expected , high - density pixels , which tend to be more centrally located , contain on average redder gcs . residuals with significance larger than 1@xmath46 , 2@xmath46 and 3@xmath46 obtained from the residual maps derived from the distribution of the whole catalog of gcs in ngc4649 . all the negative residuals ( blue pixels ) have significance between 1 and 2 @xmath46 . positive ( orange ) pixels @xmath50 are indicated with `` @xmath51 '' sign ; @xmath52 with a `` @xmath45 '' sign . the alignment of the over - density and under - density pixels along the major and minor axes of the galaxy , respectively , is not an effect of the assumption of the circularly symmetric distribution of sources in the simulations.,width=302,height=302 ] color obtained with the knn method for @xmath43 . the horizontal black line shows the color value used as threshold to separate red and blue gcs in this paper . the gray points represent the observed positions of the gcs used to reconstruct the density . the arbitrary isodensity contours are added to highlight the position of the main over - densities.,width=302,height=302 ] a slight excess ( @xmath53 ) of red gcs is found in the over - density structure observed in the n - e quadrant of the galaxy and along the east section of the major axis . figure [ fig:2dmapsngc4649_gc ] ( right ) shows the residual map , which emphasizes over - densities along the major axis , extending into the n - e quadrant . for @xmath43 , the number of all gcs located within the pixels associated to positive residuals is 438 ( @xmath54 ) , 208 ( @xmath55 ) and 127 ( @xmath56 ) for 1@xmath46 , 2@xmath46 and 3@xmath46 significance respectively ( table [ tab : numover ] ) . figure [ fig : res2dmapsngc4649sigmas_gc ] shows that the majority of the pixels associated to positive residual values with significance larger than 2@xmath46 and 3@xmath46 are located in the e and to the n - e of the main axis of ngc4649 . figure [ fig : dens2dradialgmzcolor_gc ] shows the @xmath43 density map in the radial distance vs @xmath57 plane . the bi - modality of the distribution of gcs in color is clear for @xmath58 which corresponds to the major axis of the @xmath2 isophote , in agreement with the finding of @xcite . for @xmath59 the red , metal - rich gcs become less numerous , confirming that the average @xmath57 color in pixels outside the @xmath2 is significantly bluer than inner regions of the galaxy ( see figure [ fig:2dmapsngc4649_gc ] , center ) . lccc & @xmath60&@xmath61&@xmath62 + all gcs . ] & 438(@xmath63)[190 ] & 208(@xmath55)[109 ] & 127(@xmath64)[82 ] + red gcs & 388(@xmath65)[258 ] & 184(@xmath66)[135]&111(@xmath67)[82 ] + blue gcs & 388(@xmath68)[266 ] & 161(@xmath69)[117]&74(@xmath70)[57 ] + high - l gcs & 307(@xmath71)[240 ] & 109(@xmath72)[87]&45(@xmath73)[37 ] + low - l gcs & 430(@xmath74)[243 ] & 180(@xmath75)[111]&82(@xmath76)[55 ] + all lmxbs . ] & 153(@xmath77)[74 ] & 72(@xmath78)[40 ] & 47(@xmath79)[31 ] + gc lmxbs & 60(@xmath80)[30 ] & 43(@xmath81)[25]&31(@xmath82)[20 ] + field lmxbs & 83(@xmath83)[38 ] & 39(@xmath84)[20]&19(@xmath85)[12 ] + high - l lmxbs & 49(@xmath86)[19 ] & 22(@xmath87)[11]&10(@xmath88)[5 ] + low - l lmxbs & 120(@xmath89)[57]&63(@xmath90)[37]&41(@xmath91)[29 ] + [ tab : numover ] the @xmath43 density maps of red and blue gc subsamples do not show significant differences in the location of the main over - densities ( upper panels in figure [ fig : redbluengc4649_gc ] ) , except for red gcs being more centrally concentrated ( see also figure [ fig:2dmapsngc4649_gc ] , center ) . the residuals map derived from the red gcs distribution ( lower left panel in figure [ fig : redbluengc4649_gc ] ) highlights a large contiguous region of positive residuals in the n - e quadrant , with two smaller regions of positive residuals located in the s - w quadrant . the @xmath43 residual map of blue gcs ( lower right panel in figure [ fig : redbluengc4649_gc ] ) displays a more scattered distribution of positive residuals : a marginal enhancement corresponds to the region occupied by the large n - w positive residual structure seen in the red gcs map , but small significant clusters of positive residual pixels are located also in other regions . + + figure [ fig : highlowlngc4649_gc ] gives the corresponding results for the high and low luminosity gcs subsamples . the density distributions indicate that both luminosity classes peak in the center of the galaxy , with high - l gcs more centrally concentrated than low - l ones . the residual maps ( lower panels in figure [ fig : highlowlngc4649_gc ] ) emphasize the differences between the two classes . the high - l gcs residual distribution is characterized by a significant cluster of spatially correlated pixels located along the e section of the major axis of the galaxy . two other smaller structures are visible in the s - w quadrant within and outside of the @xmath2 isophote respectively . the residual map of the low - l gcs differs significantly from the high - l map in the e side : a large group of nearby pixels with highly significant positive residuals is located almost entirely in the n - e quadrant . this structures occupies a region devoid of positive high - l pixels and overlaps only partially with the high - l main over - density structure aligned along the major galaxy axis in the e direction . + + following @xcite , we have verified that the differences between the spatial distributions of red and blue gcs , and high - l and low - l gcs are not the results of a particular choice of the thresholds values by reconstructing the density and residual maps of the two classes for several values of the thresholds , and checking that the qualitative results do not change . for the color classes , we have used ten regularly spaced threshold values around the threshold value @xmath11 in the interval @xmath92 $ ] . the results are consistent throughout this range of colors . for values outside this interval , the significance of the residual map for one of the two classes degrades rapidly because of the small number of gcs in that class . the same qualitative conclusions discussed above for the characterization of the spatial distributions of low and high luminosity classes of gcs in ngc4649 hold true using ten regularly spaced values of the magnitude threshold within the interval @xmath93 $ ] . other values of the @xmath5 magnitude threshold have not been used since they would generate luminosity classes too unbalanced to correctly estimate the significance of the results . the distribution of lmxbs is centrally concentrated ( figure [ fig:2dmapsngc4649_lmxb ] , left ) , except for an over - density located within the @xmath2 elliptical isophote of the companion spiral galaxy ngc4647 and a less significant enhancement in the s - e quadrant within the @xmath2 ngc4649 isophote . the residual map enhances these over - densities ( figure [ fig:2dmapsngc4649_lmxb ] , center , right ) . a significant association of positive residuals pixels can be seen along the major axis of ngc4649 . two other significant regions composed of multiple spatially clustered positive residual pixels are visible in the n - e and s - w corners of the footprint of the _ chandra _ observations . although the footprint of the @xcite does not include the regions of the _ chandra _ field occupied by these two structures , three x - ray sources in the n - e over - density and four in the s - w density enhancement have been associated by @xcite to gcs from the catalog of @xcite . the remaining x - ray sources in the two structures do not have an optical counterpart that could be identified by our cross - correlation with the nasa extragalactic database ( ned ) , simbad astronomical database , sloan digital sky survey ( sdss ) dr10 , two micron all sky survey ( 2mass ) and wide - field infrared survey explorer ( wise ) allwise archives . nine ultra - luminous x - ray sources ( ulxs ) with @xmath94 ergs s@xmath16 were identified by @xcite , but they are not located within the boundaries of the n - e and s - w structures . the presence of lmxbs associated to gcs @xcite suggests that these two over - densities could be the remnants of past interactions between ngc4649 and small accreted satellite galaxies . however , considering the background agn density calculated in @xcite , we can not rule out contribution from background agns . differences are found in the distribution of residuals for field vs gcs - lmxbs and low and high x - ray luminosities lmxbs . these will be discussed in section [ sec : discussion ] . we have searched for structures in the distribution of the diffuse stellar light of ngc4649 observed in the hst data used by @xcite that could be spatially overlapping with the over - density structures observed in the spatial distribution of gcs . we have fitted and subtracted elliptical isophotal models to the images of ngc4649 galaxy in both @xmath5 and @xmath6 filters , without finding significant residuals that may be indication of merging ( cp . with the results discussed by @xcite ) . also , we searched for locally enhanced stellar formation in the color map of ngc4649 obtained by combining the hst g and z , using the method based on the distribution of the pixels images in the color - magnitude ( cm ) diagram ( see * ? ? ? * ) . we do not observe significant differences in the distribution of pixels in the cm diagram derived for different azimuthal regions of the galaxy containing the gc over - density structures . the only feature in the pixel cm diagram is the `` blue '' cloud of pixels in the region of ngc4649 overlapping with ngc4647 , already observed by @xcite ) . while our results do not suggest the existence of significant deviations from a smooth model of the diffuse light and color distributions in ngc4649 , more detailed analysis will be necessary to confirm the lack of faint structures in the diffuse light 2d model of this galaxy . in figure [ fig : stellarmassdensity_ngc4649 ] we plot the isodensity contours of the over - density structures observed in the spatial distribution of lmxbs in ngc4649 over the stellar mass map from @xcite . this stellar mass map is smooth , with a steep positive gradient towards the core of ngc4649 . the only significant enhancement is associated to the central region of the companion galaxy ngc4647 , where some granularity , possibly due to the spiral arms , is visible ( see figure 10 in @xcite ) . the position of the positive residual structures on the e side of ngc4649 does not correlate with any feature in the stellar mass map , while on the w side the lmxbs over - density contours clearly follows the stellar mass density enhancement associated to the spiral galaxy ngc4647 . the different spatial resolutions of the residual maps of the distribution of lmxbs in ngc4649 and of the stellar mass density map do not permit a direct comparison of the shape of the lmxbs over - density contours to the position of the granularities of the map though . are also shown.,width=302,height=302 ] our analysis of the two - dimensional projected distribution of the gc and lmxb populations of the giant virgo elliptical ngc4649 ( m60 ) confirms a well - known feature of this gc system , that red gcs are more centrally concentrated than blue gcs , as usually observed in elliptical galaxies ( e.g. , @xcite ; see also @xcite for ngc4649 ) , but also finds unexpected significant and complex 2d asymmetries in their projected distributions . ngc4649 is the third most luminous galaxy in the virgo cluster , and resides in a galaxy - dense environment , where gravitational interactions and accretion of satellites may be frequent . kinematics evidence , suggesting a merging and accretion past for this galaxy , was recently found in studies of planetary nebulae ( pn ) and gcs @xcite . in particular , @xcite show that at galactocentric radii larger than 12 kpc , pns and gcs may belong to separate dynamical systems . a similar conclusion was reached by @xcite , who propose either tidal stripping of gcs for less massive companion galaxies , or a combination of multiple mergers and dwarf galaxy accretion events to explain the observed kinematics . besides experiencing continuing evolution at the outer radii via interaction with and accretion of companions , ngc4649 itself may be the result of a major dry merger . this is suggested by recent kinematical measurements of the diffuse stellar light that revealed disk - like outer rotation , as it may have stemmed from a massive lenticular galaxy progenitor @xcite . the anisotropies we find in the 2d distribution of gcs may be a different pointer to this complex evolution . we have found that in ngc4649 , the 2d distribution of gcs shows strong positive residuals along the eastern major axis of the galaxy , with a northward arc - like curvature beginning at a galacto - centric radius of @xmath95 kpc and extending out to @xmath96 kpc . this large - scale feature is more prominent in the distribution of red gcs , but can still be seen in the 2d distribution of blue gcs , which is overall noisier . high - luminosity gcs tend to concentrate at the southern end of this feature , while low - luminosity gcs are found in the northern portion . these trends and differences are highlighted in figure [ fig : sigma_pixels_maps_gc ] , which displays the most significant single pixels ( @xmath97 ) of each residual distribution . the extreme pixel distributions of red gcs ( upper left ) and low - luminosity gcs ( lower right ) are similar , and differ from those of high - l gcs . the reason for this similarity resides in the choice of the particular magnitude threshold used to define luminosity classes ( see discussion in section [ sec : data ] and table [ tab : summary ] ) . with the @xmath12 threshold , red and blue gcs comprise @xmath98 and @xmath99 of low - l gcs , respectively . using @xmath100 as a threshold instead , the residual maps of low - l and high - l gcs are more similar , although high - l and low - l still are spatially segregated in the n side of the `` arc '' . however , this magnitude threshold yields 140 high - l gcs and 1463 low - l gcs . + + the color and luminosity differences suggest that we may not simply be witnessing the disruption of a satellite dwarf galaxy , and observing the stream of its gc system . perhaps some additional gc formation resulted from this merger , increasing the population of red , high - metallicity gcs . gc formation during galaxy merging has been suggested as an important mechanism for the assembly of gc systems , especially in the case of large elliptical galaxies @xcite , and is supported by high - resolution simulations of galaxy mergers @xcite . numerous young massive stellar clusters , which may evolve into gcs , are for example seen in the advanced merger remnant ngc7252 @xcite . this picture is being confirmed by other observational studies . a kinematical study of the gc system of the central galaxy of the virgo cluster , m87 , shows the presence of significant sub - structures , pointing to active galaxy assembly @xcite . in the virgo elliptical ngc4365 , three separate rotating gc systems have been reported associated with the three color - families of gcs , in addition to a stream - like system @xcite . with the exception of the large - scale arc structure observed in ngc4261 @xcite , little evidence is available in the literature about the 2d spatial distribution of gcs in early - type galaxies , let alone about the differences in the spatial distributions of gc color and luminosity classes due to major mergers . @xcite discussed the existence of a phase space shell composed of gcs located in the inner halo of m87 , which is possibly due to a significant merger , but there is no evidence of similar structures in the gc spatial distribution in m87 . based on these results , we deduce that the differences in the spatial distribution of high- and low - luminosity gcs in the n section of the ngc4649 arc suggests a more complex history involving both external minor accretion events and internal mechanisms that could influence the evolution of the ngc4649 gc system ( i.e. , dynamical friction and/or disk shocking ) . overall , the emerging picture is a perturbed gc system , suggesting a still evolving ngc4649 , perhaps caught in the moment of accreting some smaller satellite galaxy . this interpretation is not straightforward because other supporting evidence is still lacking other than the overall kinematics of pn , gcs and the stellar component of ngc4649 @xcite . we have searched for the kinematical signature of gravitational interactions and accretion of satellites undergoing in ngc4649 , by comparing the gc over - densities with the radial velocities for the sample of 121 gcs discussed by @xcite with spectroscopic observations . we do not find any obvious correlation . moreover , @xcite obtained spectra of 38 gcs in ngc4649 ( all included in our sample ) and did not find evidence that red gcs are significantly younger than blue gcs . however , the gc samples of @xcite and @xcite are both sparse and do not provide the optimal coverage of the over - density feature . the similarity of the radial density profiles of red and blue gcs in ngc4649 ( per figure [ fig : radialprofiles ] , left ) would be in principle consistent with ngc4649 being the result of a major dry merger ( see * ? ? ? * ; * ? ? ? * ) . however , our results suggest that the flattening of the radial density profiles of both color classes , instead than being the result of a global process with no azimuthal dependence , could be driven by the existence of major features in the spatial distribution of gcs . in the case of ngc4649 , the arc extending from the center of the galaxy to the @xmath2 isophote and more significant for red gcs than for blue gcs , could be the responsible for the overall similarity of the radial density profiles of the two gc color classes . it has been recognized since their early discovery in the milky way that dynamical formation of lmxbs in gcs is highly efficient @xcite ; some of these binaries could then be either kicked out from the parent cluster or be left in the stellar field after cluster disruption @xcite ( see also * ? ? ? * for early - type galaxies ) . therefore at least some field lmxbs could have been formed in gcs . however , lmxbs can also evolve from native binary systems in the stellar field ( see review from * ? ? ? the effectiveness of either or both formation channels for the lmxb populations of both the milky way and external galaxies is still a matter of debate . based on statistical considerations on the properties of the lmxb populations detected with chandra in early - type galaxies , there is some indication that field evolution is a viable formation channel @xcite , and that the lmxbs detected in the stellar field may be of mixed origin @xcite . as is the case for gcs , lmxbs are also objects that can be detected individually in elliptical galaxies with the resolution of chandra ( see * ? ? therefore , their spatial distribution may provide some constraints on their origin . in particular , native field lmxbs should - within statistics - trace the stellar surface brightness distribution of their parent galaxy . although there has been some debate on the radial profiles of lmxbs and their comparison with those of the stellar surface brightness and of gcs @xcite , in ngc4649 there is clear evidence @xcite that gc - lmxbs in red and blue gcs follow the same radial profiles as their parent gc populations , while field lmxbs in ngc4649 are radially distributed like the stellar surface brightness . however , within the limited sample statistics , lmxbs in red gcs are also distributed as the stellar light , with the exception of a marked lack of both red gcs and associated lmxbs in the centermost region . therefore these results are still consistent with a mixed origin for the field lmxb population . + + we have found ( section [ subsec : lmxbdensity ] ) that the 2d distribution of lmxbs shows a positive excess over a simple radial distribution with no azimuthal dependences , in the region near to the eastern major axis of ngc4649 , where the excess feature of gcs is also observed ( see figure [ fig : sigma_pixels_maps_lmxb ] , which displays the higher significance individual pixels for the residual distribution of different classes of lmxbs ) . not surprisingly , the excess residuals of the distribution of lmxbs in gc , which are overwhelmingly associated with red gcs @xcite , resemble those of the red gcs . however , the 2d distribution of field lmxbs also shows large - scale anisotropy , which would not be expected if these sources were solely associated with native field binaries . the excess residuals appear to be adjacent to , but not overlapping , those seen in the 2d distribution of gc - lmxbs . this excess of field lmxbs is dominated by high luminosity lmxbs ( @xmath101 ) . a second excess clump of these high - luminosity lmxbs is instead associated with the eastern end of the major axis ; this over - density of high - luminosity lmxbs is composed by equal number of gc - lmxbs and field lmxbs ( see right plot in figure [ fig : positions ] ) . positive residuals in the low luminosity lmxb 2d distribution are found extending along the eastern major axis , mostly associated to the positions of field lmbxs ( compare right plot in figure [ fig : positions ] ) ; given their spatial distribution , though , these are also likely to have been originated in gcs . of the two over - densities visible in the residual map generated with all the lmxbs ( figure [ fig:2dmapsngc4649_lmxb ] ) in the n - e and s - w corners of the _ chandra _ field , only the s - w structure can be seen in the residual map generated by low - l@xmath102 lmxbs ; neither is see in the high - l@xmath102 residuals ( figure [ fig : sigma_pixels_maps_lmxb ] ) . the differences in the significance of these two lmxb over - density structures are due to the higher values of the radial density profile of high - l lmxbs at larger radii compared to that of low - l at large radii ( see figure [ fig : radialprofiles ] ) , while both over - densities are composed of similar fractions of high and low luminosity x - ray sources . the n - e structure contains a total of 23 x - ray sources ( 12 low - luminosity and 11 high - luminosity ) , while the s - w region contains 45 total sources ( 27 low - luminosity and 18 high - luminosity ) . the presence of this high - l excess of field sources , to the south of the gc ( and gc - lmxb ) excess feature , suggests that these sources can not have been formed in gcs and ejected in the field , unless the streaming motion of the gcs and of the ejected lmxbs are subject to different accelerations . the escape velocity for a typical gc can be as low as @xmath430 km s@xmath16 @xcite , where the range of line of sight velocities observed for gcs in massive elliptical galaxies extends from few to few thousands km s@xmath16 @xcite . this suggests that lacking other forces , the lmxbs ejected should settle into a cloud surrounding their parent gcs . however , dynamical friction may cause the observed positional offset between the over - density of red gcs observed along the e side of the major axis ( upper right plot in figure [ fig : sigma_pixels_maps_lmxb ] ) and the field lmxbs structure observed along the @xmath2 isophote south of the main gcs over - density ( upper left plot in figure [ fig : sigma_pixels_maps_lmxb ] ) . we assume a standard mass for the lmxbs - hosting gcs @xmath103 consistent with their average absolute magnitudes @xcite . we further assume a circular velocity @xmath104 km s@xmath16 based on the results from @xcite . using as starting radial distance half the length of the major axis of the galaxy equivalent to @xmath105 kpc , we estimate the time required to recreate the angular and radial separation of the field lmxbs and gcs lmxbs over - densities with equation ( 7 - 26 ) of @xcite , that evaluates the time required for a gc moving on a circular orbit of radius @xmath106 with velocity @xmath107 to reach the center of the galaxy ( @xmath108 is assumed , based on typical values of the gc physical parameters ) : @xmath109 we obtain a time scale of @xmath110 gyr . this timespan is shorter than the average gc age of @xmath11110 gyr estimated by @xcite , and within the range of single gcs measurements by these authors . accurate kinematic measurements for gc population in ngc4649 would be needed to test this hypothesis . also , we point out that our simple evaluation of the effects of the dynamical friction should only be considered indicative . a full set of dynamical simulations of the evolution of the gc system of ngc4649 as the results of satellite accretion and mergers is required . alternatively , one could speculate that the merging event originating the gc over - density may have also resulted in a compressed gaseous stream from the disrupted satellite galaxy , with subsequent star and x - ray binary formation . the higher luminosity of the over - dense field x - ray sources would be consistent with a younger age than the overall lmxb population ( @xcite ) . ngc4647 is a spiral galaxy in virgo that can be seen to the n - w of ngc4649 . based on its aspect and photometric properties , @xcite conclude that it is tidally interacting with ngc4649 . there is a concentration of x - ray sources ( positive residuals ) associated with this spiral galaxy . the high - luminosity x - ray sources concentrate in the southern half of ngc4647 . this is the part of the galaxy showing a prominent asymmetric arm suggesting interaction with ngc4649 , and exhibiting more intense star formation @xcite . these luminous sources are likely younger hmxbs , associated with the ongoing intense star formation ( e.g. * ? ? ? our results agree with the overall picture of tidal interaction between ngc4647 and ngc4649 . ngc4647 may well be the next accreted satellite . our analysis of the 2d projected distribution of the gc and lmxb populations of the giant virgo elliptical ngc4649 ( m60 ) has led to the following results : * there are significant 2d asymmetries in the projected distributions of both gcs and lmxbs . * red gcs are more centrally concentrated than blue gcs , as usually observed in elliptical galaxies ( e.g. , @xcite ; see also @xcite for ngc4649 ) . the same is also true for high luminosity gcs ( which tend to be red ; see @xcite ) . * the 2d distribution of gcs shows strong positive residuals along the major axis of ngc4649 , especially on the e side , with an arc - like curvature towards the n , beginning at a galactocentric radius of @xmath95 kpc and extending out to @xmath96 kpc . this large - scale feature is more prominent in the distribution of red gcs , but can still be seen in the 2d distribution of blue gc , which is overall noisier . high - luminosity gcs tend to concentrate at the s end of the `` arc '' , while low - luminosity gcs are found in the northern portion . * the 2d distribution of lmxbs also shows positive excess in the e major axis side of ngc4649 . * the anisotropies of the distribution of lmxbs in gc , which are overwhelmingly associated with red gcs , resembles that of red gcs . however , field lmxbs also show large - scale anisotropies , with positive excess to the s of the red gc feature ; this excess is composed of high luminosity lmxb ( @xmath101 ) , which are also associated with the e end of the major axis ( these are partially associated to the position of gc - lmxbs ) . positive excess in low luminosity lmxb , mostly composed of field lmxbs , is found extending along the e major axis ; given their position , we speculate that these sources are likely to have formed in gcs and subsequently ejected in the field . * there is a definite concentration of lmxbs ( positive residuals ) associated with the spiral galaxy ngc4647 in the n - w side of ngc4649 . the higher x - ray luminosity portion of these sources tends to concentrate in the southern half of ngc4647 . these results have important implications for the evolution of ngc4649 . in particular , they suggest accretion and merging of satellite galaxies , of which the gc anisotropic distribution is the fossil remnant . lmxb over - densities are found associated with those of red gcs , consistent with dynamic lmxb formation in these clusters . surprisingly , there is also an over - density in the distribution of field lmxbs , to the south of the red gc over - density . this over density suggests that these lmxbs should be somewhat connected with the gc overdensity . the displacement , however , implies either ( 1 ) gc formation followed by ejection and dynamical friction drag on the parent gcs , or ( 2 ) field formation stimulated by compression of the ism during a merging event . we also clearly detect the over - density of x - ray sources connected with the x - ray binaries ( xrb ) population of the companion spiral galaxy ngc4647 . we note that the xrbs sources located in the southern half of the ngc4647 , where the star formation rate appears to be more intense ( see @xcite ) , are significantly more luminous that the x - ray sources observed in northern region of ngc4647 , with @xmath101 . this result may suggest a younger xrb population . we speculate that this asymmetry may indicate the beginning of the tidal interaction of ngc4647 with ngc4649 . our analysis and the kinematics of ngc4649 @xcite suggest a complex assembly history for this galaxy , including one major merger and a sequence of satellite accretion events , with possibly continuous tidal stripping of gcs and pns from nearby low - luminosity galaxies . explaining the origin of the over - density structures in the spatial distribution of gcs and of the luminosity and color segregations within such features in ngc4649 will require both detailed numerical simulations with a realistic model of the galaxy potential and extensive campaign for the spectroscopic observation required to reconstruct accurately the kinematics of the whole gc population . the detailed characterization of the 2d spatial distribution of gcs and lmxbs in ngc4649 that we have achieved will provide a new benchmark for future work aimed at the understanding of the dynamical evolution of this interesting system . we thank jean brodie for comments that have helped to improve the paper . gf thanks the aspen center for physics for hospitality ( nsf grant 1066293 ) . tf acknowledges support from the cfa and the itc prize fellowship programs . sm acknowledges funding from the stfc grant st / k000861/1 . this work was partially supported by the chandra go grant go1 - 12110x and the associated hst grant go-12369.01a , and the _ chandra _ x - ray center ( cxc ) , which is operated by the smithsonian astrophysical observatory ( sao ) under nasa contract nas8 - 03060 . angelini , l. , loewenstein , m. , & mushotzky , r. f. 2001 , , 557 , l35 arnold , j. a. , romanowsky , a. j. , brodie , j. p. , et al . 2013 , arxiv:1310.2607 ashman , k. m. , & zepf , s. e. 1992 , , 384 , 50 bastian , n. , schweizer , f. , goudfrooij , p. , larsen , s. s. , & kissler - patig , m. 2013 , , 431 , 1252 belokurov , v. , zucker , d. b. , evans , n. w. , et al . 2006 , , 647 , l111 binney , j. , & tremaine , s. 2008 , galactic dynamics : second edition , by james binney and scott tremaine . isbn 978 - 0 - 691 - 13026 - 2 ( hb ) . published by princeton university press , princeton , nj usa , 2008 . , blakeslee , j. p. , jordn , a. , mei , s. , et al . 2009 , , 694 , 556 blom , c. , forbes , d. a. , brodie , j. p. , et al . 2012 , , 426 , 1959 bonfini , p. , zezas , a. , birkinshaw , m. , et al . 2012 , , 421 , 2872 bournaud , f. , duc , p .- a . , & emsellem , e. 2008 , , 389 , l8 brassington , n. j. , fabbiano , g. , kim , d .- w . , et al . 2009 , , 181 , 605 brodie , j. p. , & strader , j. 2006 , , 44 , 193 clark , g. w. 1975 , , 199 , l143 coccato , l. , arnaboldi , m. , & gerhard , o. 2013 , , 436 , 1322 dabrusco , r. , fabbiano , g. , strader , j. , et al . 2013 , , 773 , 87 das , p. , gerhard , o. , mendez , r. h. , teodorescu , a. m. , & de lorenzi , f. 2011 , , 415 , 1244 de grijs , r. , & robertson , a. r. i. 2006 , , 460 , 493 de vaucouleurs , g. , de vaucouleurs , a. , corwin , h. g. , jr . , et al . 1991 , , 82 , 621 di matteo , t. , springel , v. , & hernquist , l. 2005 , , 433 , 604 dressler , a. 1980 , , 236 , 351 fabbiano , g. 2006 , , 44 , 323 fabbiano , g. , brassington , n. j. , lentati , l. , et al . 2010 , , 725 , 1824 forbes , d. a. , & bridges , t. 2010 , , 404 , 1203 fragos , t. , lehmer , b. , tremmel , m. , et al . 2013a , , 764 , 41 fragos , t. , lehmer , b. d. , naoz , s. , zezas , a. , & basu - zych , a. r. 2013b , arxiv:1306.1405 gilfanov , m. 2004 , , 349 , 146 gilli , r. , comastri , a. , & hasinger , g. 2007 , , 463 , 79 grindlay , j. e. , hertz , p. , steiner , j. e. , murray , s. s. , & lightman , a. p. 1984 , , 282 , l13 jordn , a. , peng , e. w. , blakeslee , j. p. , et al . 2009 , , 180 , 54 juett , a. m. 2005 , , 621 , l25 humphrey , p. j. , buote , d. a. , brighenti , f. , gebhardt , k. , & mathews , w. g. 2008 , , 683 , 161 ibata , r. a. , gilmore , g. , & irwin , m. j. 1994 , , 370 , 194 irwin , j. a. 2005 , , 631 , 511 ivanova , n. , fragos , t. , kim , d .- w . , et al . 2012 , , 760 , l24 janowiecki , s. , mihos , j. c. , harding , p. , et al . 2010 , , 715 , 972 kim , d .- w . , & fabbiano , g. 2004 , , 613 , 933 kim , e. , kim , d .- w . , fabbiano , g. , et al . 2006 , , 647 , 276 kim , d .- w . , fabbiano , g. , brassington , n. j. , et al . 2009 , , 703 , 829 kim , d .- w . , & fabbiano , g. 2013 , arxiv:1304.6377 kundu , a. , maccarone , t. j. , & zepf , s. e. 2002 , , 574 , l5 kundu , a. , maccarone , t. j. , & zepf , s. e. 2007 , , 662 , 525 lanz , l. , zezas , a. , brassington , n. , et al . 2013 , , 768 , 90 lee , m. g. , hwang , h. s. , park , h. s. , et al . 2008 , , 674 , 857 luo , b. , fabbiano , g. , strader , j. , et al . 2013 , , 204 , 14 mcconnachie , a. w. , irwin , m. j. , ibata , r. a. , et al . 2009 , , 461 , 66 mineo , s. , gilfanov , m. , & sunyaev , r. 2012 , , 419 , 2095 mineo , s. , fabbiano , g. , dabrusco , r. , et al . 2013 , arxiv:1306.6490 morscher , m. , umbreit , s. , farr , w. m. , & rasio , f. a. 2013 , , 763 , l15 navarro , j. f. , ludlow , a. , springel , v. , et al . 2010 , , 402 , 21 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1994 , , 267 , l1 paolillo , m. , puzia , t. h. , goudfrooij , p. , et al . 2011 , , 736 , 90 pearrubia , j. , walker , m. g. , & gilmore , g. 2009 , , 399 , 1275 pierce , m. , bridges , t. , forbes , d. a. , et al . 2006 , , 368 , 325 romanowsky , a. j. , strader , j. , brodie , j. p. , et al . 2012 , , 748 , 29 salinas , r. , jlkov , l. , carraro , g. , catelan , m. , & amigo , p. 2012 , , 421 , 960 schweizer , f. , & seitzer , p. 1998 , , 116 , 2206 shin , m .- s . , & kawata , d. 2009 , , 691 , 83 strader , j. , romanowsky , a. j. , brodie , j. p. , et al . 2011 , , 197 , 33 strader , j. , fabbiano , g. , luo , b. , et al . 2012 , , 760 , 87 tal , t. , van dokkum , p. g. , nelan , j. , & bezanson , r. 2009 , , 138 , 1417 teodorescu , a. m. , mndez , r. h. , bernardi , f. , et al . 2011 , , 736 , 65 tzanavaris , p. , fragos , t. , tremmel , m. , et al . 2013 , , 774 , 136 verbunt , f. , & van den heuvel , e. p. j. 1995 , x - ray binaries , 457 white , s. d. m. , & rees , m. j. 1978 , , 183 , 341 zibetti , s. , charlot , s. , & rix , h .- w . 2009 , , 400 , 1181

observations with merlin at 408 mhz have been used to establish the low - frequency part of the spectra of more than a hundred compact radio sources taken from the part of the jodrell bank vla astrometric survey limited by . these sources were selected from jvas and other catalogues to have convex spectra between 1.4 and 8.4 ghz , characteristic of gigahertz peaked spectrum ( gps ) sources . we have confirmed convex shapes of the spectra of 76 objects ( one half of our initial candidates ) thereby yielding the largest genuine sample of gps sources compiled so far . seven of 17 identified quasars in the sample have large ( ) redshifts .

astro-ph9811128

by definition , gigahertz peaked spectrum ( gps ) sources have convex spectra with a turnover frequency @xmath2 ghz . it is generally accepted that such a spectral shape results from synchrotron self - absorption due to a high compactness . indeed , the linear sizes of gps sources are very small ( 101000 pc ) and their radio luminosities are large ( @xmath3 erg / s ) . an important feature which makes the gps class particularly interesting , is that gps objects identified with quasars often have very large redshifts . a review on the properties of gps sources has been given by odea et al . ( 1991 ) hereinafter obs91 and odea ( 1998 ) . de vries et al . ( 1997 ) compiled spectra of 72 gps sources and constructed a canonical gps radio spectrum . they found it to have a constant shape independent of agn type , redshift or radio luminosity . as gopal - krishna & spoelstra ( 1993 ) pointed out , _ the great potential of gps sources for discovering high-@xmath4 objects continues to be the major motivation factor for enlarging the sample of gps sources_. a second motivation for increasing the number of known gps sources comes from the discovery ( wilkinson et al . 1994 ) of a new class of compact symmetric objects ( cso ) . five archetypal csos ( cf . readhead et al . 1996a ) are all acknowledged gps sources . readhead et al . ( 1996b ) argue that csos are the young precursors of classical double radio sources . increasing the list of known gps sources offers a promising way to find more csos . the search for gps sources can be difficult because it requires both high resolution and high sensitivity observations at frequencies well below 1 ghz and high sensitivity observations at @xmath5 5 ghz , well above the potential peak of the spectrum . therefore , unlike in the case of flat spectrum sources , only a modest number of gps sources can simply be extracted from classical large catalogues e.g. the green bank ( gb ) surveys at 1.4 ghz and 4.85 ghz ( white & becker 1992 and references therein hereinafter wb92 ) and the texas survey at 365 mhz ( douglas et al . for example , the sample of stanghellini et al . ( 1990 ) was derived from the texas survey and the 1 jy catalogue of khr et al . ( 1981a ) . out of 55 sources and candidates they listed , 41 are currently acknowledged as gps sources ( odea , priv . 1996 ) and 33 make a complete sample ( stanghellini et al . 1996 ) . the lists published by gopal - krishna et al . ( 1983 ) and spoelstra et al . ( 1985 ) resulting from dedicated observations , e.g. with the westerbork synthesis radio telescope ( wsrt ) or the ooty telescope , are not very large : each one contains 25 sources ( 5 and 2 out of these two samples respectively had been retracted later ) . the sources gathered by the authors mentioned above have been collected in obs91 in the so called `` working sample '' encompassing 95 objects . gopal - krishna & spoelstra ( 1993 ) confirmed the existence of 10 gps sources and cersosimo et al . ( 1994 ) found 7 more . on the other hand 6 sources from the list in obs91 , namely 0218 + 357 , 0528 + 134 , 0902 + 490 , 1851 + 488 , 2053@xmath6201 and 2230 + 114 proved to be `` not so good '' examples of the class and have been retracted . by the end of 1997 very few other gps sources had been discovered . there are two ways to achieve a `` bulk '' increase of this number . the first one has been followed by snellen et al . ( 1998 ) hereinafter ssb98 the second is the basis of this paper . the former one is based on the westerbork northern sky survey ( wenss ) being carried out with wsrt at 325 and 609 mhz . naturally , the sources with inverted spectra in the wenss are gps candidates , which are followed up with observations at higher frequencies . this survey is particularly useful since it is the most sensitive survey at low frequencies ( more than an order of magnitude better than the texas survey ) . the published part of wenss called `` mini survey '' ( rengelink et al . 1997 ) is limited to @xmath7 , @xmath8 and covers @xmath9 square degrees only , so there are still large parts of the sky where this method can not be applied . nevertheless , ssb98 using the mini survey plus an unpublished part of wenss ( @xmath10 , the same declination range as the mini survey ) were able to establish a list of 47 `` new '' gps sources . since wenss was not available when our programme started in 1994 , we adopted a different approach which can be regarded as `` opposite '' to that of ssb98 because we started at the high frequencies end instead of the low one . hence , to determine candidate gps sources , we used 8.4 ghz fluxes from the first part of the jodrell bank vla astrometric survey ( jvas ) ( patnaik et al . 1992 ) containing all compact radio sources in the range @xmath0 with fluxes @xmath11 mjy at 5 ghz . we combined them with 1.4 ghz and 4.85 ghz fluxes from the gb catalogues ( wb92 ) . for 32 sources 1.4 ghz fluxes were missing in wb92 so we substituted them with the catalogue flux density limit . then we fitted the data with a second order polynomial of the form : @xmath12 where @xmath13 is the observed frequency in ghz and @xmath14 the flux density in mjy and selected the sources with convex spectral shapes . the criterion for selecting an initial list of rrrc iauname & & & opt . + & & i d + 0059 + 581 & 01 02 45.7630 & 58 24 11.139 & cw + 0102 + 480 & 01 05 49.9295 & 48 19 03.183 & ef + 0627 + 532 & 06 31 34.6860 & 53 11 27.754 & bs + 0652 + 426 & 06 56 10.6629 & 42 37 02.751 & g + 0652 + 577 & 06 57 12.5027 & 57 41 56.740 & ef + 0750 + 535 & 07 54 15.2177 & 53 24 56.450 & ef + 0828 + 493 & 08 32 23.2171 & 49 13 21.036 & bl + 1107 + 485 & 11 10 36.3237 & 48 17 52.446 & ef + 1239 + 552 & 12 41 27.7043 & 54 58 19.040 & ef + 1256 + 546 & 12 58 15.6078 & 54 21 52.112 & ef + 1311 + 552 & 13 13 37.8518 & 54 58 23.894 & ef + 1428 + 422 & 14 30 23.7418 & 42 04 36.503 & ef + 1532 + 680 & 15 32 43.3426 & 67 55 13.992 & ef + 1602 + 576 & 16 03 55.9311 & 57 30 54.415 & bs + 1627 + 476 & 16 28 37.5064 & 47 34 10.414 & bs + 1630 + 358 & 16 32 31.2578 & 35 47 37.740 & ef + 1745 + 670 & 17 45 54.3577 & 67 03 49.302 & ef + 1755 + 578 & 17 56 03.6285 & 57 48 47.990 & bs + 1815 + 614 & 18 15 36.7920 & 61 27 11.641 & ef + 1946 + 708 & 19 45 53.5197 & 70 55 48.723 & ? + 2253 + 417 & 22 55 36.7082 & 42 02 52.535 & bs + 2310 + 385 & 23 12 58.7950 & 38 47 42.668 & bs + 2323 + 478 & 23 25 44.9131 & 48 06 25.280 & bs + 2356 + 385 & 23 59 33.1809 & 38 50 42.322 & bs + candidate gps sources was that the curvature @xmath15 had to be greater than 1 with a peak in the spectrum in the observable range ( the limit of @xmath16 is rather conservative and includes also spectra which are much flatter than those of typical gps sources ) . admittedly , although this criterion seems to be conservative enough , a few sources from the obs91 `` working sample '' which happen to be jvas members , and therefore have been fitted with polynomials ( 0018 + 729 , 0248 + 430 , 0552 + 398 , 2015 + 657 , 2021 + 614 , 2352 + 495 ) , _ did not _ fulfill it . in the end , the @xmath16 threshold was chosen as a good compromise , to produce a reasonably sized sample for future observations . the number of the initial candidates selected with the above procedure was 163 . this list contained 14 known gps sources : 0108 + 388 , 0153 + 744 , 0636 + 680 , 0646 + 600 , 0710 + 439 , 0711 + 356 , 1031 + 567 , 1225 + 368 , 1333 + 459 , 1843 + 356 and 2050 + 364 listed in obs91 ; 0903 + 684 from gopal - krishna & spoelstra ( 1993 ) and two discovered by snellen et al . ( 1995 ) 0700 + 470 and 1324 + 574 . these were not considered for further observations . we divided the remaining sample of 149 sources into two parts . the first one hereinafter `` subsample one '' contains those which are found in the 365 mhz texas survey ( 24 sources ) . they are listed in table 1 along with jvas positions rounded to 1 milliarcsecond and our identifications made using poss . nine of these are also listed in the 6c catalogue which has a flux density limit of 200 mjy at 151 mhz ( hales et al . 1993 and references therein ) . for the remaining 125 sources hereinafter `` subsample two '' we had virtually no low frequency data . although 19 of these sources were present in the b3 survey at 408 mhz ( ficarra et al . 1985 ) , its overlap in sky coverage with the first part of the jvas survey we use here is small and , furthermore , the b3 survey has significant errors in flux density near the catalogue limit of 100 mjy . the aforementioned crude extrapolation of the spectra applied to medium and high frequency data ( 1.4 ghz @xmath17 8.4 ghz ) of subsample two plus the flux density limit of the texas survey gave us only an unreliable estimate of the frequency turnover and even the convex shape of the spectrum remained uncertain in some cases . particularly some weaker sources could either be gps or mere flat spectrum depending on whether their 365 mhz fluxes were far below or _ just _ below the texas catalogue limit respectively . another effect that will produce a spurious convex shape in our non - simultaneous data is the variability typically observed in flat spectrum sources . in order to investigate the low frequency part of the spectra of subsample two sources we set up a programme of flux density measurements at low frequency with a high resolution facility . we used merlin at 408 mhz because of its superior resolving capability @xmath18 . the observations were carried out in the period from november 1994 until january 1995 and each source was typically observed twice ( with different hour angles ) for about 15 minutes per scan . such short snapshots can not be used to produce reliable maps with merlin , since the aperture coverage is too sparse . confusion is a significant effect at this low frequency and therefore fringe - frequency vs. delay ( ffd ) plots were used to separate confusing sources from the central target source ( see eg . walker 1981 ) . since the data were not phase - calibrated the coherence time is limited to a few minutes . by taking @xmath19 second sub - samples of the data and fourier transforming them in both time and frequency , a `` map '' of the field surrounding the source can be produced whose axes are fringe - frequency and delay . the target source will have close to zero fringe - frequency and delay , since its position is well known from the jvas survey , and appear at the centre of the `` map '' , while confusing sources will be offset from the centre . the flux density of the target source was determined from the central peak in the ffd map . the ffd plots were inspected visually ; the detection threshold in a single plot was set at @xmath20 and flux density values were only listed if the target was detected in two or more plots . in all cases , the values determined from the ffd plots were consistent with simple averages of phase - calibrated data on individual baselines calculated using the astronomical image processing system ( aips ) . the flux density scale was determined using observations of 3c286 , for which the baars et al . ( 1977 ) value was used . in table 2 we report the results of successful measurements of 408 mhz fluxes for 98 sources from subsample two along with their jvas positions rounded to 1 milliarcsecond and our identifications made using poss . interference and other problems at other sites resulted in the loss of data for the remaining 27 sources . they are listed in table 3 . the flux density values given in table 2 resulted from averaging the measurements over all baselines involving the defford or knockin telescopes ( 6 or 8 baselines , depending on whether the lovell telescope was used or not ) and in both ll and rr polarisations . shorter baselines were not used because they provide insufficient resolution in delay or fringe - frequency to separate the confusing sources ; additionally the use of longer baselines reduces any contribution from large - scale halos seen around some gps sources . the longest baselines involving the cambridge telescope were not used because interference limited the useful bandwidth to 1.5 mhz , rather than the 4 mhz used elsewhere . because baseline combinations and the volume of data per source varied form source to source the errors range form 10 to 30 mjy . rrrrrc iauname & & & & opt . + & & & i d + 0001 + 478 & 00 03 46.0413 & 48 07 04.134 & 104 & bs + 0015 + 529 & 00 17 51.7596 & 53 12 19.126 & 100 & bs + 0046 + 511 & 00 49 37.9901 & 51 28 13.700 & 123 & ns + 0051 + 679 & 00 54 17.6237 & 68 11 11.175 & 127 & ob + 0051 + 706 & 00 54 17.6884 & 70 53 56.625 & 260 & ns + 0058 + 498 & 01 01 16.9988 & 50 04 44.991 & 125 & bs + 0102 + 511 & 01 05 29.5588 & 51 25 46.576 & 262 & ns + 0123 + 731 & 01 27 04.7169 & 73 23 12.676 & 330 & ef + 0129 + 431 & 01 32 44.1273 & 43 25 32.667 & 244 & bs + 0129 + 560 & 01 32 20.4503 & 56 20 40.372 & 113 & cw + 0140 + 490 & 01 43 46.8791 & 49 15 41.586 & 105 & cw + 0148 + 546 & 01 51 36.2876 & 54 54 37.688 & 100 & ef + 0153 + 389 & 01 56 31.4088 & 39 14 30.929 & 119 & bs + 0213 + 444 & 02 16 17.1707 & 44 37 43.405 & 154 & ef + 0251 + 393 & 02 54 42.6316 & 39 31 34.714 & 114 & bs + 0307 + 380 & 03 10 49.8805 & 38 14 53.845 & 148 & ns + 0335 + 599 & 03 39 09.3942 & 60 08 56.960 & 134 & rs + 0336 + 473 & 03 40 10.7897 & 47 32 27.328 & 103 & bs + 0338 + 480 & 03 42 10.3522 & 48 09 46.948 & 113 & ns + 0412 + 447 & 04 15 56.5246 & 44 52 49.676 & 110 & ob + 0424 + 414 & 04 27 46.0455 & 41 33 01.091 & 272 & ef + 0454 + 550 & 04 58 54.8417 & 55 08 42.042 & 192 & bs + 0513 + 714 & 05 19 28.8835 & 71 33 03.740 & 153 & ef + 0514 + 474 & 05 18 12.0899 & 47 30 55.536 & 220 & cw + 0533 + 446 & 05 37 30.0630 & 44 41 03.533 & 155 & ? + 0559 + 422 & 06 02 58.9438 & 42 12 09.999 & 259 & bs + 0610 + 510 & 06 14 49.1589 & 51 02 13.124 & 143 & bg + 0621 + 446 & 06 25 18.2652 & 44 40 01.628 & 101 & bs + 0630 + 497 & 06 33 52.2068 & 49 43 45.939 & 149 & bs + 0651 + 410 & 06 55 10.0243 & 41 00 10.148 & 130 & g + 0655 + 696 & 07 01 06.6159 & 69 36 29.414 & 293 & bs + 0708 + 742 & 07 14 36.1236 & 74 08 10.142 & 105 & bs + 0713 + 669 & 07 18 05.6314 & 66 51 53.332 & 145 & ef + 0718 + 374 & 07 22 01.2600 & 37 22 28.628 & 64 & bs + 0732 + 755 & 07 39 13.1962 & 75 27 47.702 & 141 & ef + 0753 + 373 & 07 56 28.2513 & 37 14 55.647 & 59 & bs + 0753 + 519 & 07 56 59.5457 & 51 51 00.237 & 98 & bs + 0758 + 594 & 08 02 24.5932 & 59 21 34.800 & 102 & bs + 0849 + 675 & 08 53 34.3220 & 67 22 15.665 & 212 & bs + 0851 + 719 & 08 56 54.8695 & 71 46 23.894 & 105 & bs + 0900 + 520 & 09 03 58.5758 & 51 51 00.658 & 354 & ? + 0924 + 732 & 09 29 42.1565 & 73 04 04.553 & 114 & g + 0925 + 745 & 09 30 53.7823 & 74 20 05.930 & 112 & ef + 0939 + 620 & 09 43 14.5025 & 61 50 33.343 & 99 & bs + 1017 + 436 & 10 20 27.2021 & 43 20 56.342 & 51 & bs + 1019 + 429 & 10 22 13.1324 & 42 39 25.618 & 138 & rs + 1032 + 509 & 10 35 06.0176 & 50 40 06.087 & 198 & ef + 1035 + 430 & 10 38 18.1899 & 42 44 42.766 & 194 & rs + 1043 + 541 & 10 46 24.0372 & 53 54 26.220 & 181 & bs + 1055 + 433 & 10 58 02.9208 & 43 04 41.505 & 182 & ef + 1101 + 609 & 11 04 53.6946 & 60 38 55.287 & 151 & bs + 1125 + 366 & 11 27 58.8707 & 36 20 28.352 & 86 & bs + 1138 + 644 & 11 41 12.2283 & 64 10 05.484 & 180 & ef + continued rrrrrc iauname & & & & opt . + & & & i d + 1157 + 532 & 12 00 06.0107 & 53 00 37.118 & 188 & q + 1206 + 415 & 12 09 22.7884 & 41 19 41.369 & 152 & bs + 1226 + 638 & 12 29 06.0256 & 63 35 00.986 & 153 & rs + 1232 + 366 & 12 35 05.8076 & 36 21 19.308 & 121 & bs + 1239 + 606 & 12 41 29.5907 & 60 20 41.320 & 136 & bs + 1245 + 676 & 12 47 33.3300 & 67 23 16.457 & 107 & g + 1300 + 485 & 13 02 17.1974 & 48 19 17.572 & 163 & bs + 1308 + 471 & 13 10 53.5906 & 46 53 52.219 & 160 & ef + 1320 + 394 & 13 22 55.6615 & 39 12 07.984 & 119 & bs + 1321 + 410 & 13 24 12.0940 & 40 48 11.773 & 92 & bs + 1337 + 637 & 13 39 23.7812 & 63 28 58.425 & 185 & bs + 1338 + 381 & 13 40 22.9519 & 37 54 43.839 & 105 & rs + 1357 + 404 & 13 59 38.0943 & 40 11 38.260 & 129 & ef + 1403 + 411 & 14 05 07.7949 & 40 56 57.847 & 134 & bs + 1454 + 447 & 14 55 54.1361 & 44 31 37.668 & 115 & bs + 1533 + 487 & 15 35 14.6540 & 48 36 59.697 & 122 & bs + 1534 + 501 & 15 35 52.0395 & 49 57 39.084 & 103 & bs + 1544 + 398 & 15 45 53.2331 & 39 41 46.857 & 168 & g + 1607 + 563 & 16 08 20.7518 & 56 13 56.373 & 129 & bs + 1614 + 466 & 16 16 03.7667 & 46 32 25.231 & 98 & bs + 1722 + 611 & 17 22 40.0578 & 61 05 59.801 & 213 & bs + 1724 + 609 & 17 24 41.4142 & 60 55 55.731 & 179 & ef + 1753 + 648 & 17 54 07.5904 & 64 52 02.642 & 89 & bs + 1801 + 459 & 18 02 25.1427 & 45 57 34.645 & 192 & bs + 1812 + 560 & 18 12 57.6692 & 56 03 49.198 & 193 & bs + 1820 + 397 & 18 21 59.6991 & 39 45 59.647 & 743 & bs + 1822 + 682 & 18 21 59.4951 & 68 18 43.003 & 192 & bs + 1828 + 399 & 18 29 56.5203 & 39 57 34.690 & 116 & ef + 1839 + 389 & 18 40 57.1550 & 39 00 45.712 & 133 & bs + 1839 + 548 & 18 40 57.3780 & 54 52 15.920 & 121 & bs + 1937 + 630 & 19 38 16.1680 & 63 07 17.803 & 254 & ? + 1939 + 429 & 19 40 49.3198 & 43 04 24.671 & 197 & bs + 1941 + 413 & 19 42 58.6385 & 41 29 23.073 & 167 & bs + 2000 + 472 & 20 02 10.4183 & 47 25 28.777 & 150 & cw + 2005 + 642 & 20 06 17.6949 & 64 24 45.423 & 137 & rs + 2013 + 508 & 20 14 28.5899 & 50 59 09.532 & 174 & ef + 2014 + 463 & 20 15 39.9865 & 46 28 50.886 & 122 & cw + 2112 + 374 & 21 14 44.1230 & 37 42 25.719 & 181 & cw + 2119 + 709 & 21 19 54.1676 & 71 10 36.091 & 147 & ef + 2151 + 431 & 21 53 50.9585 & 43 22 54.497 & 110 & ns + 2202 + 716 & 22 03 30.4694 & 71 51 08.527 & 200 & rs + 2248 + 555 & 22 50 42.8496 & 55 50 14.608 & 212 & g + 2300 + 638 & 23 02 41.3165 & 64 05 52.858 & 117 & ef + 2310 + 724 & 23 12 19.6998 & 72 41 26.924 & 231 & ef + 2341 + 697 & 23 43 43.7360 & 70 03 19.398 & 160 & ns + rrrc iauname & & & opt . + & & i d + 0310 + 435 & 03 14 08.0539 & 43 45 19.770 & ef + 0314 + 696 & 03 19 22.0734 & 69 49 25.603 & ef + 0418 + 437 & 04 21 52.0619 & 43 53 04.216 & cw + 0537 + 392 & 05 40 44.4377 & 39 16 12.236 & rs + 0601 + 578 & 06 05 42.2275 & 57 53 16.351 & bs + 0638 + 528 & 06 42 27.8215 & 52 47 59.282 & bs + 0644 + 491 & 06 48 47.1190 & 49 07 20.736 & bs + 0651 + 428 & 06 54 43.5263 & 42 47 58.728 & g + 0903 + 669 & 09 07 23.5240 & 66 44 46.942 & ef + 1238 + 702 & 12 40 34.6989 & 69 58 30.616 & bs + 1245 + 716 & 12 47 09.3270 & 71 24 20.018 & ef + 1341 + 691 & 13 43 00.5520 & 68 55 17.160 & bs + 1406 + 564 & 14 08 12.9466 & 56 13 32.488 & bs + 1436 + 445 & 14 38 28.5048 & 44 18 12.085 & bs + 1447 + 536 & 14 48 59.1739 & 53 26 09.282 & ef + 1456 + 375 & 14 58 44.7949 & 37 20 21.627 & bs + 1526 + 670 & 15 26 42.8732 & 66 50 54.617 & ns + 1550 + 582 & 15 51 58.2077 & 58 06 44.466 & bs + 1611 + 425 & 16 13 04.8038 & 42 23 18.903 & bs + 1622 + 665 & 16 23 04.5221 & 66 24 01.084 & g + 1924 + 420 & 19 26 31.0504 & 42 09 58.991 & g + 2119 + 664 & 21 20 46.2045 & 66 42 20.216 & ef + 2132 + 406 & 21 34 24.1053 & 40 50 11.345 & ef + 2230 + 625 & 22 32 22.8655 & 62 49 36.436 & ob + 2236 + 678 & 22 38 15.0284 & 68 04 59.758 & ob + 2249 + 402 & 22 51 59.7715 & 40 30 58.155 & bs + 2351 + 550 & 23 53 42.3011 & 55 18 40.670 & bs + finally we merged subsample one ( 24 sources ) with those 98 sources from subsample two whose flux densities we had successfully measured with merlin at 408 mhz . since the selection process for this project was carried out , the 5 ghz green bank survey of wb92 has been superseded by the gb6 survey ( gregory et al . 1996 ) and at 1.4 ghz we can now use the the nrao vla sky survey ( nvss ) ( condon et al . 1998 ) which has a resolution of @xmath21 and an rms sensitivity of approximately 1.5 mjy . unfortunately , at the time of writing , nvss although almost complete did not cover all areas of the sky within its declination limits . many nvss maps ( @xmath22 each ) appear to be `` patchy '' and the `` holes '' can sometimes be quite large . our survey suffered considerably from this shortcoming of the current edition of nvss 9 sources out of those 122 sources we wanted to study were simply not present in the nvss catalogue . ( one source out of these nine was also unavailable in gb6 . ) additionally we decided to remove 2 other sources from the further processing : one of these is blended with a nearby source and the second one has an extended structure which should be studied in more detail . at 1.4 ghz we also tried to use the faint images of radio sky at twenty ( first ) catalogue ( white et al . 1997 ) 24 our sources could be found there . for 20 objects out of these we noted a very good compatibility between first and nvss based fluxes ; the 4 objects which showed discrepancy are indicated in table 6 . the selection process described above gave us finally 111 objects for which we arrayed the flux density values at each frequency . then we attempted to fit model spectra to the available data using a broken power - law with the following formula ( moffet 1975 ) : @xmath23 here @xmath24 and @xmath25 are the spectral indices of the rising and declining parts of the spectrum as often used in radio astronomy , while @xmath26 and @xmath27 are just fitting parameters which are _ not _ equal to the maximum flux density ( @xmath28 ) and the peak frequency ( @xmath29 ) of the fitted spectrum . even though the broken power - law seems to be the physically more sensible choice for a model spectrum compared to a simple second - order polynomial , it has the disadvantage that it is unconstrained if the peak of the spectrum falls beyond the 2nd highest or below the 2nd lowest available frequency . in these cases we fixed the peak of the model spectrum ( i.e. @xmath28 and @xmath29 ) at the peak of the measured data this was usually the measurement at 4.85 ghz and marked the fit as unconstrained in table 4 . this means that the values for @xmath24 or @xmath25 have to be considered as a lower or upper limit respectively ( i.e. in reality the spectrum will be more inverted at low frequencies or steeper at high frequencies ) . the spectra of 35 sources could not be fitted with such a convex - shaped curve ( fig 1 . ) and we claim that those sources can not be termed `` gps sources '' at all and most likely are just variable flat - spectrum sources . the 76 spectra that could be fitted with our algorithm are presented in fig . 2 and the fitting parameters are given in table 4 . as can be seen from fig . 2 and table 4 , some of the sources with unconstrained model spectra , fit the data relatively poorly at low frequencies or have relatively flat spectral indices ( e.g. 0307 + 380 and 0610 + 510 ) and thus are less probable gps candidates . in table 5 we specified some parameters of our `` new '' gps sources gathered from the literature : the names of other catalogues a particular source is a member , the optical identification according to the nasa / ipac extragalactic database ( ned ) and the redshift . at the time of writing 21 objects from our collection have been identified ( 3 galaxies , 18 qsos ) and their redshifts are known . those 3 galaxies have low redshifts ( @xmath30 ) ; on the other hand as expected the majority of quasars have large redshifts : for 6 qsos @xmath31 , for 7 other qsos @xmath1 . one qso , 1338 + 381 , is extremely redshifted : @xmath32 . most of the gps sources studied so far hardly show any variability , therefore we checked our sources against possible flux variations . firstly , because a significant variability of the flux density would mean that the source in question is likely not to be a gps and secondly since our data are not simultaneous any variability makes derivation of spectra questionable . the part of sources spectra around 1.4 ghz is obviously the most `` sensitive '' with regard to the gps phenomenon so we compared fluxes at this frequency given in wb92 to those from nvss . we applied corrections for the different beam sizes of these two measurements . if a particular source had changed its flux between epochs of the gb surveys and nvss / first more than 25 % or the 1.4 ghz gb flux was missing in wb92 we treated such a source as potentially variable , unless we could find a second epoch flux density measurement elsewhere . we assigned a `` candidate '' status for such objects and listed them in table 6 . among these there are 4 sources ( 0412 + 447 , 1125 + 366 , 1357 + 404 , 2005 + 642 ) with inverted spectra only , i.e. apparently having turnovers in their spectra at frequencies larger than 8.4 ghz . this feature was yet another reason to assign them a candidate status . apart from the information in table 5 we note that : * 0627 + 532 , 0652 + 577 , 1107 + 485 , 1256 + 546 and 1311 + 552 are 6c sources ( hales et al . 1993 and references therein ) * 1107 + 485 is a member of the drao 408 mhz survey ( green & riley 1995 ) * 1745 + 670 and 1753 + 648 are members of the nep survey ( kollgaard et al . 1994 ) * 1607 + 563 , 1755 + 578 and 1815 + 614 are members of the 7c survey ( visser et al . 1995 ) * 0514 + 474 was observed by leahy & roger ( 1996 ) * 0102 + 480 and 2253 + 417 are members of the cj1 survey ( polatidis et al . 1995 ) * 1245 + 676 is a giant radio galaxy with a gps core ( odea priv . * 0140 + 490 has a large scale symmetric structure @xmath33 across . * 1839 + 548 and 2119 + 709 are marked as `` quasi - point '' sources in jvas . all other sources are pointlike i.e. they are unresolved by the vla in `` a '' configuration at 8.4 ghz . * 1815 + 614 and 1946 + 708 are csos ( taylor et al . 1996 , taylor & vermeulen 1997 ) gigahertz peaked spectrum objects are an astrophysically significant and important class yet they are still not well understood ( odea 1998 ) . they are not necessarily a uniform class , and one can easily name subclasses among the whole gps ensemble . for example csos make one well defined group all of them are gps galaxies with characteristic vlbi morphologies . it is claimed by readhead et al . ( 1996b ) that csos play a key role as an initial stage in the evolutionary scenario of radio - loud agns . another fascinating subset of gps class are objects with extreme ( @xmath34 ) redshifts . there are 9 such objects in the `` working sample '' ( odea , priv . additionally there are @xmath35 objects with high ( @xmath36 ) redshifts . all the objects with @xmath37 are identified with quasars . the above two issues alone are already a good argument to extend the number of known gps radio sources . with such a goal in mind we have made a search for candidate gps sources in the part of the jodrell bank vla astrometric survey limited to @xmath0 , namely we compared 8.4 ghz flux densities derived from jvas with respective 1.4 ghz and 5 ghz fluxes in available catalogues . we treated a source as a plausible candidate if its spectrum seemed to be convex according to the criterion we had arbitrarily assumed . quite expectedly some of the candidates selected in this manner are already recognised as gps objects so we did not deal with them here . using flux densities at low frequencies i.e. 365 mhz and sometimes even 151 mhz in available catalogues we were able to classify 24 objects as gps sources without any further measurements . for the majority of selected objects the flux densities at frequencies well below 1 ghz were not available . in these cases we performed observations with merlin at 408 mhz to establish the low - frequency part of the spectra . our final decision on which of our candidates are and which are not gps sources has been made based on our merlin data ( or texas catalogue when available ) , nvss , gb6 and jvas catalogues . combining flux density measurements made with high resolution both at low ( merlin , 408 mhz ) and high ( vla , 8.4 ghz ) frequencies enabled us to eliminate the effects of confusion and any contribution from possible extended `` halos '' . we regarded a source as a gps if it had fitted well a `` broken power - law '' function and was not variable . the sample we present here is the largest single contribution to the pool of known gps sources collected so far . only 3 of our sources ( 0513 + 714 , 0758 + 594 , 1946 + 708 ) are overlapping with another large collection of gps sources , namely with the wenss based sample ( ssb98 ) . seven sources in our sample have large redshifts ( @xmath1 ) ; the largest one is @xmath32 . our approach to finding gps sources was to search from high frequency to lower ones . wenss has been equally successful in defining gps sources but searching from low frequencies to higher ones . we want to stress though that our approach can successfully be applied to the areas of the sky _ not _ planned to be covered by wenss ( @xmath38 ) but already covered by the other catalogues we used ( nvss , gb6 , jvas ) . for the part of jvas we used so far ( patnaik et al . 1992 ) we found that 9% of jvas sources are gps ; therefore the whole jvas encompassing around 3000 sources could easily yield 250 300 of such objects . the initial stages of the programme described in this paper were completed when am stayed at mpifr in bonn . the max - planck - gesellschaft stipend which supported him in that time is gratefully acknowledged . 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 . baars j.w.m . , genzel r. , pauliny - toth i.i.k . , witzel a. , 1977 , a&a 61 , 99 cersosimo j.c . , lebron santos m. , cintrn s.i . , quiniento z.m . , 1994 , apjs 95 , 157 condon j.j . , cotton w.d . , greisen e.w . , et al . , 1998 , aj 115 , 1693 douglas j.n . , bash f.n . , arakel bozyan f. , torrence g.w . , 1996 , aj 111 , 1945 ficarra a. , grueff g. , tomasetti g. , 1985 , a&as 59 , 255 gopal - krishna , patnaik a.r . , steppe h. , 1983 , a&a 123 , 107 gopal - krishna , spoelstra t.a.th . , 1993 , a&a 271 , 103 green d.a . , riley j.m . , 1995 , mnras 274 , 324 gregory p.c . , scott w.k , douglas k. , condon j.j . , 1996 , apjs 103 , 427 hales s.e.g . , baldwin j.e . , warner p.j . , 1993 , mnras 263 , 25 henstock d.r . , browne i.w.a . , wilkinson p.n . , mcmahon r.g . , 1997 , mnras 290 , 380 hewitt a. , burbidge g. , 1989 , magnetic tape hook i.m . , mcmahon r.g . , patnaik a.r . , et al . , 1995 , mnras 273 , 63 hook i.m . , mcmahon r.g . , irwin m.j . , hazard c. , 1996 , mnras 282 , 1274 kollgaard r.i . , brinkmann w. , mcmath chester m. , et al . , 1994 , apjs 93 , 145 khr h. , witzel a. , pauliny - toth i.i.k . , nauber u. , 1981a , a&as 45 , 367 khr h. , pauliny - toth i.i.k . , witzel a. , schmidt j. , 1981b , aj 86 , 854 leahy d.a . , roger r.s . , 1996 , a&as 115 , 345 marzke r.o . , huchra j.p . , geller m.j . , 1996 , aj 112 , 1803 moffet a.t . , 1975 , in : sandage a. , sandage m. , kristian j. ( eds . ) stars and stellar systems , vol . ix , p.211 moran e.c . , helfand d.j . , becker r.h . , white r.l . , 1996 , apj 461 , 127 odea c.p . , baum s.a . , stanghellini c. , 1991 , apj 380 , 66 ( obs91 ) odea c.p . , 1998 , pasp 110 , 493 patnaik a.r . , browne i.w.a . , wilkinson p.n , wrobel j.m . , 1992 , mnras 254 , 655 pauliny - toth i.i.k . , witzel a. , preuss e. , et al . , 1978 , aj 83 , 451 polatidis a.g . , wilkinson p.n . , xu w. , et al . , 1995 , apjs 98 , 1 rengelink r.b . , tang y. , de bruyn a.g . , et al . , 1997 , a&a 124 , 259 readhead a.c.s . , taylor g.b . , xu w. , et al . , 1996a , apj 460 , 612 readhead a.c.s . , taylor g.b . , pearson t.j . , wilkinson p.n . , 1996b , apj 460 , 634 snellen i.a.g . , zhang m. , schilizzi r.t . , et al . , 1995 , a&a 300 , 359 snellen i.a.g . , schilizzi r.t . , de bruyn a.g . , et al . , 1998 , a&as 131 , 435 ( ssb98 ) spoelstra t.a.th . , patnaik a.r . , gopal - krishna , 1985 , a&a 152 , 38 stanghellini c. , odea c.p . , baum s.a . , fanti r. , 1990 , _ compact steep spectrum & gigahertz - peaked spectrum sources _ , dwingeloo workshop , 1819 june 1990 , fanti c. , fanti r. , odea c.p . , schilizzi r.t . stanghellini c. , dallacasa d. , odea c.p . , et al . , 1996 , _ 2nd workshop on gps and css radio sources _ , leiden , 30 sept. 2 oct . 1996 , eds . snellen i.a.g . , schilizzi r.t . , rttgering h.j.a . , bremer m.n . stickel m. , khr h. , 1993 , a&as 100 , 395 stickel m. , khr h. , 1994 , a&as 105 , 67 taylor g.b . , vermeulen r.c . , 1997 , apj 485 , 9 taylor g.b . , vermeulen r.c . , pearson t.j . , et al . , 1994 , apjs 95 , 345 taylor g.b . , vermeulen r.c . , readhead a.c.s . , et al . , 1996 , _ 2nd workshop on gps and css radio sources _ , leiden , 30 sept. 2 oct . 1996 , eds . snellen i.a.g . , schilizzi r.t . , rttgering h.j.a . , bremer m.n . vermeulen r.c . , taylor g.b . , 1995 , aj 109 , 1983 vermeulen r.c . , taylor g.b . , readhead a.c.s . , browne i.w.a . , 1996 , aj 111 , 1013 visser a.e . , riley j.m . , rttgering h.j.a . , waldram e.m . , 1995 , a&as 110 , 419 de vries w.h . , barthel p.d . , odea c.p . , 1997 , a&a 321 , 105 walker r.c . , 1981 , aj 86 , 1323 wilkinson p.n . , polatidis a.g . , readhead a.c.s . , xu w. , pearson t.j . , 1994 , apj 432 , l87 white r.l . , becker r. h. , 1992 , apjs 79 , 331 ( wb92 ) white r.l . , helfand d.j . , becker r.h . , gregg m.d . , 1997 , apj 475 , 479 1 & 0001 + 478 & 286 . & 3.66 & @xmath410.671 & @xmath62.1 & 332 . & 2.68 & + 2 & 0015 + 529 & 657 . & 3.27 & @xmath411.13 & @xmath60.381 & 687 . & 4.82 & + 3 & 0046 + 511 & 246 . & 7.14 & @xmath410.549 & @xmath61.63 & 284 . & 4.85 & @xmath42 + 4 & 0051 + 679 & 324 . & 6.97 & @xmath410.577 & @xmath61.85 & 379 . & 4.85 & @xmath42 + 5 & 0058 + 498 & 187 . & 5.94 & @xmath410.322 & @xmath61.76 & 237 . & 3.49 & + 6 & 0102 + 480 & 1110 . & 3.03 & @xmath411.06 & @xmath60.774 & 1110 . & 3.01 & + 7 & 0102 + 511 & 530 . & 1.85 & @xmath410.775 & @xmath61.33 & 564 . & 1.42 & @xmath42 + 8 & 0123 + 731 & 361 . & 3.25 & @xmath410.264 & @xmath61.27 & 449 . & 1.63 & + 9 & 0129 + 560 & 659 . & 3.42 & @xmath411.05 & @xmath60.956 & 662 . & 3.14 & + 10 & 0140 + 490 & 228 . & 4.28 & @xmath410.524 & @xmath61.53 & 262 . & 2.85 & + 11 & 0148 + 546 & 137 . & 8.39 & @xmath410.256 & @xmath61.86 & 181 . & 4.73 & + 12 & 0153 + 389 & 203 . & 7.44 & @xmath410.404 & @xmath62.17 & 257 . & 4.85 & @xmath42 + 13 & 0307 + 380 & 581 . & 6.21 & @xmath410.862 & @xmath61.66 & 628 . & 4.85 & @xmath42 + 14 & 0335 + 599 & 192 . & 7.92 & @xmath410.309 & @xmath62.12 & 251 . & 4.85 & @xmath42 + 15 & 0336 + 473 & 295 . & 2.96 & @xmath410.762 & @xmath60.886 & 301 . & 2.43 & + 16 & 0412 + 447 & 277 . & 7.84 & @xmath410.349 & @xmath410.349 & & & + 17 & 0454 + 550 & 257 . & 7.85 & @xmath410.302 & @xmath62.23 & 340 . & 4.85 & @xmath42 + 18 & 0513 + 714 & 219 . & 3.26 & @xmath410.394 & @xmath61.19 & 253 . & 1.91 & + 19 & 0514 + 474 & 863 . & 2.05 & @xmath411.13 & @xmath60.842 & 863 . & 2.02 & + 20 & 0610 + 510 & 143 . & 9.9 & @xmath410.157 & @xmath61.74 & 198 . & 4.85 & @xmath42 + 21 & 0627 + 532 & 832 . & 0.767 & @xmath410.996 & @xmath60.593 & 835 . & 0.836 & + 22 & 0630 + 497 & 183 . & 7.51 & @xmath410.245 & @xmath62.89 & 255 . & 4.85 & @xmath42 + 23 & 0651 + 410 & 408 . & 6.57 & @xmath410.661 & @xmath61.06 & 431 . & 4.85 & @xmath42 + 24 & 0652 + 577 & 528 . & 0.564 & @xmath411.01 & @xmath60.649 & 528 . & 0.592 & + 25 & 0655 + 696 & 481 . & 1.63 & @xmath410.69 & @xmath60.644 & 484 . & 1.42 & @xmath42 + 26 & 0718 + 374 & 277 . & 6.62 & @xmath410.692 & @xmath61.43 & 302 . & 4.85 & @xmath42 + 27 & 0750 + 535 & 602 . & 2.26 & @xmath410.57 & @xmath61.44 & 678 . & 1.55 & + 28 & 0753 + 373 & 287 . & 2.85 & @xmath411.05 & @xmath61.15 & 292 . & 2.5 & + 29 & 0753 + 519 & 221 . & 3.66 & @xmath410.58 & @xmath61.22 & 242 . & 2.52 & + 30 & 0758 + 594 & 221 . & 2.84 & @xmath410.635 & @xmath60.711 & 225 . & 2.27 & + 31 & 0849 + 675 & 308 . & 2.72 & @xmath410.44 & @xmath61.11 & 347 . & 1.66 & + 32 & 0925 + 745 & 284 . & 4.95 & @xmath61.07 & @xmath410.554 & 287 . & 4.26 & + 33 & 1017 + 436 & 212 . & 6.49 & @xmath410.728 & @xmath62.22 & 246 . & 4.85 & @xmath42 + 34 & 1019 + 429 & 277 . & 7.5 & @xmath410.496 & @xmath61.3 & 314 . & 4.85 & @xmath42 + 35 & 1032 + 509 & 180 . & 7.71 & @xmath410.124 & @xmath61.16 & 245 . & 2.81 & + 36 & 1055 + 433 & 267 . & 0.69 & @xmath411.44 & @xmath60.237 & 308 . & 1.42 & @xmath42 + 37 & 1107 + 485 & 712 . & 0.69 & @xmath410.645 & @xmath60.845 & 734 . & 0.528 & + 38 & 1125 + 366 & 70.4 & 0.347 & @xmath410.292 & @xmath412.14 & & & + 39 & 1138 + 644 & 228 . & 9.59 & @xmath410.192 & @xmath61.68 & 307 . & 4.85 & @xmath42 + 40 & 1206 + 415 & 449 . & 10.4 & @xmath62.04 & @xmath410.475 & 492 . & 7.29 & + 41 & 1226 + 638 & 422 . & 2.89 & @xmath410.751 & @xmath61.24 & 446 . & 2.21 & + 42 & 1232 + 366 & 198 . & 8.41 & @xmath410.372 & @xmath61.29 & 234 . & 4.85 & @xmath42 + 43 & 1239 + 606 & 435 . & 1.41 & @xmath411.31 & @xmath60.93 & 435 . & 1.42 & @xmath42 + 44 & 1245 + 676 & 258 . & 1.16 & @xmath411.29 & @xmath60.561 & 264 . & 1.42 & @xmath42 + 45 & 1256 + 546 & 1090 . & 0.539 & @xmath410.901 & @xmath60.925 & 1110 . & 0.471 & + 46 & 1311 + 552 & 1980 . & 0.555 & @xmath411.33 & @xmath60.851 & 1980 . & 0.578 & + 47 & 1321 + 410 & 476 . & 2.83 & @xmath411.09 & @xmath60.98 & 478 . & 2.62 & + 48 & 1338 + 381 & 330 . & 3.27 & @xmath410.771 & @xmath61.21 & 347 . & 2.53 & + 49 & 1357 + 404 & 356 . & 12.1 & @xmath410.349 & @xmath410.349 & & & + 50 & 1403 + 411 & 223 . & 6.71 & @xmath410.346 & @xmath61.58 & 276 . & 3.92 & + 51 & 1454 + 447 & 206 . & 2.36 & @xmath410.588 & @xmath60.407 & 206 . & 2.43 & + 52 & 1532 + 680 & 455 . & 2.57 & @xmath410.329 & @xmath61.32 & 551 . & 1.42 & @xmath42 + 53 & 1534 + 501 & 341 . & 6.97 & @xmath410.584 & @xmath61.09 & 367 . & 4.85 & @xmath42 + 54 & 1544 + 398 & 221 . & 6.4 & @xmath410.363 & @xmath64.67 & 310 . & 4.85 & @xmath42 + 55 & 1607 + 563 & 261 . & 3.54 & @xmath410.538 & @xmath60.797 & 273 . & 2.48 & + 56 & 1627 + 476 & 219 . & 3.06 & @xmath410.241 & @xmath61.11 & 271 . & 1.42 & @xmath42 + 57 & 1630 + 358 & 466 . & 2.5 & @xmath410.377 & @xmath61.11 & 537 . & 1.42 & @xmath42 + 58 & 1745 + 670 & 548 . & 2.7 & @xmath410.266 & @xmath61.46 & 696 . & 1.42 & @xmath42 + 59 & 1753 + 648 & 220 . & 7.26 & @xmath410.519 & @xmath61.66 & 257 . & 4.85 & @xmath42 + 60 & 1755 + 578 & 785 . & 2.07 & @xmath411.06 & @xmath61.07 & 793 . & 1.86 & + 61 & 1801 + 459 & 231 . & 2.78 & @xmath410.337 & @xmath60.77 & 256 . & 1.46 + 62 & 1815 + 614 & 580 . & 3.9 & @xmath410.0731 & @xmath61.69 & 852 . & 1.61 & + 63 & 1820 + 397 & 753 . & 0.457 & @xmath60.519 & @xmath410.527 & 761 . & 0.575 & + 64 & 1839 + 548 & 240 . & 5.62 & @xmath410.436 & @xmath60.983 & 265 . & 3.42 & + 65 & 1946 + 708 & 965 . & 1.92 & @xmath410.818 & @xmath60.755 & 970 . & 1.72 & + 66 & 2000 + 472 & 946 . & 4.11 & @xmath410.996 & @xmath60.588 & 949 . & 4.51 & + 67 & 2005 + 642 & 227 . & 0.502 & @xmath410.346 & @xmath413.4 & & & + 68 & 2013 + 508 & 363 . & 1.67 & @xmath410.849 & @xmath60.927 & 369 . & 1.42 & @xmath42 + 69 & 2014 + 463 & 153 . & 8.03 & @xmath410.231 & @xmath61.88 & 205 . & 4.44 & + 70 & 2119 + 709 & 167 . & 9.07 & @xmath410.217 & @xmath61.79 & 224 . & 4.85 & @xmath42 + 71 & 2151 + 431 & 257 . & 3.25 & @xmath410.63 & @xmath60.917 & 268 . & 2.41 & + 72 & 2248 + 555 & 469 . & 4.72 & @xmath410.512 & @xmath60.633 & 481 . & 3.43 & + 73 & 2253 + 417 & 1840 . & 1.83 & @xmath410.706 & @xmath60.909 & 1900 . & 1.44 & + 74 & 2310 + 385 & 682 . & 3.2 & @xmath410.6 & @xmath61.14 & 735 . & 2.25 & + 75 & 2341 + 697 & 155 . & 7.5 & @xmath410.147 & @xmath61.59 & 213 . & 3.45 & + 76 & 2356 + 385 & 560 . & 3.85 & @xmath410.524 & @xmath61.34 & 632 . & 2.55 & + 1 & 0001 + 478 & & & & & & & & + 2 & 0015 + 529 & & & & & & & & + 3 & 0046 + 511 & & & & & & & & + 4 & 0051 + 679 & & & & & & & & + 5 & 0058 + 498 & & & & & & & & + 6 & 0102 + 480 & @xmath42 & & & & & & & + 7 & 0102 + 511 & & & & & & & & + 8 & 0123 + 731 & @xmath42 & & & & & & & + 9 & 0129 + 560 & & & & & & & & + 10 & 0140 + 490 & & & & & & & & + 11 & 0148 + 546 & & & & & & & & + 12 & 0153 + 389 & & & & & & & & + 13 & 0307 + 380 & & & @xmath42 & & @xmath42 & qso & 0.816 & vermeulen & taylor , 1995 + 14 & 0335 + 599 & & & & & & & & + 15 & 0336 + 473 & & & & & & & & + 16 & 0412 + 447 & & & & & & & & + 17 & 0454 + 550 & & & & & & & & + 18 & 0513 + 714 & & & & & & & & + 19 & 0514 + 474 & & & @xmath42 & & & & & + 20 & 0610 + 510 & & & & & & qso & 1.59 & hook et al . , 1996 + 21 & 0627 + 532 & @xmath42 & & & & @xmath42 & qso & 2.204 & henstock et al . , 1997 + 22 & 0630 + 497 & @xmath42 & & & & & & & + 23 & 0651 + 410 & & & & & @xmath42 & g & 0.02156 & marzke et al . , 1996 + 24 & 0652 + 577 & & & & & & & & + 25 & 0655 + 696 & & & & & & qso & 1.971 & moran et al , 1996 + 26 & 0718 + 374 & & @xmath42 & & & & & & + 27 & 0750 + 535 & & @xmath42 & & & & & & + 28 & 0753 + 373 & & @xmath42 & & & & & & + 29 & 0753 + 519 & & @xmath42 & & & & qso & 1.33 & hook et al . , 1996 + 30 & 0758 + 594 & & & & & & & & + 31 & 0849 + 675 & & & & & & & & + 32 & 0925 + 745 & @xmath42 & & & & & & & + 33 & 1017 + 436 & & @xmath42 & & & & qso & 1.96 & hook et al . , 1996 + 34 & 1019 + 429 & @xmath42 & @xmath42 & @xmath42 & & & & & + 35 & 1032 + 509 & & @xmath42 & & & & & & + 36 & 1055 + 433 & @xmath42 & @xmath42 & @xmath42 & & & & & + 37 & 1107 + 485 & @xmath42 & @xmath42 & & & & qso & 0.74 & hook et al . , 1996 + 38 & 1125 + 366 & & @xmath42 & & & & & & + 39 & 1138 + 644 & & & & & & & & + 40 & 1206 + 415 & & @xmath42 & & & @xmath42 & & & + 41 & 1226 + 638 & & & & & & & & + 42 & 1232 + 366 & @xmath42 & @xmath42 & & & & qso & 1.60 & hook et al . , 1996 + 43 & 1239 + 606 & & & & & & & & + 44 & 1245 + 676 & & & & & & g & 0.103 & marzke et al , 1996 + 45 & 1256 + 546 & & @xmath42 & & & & & & + 46 & 1311 + 552 & @xmath42 & @xmath42 & & & @xmath42 & qso & 0.613 & vermeulen et al . , 1996 + 47 & 1321 + 410 & @xmath42 & @xmath42 & & & @xmath42 & qso & 0.496 & vermeulen et al . , 1996 + 48 & 1338 + 381 & @xmath42 & @xmath42 & & & & qso & 3.103 & hook et al . , 1995 + 49 & 1357 + 404 & & @xmath42 & & & & & & + 50 & 1403 + 411 & & @xmath42 & @xmath42 & & & & & + 51 & 1454 + 447 & & @xmath42 & & & & & & + 52 & 1532 + 680 & & & & @xmath42 & & & & + 53 & 1534 + 501 & @xmath42 & @xmath42 & & & @xmath42 & qso & 1.119 & vermeulen & taylor , 1995 + 54 & 1544 + 398 & @xmath42 & @xmath42 & @xmath42 & & & & & + 55 & 1607 + 563 & & @xmath42 & & & & & & + 56 & 1627 + 476 & & @xmath42 & & & & & & + 57 & 1630 + 358 & & @xmath42 & & & & & & + 58 & 1745 + 670 & & & & @xmath42 & & & & + 59 & 1753 + 648 & & & & @xmath42 & & & & + 60 & 1755 + 578 & @xmath42 & & & @xmath42 & @xmath42 & qso & 2.110 & henstock et al . , 1997 + 61 & 1801 + 459 & & & @xmath42 & & & & & + 62 & 1815 + 614 & & & & @xmath42 & @xmath42 & qso & 0.601 & vermeulen & taylor , 1995 + 63 & 1820 + 397 & @xmath42 & & @xmath42 & & & & & + 64 & 1839 + 548 & & & & & & & & + 65 & 1946 + 708 & @xmath42 & & & @xmath42 & @xmath42 & g & 0.101 & stickel & khr , 1993 + 66 & 2000 + 472 & & & & & & & & + 67 & 2005 + 642 & & & & @xmath42 & @xmath42 & qso & 1.574 & henstock et al . , 1997 + 68 & 2013 + 508 & & & & & & & & + 69 & 2014 + 463 & & & & & & & & + 70 & 2119 + 709 & & & & & & & & + 71 & 2151 + 431 & & & & & & & & + 72 & 2248 + 555 & & & & & & & & + 73 & 2253 + 417 & @xmath42 & & & & & qso & 1.476 & hewitt & burbidge , 1989 + 74 & 2310 + 385 & & & & & @xmath42 & qso & 2.17 & hewitt & burbidge , 1989 + 75 & 2341 + 697 & & & & & & & & + 76 & 2356 + 385 & @xmath42 & & & & @xmath42 & qso & 2.704 & stickel & khr , 1994 + 0001 + 478 & gb flux is 73% greater than nvss flux + 0046 + 511 & no 2nd epoch data + 0051 + 679 & gb flux is 36% greater than nvss flux + 0058 + 498 & no 2nd epoch data + 0307 + 380 & poor fit to the broken power - law curve + 0412 + 447 & gb flux is 4.7 times greater than nvss flux , inverted spectrum only + 0651 + 410 & no 2nd epoch data + 0655 + 696 & gb flux is 55% less than nvss flux + 0718 + 374 & no 2nd epoch data + 0753 + 519 & no 2nd epoch data + 0758 + 594 & gb flux is 34% less than nvss flux + 1017 + 436 & no 2nd epoch data + 1055 + 433 & b3 flux ( 420 mjy ) and merlin 408 mhz flux ( 182 mjy ) do not match + 1125 + 366 & first flux does not match nvss and gb fluxes , inverted spectrum only + 1206 + 415 & first flux does not match nvss and gb fluxes + 1232 + 366 & first flux matches nvss flux but they both do nt match gb and s4 fluxes + & extended component ? + 1357 + 404 & first flux does not match nvss and gb fluxes , inverted spectrum only + 1544 + 398 & b3 flux matches merlin flux but first flux does not match nvss and gb fluxes + 1839 + 548 & no 2nd epoch data + 2005 + 642 & inverted spectrum only + 2013 + 508 & gb flux is 48% less than nvss flux + 2014 + 463 & no 2nd epoch data + 2248 + 555 & no 2nd epoch data + 2341 + 697 & no 2nd epoch data +

in the leptonic scenario for tev emission from a few well - observed shell - type tev supernova remnants ( sttsnrs ) , very weak magnetic fields are inferred . if fast - mode waves are produced efficiently in the shock downstream , we show that they are viable agents for acceleration of relativistic electrons inferred from the observed spectra even in the subsonic phase , in spite that these waves are subject to strong damping by thermal background ions at small dissipation scales . strong collisionless non - relativistic astrophysical shocks are studied with the assumption of a constant aflvn speed in the downstream . the turbulence evolution is modeled with both the kolmogorov and kraichnan phenomenology . processes determining the high - energy cutoff of nonthermal electron distributions are examined . the kraichnan models lead to a shallower high - energy cutoff of the electron distribution and require a lower downstream density than the kolmogorov models to fit a given emission spectrum . with reasonable parameters , the model explains observations of sttsnrs , including recent data obtained with the _ fermi _-ray telescope . more detailed studies of the turbulence generation and dissipation processes , supernova explosions and progenitors are warranted for better understanding the nature of supernova shocks . [ firstpage ] acceleration of particles mhd plasmas shock waves turbulence ism : supernova remnants .

0909.3349

the acceleration of cosmic rays up to @xmath1 ev has been attributed to supernova explosions and tev emission is expected from the remnants @xcite . the standard diffusive shock particle acceleration ( dsa ) model has been successful in explaining emissions from most supernova remnants @xcite . investigations of acceleration by a spectrum of turbulent plasma waves , the so - called stochastic particle acceleration ( sa ) , also have a long and resilient history @xcite . most authors prefer the use of relativistic leptons to account for the nonthermal radio , x - rays , and tev emissions from the remnants . the tev emission has also been attributed to energetic protons and ions @xcite . although the dsa can naturally give a universal power - law energetic particle distribution with the spectral index determined by the shock compression ratio for a linear model , it requires well - defined shock structure and efficient scatter of high - energy particles by small - scale turbulence @xcite . the initial acceleration of low - energy particles to a high enough energy for the shock to be effective , the so - called injection problem , is likely due to the sa by turbulence . @xcite showed that the sa by small - scale alfvn waves can be more efficient than the first order fermi acceleration by shocks . @xcite derived approximate diffusion coefficients and showed that stochastic interactions of particles with a spectrum of plasma waves can lead to efficient particle acceleration . over the past few decades , the sa has also been explored for broader astrophysical applications ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the essential challenges to the sa model are a self - consistent treatment of the nonlinear turbulence spectral evolution and the requirement of the same energy dependence of the acceleration and escape timescales for the production of a power - law particle distribution @xcite . recently gibbsian theory has been generalized to account for power - law distributions in marginally stable gibbsian equilibria @xcite . it remains to be shown how the physical processes of the sa are related to the ordering parameter @xmath2 of this statistics . most previous studies assume isotropic magnetohydrodynamic ( mhd ) waves for the turbulence @xcite . the anisotropy of the turbulence caused by cascade and damping processes has been considered recently @xcite . in particular , @xcite show that the scatter and acceleration rates of charged relativistic particles by fast - mode waves in a high-@xmath3 plasma may be much higher than those given by the standard quasi - linear theory with an isotropic wave power spectrum . over the past few years , detailed radio , x - ray , @xmath0-ray , and tev observations of a few shell - type tev supernova remnants ( sttsnrs ) pose several challenges to the classical dsa model in the hadronic scenario , where the tev emission is produced through neutral pion decays induced by proton - proton scatter @xcite . besides requiring efficient amplification of the magnetic field in the upstream plasma and a good correlation between the magnetic field and background plasma density @xcite , the model also implies a cosmic ray energy of @xmath4 ergs for each remnant and an electron acceleration efficiency more than 4 orders of magnitude lower than the ion acceleration efficiency @xcite . the high density of the upstream plasma in the model also implies significant thermal x - ray emission from the downstream , which may exceed the observed upper limit @xcite . a very hard proton spectrum is also required to fit the @xmath0-ray spectrum obtained recently with the _ fermi _ @xmath0-ray telescope @xcite . although these kinds of remnants may be atypical , detailed modeling can still have profound implications for our understanding of supernova shocks @xcite . in a previous paper , we showed that the sa of electrons by turbulent plasma waves in the shock downstream might naturally explain these observations @xcite . turbulence is expected given that the size of these remnants are many orders of magnitude larger than the dissipation scale of the ion inertial length @xcite . the dsa model proposes that particle acceleration occurs directly and predominantly at the ion inertial length or gyro - radius @xcite . this requires the absence of instabilities over a large range of spatial scales , which is highly idealized . @xcite showed that density fluctuations in the upstream can be amplified significantly by shock waves , resulting in strong turbulence in the downstream . magnetized turbulence appears to be a more generic and natural energy dissipation channel than the short - length - scale shock fronts ( sf ) . with the leptonic model , the tev emission is mostly produced by the inverse comptonization of the cosmic microwave background radiation by tev electrons @xcite . the magnetic field required to reproduce the observed x - ray flux by the same tev electrons through the synchrotron process implies a spectral cutoff in the hard x - ray band , which is in agreement with observations . the sa model also requires a much lower gas density than the dsa model , which not only explains the lack of thermal x - ray emission from the shell of the remnants , but also reduces the energetics of the supernovae . the required turbulence generation scale is comparable to the size of the observed x - ray filaments as well @xcite . the fast variability of small x - ray features may be attributed to rapid spatial diffusion of high energy electrons @xcite . as we will show in this paper , the agreement between preliminary results from _ fermi _ observations and the model prediction is also impressive @xcite . the sa by fast - mode waves has been studied by several authors @xcite . both resonant and nonresonant interactions have been considered . in this paper , we consider the nonresonant acceleration by compressional waves first studied by @xcite . compared with these original studies , our model has several distinct features : 1 ) most of the dissipated fast - mode turbulence energy is absorbed by thermal background ions ; 2 ) the residual fast - mode waves in the dissipation scales propagate along local magnetic fields and preferentially accelerate electrons in the background plasma ; 3 ) the plasma physics processes in the dissipation range in principle may lead to a self - consistent treatment of the electron injection process at low energies ; 4 ) the scatter mean - free - path of relativistic electrons , which is a free parameter in most of the previous studies , is determined by the characteristic length of the magnetic field , which in a high-@xmath3 plasma is reduced by strong turbulence motions significantly ; 5 ) the high - energy cutoff of the particle distribution is tied to the characteristic length of the magnetic field . in this paper , we first discuss the sa by decaying turbulence in general and show that fast - mode waves may account for observations of a few sttsnrs ( section [ ts ] ) . in section [ shock ] , we present the structure of the downstream turbulence with both the kolmogorov and kraichnan phenomenology for the turbulence cascade . the transit - time damping ( ttd ) by the thermal background particles of compressional fast - mode turbulence is considered in the dissipation range . the stochastic acceleration of relativistic electrons by fast - mode wave turbulence in the subsonic phase is discussed in section [ sa ] , where physical processes determine the acceleration of the highest energy electrons are discussed . the models are applied to the well - observed tev snr rx j1713.7 - 3946 in section [ application ] . although the inferred plasma density is much lower than that in the hadronic scenario , some models still have significantly higher densities than that derived from _ xmm - newton _ observations @xcite . in section [ disc ] , we discuss how the density may be further reduced by considering the turbulence generation processes and first - order fermi acceleration in the supersonic phase . conclusions are drawn in section [ con ] . according to the classical fermi mechanism @xcite , the particle acceleration rate is determined primarily by the scatter mean - free - path @xmath5 and the velocity of the scatter agents @xmath6 @xcite . very general constraints can be obtained on the nature of these processes when applying this mechanism to specific observations . for example , @xcite showed that the acceleration process must be selective in the sense that only a fraction of the background particles are accelerated to very high - energies . otherwise , these stochastic interactions likely lead to plasma heating instead of a very broad energy distribution of accelerated particles as frequently observed in dynamically evolving collisionless astrophysical plasmas . for particle acceleration in solar flares , @xcite argued that this selective acceleration could be achieved in the energy domain ( i.e. , the frequency domain for the waves ) through cyclotron resonances of particles with a spectrum of cascading plasma waves . in the dsa model , the particle acceleration at low energies and the shock structure determine the efficiency of different particle species @xcite . in the presence of a magnetic field , selective energization may also be realized in the domain of the particle pitch - angle and/or wave direction angle with respect to the magnetic field @xcite . further constraints can be put on the sa by a spectrum of turbulence . given the small gyro - radii of charged particles in magnetized astrophysical plasmas , charged particles couple strongly through the magnetic field . as a result , the turbulence responsible for the sa will decay as the energy carrying plasma being carried away from the source region of the turbulence by large scale flows and/or magnetic fields for a high and/or low @xmath3 plasma , respectively . this is the case for the sa in a shock downstream with a high value of the plasma @xmath3 , where the turbulence is generated at the sf and its intensity decreases as the flow moves away from the sf @xcite . we next discuss constraints on such a particle acceleration scenario . in the _ kolmogorov _ phenomenology for the turbulence cascade @xcite , the free energy dissipation rate is given by @xmath7 where @xmath8 is a dimensionless constant , @xmath9 is the mass density , and @xmath6 and @xmath10 are the eddy speed and the turbulence generation scale , respectively . the eddy turnover speed and time at smaller scales are given respectively by @xmath11 where @xmath12 is the isotropic turbulence power spectrum , @xmath13 is the wave number and @xmath5 is the eddy size . from the three - dimension kolmogorov constant @xmath14 @xcite , we obtain @xmath15 at the turbulence generation scale @xmath16 , @xmath17 , @xmath18^{1/2}= c_1\rho v^2_{edd}(k)/ \tau_{edd}(k)\,,$ ] and the total turbulence energy is given by @xmath19 the turbulence decay time is therefore given by @xmath20 where @xmath21 indicates the time , i.e. , eddies decay after making @xmath22 turn . we are interested in the acceleration of particles through scattering randomly with heavy scatter centers with the corresponding acceleration time given by @xcite @xmath23\ , , \label{taukol}\ ] ] where @xmath24 is the scatter time , @xmath25 is the particle speed , and we have assumed that the scatter mean free - path is equal to @xmath5 . for the above isotropic kolmogorov turbulence spectrum , @xmath26 to have significant stochastic particle acceleration , the acceleration time @xmath27 should be shorter than the turbulence decay time @xmath28 , which implies @xmath29 so , in general , the sa is more efficient at smaller scales . the onset scale of the sa is given by @xmath30 , where @xmath31 . therefore , to produce energetic particles with a speed of @xmath25 by a kolmogorov spectrum of scatter centers , the turbulence must have a dynamical range greater than @xmath32 this acceleration process corresponds to the acceleration by incompressional motions studied by @xcite . in the _ iroshnikov kraichnan _ phenomenology @xcite , the turbulence cascade rate is suppressed by the wave propagation effect by a factor of @xmath33 , where @xmath34 ( independ of @xmath35 ) is the wave group speed : @xmath36 then we have @xmath37 and the turbulence decay time is given by @xmath38 where the wave speed @xmath39 . to have significant acceleration through scatter with the eddies , the dynamical range of the turbulence must be greater than @xmath40 which is much less than @xmath41 . the resonant interactions of particles with waves may be more effective in accelerating particles in this case than the interactions with eddies . for a wave phase speed @xmath34 independent of @xmath35 and a wave spectrum of @xmath42 , where @xmath43 is the wave spectral index , the standard quasi - linear theory gives a scatter time of @xmath44 r_g / v$ ] for particles with a gyro - radius of @xmath45 . and the acceleration time is given by @xmath46 , which is essentially the same as equations ( [ taukol ] ) and ( [ tsckol ] ) . although the acceleration rate depends on the phase speed of these waves , the scatter rate is proportional to the intensity of waves in resonance with the particles . so the acceleration is not enhanced by these kinds of resonant interactions . in the presence of magnetic fields , the stochastic acceleration of charged particles can be very efficient @xcite . @xcite reviewed several mechanisms by mhd waves , which can accelerate particles within the turnover time of large scale eddies for appropriate particle spatial diffusion coefficients . some of these mechanisms likely result in plasma heating instead of particle acceleration as the particle energization processes are not selective . the corresponding waves are also likely subject to efficient damping by the thermal background particles . there are other mechanisms , which preferentially accelerate energetic particles . for isotropic acoustic wave turbulence with a wave phase speed @xmath34 , @xcite showed that the stochastic acceleration timescale of energetic particles with a spatial diffusion coefficient @xmath47 is given by ( see also @xcite ) @xmath48d / u^2\ , , \label{tac0}\ ] ] where @xmath49 is the overall wave intensity , @xmath50 indicates the wave generation scale , and below the turbulence dissipation scale @xmath51 the wave intensity is negligible . @xmath52 depends on @xmath43 and is on the order of unity ( e.g. , @xmath53 ) . the results for fast and slow diffusion have been combined here approximately to give a unified expression @xcite . for @xmath54 , @xmath55 increases monotonically with @xmath56 . to have significant acceleration , the minimum acceleration timescale @xmath57 must be less than @xmath58 . then the scatter mean - free - path of the particles @xmath59 must be shorter than @xmath60 it is also possible that the dynamics of the turbulence cascade is not affected by the wave propagation , which only enhances the particle acceleration rate . then @xmath61 , the dynamical range required for the sa to be significant is given by @xmath62 several sttsnrs have been observed extensively in the radio , x - ray , and tev bands . x - ray observations with the _ chandra _ , _ xmm - newton _ , and _ suzaku _ , and tev observations with the hess have made several surprising discoveries that challenge the dsa model in the hadronic scenario @xcite . the leptonic scenario , on the other hand , is relative simple except that the electron acceleration mechanism needs to be addressed @xcite . the tev snr rx j1713.7 - 3946 is about @xmath63 years old @xcite with a radius of @xmath64 pc and a distance of @xmath65 kpc . by fitting its broadband spectrum with an electron distribution of @xmath66 ( the dashed lines in fig . [ spec ] ) , we find that @xmath67 , @xmath68 g , @xmath69 tev , and the total energy of relativistic electrons with the lorentz factor @xmath70 @xmath71 erg . the x - ray emitting electrons have a gyro - radius of @xmath72 cm , which should be shorter than the particle scatter mean - free - path . to produce these electrons through the sa , the turbulence must be generated on scales greater than @xmath73 , @xmath74 , @xmath75 , and @xmath76 for the non - resonant kolmogorov , kraichnan , acoustic kolmogorov , and kraichnan phenomenology , respectively . for sttsnrs , @xmath77 , @xmath78 kpc , which is much greater than the radii of the remnants . the sa by eddies with a kolmogorov spectrum is therefore insignificant . the standard quasi - linear theory also predicts negligible sa . @xmath79 pc , which is also too thick . @xmath80 pc , which is much greater than the ion inertial length and may result from kelvin - helmholtz instabilities or cosmic ray drifting upstream @xcite . therefore if relativistic electrons from the sttsnrs are accelerated through the sa , they may be energized by high - speed compressional plasma waves . previously @xcite showed that the non - selective acceleration as given by equations ( [ taukol ] ) and ( [ taukra ] ) leads to plasma heating instead of particle acceleration . here we give a different argument against these acceleration processes . the acceleration timescale in the following is given by equation ( [ tac0 ] ) with @xmath56 depending on the wave spectrum and particle energy . the dsa models are usually favored over the sa models for two main reasons : 1 ) the dsa can naturally produce a power - law particle distribution ; 2 ) the dsa corresponds to a first - order fermi mechanism and is presumably more efficient than the sa , which corresponds to a second - order fermi mechanism . while how shocks produce power - law high - energy particle distributions is relatively well understood , the second reason appears to be a misconception . it is true that the ratios of the acceleration and scatter rates are proportional to the first and second power of the speed ratio of the scatter agent and the particle for the dsa and sa , respectively . but in the dsa models , there are two scatter processes : scatter of particles by turbulence , which causes the particle diffusion , and the particle crossing of the sf due to this diffusion . the sf crossing rate is about a factor of @xmath81 lower than the particle scatter rate by the turbulence for a shock speed of @xmath82 @xcite . as a result , the acceleration rates of the dsa and sa models are both proportional to @xmath83 . therefore the sa is not necessarily less efficient than the dsa . the sa at small scales can also be enhanced by high speed kinetic plasma waves @xcite . of course , the speed of the scatter agent accessible to a particle may depend on the scatter mean - free - path @xmath5 in the sa . the stochastic acceleration rate can be very low if @xmath84 decreases quickly with the decrease of @xmath5 . for the dsa and sa by large - scale acoustic waves in the shock downstream , the speed of the scatter agent is proportional to the shock speed . one just needs to reduce the scatter mean - free - path @xmath5 to enhance the acceleration rate . as we will show in sections [ shock ] and [ sa ] , considering the competition of isotropic cascade with anisotropic damping of fast - mode waves through the transit - time - damping process , the waves in the dissipation range propagate along local magnetic field lines with a spectrum index of @xmath85 . in a high-@xmath3 plasma , the damping is mostly caused by ions . the residual parallel propagating waves however preferentially accelerate electrons . enegetic particles are also subject to efficient acceleration by large - scale fast - mode ( acoustic ) waves with energy independent acceleration and scatter rates , giving rise to power - law particle distributions in the steady - state . at even higher energies , the gyro - radius of the particles exceeds the characteristic length of the magnetic field , which also corresponds to the dissipation scale , the spatial diffusion coefficient @xmath56 increases quickly with energy . however , due to interactions with large - scale turbulence , both the acceleration and spatial diffusion timescales vary gradually with @xmath56 , implying a gradual high - energy cutoff . we consider the relatively simple case , where both the thermal pressure and magnetic field are negligible in the upstream and the shock normal is parallel to the plasma flow . the mass , momentum , and energy fluxes are given , respectively , by @xmath86 , @xmath87 , and @xmath88 , where @xmath89 , @xmath90 , and @xmath91 are the speed , pressure , and energy density of the plasma flow , respectively . based on the parameters inferred from the leptonic scenario for the tev emission from sttsnrs shown in section [ ts ] , the plasma @xmath3 in the shock downstream is likely high and the fast - mode wave damping by thermal background is then dominated by protons and ions @xcite . the alfvn speed given by @xmath92 therefore must be much smaller than the turbulence speed @xmath6 near the shock front , where @xmath93 is the magnetic field intensity . for strong non - relativistic shocks with the shock frame upstream speed @xmath94 much higher than the speed of the parallel propagating fast mode waves in the upstream @xmath95 , where @xmath96 is the isothermal sound speed and @xmath97 is the thermal pressure of the gas , @xmath98 and @xmath99 in the upstream . in the downstream , the pressure and energy density have contributions from the thermal gas and turbulence : @xmath100 and @xmath101 , where we have assumed that the turbulence behaves as an ideal gas and ignored possible dynamical effects of the wave propagation . then we have @xcite @xmath102 where the subscripts @xmath6 and @xmath103 denote the upstream and downstream , respectively , and we have ignored the effects of the electromagnetic fields and the thermal energy and pressure in the upstream . then we have @xmath104 this is slightly different from that given by @xcite , where we assumed that the pressure and enthalpy of the turbulent magnetic field are given by @xmath105 and @xmath106 , respectively . the shock structure can be complicated due to the present of turbulence . we assume that the turbulence is generated isotropically and has a generation scale of @xmath10 , which does not change in the downstream . the speeds @xmath107 , @xmath108 , and @xmath6 therefore should be considered as averaged quantities on the scale @xmath10 . @xmath108 depends on the upstream conditions and/or the dynamo process of magnetic field amplification @xcite . here we assume it a constant in the downstream . one can then quantify the evolution of other speeds in the downstream . for the kolmogorov phenomenology @xcite , @xmath109 near the sf , we denote the isothermal sound speed and aflvn speed by @xmath110 and @xmath111 , respectively . then the eddy speed at the sf is given by @xmath112 with @xmath113 . integrate equation ( [ ux ] ) from the sf @xmath114 to downstream ( @xmath115 ) , we then have @xmath116^{1/2}\ , , \\ { v_f(x)\over u}&= & \left[{5\over 16}-{5\over 48\left(c_1x/3l+a^{-1/2}\right)^2 } + { v_a^2\over u^2}\right]^{1/2}\,.\end{aligned}\ ] ] and the speed of parallel propagating fast mode waves @xmath34 in the downstream . the thick and thin lines are for the kraichnan with @xmath117 ( left ) and kolmogorov with @xmath118 ( right ) phenomenology , respectively . [ vs ] , title="fig:",scaledwidth=48.0% ] and the speed of parallel propagating fast mode waves @xmath34 in the downstream . the thick and thin lines are for the kraichnan with @xmath117 ( left ) and kolmogorov with @xmath118 ( right ) phenomenology , respectively . [ vs ] , title="fig:",scaledwidth=48.0% ] as mentioned in section [ ts ] , to produce the observed x - ray emitting electrons in the sttsnrs through the sa processes , fast - mode waves needs to be excited efficiently . the mhd wave period is given by @xmath119 . then the transition from the kolmogorov to kraichnan phenomenology occurs at the scale , where @xmath120 or @xmath121 @xcite . we then have @xmath122 for @xmath123 , the turbulence spectrum is kolmogorov like given by equation ( [ kol ] ) . for @xmath124 , the turbulence spectrum in the inertial range is given by @xmath125 although the turbulence energy exceeds @xmath126 when the wave propagation effect is considered , we still assume that the enthalpy of the turbulence is given by @xmath127 for @xmath128 so that equation ( [ energy ] ) and the above solutions for the speed profiles remain valid . in the subsonic phase with @xmath129 , @xmath130 and @xmath131 where from equation ( [ energy ] ) one has @xmath132^{1/2}.$ ] equation ( [ ik ] ) can be solved numerically to get the speed profiles in the subsonic phase . figure [ vs ] shows the @xmath34 and @xmath6 profiles with @xmath133 in the downstream . the thick and thin lines are for the kraichnan and kolmogorov phenomenology , respectively . in summary , for @xmath134 , the turbulence spectrum in the inertial range is given by @xmath135 and for @xmath123 , @xmath136 . the transit - time damping ( ttd ) of compressional fast - mode waves starts at the characteristic length of the magnetic field @xmath137 where the alfvn speed is comparable to the eddy speed , i.e. , @xmath138 at even larger scales , the vortex motions produce random magnetic fields comparable with the mean field reducing the scatter mean - free - path of charged background particles to @xmath139 or even shorter scales . this trapping of charged particles within a scale of @xmath139 prevents the ttd on scales above @xmath139 . the incompressional modes are not subject to the ttd and have different spectra @xcite . in what follows , we only consider the compressional ( fast magnetosonic ) wave modes . then we have @xmath140 since @xmath141 , one has @xmath142 , and damping is negligible in the regime , where @xmath123 . for a fully ionised hydrogen plasma with isotropic particle distributions , which is reasonable in the absence of strong large - scale magnetic fields , the ttd rate is given by @xcite @xmath143 \label{d1}\end{aligned}\ ] ] where @xmath144 , @xmath145 , @xmath146 , @xmath147 , @xmath148 , @xmath149 , @xmath150 , and @xmath151 are the boltzmann constant , electron and proton temperatures and masses , angle between the wave vector and mean magnetic field , wave frequency , and parallel component of the wave vector , respectively . the first and second terms in the brackets on the right hand side correspond to damping by electrons and protons , respectively . for weakly magnetized plasmas with @xmath152 , proton damping always dominates the ttd for @xmath153 . , the electron damping term dominates when @xmath154 . ] if @xmath108 does not change dramatically in the downstream , the continuous heating of background particles through the ttd processes makes @xmath155 since the heating rates are proportional to @xmath156 , where @xmath157 and @xmath158 represent the mass and temperature of the particles , respectively . we see that parallel propagating waves ( with @xmath159 ) are not subject to the ttd processes and can accelerate some particles to relativistic energies through resonant interactions . obliquely propagating waves are damped efficiently by the background particles . although the damping rates for waves propagating nearly perpendicular to the magnetic field ( @xmath160 ) are also low , these waves are subject to damping by magnetic field wandering @xcite . at several locations in the downstream indicated in the legend for the speed profiles in figure [ vs ] . in the dissipation range , it corresponds to @xmath161 instead of @xmath162 . the onset of the ttd at @xmath163 causes the discontinuity for the former . the left panel is for the kraichnan phenomenology . the kolmogorov , kraichnan , and damping ranges are indicated for the supersonic phase spectrum with @xmath164 . at the other locations , the turbulence is subsonic and there are only kraichnan and damping ranges . the right panel is for the kolmogorov phenomenology . [ vedd],title="fig:",scaledwidth=48.0% ] at several locations in the downstream indicated in the legend for the speed profiles in figure [ vs ] . in the dissipation range , it corresponds to @xmath161 instead of @xmath162 . the onset of the ttd at @xmath163 causes the discontinuity for the former . the left panel is for the kraichnan phenomenology . the kolmogorov , kraichnan , and damping ranges are indicated for the supersonic phase spectrum with @xmath164 . at the other locations , the turbulence is subsonic and there are only kraichnan and damping ranges . the right panel is for the kolmogorov phenomenology . [ vedd],title="fig:",scaledwidth=48.0% ] the turbulence power spectrum cuts off sharply when the damping rate becomes comparable to the turbulence cascade rate @xcite @xmath165 one can define a critical propagation angle @xmath166 , where @xmath167 . equations ( [ ik ] ) and ( [ d1 ] ) then give @xmath168 where the electron damping term has been ignored . the angular - integrated turbulence spectrum in the dissipation range is therefore given by @xmath169 interestingly the turbulence spectrum is inversely proportional to @xmath170 in both scenarios . the angular - integrated turbulence spectra @xmath171 for the velocity profiles in figure [ vs ] at several locations in the downstream are shown in figure [ vedd ] . the discontinuities of the angular - integrated turbulence spectra are caused by the abrupt onset of thermal damping at the characteristic length @xmath139 of the magnetic field . obliquely propagating fast - mode waves are absorbed by the thermal background ions at this scale . in a magnetized plasma , fast - mode waves are likely the agent responsible for efficient sa of electrons @xcite . the resonant interactions of particles with fast - mode waves have been studied by several authors @xcite . in these studies , the authors prescribed the wave spectrum with several parameters and calculated the corresponding fokker - planck coefficients . a self - consistent treatment of the turbulence spectral evolution in the dissipation range was presented by @xcite for a variety of astrophysical plasmas . @xcite considered the damping of fast - mode waves in a low-@xmath3 plasma and the application of the sa of relativistic protons in magnetic field dominated funnels derived from general relativistic mhd simulations of non - radiative accretion flow around black holes . section [ shock ] shows that only fast - mode waves propagating parallel to the local magnetic field can survive the ttd by the thermal ions in the high-@xmath3 downstream plasma . these waves are right - handed polarized and resonate preferentially with electrons in the thermal background and therefore may selectively accelerate electrons to relativistic energies @xcite . we next explore the sa of electrons in the shock downstream in this scenario . we assume that the turbulence is isotropic from the turbulence generation scale @xmath10 down to the dissipation scale @xmath172 , which depends on the turbulence speed @xmath6 ( eq . [ [ kd ] ] ) . fast - mode waves are excited at the scale @xmath173 and are also isotropic down to the dissipation scale . due to the ttd , the energy density of fast - mode waves in the dissipation range is less than the magnetic field energy density . the resonant scatter rate of energetic particles by mhd waves is therefore smaller than @xmath174 . however , for this high-@xmath3 plasma , particles with a gyro - radius @xmath175 less than the characteristic length of the magnetic field @xmath139 draft along magnetic field lines with a scatter mean - free - path @xmath176 . the scatter mean - free - path of particles with @xmath177 should be comparable to @xmath178 since particles with even higher energies scatter with the magnetic field randomly instead of performing gyro - motions . this efficient scatter of low - energy particles by the turbulence magnetic field results from bending of magnetic field lines by strong turbulence in the inertial range beyond @xmath139 . the resonant wave - particle interactions in the dissipation range give a much lower scatter rate due to the strong damping of obliquely propagating fast - mode waves by thermal ions . although this scatter through the particle gyro - motion and chaotic magnetic field struture is not caused by resonances with fast - mode waves , it determines the spatial diffusion coefficient @xmath179 , which plays essential roles in the sa by large - scale fast - mode waves @xcite . the sa in the supersonic phase with @xmath180 is not well understood @xcite . if we assume the turbulence speed @xmath6 as the characteristic speed of the scatter agents and a scatter rate of @xmath181 for relativistic particles with @xmath182 , where @xmath183 is the speed of relativistic particles , the acceleration rate will be comparable to that of the dsa . we also note that the turbulence speed @xmath6 is higher than @xmath34 in a narrow region near the sf ( fig . [ vs ] ) corresponding to the turbulence generation . such an sa will be difficult to distinguish from the first - order dsa . further downstream , the turbulence speed is lower than @xmath34 . we will ignore the particle acceleration in the supersonic phase and only consider the acceleration by acoustic waves in the subsonic phase , where equation ( [ tac0 ] ) is approximately applicable @xcite . we assume @xmath184 in the following for relativistic particles with @xmath185 . particles with even higher energies have a scatter time @xmath186 . the corresponding spatial diffusion coefficient is given by @xmath187 @xcite . following @xcite , we have the acceleration timescale of these particles by a spectrum of fast - mode waves given by @xmath188^{-1}\ , . \label{tac}\ ] ] the fast - mode turbulence also enhances the spatial diffusion of these particles . the incompressional modes are more efficient in enhancing the spatial diffusion than compressional modes @xcite . to include these effects and partially take into the effects of energy loss due to adiabatic expansion @xcite , we adopt an effective diffusion coefficient : @xmath189 where @xmath190 is a dimensionless parameter . the escape time of relativistic electrons from the acceleration region is then given by @xmath191 physically , @xmath190 needs to be less than 1 , which corresponds to maximum diffusion caused by turbulence . however , considering the possible presence of incompressional modes and energy loss due to adiabatic expansion , the acceleration timescale by compressional modes will increase since the incompressional modes will carry part of the turbulence energy . the effect of this increase of acceleration timescale on the electron distribution can be partially taken into account by reducing the escape timescale , i.e. , by increasing the effective spatial diffusion coefficient @xmath192 . in what follows , we will treat @xmath190 as a free parameter with @xmath193 indicating the presence of incompressional modes and reduction of acceleration by fast - mode waves . the presence of incompressional modes will also bring the transonic point closer to the sf . the acceleration by compressible modes in the subsonic phase can still be efficient . the details of these processes will depend on the coupling between compressible and incompressible modes @xcite and is beyond the scope of this paper . the dependence of acceleration and escape timescales on @xmath56 at the transonic point in the downstream for typical conditions of snr rx j1713.7 - 3946 are shown in figure [ times ] . the approximate acceleration timescale given by equation ( [ tac0 ] ) and the escape timescale due to the diffusion coefficient @xmath56 alone , @xmath194 , are indicated with the thin lines . at high values of @xmath56 , the slightly high discrepancy in the exact and approximate acceleration timescales of the kraichnan phenomenology is due to the fact that the overall wave intensity is given by @xmath195 , instead of @xmath49 as is for the kolmogorov phenomenology . for conditions at the transonic point @xmath196 in the downstream , where @xmath197 , for the kraichnan ( left with @xmath198 , @xmath199 cm and @xmath200 ) and kolmogorov ( right with @xmath201 , @xmath202 cm and @xmath118 ) phenomenology in figure [ vs ] . @xmath203 km@xmath204s for both cases . the vertical line indicates the diffusion coefficient @xmath205 . the thick lines are the exact results given by equations ( [ tac ] ) and ( [ tesc ] ) . the thin solid line indicates the approximate acceleration timescale given by equation ( [ tac0 ] ) . the thin dashed line indicates @xmath194 . for a magnetic field of @xmath206 g , the diffusion coefficient for tev electrons is about @xmath207 cm . the corresponding diffusion timescale over a length of @xmath208 cm , the half - width of the variable x - ray filements observed with _ @xcite , is about one years . [ times],title="fig:",scaledwidth=48.0% ] for conditions at the transonic point @xmath196 in the downstream , where @xmath197 , for the kraichnan ( left with @xmath198 , @xmath199 cm and @xmath200 ) and kolmogorov ( right with @xmath201 , @xmath202 cm and @xmath118 ) phenomenology in figure [ vs ] . @xmath203 km@xmath204s for both cases . the vertical line indicates the diffusion coefficient @xmath205 . the thick lines are the exact results given by equations ( [ tac ] ) and ( [ tesc ] ) . the thin solid line indicates the approximate acceleration timescale given by equation ( [ tac0 ] ) . the thin dashed line indicates @xmath194 . for a magnetic field of @xmath206 g , the diffusion coefficient for tev electrons is about @xmath207 cm . the corresponding diffusion timescale over a length of @xmath208 cm , the half - width of the variable x - ray filements observed with _ @xcite , is about one years . [ times],title="fig:",scaledwidth=48.0% ] when @xmath209 , corresponding to the fast diffusion limit , @xmath210 and @xmath211 . then @xmath212 . the acceleration is negligible . with these asymptotic expressions for these timescales , the ratio of the acceleration to escape timescales decreases with the decrease of @xmath56 and becomes close to unity near @xmath213 . since deviations from these asymptotic expressions occur near @xmath214 , efficient particle acceleration is only possible in the regime where @xmath215 . when @xmath216 , corresponding to the slow diffusion limit , @xmath217 $ ] . @xmath218^{-1 } \simeq \left[{(\delta-3)u^2 \over 3d}\left({dk_m\over v_f}\right)^{\delta-3 } \int_{0}^{\infty}dx { x^{4-\delta}\over 1 + x^2}\right]^{-1 } \sim u^{-2}d(dk_m / v_f)^{3-\delta}\ , , \label{tac1}\ ] ] where @xmath219 , which ensures the convergence of the integration . these results are in agreement with figure [ times ] . the acceleration and escape timescales are determined by @xmath56 and the turbulence spectrum . since we assume that @xmath220 for electrons with @xmath182 , the electron acceleration and escape timescales are independent of the energy for @xmath182 . when the ratio of acceleration to escape timescale is independent of the particle energy , a power - law particle distribution is expected in the steady - state with the spectral index given by @xcite @xmath221 since electrons with an energy less the proton rest energy can be scattered by whistler waves at small scales , the diffusion coefficient @xmath56 for these particles can be much less than @xmath222 and the corresponding spectral index @xmath223 @xcite . the scatter will be dominated by the turbulent magnetic field before electrons reach the proton rest mass energy , the spectral break may appear below the proton rest mass energy . without detailed treatment of these processes , we will assume that the electron distribution follows a power law with @xmath224 for @xmath225 in the following . will lead to a spectral bump near this break energy , in conflict with radio observations . this suggests that , when the acceleration time is long and @xmath226 is high , one needs better treatment of the electron acceleration at low energies . ] above @xmath227 , the steady - state spectral index of the electron distribution is given by equation ( [ index ] ) since the diffusion coefficient @xmath228 is assumed to be independent of the electron energy . for relativistic electrons with @xmath229 , @xmath230 . the increase of @xmath56 with the electron energy @xmath91 will lead to softening of the electron distribution toward higher energy dictated by the energy dependence of @xmath231 . ( dotted ) , cutoff lorentz factor @xmath232 ( dotted - dashed ) , spectral index @xmath226 ( dashed ) , and @xmath233 ( thin solid ) in the downstream for the kraichnan phenomenology with @xmath234 and @xmath198 ( left ) and kolmogorov phenomenology with @xmath235 and @xmath201 ( right ) . @xmath203 km / s . the particle acceleration is significant for @xmath236 . we only consider acceleration between the two vertical dashed lines indicating @xmath196 and @xmath237 . for @xmath232 , we have assumed that @xmath238 g , @xmath239 cm ( left ) and @xmath238 g , @xmath240 cm ( right ) . [ paras],title="fig:",scaledwidth=48.0% ] ( dotted ) , cutoff lorentz factor @xmath232 ( dotted - dashed ) , spectral index @xmath226 ( dashed ) , and @xmath233 ( thin solid ) in the downstream for the kraichnan phenomenology with @xmath234 and @xmath198 ( left ) and kolmogorov phenomenology with @xmath235 and @xmath201 ( right ) . @xmath203 km / s . the particle acceleration is significant for @xmath236 . we only consider acceleration between the two vertical dashed lines indicating @xmath196 and @xmath237 . for @xmath232 , we have assumed that @xmath238 g , @xmath239 cm ( left ) and @xmath238 g , @xmath240 cm ( right ) . [ paras],title="fig:",scaledwidth=48.0% ] produced at several locations in the downstream for the kraichnan ( left ) and kolmogorov ( right ) phenomenology in figure [ paras ] . [ f],title="fig:",scaledwidth=48.0% ] produced at several locations in the downstream for the kraichnan ( left ) and kolmogorov ( right ) phenomenology in figure [ paras ] . [ f],title="fig:",scaledwidth=48.0% ] the exact electron distribution can be obtained numerically . for the sake of simplicity , we assume that the electron distribution has a high - energy cutoff at lorentz factor @xmath241 where @xmath242 is the elementary charge units . then for a steady - state treatment , the distribution of electrons escaping from the acceleration site may be approximated reasonably well with @xcite @xmath243^{1/2}\}/\tau_{esc}(d)\ , , \label{f1}\ ] ] where @xmath244 we note that the shape of the distribution function near @xmath232 can also be affected by the energy loss processes @xcite . in our model , it is mostly determined by the balance between the acceleration and escape processes @xcite . the shape of this high - energy cutoff has significantly effect on the fitting parameter , especially the value of @xmath232 . we are carrying out detailed numerical investigation of the particle distribution and the results will be reported in a separate publication . although the parallel propagating waves in the dissipation range may not contribute to the particle scatter significantly , energies carried by these waves are only accessible to relativistic electrons . at very small scales , these waves resonate with thermal background electrons giving rise to a preferential acceleration of electrons . one therefore may assume that the ratio of the dissipated energy carried by non - thermal electrons to that of the thermal ions is proportional to the ratio of the energy density of parallel propagating fast - mode waves to that of the magnetic field : @xmath245 where @xmath246 is the base of the natural logarithm . a more quantitatively treatment of this issue may address the electron injection processes self - consistently @xcite . to have efficient acceleration of relativistic electrons , the turbulence decay time @xmath247 and the remnant lifetime @xmath248 should be longer than the acceleration time . as we will see below , the turbulence decay time is always longer than @xmath249 in the subsonic phase for snr rx j1713.7 - 3946 . there are at most two locations @xmath250 and @xmath237 with @xmath251 in the downstream , where @xmath252 . figure [ paras ] shows the evolution of @xmath253 , @xmath232 , @xmath226 , @xmath254 and @xmath255 in the downstream for @xmath203 km / s . the profiles of @xmath256 and @xmath257 only depend on @xmath258 . so is the profile of @xmath253 . the profiles of @xmath259 and @xmath226 also depend on the absolute value of @xmath94 . to obtain @xmath232 , one needs to know @xmath10 and @xmath93 as well . most sa occurs near the sonic point @xmath196 , where @xmath197 . in the late subsonic phase , @xmath260 , the sa is insignificant since most of the free energy of the system has been converted into heat . the characteristic length of the magnetic field is also long far downstream due to the weak turbulence , which implies long electron scatter and acceleration timescales . at several locations in the downstream for the kraichnan ( left ) and kolmogorov ( right ) phenomenology in figure [ paras ] . [ fy],title="fig:",scaledwidth=48.0% ] at several locations in the downstream for the kraichnan ( left ) and kolmogorov ( right ) phenomenology in figure [ paras ] . [ fy],title="fig:",scaledwidth=48.0% ] then the distribution of non - thermal electrons in the downstream @xmath261 where @xmath262 , and @xmath263 gives the energy density of non - thermal particles at @xmath264 . figures [ f ] and [ fy ] show the normalized electron distribution @xmath265 and @xmath266 for parameters in figure [ paras ] at several locations in the downstream , respectively . we note that the electron distribution has a rather gradual high - energy cutoff due to the weak dependence of the acceleration and escape timescales on the spatial diffusion coefficient @xmath56 , which is proportional to the electron energy above the cutoff energy @xmath267 . this gradual cutoff results in a broad emission component due to inverse comptonization of the low - energy background photons by high - energy electrons , which can fit the observed broad tev emission spectrum from a few snrs . , respectively . the dashed line is for a simple power - law model with a gradual high - energy cutoff @xmath268 , where @xmath269 tev . the solid lines are for the fiducial models . the low and high energy spectral peaks are produced through the synchrotron and inverse compon scatter of the background photons @xcite , respectively . the preliminary data from fermi were not considered in the fit and are included here to show the agreement between model predictions and this observation @xcite . the fermi observation seems to favor the kraichnan model . [ spec],title="fig:",scaledwidth=48.0% ] , respectively . the dashed line is for a simple power - law model with a gradual high - energy cutoff @xmath268 , where @xmath269 tev . the solid lines are for the fiducial models . the low and high energy spectral peaks are produced through the synchrotron and inverse compon scatter of the background photons @xcite , respectively . the preliminary data from fermi were not considered in the fit and are included here to show the agreement between model predictions and this observation @xcite . the fermi observation seems to favor the kraichnan model . [ spec],title="fig:",scaledwidth=48.0% ] here we use the snr rx j1713.7 - 3946 as an example to demonstrate how the sa by fast - mode waves accounts for the observed broadband spectrum . the electron distribution produced by the sa is given by equation ( [ f0 ] ) , where the integration over @xmath270 should stop at @xmath237 . given the evolution history of the snr , the nonthermal electron distribution will vary along the radial direction . a detailed modeling of the explosion is necessary to take this effect into account properly @xcite . here we treat the volume of the emission region @xmath271 as a free parameter to control the normalization of the emission spectrum , which is appropriate for snrs where the nonthermal particles appear to be concentrated near the sf . by adjusting @xmath94 , @xmath93 , @xmath108 , @xmath190 , @xmath10 , and @xmath271 , one can use the corresponding electron distribution @xmath272 to fit the observed spectrum of the snr rx j1713.7 - 3946 . figure [ spec ] shows the best fit with the corresponding parameters listed in table [ t1 ] , where @xmath273 is the enclosed volume of the snr sf . comparing to the thin - dashed line , which is derived by assuming an electron distribution @xmath274 , the high - energy cutoff of the kraichnan model is more gradual than the kolmogorov model , which makes the former spectrum broader . the emission volume is @xmath275 ( @xmath276 ) times the volume of the snr for the kraichnan ( kolmogorov ) phenomenology , which is compatible with observations . the values of @xmath190 are greater than 1 , which suggests that the particle spatial diffusion coefficient might be enhanced significantly by incompressible turbulence motions . the total energies of nonthermal electrons are also comparable to the magnetic field energy for a uniform magnetic field within the remnant , suggesting near energy equipartition between the nonthermal electrons and the magnetic field . [ cols="<,^,^,^,^,^,^,^,^ " , ] [ t1 ] there are six parameters in the model : @xmath93 , @xmath94 , @xmath277 , @xmath190 , @xmath10 , and @xmath271 . the observed radio to x - ray spectral index , x - ray to tev flux ratio , location of the x - ray cutoff , and bolometric luminosity of the source give four constraints , which leads to two more degree of freedom . our model fit to the spectrum therefore is not unique . @xmath190 is determined by the coupling between incompressible and compressible turbulence motions . @xmath93 is well constrained by the ratio of the x - ray to tev flux . to reproduce the observed spectral shape , the profiles of @xmath226 , @xmath232 , and @xmath253 should not change significantly , which implies that @xmath278 and @xmath279 at the transonic point for the kraichnan phenomenology . for the kolmogorov model , @xmath280 and @xmath281 . for @xmath282 , @xmath6 is proportional to @xmath94 , we find that nearly identical emission spectra can be obtained by adjusting @xmath94 and @xmath108 @xcite . since the turbulence decay time is longer than the supernova lifetime , the steady - state treatment is also justified . a time - dependent treatment gives identical parameters @xcite . the primary discrepancy between these models and the observations are the relatively high densities of the downstream plasma . from x - ray observations , @xcite inferred an upper limit for the electron density of 0.02 @xmath283 . the corresponding mass density is about @xmath284 g @xmath283 , which is comparable to the densities inferred with the kraichnan phenomenology but lower than those inferred with the kolmogorov phenomenology . on the other hand , the electron temperature could be much lower than the ion temperature in the shock downstream @xcite . @xcite and @xcite have argued that the density in the downstream can be as high as @xmath285 @xmath283 , corresponding to a mass density of @xmath286 g @xmath283 . given the age of @xmath287 years and a radius of @xmath288 pc at a distance of @xmath289 kpc , the corresponding average speed of the shock front is @xmath290 km s@xmath291 . with the self - similar solution of @xcite , we infer a shock speed of @xmath2926100 $ ] km s@xmath291 , where @xmath293 and @xmath294 are the power - law exponents of the density profile for the ejecta and the ambient medium , respectively . observations give an upper limit of @xmath295 km s@xmath291for the shock speed @xcite . from the self - similar solution , the shock speed must be higher than @xmath296 km s@xmath291 ( for @xmath297 and @xmath298 ) . if ions are preferentially heated by the shock , the corresponding ion temperature @xmath299 will be higher than @xmath300 k. the electron temperature should be higher than that given through coulomb collisional energy exchange with ions @xcite : @xmath301 where @xmath302 is the electron number density . the corresponding bremsstrahlung luminosity is @xmath303 erg s@xmath291 , which is comparable to the luminosity of the observed nonthermal x - ray emission . we therefore expect strong thermal emission with such a high density . @xcite obtained a very low thermal bremsstrahlung luminosity by arbitrarily adopting an electron temperature 100 times lower than the ion temperature . as shown above , considering the electron ion coulomb collisional energy exchange , the electron temperature will not be that low and significant thermal x - ray is expected with a density of @xmath285 @xmath283 , except that cooling of the shock front by cosmic ray ions dominates @xcite . the highest electron density given by our models is about @xmath304 @xmath283 . the corresponding thermal x - ray luminosity will be reduced by nearly one order of magnitude and should be in agreement with observations . detailed modeling of the supernova explosion and the thermal emission is needed to see the validity of these models . the model inferred density may also be reduced by considering the acceleration of electrons by large - scale structures in the downstream and the acceleration in the supersonic phase , where the first - order fermi acceleration is also possible . in this paper , we consider the electron acceleration by the fully developed turbulence in the subsonic phase . it assumes that once the electrons diffuse over a scale of the turbulence generation length @xmath10 , the acceleration stops . as shown above , the turbulence evolves in the downstream . in a more self - consistent treatment , one may use the turbulence properties to derive nonthermal electrons injected into the downstream flow by small scale plasma waves and consider the further acceleration of these electrons as they diffuse spatially in the downstream . the scatter mean - free - path of these particles are determined by the properties of turbulence . the electron acceleration stops only after they diffuse into upstream or far downstream , where the turbulence becomes insignificant . if these effects lead to a harder overall electron distribution , the alfvn speed needs to be increased to fit the observations , leading to a lower density . from these models studied here , we see that , to have efficient sa , both high - speed waves ( @xmath305 ) and short scatter mean - free - path are required . quantitatively , one needs @xmath306 to be on the order of unity so that the acceleration and escape time scales of relativistic particles are comparable . @xmath6 is constrained by the shock speed . a short scatter mean - free - path is achieved by the reduction of the characteristic length of the magnetic field , which also determines the maximum energy of the accelerated particles . in these models , turbulence motions are invoked to reduce the characteristic length of the magnetic field . it is obvious , such a mechanism is only possible for strong turbulence , where the turbulence speed is higher than the alfvn speed . the turbulence speed is determined by the shock speed @xmath307 , which is less than 1949 km s@xmath291 for snr rx j1713.7 - 394 . therefore @xmath308 km s@xmath291 and we obtain a low limit for the mass density from the inferred magnetic field of 14 @xmath309 g : @xmath310 g @xmath311 , which corresponds to an electron density of @xmath312 @xmath313 . therefore further reduction of the density can be achieved by considering the generation of the turbulence and its effect on the turbulence spectrum , i.e. , the dependence of the eddy velocity on the spatial scale . indeed , in the models considered above , we assume that the turbulence is generated in a very narrow spatial range instantaneously at the sf and an inertial range develops . since the large - scale eddy speed is comparable to the bulk velocity of the downstream flow moving away from the shock front , the region with @xmath314 should be considered as the turbulence generation phase . this is an intrinsic limitation of the above treatments , which focus on the averaged properties of the downstream flow without addressing the turbulence generation process . significant particle acceleration occurs within @xmath314 , i.e. , the turbulence generation phase . it is also possible that the turbulence is generated over a broad spatial range and/or the turbulence is not isotropic at large scales . one then expects a turbulence spectrum shallower than the initial range spectrum . for example , for a turbulence spectrum of @xmath315 ( 2\pi / l)^{g-3 } k^{-g-2}\ ] ] with @xmath316 , , the eddy speed increases with the decrease of the spatial scale and the turbulence energy is not dominated by the large scale eddies . one then needs to introduce a spectral break at certain small scale . such a complex scenario is not well justified from both observational and theoretical point of views . ] where the normalization is chosen so that the turbulence energy density is given by @xmath317 , the eddy speed can be redefined as @xmath318 k^3 w\}^{1/2 } = u ( kl/2\pi)^{(1-g)/2}\,.\ ] ] the eddy speed is comparable to the alfvn speed at the characteristic length of the magnet field @xmath5 , we then have @xmath319 . therefore @xmath320 . since @xmath321 , we have @xmath322 . the alfvn speed can be comparable to the turbulence speed for a shallow turbulence spectrum with @xmath323 approaching 1 . thus stochastic electron acceleration can account for observations of snr rx j1713.7 - 394 as far as the mass density of the shocked plasma is greater than @xmath324 g @xmath283 . detailed studies of the turbulence generation and the associated particle acceleration are warranted @xcite . in the paper , we study the sa of electrons by a decaying turbulence as produced and/or enhanced by strong non - relativistic shocks and carried away from the sf with the downstream flow . it is shown that , to have significant particle acceleration , the turbulence must cover a large spatial scale so that the particle acceleration time may be shorter than the turbulence decay time . to account for observations of a few sttsnrs with the leptonic scenario for the tev emission , fast - mode waves need to be excited in the subsonic phase . given the turbulent nature of the downstream flow , fast - mode waves may prevail in the downstream . we show that the sa by large - scale acoustic ( fast - mode ) waves can account for the observations . there are four basic model parameters , namely , the magnetic field , mass density , shock speed , and the turbulence generation scale . observations of a few snrs and radio galaxies have shown that the particle acceleration may change dramatically along the sf @xcite . this variation has been attributed to a large - scale magnetic field in the dsa model as the acceleration efficiency varies with the angle between the magnetic field and the shock normal . with our model , this variation is likely caused by a quite different mechanism . detailed comparative studies should be able to distinguish these models . the particle acceleration is very sensitive to the magnetic field . to produce nonthermal particle distribution in compatible with observations , @xmath325 and @xmath326 , where @xmath6 is the large - scale eddy speed near the shock front , should be on the order of 1 for the kraichnan and kolmogorov phenomenology , respectively . the high - energy cutoff of the particle distribution is determined by the magnetic field and the turbulence generation scale . weaker fields will lead to lower cutoff energies and harder spectra . the particle acceleration may be turned off completely for strong fields due to the increase of the characteristic length of the magnetic field , and therefore the particle scatter mean - free - path . if the magnetic field is predominantly generated by the streamline of nonthermal particles upstream , the models then imply @xmath327 and @xmath328 for this dynamo process and for the kraichnan and kolmogorov phenomenology in the downstream turbulence , respectively . assuming that the turbulence is isotropic and generated in a narrow spatial scale , the model inferred densities of the downstream flow may be so high that thermal x - ray emission becomes observable , in conflict with observations . although the thermal x - ray emission may be suppressed by a lower electron temperature due to dominance of the cooling by cosmic ray ions , we find that a low density is also possible if the turbulence is not isotropic or generated over a broad spatial scale so that the eddy speed has very weak dependence on the spatial scales . detailed modeling of the progenitor and the evolution history of the remnant may help to constrain the density @xcite . the application of the models to the snr rx j1713.7 - 3946 also suggests energy equipartition between the magnetic field and the acceleration electrons . if the rest of the shock energy is dissipated as heat in the downstream , then the overall electron acceleration efficiency will be @xmath329 , which is inversely proportional to the plasma @xmath3 of the downstream flow . for the kraichnan phenomenology , @xmath330 and for the kolmogorov phenomenology @xmath331 , acceleration is more efficient for stronger shocks , which also produce hotter downstream plasma . however , the dependence of the acceleration efficiency on the shock speed is rather weak . these may have significant implications on the origin of cosmic rays and their connection to the properties of the interstellar medium @xcite . with the fast - mode wave turbulence model studied in this paper , relativistic electrons may also be accelerated through the first - order fermi mechanism as in the dsa model , especially in the supersonic phase . the high - energy cutoff can still result from decoupling of higher energy particles with the background magnetic field as their gyro - radius exceeds the characteristic length of the magnetic field . this is quite different from the dsa models , where the high - energy cutoff is related to a finite lifetime of the shock , shock curvature , or efficient energy loss processes @xcite . we thank the referee for critical reviews of the paper , which made us examine the acceleration timescale and the shape of the high - energy cutoff of the electron distributions carefully . sl acknowledges support from the eu s solaire research and training network at the university of glasgow ( mtrn - ct-2006 - 035484 ) and thanks lyndsay fletcher for constant support and john kirk and christian fendt for invitation to the workshop entitled `` the high - energy astrophysics of outflows from compact object '' , where some ideas in this paper were conceived . this work is supported in part under the auspices of the us department of energy by its contract w-7405-eng-36 to los alamos national laboratory and by the national science foundation of china ( grants 10963004 and 10778702 ) , yunnan provincial science foundation of china ( grant 2008cd061 ) and srfdp of china ( grant 20095301120006 )

we estimate the maximum contribution to reionization from the first generation of massive stars , with zero metallicity , under the assumption that one of these stars forms with a fixed mass in every collapsed halo in which metal - free gas is able to cool . we assume that any halo that has already had stars previously formed in one of their halo progenitors will form only stars with metals , which are assigned an emissivity of ionizing radiation equal to that determined at from the measured intensity of the ionizing background . we examine the impact of molecular hydrogen photodissociation ( which tends to reduce cooling when a photodissociating background is produced by the first stars ) and x - ray photoheating ( which heats the atomic medium , raising the entropy of the gas before it collapses into halos ) . we find that in the cdm model supported by present observations , and even assuming no negative feedbacks for the formation of metal - free stars , a reionized mass fraction of 50% is not reached until redshift . the combination of ordinary metal - enriched stars and early metal - free stars can yield a cmb optical depth to electron scattering of at most . the contribution of metal - free stars to the present cosmic infrared background is negligibly small .

astro-ph0511489

in the last few years , our understanding of the reionization history of the universe has greatly improved thanks to the detection of quasars at increasingly high redshift ( e.g. , becker et al . 2001 ; fan et al . 2002 ; hu et al . 2002 ; white et al . 2003 ) and the new data on the cosmic microwave background ( cmb ) radiation from wmap ( bennet et al . observations from high redshift quasars indicate a fast increase of the intensity of the ionizing background with cosmic time occurring at @xmath4 , probably signalling the end of reionization ( fan et al . 2002 ) , or the epoch when the low - density intergalactic medium ( hereafter , igm ) became fully ionized . reionization is expected to have occurred over an extended period of time , during which sources gradually reionized every atom in the igm ( e.g. , gnedin 2000 ) . indeed , the detection of ly@xmath5 emission from sources at @xmath6 ( hu et al . 2002 ; kodaira et al . 2003 ; cuby et al . 2003 ; kneib et al . 2004 ; rhoads et al . 2004 ) that are not highly luminous to have produced large regions around them implies that the igm could not be fully neutral at the redshift of these sources , otherwise such photons would have been scattered out of the line of sight ( miralda - escud & rees 1998 ; madau & rees 2000 ) . another important probe to the epoch of reionization is the optical depth to electron scattering of cmb photons , @xmath7 . the first measurement of @xmath7 was reported by the _ wilkinson microwave anisotropy probe _ mission ; although its value is still rather uncertain ( kogut et al . 2003 ; spergel et al . 2003 ) , the measurement favored an early start to reionization , such that the fractional ionization of the igm would have reached 50% at @xmath8 . it is worth noting that this does not by itself contradict the appearance of the gunn - peterson trough ( which marks the end of reionization ) at @xmath4 , because reionization may advance gradually over a long period of time . however , an early start of reionization presents a problem in the cold dark matter model of structure formation , in which only a small fraction of matter has collapsed into halos that can form stars at @xmath9 , and therefore one needs to assume a very high rate of emission of ionizing photons from this first generation of stars ( e.g. , haiman & holder 2003 ; chiu , fan , & ostriker 2003 ; onken & miralda - escud 2004 ) . one possibility is that the first collapsed halos in which gas was able to cool were efficient in driving gas to accrete onto central black holes that were formed by the core collapse of the first massive stars . however , as argued by dijkstra et al . ( 2004a ) , any such population of high-@xmath10 active galactic nuclei ( agn ) would likely contribute a smooth component to the present x - ray background at energies near @xmath11 kev that may be higher than is allowed by observational constraints ( although significant calibration uncertainties of the x - ray background intensity remain which could increase the upper limit on the fraction of the x - ray background in a smooth component ) . the other possibility is that the first massive stars , which were metal - free and highly efficient at producing ionizing photons , were responsible for an early reionization ( oh et al . 2001 , yoshida et al . 2003 , sokasian et al . 2004 ) . much work has been done to characterize the properties of these metal - free stars ( abel et al . 1998 , 2002 ; bromm , coppi , & larson 1999 , 2002 ; oh et al . 2001 ) . the formation of the first stars was governed by molecular hydrogen radiative cooling , which becomes effective at gas temperatures above @xmath12 k ( yoshida et al . this virial temperature is first reached in halos of total mass @xmath13 m@xmath14 , which start becoming substantially abundant at @xmath15 . metal - free stars have higher effective surface temperatures than their metal - rich counterparts during their main - sequence phase . in addition , they are nearly fully convective and can burn most of their hydrogen content during their main - sequence lifetime . this makes them highly efficient as producers of ionizing radiation ( emitting @xmath16 ionizing photons per baryon ; schaerer 2002 ) , most of which can escape into the igm due to the small amount of gas in the host halos . despite these advantages of metal - free stars as sources of ionizing radiation , the idea that these stars may have been responsible for an early reionization faces a number of difficulties . first , it is unlikely that many massive stars might form simultaneously in the first halos where star formation took place , which have low velocity dispersion and can therefore easily lose their gas after it is heated by photoionization and supernova explosions . numerical simulations suggest that one central massive star will initially form , with a mass of @xmath17 m@xmath14 ( e.g. , abel et al . 2002 ; bromm et al . 2002 ) , containing a fraction of only @xmath18 of the baryonic mass of the halo from which it formed . the ionization and the supernova explosion resulting from this star can then eject the rest of the gas from the halo ( bromm , yoshida , & hernquist 2003 ; whalen , abel , & norman 2004 , kitayama et al . 2004 ) , yielding a very low efficiency to form stars . later , when the gas falls back into a halo with a total mass increased by the merging process , it will already be enriched by metals and will form the next generation of stars with properties that are presumably similar to present stellar populations in normal galaxies . if the metal - free star releases about @xmath19 of metals , a total mass of @xmath20 of baryons may be polluted to a metallicity of @xmath21 , above which cooling is already modified by the presence of metals , while the emission properties of massive stars are also modified for even lower metallicities . in cases where a metal - free star collapses into a black hole releasing no metals , any cold gas reaching the halo center at a later time will also not form another metal - free star because the black hole will be in the center . the second difficulty is that the radiation emitted by these first stars can act as a negative feedback mechanism to slow down the formation of other metal - free stars in new collapsed halos with pristine gas . a widely studied effect is photodissociation of molecular hydrogen ( haiman , abel , & rees 2000 ) . another feedback effect is the heating of the igm by x - rays emitted by supernovae or by the metal - free main - sequence stars themselves ( oh & haiman 2003 ) , which raises the entropy of gas accreting into halos , resulting in a more stringent requirement for the gas to cool . this effect has not been so widely studied as the molecular photodissociation one , but as we shall see below it may be even more important . the aim of this paper is to determine the maximum possible contribution to reionization from zero metallicity stars , under the assumption that only one such star forms in each halo , with a fixed stellar mass . special attention will be paid to the way that the radiative feedback mechanisms of photodissociation and x - ray heating limit the formation rate of these metal - free stars . we shall use a semianalytic model based on merger trees ( similar to the approach of , e.g. , haiman et al . 2000 and wise & abel 2005 ) , including a new ingredient to compute the halo age distribution and derive the fraction of halos that contain a metal - free main - sequence star at any given time . the paper is organized as follows . in [ smethod ] , we present our method to calculate the formation rate of metal - free stars , the photodissociating and x - ray background they create , and how these radiation backgrounds affect the formation rate by altering the entropy profile in collapsed halos and the molecular hydrogen abundance . the results are presented in [ res ] , showing how each of the two feedback effects we consider impact the rate at which metal - free stars can form and their contribution to reionization . a discussion of these results follows in [ sdisc ] . our central assumption in this paper is that when a halo that contains metal - free gas ( i.e. , a halo for which no star has previously formed in any of its merger - tree progenitors ) collapses , a single massive star forms at the halo center after a cooling time . the mass of this star is assumed to be constant for all halos with metal - free gas . as explained in detail later ( [ sstform ] ) , the star forms in our model if the cooling time of the gas at the center of the halo is shorter than the lifetime of the halo . we develop an algorithm to define the lifetime of halos from the merger tree framework based on the extended press - schechter model , which is fully described in the appendix . the cooling time depends on the central gas density and temperature , which depends at the same time on the initial entropy of the gas prior to the collapse of the halo ( e.g. , oh & haiman 2003 ) . we compute the initial central gas density ( prior to the onset of cooling ) in halos at each redshift self - consistently , assuming hydrostatic equilibrium in a halo with the navarro , frenk , & white ( 1997 ) density profile for the dark matter ( hereafter , nfw ) . we calculate the ionizing radiation emitted by the metal - free stars to infer the history of reionization ( taking into account an escape fraction of ionizing photons from halos determined by recombinations in the halo gas ) , as well as the feedback effects of x - ray heating ( and consequent entropy raise of the atomic medium , which affects the gas density profiles in halos ) and molecular photodissociation . we also include a component of ionizing radiation from a normal stellar population assumed to form in metal - enriched halos , which have had previous star formation and ionization . this section explains in detail how all these ingredients are implemented in our model . our goal will be to find the maximum contribution that these metal - free stars could plausibly make to an early reionization , and how this is limited by the minimal expected feedback effects . we use the halo mass function of sheth & tormen ( 2002 ) . we assume that a first - generation star forms when the halo age exceeds the cooling time at the halo center . we adopt for this purpose the definition of halo formation time of lacey & cole ( 1993 ) , which is the time when the halo mass was half the present one . the probability distribution of the age of a halo as a function of its mass and redshift is calculated following lacey & cole , which we then use to compute the probability that a halo contains a metal - free star as described in the appendix . to determine the cooling time at the halo center , we assume the gas is in hydrostatic equilibrium in the halo gravitational potential , and that the dark matter follows a nfw density profile with concentration parameter @xmath22 . in reality , @xmath23 depends on the halo mass and redshift ( bullock et al . 2001 ; eke et al . 2001 ) , but here we approximate it as constant and with the same value used by oh & haiman ( 2003 ) . under the assumption of hydrostatic equilibrium , the gas density profile depends on its temperature . shock - heating of the gas during the collapse heats the gas to a roughly constant temperature , about equal to the halo virial temperature . this results in a final gas density profile similar to the dark matter , although with a small , constant density core , with a size that is determined by the minimum entropy @xmath24 of the gas in the halo . the gas with the lowest entropy is generally the one that is shock - heated to the virial temperature only after reaching a high density , and ends up near the halo center . however , if the gas had already been heated prior to the collapse , acquiring a high initial entropy , the value of @xmath24 may be increased , resulting in a much more extended gas profile and reduced central gas density , and therefore a much longer central cooling time . to take this into account we follow the procedure explained in oh & haiman ( 2003 ) . we define the function @xmath25 as @xmath26 where @xmath27 is the adiabatic index ( equal to @xmath28 for monatomic gas ) , and @xmath29 and @xmath30 the gas temperature and number density . note that the entropy of a gas is actually the logarithm of @xmath25 , although in this paper we refer to the function @xmath25 as `` entropy '' for convenience . the entropy before the collapse is what we call the entropy floor , @xmath31 , which we calculate as a function of redshift as described in [ xr ] . as an example , we show in figure [ fprof ] the gas density profile in a halo with @xmath32 m@xmath14 formed at different redshifts . the initial entropy is fixed to the primordial value given in oh & haiman ( 2003 ) , determined by early compton heating by the cmb on the adiabatically cooling gas , with the residual ionized fraction left over from recombination . at @xmath33 , the entropy of the igm is practically constant and the entropy floor decreases in importance with decreasing redshift because of the decreasing gas density , hence the gas density profile becomes steeper at low redshift . we show also the effect of raising the igm temperature to @xmath34 k at @xmath35 , which leads to a large reduction in the central gas density . we define the cooling timescale as the ratio of the internal energy of the gas to the cooling rate per unit of volume , @xmath36 : @xmath37 where @xmath38 is the boltzmann constant . we include only molecular hydrogen cooling for metal - free gas . in practice , atomic cooling can be neglected because metal - free stars form very rarely in halos with virial temperature above @xmath39 k in our model . the reason is that these halos usually have some progenitor of lower virial temperature that has already formed a star previously ( this will be seen in fig . 2 ) . in this case , we consider that the gas is already enriched with metals and that the emissivity per unit mass is constant , as described in [ semis_met_rich ] . the cooling rate ( due to collisional excitation of molecular hydrogen roto - vibrational lines ) is @xmath40 , where @xmath41 is the molecular fraction . we use the cooling function @xmath42 of galli & palla ( 1998 ) . the molecular fraction @xmath41 is determined by the continuous formation of molecules in the gas from @xmath43 , which is at the same time made by the primordial concentration of electrons from the recombination epoch ( see tegmark et al . we use equation ( 16 ) of tegmark et al . , although simplified and modified to consider h@xmath44 dissociation in the following way : @xmath45 where @xmath46 is the dissociation rate ( which we discuss later in [ sdiss ] ) and @xmath47 is the electron fraction , given by equation ( [ xr : fe ] ) . the constant @xmath48 determines the formation rate of @xmath43 ( which determines the formation rate of molecules ) , and is given by @xmath49 @xmath50 s@xmath51 . the initial , primordial value of @xmath41 is taken as @xmath52 ( galli & palla 1998 ) . note that the fraction of hydrogen that forms molecules from @xmath43 within collapsed halos is much larger than this primordial value , and therefore the long - term evolution of the molecular hydrogen fraction does not strongly depend on its initial value . as mentioned earlier , the main assumption in this work is that only one metal - free star is formed in each halo that collapses from progenitors that have never formed a star previously . the mass of the star , @xmath53 , is much lower than the baryonic mass of the halo , and is in fact lower than the gas mass in the core of the gas distribution . consequently , we can evaluate the cooling time at the halo center and consider that the star will form when the age of the halo is equal to this cooling time . the star forms in a halo of mass @xmath54 with formation time @xmath55 at a time @xmath56 that obeys @xmath57 the star will then live up to a time @xmath58 where @xmath59 is the main - sequence lifetime of a metal - free star with mass @xmath60 . the probability that a halo of mass @xmath54 at time @xmath61 has already formed a star can be evaluated by computing first numerically the formation time @xmath62 that obeys equation ( [ cool_c1 ] ) with @xmath63 , and then calculating the probability @xmath64 that the halo formation time is earlier than @xmath62 ( see eq . [ 2.26 ] of lacey & cole 1993 ) . similarly , the probability that the halo contains a metal - free star on the main - sequence at time @xmath61 is evaluated as the probability @xmath65 , where @xmath66 is the solution of @xmath55 of equation ( [ cool_c2 ] ) when @xmath67 . this condition , however , would be valid only if the halo gas were always metal - free . in reality , if a star had already formed in one of the halo progenitors , the supernova explosion from that star would already have polluted the gas with metals and blown it away . later on , as the sequence of halo mergers continues , the final halo present at time @xmath61 will accrete this gas , but since the gas is metal enriched it will form metal - rich stars . the emission from metal - rich stars will be treated differently in our model , as we describe later in [ semis_met_rich ] . here , our main assumption is that the very massive stars we consider that are highly efficient at emitting ionizing radiation are formed only when the gas is totally free of heavy elements . to take this into account , we evaluate at each time @xmath61 the average number of metal - free stars that have formed in a halo of mass @xmath54 , as ( see the appendix ) @xmath68 where @xmath69 is the number of halos of mass between @xmath70 and @xmath71 at time @xmath72 that have been incorporated into a halo of mass @xmath54 at time @xmath61 ( see eq . [ 2.25 ] in lacey & cole 1993 ) , and @xmath73 is the probability per unit time that a star forms at @xmath72 in a halo of mass @xmath70 . the way the probability @xmath74 is calculated is described in the appendix . we then assume that there is a poisson distribution of the number of stars that have formed within any halo , so that the probability that none has formed is @xmath75 . therefore , the probability that a halo of mass @xmath54 harbours a main - sequence star at time @xmath61 is @xmath76 where @xmath77 is the differential probability that a halo at time @xmath61 was formed at time @xmath55 and is given by equation ( 2.19 ) in lacey & cole ( 1994 ) . in practice , the lifetime of the main - sequence star ( @xmath78 years ) is short enough that the above integral can be approximated as @xmath79 figure [ fnstarm ] shows the probability that a halo with mass @xmath54 harbours a metal - free main - sequence star . different plots correspond to different redshifts ( @xmath80 , 15 , 10 , and 6 ) , and different lines to different models that will be described later . the model a1 corresponds to no feedback effects on the formation of metal - free stars , and the other models incorporate feedback effects that are described in 2.5 . the probability to contain a star tends to peak at halo masses of several times @xmath81 . at lower masses the molecular cooling rate is too slow and stars have not yet formed , while in halos of higher mass a metal - free star has typically already formed in one of their lower - mass progenitors ( i.e. , the second term in eq.is very small ) . as a byproduct , we can also calculate the global formation rate of metal - free stars ( see the appendix ) as @xmath82}{{{\rm d}}t}\ , . \label{dndt}\ ] ] where @xmath83 is the metal - free star formation rate and @xmath84 is the sheth - tormen halo mass function . this last formula is useful to calculate the preheating due to x - rays ( see section [ xr ] ) . we consider three feedback processes that affect metal - free star formation . the first is halo photoevaporation due to the ionizing background ( section [ sevap ] ) . the second , photodissociation of molecules due to a soft ultraviolet background ( section [ sdiss ] ) . finally , the third is reheating of the igm caused by x - rays ( section [ xr ] ) . the ionizing background reheats the gas in halos and causes a large increase in the entropy floor . gas trapped in minihalos escapes ( e.g. , shapiro et al . 2004 ) , quenching star formation . cooling and star formation in reionized regions can only occur in halos that are massive enough for the gas to undergo dissipation by atomic cooling . we neglect any possible formation of metal - free stars in these halos . in practice , once these halos collapse they should almost always have a progenitor that already formed a metal - free star in the past , because the number of halo progenitors increases with halo mass ; moreover , even if one metal - free star may form in one of these more massive halos cooling mainly by atomic processes , their contribution to the total number of metal - free stars formed should be negligible because the abundance of high - mass halos is greatly reduced compared to halos of low mass . we therefore neglect any possible formation of metal - free stars in reionized regions . to take this into account , equation ( [ prob_sh_star ] ) is multiplied by a factor @xmath85 , equal to the fraction of the igm that is neutral at every redshift . the calculation of the ionized fraction of the igm as a function of time using the global rate of star formation that we compute is described below in 2.6 . actually , the factor @xmath86 represents a lower limit to the suppression of the star formation by reionization , because in practice halos are correlated , or biased relative to the mass , so that more halos should be present in the reionized regions where other high - mass halos have already formed . is easily photodissociated by soft ultra - violet photons ( haiman et al . the rate at which dissociation is produced can be approximated by ( abel et al . 1997 ) @xmath87 being @xmath88 the flux at frequency @xmath89 in units of @xmath90 and @xmath91 the self - shielding factor given by draine & bertoldi ( 1996 ) . the mean free path of dissociating photons in the igm is typically larger than the mean distance between minihalos . therefore , this flux can be approximated as homogeneous and isotropic . to calculate this background we apply equation ( 7 ) of haiman et al . the emissivity of dissociating photons ( @xmath92 in their formula ) is computed similarly to that of ionizing ones ( see [ semis_met_free ] ) , although we take into account only the emission by the metal - free stars . emission from the metal - rich stars has a small effect at the high redshifts of interest , which we have neglected . as proposed by oh & haiman ( 2003 ) , soft x - ray photons raise the entropy floor , greatly reducing cooling and star formation in low - mass halos . x - rays can have a long mean free path through the atomic igm and can therefore heat this medium in an approximately uniform way , whereas the more abundant ultraviolet ionizing photons heat the medium only near the boundaries of hii regions . the mean free path in the atomic igm is roughly @xmath93 when this mean free path is in the range that goes from the mean distance between neighboring halos containing first - generation stars ( which is about @xmath94 kpc at @xmath95 ; see yoshida et al . 2003 ) to the horizon radius ( @xmath96 mpc at @xmath95 ) , the photons are effective at heating the atomic igm . this corresponds to a range of frequencies from 200 ev to 2 kev , or the soft x - ray band . soft x - ray photons can be produced by metal - free stars during two phases . first , these stars have a high enough effective temperature during the main sequence lifetime to emit substantially in soft x - rays ( schaerer 2002 ) . second , x - rays can be emitted by the hot gas in a supernova remnant after the death of the star . for simplicity we consider that all the emission is produced at the end of the metal - free star lifetime ( a good approximation because of the short lifetime ) . therefore , the rate of increase of the entropy floor is @xmath97 where @xmath98 is the rate of energy increase per particle , @xmath99 and @xmath100 is the rate of sne per unit of volume , equal to the star formation rate @xmath101 ( eq . [ [ dndt ] ] ) at the time @xmath102 . the constant @xmath103 is the total energy per star that is used to reheat the medium . note that this is not the same as the total energy radiated in x - rays by the star since some of that energy goes into collisional ionizations or emission of recombination photons . we leave the value of @xmath103 as a free parameter in our model ( the range of possible values is discussed in [ sres_xr ] ) . even though the reheating by x - rays is roughly homogeneous , the increase of the entropy is of course lower in regions of high gas density . to compute the entropy floor for a specific halo with formation at @xmath55 , we follow the evolution of the gas density during the collapse assuming the spherical top - hat model , and we compute the entropy increase from the previous history of the intensity of the x - ray background generated by metal - free stars . the evolution of @xmath100 is computed self - consistently with the model from the star formation rate . heating by x - rays is no longer taken into account after virialization ( this would be negligible in any case once the gas has reached a high density ) . x - rays have yet another effect : owing to secondary ionizations , they can produce a higher ionized fraction than the residual value after recombination ( which is @xmath104 , see galli & palla 1998 ) . this increases the formation rate of molecular hydrogen , enhancing the cooling rate ( haiman , rees , & loeb 1996 ) . according to shull & van steenberg ( 1985 ) , around 30 % of the total energy in x - rays is used in ionizations when the neutral fraction is high . the rate of ionizations is then found by dividing expression ( [ e : reh ] ) by 13.6 ev ( the ionization energy of hydrogen ) . considering this effect , the evolution of @xmath105 is determined by @xmath106 where @xmath107 is the recombination rate , which can be approximated by @xmath108 s@xmath51 ( hutchins 1976 ) and @xmath30 is the gas number density in the halo center after collapse . before collapse , the first term is negligible because the density is very low . the initial value of @xmath105 is given by the residual ionized fraction after recombination . to summarize , for each halo mass at a fixed time , stars can form only when the formation time satisfies equation ( [ cool_c1 ] ) . when considering feedback effects , cooling is delayed and therefore halo formation has to take place earlier . consequently , the probability that a halo hosts a main sequence star in equation ( [ prob_sh_star ] ) decreases , reflecting the feedback effects . to calculate the reionization history of the igm we must first calculate the emissivity @xmath109 of ionizing photons per igm baryon . we consider two contributions to the emissivity , one from metal - free stars which we discuss in [ semis_met_free ] , and another from stars with metals , discussed in [ semis_met_rich ] . finally , we explain the procedure to calculate the reionization history in [ semis_met_free ] . metal - free stars emit a higher quantity of ionizing photons than enriched stars ( see schaerer ( 2002 ) ) . we use here the fitting formula in table 6 of schaerer ( 2002 ) to compute the ionizing luminosity of each star as a function of its mass , @xmath110 . for the values of @xmath53 we consider here , this yields @xmath111 s@xmath51 and @xmath112 s@xmath51 . in addition , we take into account that some of the stellar photons may be absorbed locally in the halo gas surrounding the star , and only a fraction @xmath113 of these photons will be able to escape and ionize the igm . we use the following simple model to compute @xmath113 ( see whalen et al . 2004 for a more complete discussion and several examples of the evolution of the regions formed around a metal - free star ) : we assume that the star is initially surrounded by a distribution of atomic gas following a singular isothermal profile . an region grows around the star , and the heated gas acquires a sound speed of @xmath114 and starts expanding . this speed is typically larger than the halo circular velocity , but still smaller than the speed of the ionization front as long as the escape fraction is not very small and the star is capable of ionizing the halo ( the stellar lifetime is @xmath115 3 million years , which is short compared to the crossing time of the halo at this sound speed . hence , the ionization front advances supersonically . we use the simple approximation where all the gas moves out by the same distance @xmath116 . the gas distribution then has a central hole of radius @xmath117 , and at @xmath118 the density is @xmath119\frac{[r - r{_{\rm c}}(t)]^2}{r^2},\ ] ] where @xmath120 is the initial gas number density profile of the halo . the total rate of recombinations is computed from this density profile assuming the halo gas has been fully ionized by the star , and is subtracted from the photon emission rate of the star . the escape fraction of ionizing photons as a function of halo mass ( @xmath121 ) is plotted for different redshifts in figure [ ffesc ] . we consider two possible masses of metal - free stars , @xmath34 m@xmath14 ( left panel ) and @xmath122 m@xmath14 ( right panel ) . naturally , the escape fraction increases with stellar mass and decreases with halo mass , because the number of recombinations in a halo with the gas fully ionized increases with halo mass and does not depend on the luminosity of the ionizing star . these results agree with the numerical simulations of kitayama et al . ( 2004 ) , suggesting that our simple model is able to capture the main physics of the problem . the escape fraction increases when reheating is considered , being more significant in low mass halos , which have their density profiles flattened . note that there is a halo mass at which the escape fraction falls to zero ; for halos above this mass , the single central star is not able to ionize all the halo gas and the region reaches a strmgren radius ( the assumption that all the halo gas is ionized is then obviously incorrect , but this affects the calculation of @xmath113 only when @xmath113 is already very small in any case ) . the global emissivity from metal - free stars , @xmath123 , is finally calculated by integrating over all halo masses the number of photons emitted per igm particle corrected by absorption , i.e. , @xmath124 where @xmath125 is the baryonic number density of the igm at @xmath61 , and @xmath126 is the probability that a halo harbors a main sequence metal - free star ( see eq . [ prob_sh_star ] ) . for the emissivity of dissociating photons , we also use equation ( [ eq : emis : mf ] ) with @xmath127 . the emission of dissociating photons from metal - free stars , @xmath128 , is also computed from the fit given in table 6 in schaerer ( 2002 ) which yields @xmath129 s@xmath51 and @xmath130 s@xmath51 . after the formation of the first metal - free star in any halo , the surrounding gas is polluted with the metals released by this star . the metal - enriched gas can recombine and fall back into the halo . we assume the metallicity of this infalling gas is then sufficiently high to cool by the usual mechanisms that are prevalent in present galaxies ( bromm et al . however , halos with metal - enriched gas below a critical virial temperature may still not form stars because the cooling rates are too small ( e.g. , dijkstra et al . 2004b ) , and gas may remain close to @xmath131 k in a spherical or thick disk distribution where the gas is unable to cool further and fragment . in this paper , we adopt the simple assumption that in halos were metal - rich star formation takes place , the emissivity per unit baryonic mass is constant at all redshifts . then , the global emissivity due to enriched stars , @xmath132 , is calibrated to the observations of the ionizing background intensity and mean free path of ionizing photons at redshift @xmath0 ( see onken & miralda - escud 2004 ) : @xmath133 where @xmath134 is the emissivity at redshift 4 in units of ionising photons per particle and per hubble time , @xmath135 is the fraction of mass in halos where metal - rich star formation takes place , and @xmath136 is the same fraction of mass at @xmath0 . we assume that halos with circular velocity above @xmath137 km / s are the ones forming stars at @xmath0 ( as in onken & miralda - escud 2004 ; this affects only the normalization of the emissivity ) , but consider different possibilities for the emissivity at higher redshift , as explained below . in order to study the effect of varying the emission from normal , metal - rich galaxies , we consider several models listed in table [ tcases ] . these models are labeled by a number , and in general any model will be referred to by the letter specifying the feedback processes assumed ( see table 1 ) and this number . in models 1 and 2 , we compute @xmath135 as the fraction of mass in metal - enriched halos with a circular velocity greater than @xmath138 km / s , and in models 3 to 6 we assume that all halos that have been polluted by metals form enriched stars , with no minimum halo circular velocity . note that even by assuming a minimum halo circular velocity that is lower than at the calibration redshift @xmath0 , we are maximizing the emission at high redshift under the constraint that the emission per unit mass is not greater than in halos at @xmath0 . the value of the normalization constant @xmath134 is also subject to uncertainties arising from the process by which the intensity of the ionizing background is inferred from modelling the ly@xmath5 forest transmitted flux , and the measurement of the photon mean free path from the abundance of lyman limit systems . we adopt the values @xmath139 for models 1 to 4 ( as in miralda - escud 2003 ) , and twice this value , @xmath140 ( which would be favored by the measurement of the ionizing background intensity by bolton et al . 2005 ) , in models 5 and 6 . finally , the emissivity from the metal - free stars is varied by considering all stars to have a mass @xmath141 in models with odd number , and @xmath142 in models with even number . the evolution of the ionized fraction @xmath143 is calculated according to @xmath144 where @xmath145 is the total emissivity of ionizing photons , and @xmath146 is the recombination rate which we evaluate at @xmath147 k. the cosmic history of @xmath143 then yields the electron scattering optical depth @xmath148 . in this section , we present the results on metal - free star formation and the reionization history of the igm obtained from the approach given in [ smethod ] . in this model , there are four free parameters : the stellar mass , @xmath53 , the reheating of igm per metal - free star , @xmath103 , and the emissivity of metal - rich stars , @xmath134 . in addition , there is some uncertainty in the physical processes governing feedback effects . for this reason , we play with the possibility to switch on or off the effect of photodissociating photons in order to check how important it is . finally , to compute @xmath135 in equation ( [ emisp12 ] ) we may consider or not halos with circular velocity @xmath149 km / s ( minihalos ) to see whether the value of @xmath7 experiences a big change . the models including all the different possibilities are labelled xy , whith x running from a to d and y running from 1 to 6 ( see tables [ tcases ] and [ tsub_cases ] ) . we now present the results on the rate of metal - free star formation and the reionization history of the igm for the models described in 2 . the models and their nomenclature are described in tables 1 and 2 . the model number ( 1 to 6 ) refers to properties of the radiation sources : the mass of metal - free stars , and the normalization and low halo mass cutoff for the emission from metal - enriched stars . the model letter ( a to d ) refers to the presence or absence of the molecular photodissociation and x - ray heating feedback effects . results are first presented for the case of no feedback effects ( models a , [ sresultsnofeed ] ) , then including molecular photodissociation ( models b , [ sresultsj21 ] ) , and finally including x - ray photoheating ( models c and d , [ sres_xr ] ) . then , the reionization histories are reviewed ( [ sresreion ] ) . all models are run with the following set of cosmological parameters : @xmath150 , @xmath151 , @xmath152 , @xmath153 , @xmath154 and @xmath155 . cccc 1 & 7 & no & 100 + 2 & 7 & no & 300 + 3 & 7 & yes & 100 + 4 & 7 & yes & 300 + 5 & 14 & yes & 100 + 6 & 14 & yes & 300 + [ tcases ] ccc a & no & 0 + b & yes & 0 + c & yes & 10@xmath156 + d & yes & 10@xmath157 + [ tsub_cases ] the evolution of the emissivity for model a1 is shown in figure [ femis ] , both for metal - free and metal - rich stars . metal - free stars dominate at high redshift , and enriched stars start dominating below @xmath158 . this behaviour is easily understood . initially there are only metal - free stars and their contribution grows with time as more halos form . this results in an increasing fraction of halos being polluted with metals , which at some point causes a decline in the rate of new metal - free stars that are formed . the halos collapsing with enriched gas start forming enriched stars which eventually dominate the emissivity . the formation of metal - free stars is also reduced as redshift decreases due to the increased fraction of the igm that is ionized . in ionized regions , the gas is less able to cool and low - mass halos would continue merging into more massive ones before forming any stars , implying a great reduction in the number of metal - free stars that can be formed . this effect is less important because the igm ionized fraction remains relatively low at the epoch where most metal - free stars form . comparing models 1 ( with metal - free stellar mass of @xmath159 ) with models 2 ( @xmath142 ) , figure 4 shows that the emissivity of metal - free stars increases by a factor of @xmath160 as expected at @xmath161 , but at lower redshift it increases by a smaller factor because the fraction of ionized igm increases faster for @xmath162 , reducing the formation of metal free stars to a greater extent than for models 1 . we note here that a high bias in the distribution of metal - free stars relative to the mass could further increase the effect of ionization in suppressing the number of metal - free stars that can be formed . we now examine how the inclusion of a photodissociating background , computed self - consistently from the star formation history , alters the results . figure 5 shows the evolution of the photodissociating background intensity in models b1 and b2 . the maximum is reached around @xmath8 as the effects of increased metal pollution and ionized igm fraction suppress new metal - free star formation ( note that emission from enriched stars is not included in this figure ) . figure 4 shows that the maximum of formation rate of metal - free stars occurs at higher redshift when the photodissociating feedback is included , because of the additional suppression of star formation . we also see in figure 2 how the formation of metal - free stars is shifted to higher mass halos in models a compared to models b at @xmath163 , because the negative feedback delays the formation of the stars to a later stage at which halos have grown to a higher mass . we consider two models for the soft x - ray energy produced by a metal - free star that is converted into igm heat , @xmath164 ( table 2 , models c and d ) . during the main - sequence , and at effective surface temperatures near @xmath165 k , the fraction of radiative energy emitted above @xmath166 is @xmath115 0.1% of the total energy output of the star , or @xmath167 erg . the supernova explosion at the end of the life of a metal - free star can produce as much as @xmath168 erg of kinetic energy for some stellar masses ( see , e.g. , bromm et al . 2003 ) , and the fraction of these kinetic energy that is eventually reprocessed into soft x - rays may be @xmath169 ( oh & haiman 2003 ) , of which a fraction of 50% may be converted into igm heat when the soft x - rays are absorbed in a mostly neutral medium ( shull & van steenberg 1985 ) . given the substantial degree of uncertainty in the soft x - ray energy emitted per star , we consider the two values in models c and d bracketing our estimate . the increase in igm temperature caused by the absorption of soft x - rays is shown in figure 6 for models c1 and d1 . the onset of formation of metal - free stars causes a rise in temperature of the neutral medium starting at @xmath170 , and at lower redshift the temperature ceases to increase when the formation of metal - free stars is reduced ( in practice there would probably be an important contribution to the x - ray emission from the enriched population , which has been considered here ) . the establishment of an entropy floor causes a large reduction of the central density in low mass halos ( see fig . 1 ) , which greatly reduces the cooling rate . this implies that metal - free stars tend to be formed in models c and d in much more massive halos than in models a and b : the formation of these stars is suppressed in low - mass halos by the feedback effect , so metal - free stars are not formed until later when a more massive halo has formed from the mergers of several small ones ( see figure 2 ) . as long as our assumption that only one metal - free star forms in each halo continues to hold , the metal - free stars will be much less abundant . this is seen clearly in figure 4 , which shows the large reduction in the emissivity of metal - free stars caused by the soft x - ray heating . cccccc 1 & & 0.084 & 0.077 & 0.066 & 0.057 + 2 & & 0.116 & 0.103 & 0.087 & 0.068 + 3 & & 0.094 & 0.089 & 0.079 & 0.066 + 4 & & 0.121 & 0.110 & 0.096 & 0.076 + 5 & & 0.105 & 0.102 & 0.094 & 0.082 + 6 & & 0.127 & 0.119 & 0.108 & 0.090 + finally , we examine the reionization history and electron scattering optical depth to the cmb of electron scattering in all the models . in figure [ fionf ] we see that , as expected , feedback delays the ionization of the igm . reionization always occurs gradually over a relatively wide redshift range . naturally , models with the highest stellar mass and emission from enriched stars achieve a higher ionized fraction at every redshift , and hence a higher optical depth . even though the total emissivity history may be double - peaked owing to the presence of the metal - free and the enriched populations of stars ( see fig . 4 ) , there is never a double reionization because the recombination rate is slow enough that the ionized fraction continues to increase when the emission from metal - free stars starts decreasing . the redshift at which reionization is completed stays in the range of @xmath171 . this prediction of the redshift of the end of reionization is a natural consequence of the value of the emissivity assumed for the enriched stars that is derived from observations of ly@xmath5 absorption systems at @xmath172 , and of assuming that the clumping factor of ionized gas is not much larger than unity ( miralda - escud 2003 ) . the exact redshift of the end of reionization predicted by each model is not an independent test of our models , because this redshift can be modified and adjusted to agree with observations of the gunn - peterson trough in the highest redshift quasars ( white et al . 2003 ) by introducing a moderate clumping factor with an adequate dependence on redshift ( note that we have assumed a clumping factor equal to one ) . the values of the electron scattering optical depth for each model are presented in table 3 . the presence of metal - free stars has an important effect in creating a small ionized fraction at very high redshifts ( @xmath173 ) . however , the increase of the optical depth that metal - free stars can cause is modest and , even for @xmath174 and in the absence of any negative feedback effects , they do not increase the optical depth beyond @xmath175 . models d1 , d3 and d5 represent cases where the contribution from metal - free stars is minimal ( due to the very strong x - ray heating assumed ; see fig . 4 ) . comparing them with models a2 , a4 and a6 where the emission from metal - free stars is maximum , we see that the early ionization caused by metal - free stars can cause only a relatively modest increase of the optical depth , under our central assumption that only one metal - free star is made per halo , and no metal - free stars are made in halos that have already merged with a halo in which star had previously formed . we have considered in this paper the maximum emissivity and the contribution to reionization that metal - free stars could make . our central assumption is that only one metal - free star is made in every halo where pristine gas is able to cool via molecular hydrogen rotovibrational lines . this is a reasonable assumption in view of the simulations that have been made of the formation of the first star and the effects of the subsequent ionizing radiation emitted and supernova explosion : the gas cools hydrostatically towards the center , avoiding fragmentation and forming a central star which can then ionize and heat all the gas in its host halo , and probably expel it in the supernova explosion ( e.g. , abel et al . 2002 ; bromm et al . 2002 , 2003 ) . once a star has exploded , it pollutes its host halo as well as every larger halo into which its host halo merges in the future . any halo that has already formed a metal - free star in any of its halo progenitors in the past is assumed to form a metal - enriched population of stars that has an emissivity of ionizing photons similar to the emissivity that is observationally determined at @xmath176 . under this basic assumption , the contribution of the metal - free stars to reionization is relatively modest . metal - free stars may dominate the emissivity at very high redshifts ( @xmath177 ) , but before they can manage to ionize much of the universe , the metal pollution of halos rapidly reduces their formation rate . moreover , the effects of the ionization itself prevent cooling in the remaining low - mass halos with pristine gas , delaying the formation of a central metal - free star to the time of formation of more massive halos that collapse later , and thereby reducing the number of metal - free stars that are formed per unit of mass in the universe . reionization was likely completed by a population of stars and quasars in galaxies forming from pre - enriched gas . we have analyzed the effects of molecular photodissociation and of increasing the entropy floor by x - ray heating . these can further reduce the amount of ionizing radiation that is emitted by the population of metal - free stars . in agreement with oh & haiman ( 2003 ) , we have found that the negative feedback induced by x - ray heating can in principle be very large . however , as pointed out by kuhlen & madau ( 2005 ) , this negative feedback might be much less important than suggested by the simple calculation of oh & haiman ( 2003 ) and the one we have presented here . in our calculation , we consider only the cooling time of the gas at the halo center in the hydrostatic equilibrium configuration obtained at the end , by assuming an increased initial entropy produced by the x - ray heating . in reality , the gas should be able to gradually cool and lose entropy as it collapses in dense regions of halos , and these halos later merge to form the final one where a star is formed . the same x - rays that heat the gas would also increase the ionization , allowing faster molecular cooling as the gas density increases , which might greatly reduce the x - ray negative feedback effect . nevertheless , our main conclusion is that even without feedback effects the emission from metal - free stars is strongly limited by the metal - pollution and the ionization itself . our models are generally in agreement with previous work . the value of @xmath148 we infer is always lower than @xmath3 , even when we increase the emissivity of the enriched stellar population and we minimize any negative feedback effects on the metal - free population . for a smaller contribution from metal - free stars , values of @xmath148 are closer to @xmath178 ( see table [ ttaue ] ) , in agreement with the results of ricotti & ostriker ( 2004 ) . our results are also compatible with those of sokasian et al . ( 2004 ) , who find that high cmb optical depths to electron scattering can be achieved only when many stars are assumed to be formed within halos that cool by molecular hydrogen . the model m9 of sokasian et al . is the only one that assumes that only one star is formed per halo , and in this case their results for the maximum optical depth are in good agreement with ours . similarly , wise & abel ( 2005 ) can also produce high optical depths only by assuming that many metal - free stars form in each halo with a total mass much larger than in our model . naturally , if one is willing to assume a mass for metal - free stars even larger than @xmath179 , their contribution to the cmb optical depth , @xmath148 , can be further increased , but their total emission increases by a factor smaller than the increase in the mass because of the enhanced negative feedbacks . at the same time , if miniquasars emitting copious amounts of ionizing radiation where produced by the black holes created by the metal - free stars ( which could occur when a halo containing a central clump of cooling gas merges with another halo that has already formed a star and contains a black hole in its center ) , much more ionizing radiation could be produced , although the x - rays emitted by these mini - quasars might then be in conflict with limits on any unresolved component of the present soft x - ray background ( dijkstra et al . 2004a ) . our work can also be used to estimate the contribution of metal - free stars to the cosmic infrared background , which might include the ultraviolet light of these stars from high redshift . the total number of ionizing photons that are ever emitted by metal - free stars can be computed straightforwardly by integrating the curves shown in figure 4 . the results are shown in table [ temis ] for each one of our models . even in the absence of negative feedbacks , and for stellar masses @xmath180 , metal - free stars do not emit more than @xmath115 one photon per baryon in the universe . this conclusion is not surprising , since in the absence of many recombinations one does not need a large number of ionizing photons to be emitted per baryon to complete the reionization . because of the very high effective temperatures of metal - free stars ( @xmath181 k ) , the number of photons emitted in ultraviolet light at wavelengths longer than the lyman limit is similar . hence , the contribution of metal - free stars to the present cosmic infrared background is not more than @xmath115 one photon per baryon under the assumptions we have made . recently , kashlinsky et al . ( 2005 ) detected brightness fluctuations in the cosmic infrared background after subtracting all contributions from known galaxies . the measured fluctuations are at a level of @xmath182 . this background intensity at a wavelength of @xmath183 corresponds to @xmath184 photons per baryon in the universe ; as discussed by kashlinsky et al . , in order to account for the observed fluctuations the absolute brightness of the infrared background contributed by metal - free stars would have to be as high as @xmath185 ( or @xmath115 5000 photons emitted per baryon ) , after taking into account the expected high bias in the spatial distribution of these stars . it is clear that very extreme assumptions about the number of metal - free stars that were formed need to be made if the unaccounted fluctuations in the cosmic infrared background are related in any way to these metal - free stars . as discussed by santos et al . ( 2002 ) , if metal - free stars are to make an important contribution to the cosmic infrared background , one needs to assume that a large fraction of all the gas in each halo where pristine gas cools by molecular hydrogen ( typically more than @xmath81 ) forms metal - free stars , and that most of the radiation from the stars is somehow internally absorbed in the halos in order to prevent an excessively early reionization of the igm . both of these requirements are not realistic : once a central star is formed in a @xmath186 halo , its ionizing radiation will ionize and push all the halo gas out over the short main - sequence lifetime of the star , and the escape fraction of the ionizing photons is high except in rare cases where a single star forms in a very massive halo ( see figure [ ffesc ] ) . the detected cosmic infrared background fluctuations have other more likely possible sources , such as an incomplete accounting of the faint - end of the galaxy luminosity function , or a normal population of low - luminosity galaxies with metal - enriched gas and stars at high redshift . jm acknowledges helpful conversations with mark kuhnen , piero madau , and daniel wang . jm would also like to thank the institute for advanced study for their hospitality , where part of this work was completed . this work was supported in part by the direccin general de investigacin cientfica y tcnica of spain , under contract aya2003 - 07468-c03 - 01 . jmr was supported by a fellowship of the ministerio de educacin , cultura y deporte of spain . abel , t. , anninos , p. , zhang , y. , & norman , m. l. 1997 , new astronomy , 2 , 181 abel , t. , bryan , g. l. , & norman , m. l. 2002 , science , 295 , 93 becker , r. h. , et al . 2001 , , 122 , 2850 bennett , c. l. , et al . 2003 , , 148 , 1 bolton , j. s. , haehnelt , m. g. , viel , m. , & springel , v. 2005 , , 357 , 1178 bromm , v. , coppi , p. s. , & larson , r. b. 2002 , , 564 , 23 bromm , v. , ferrara , a. , coppi , p. s. , & larson , r. b. 2001 , , 328 , 969 bromm , v. , yoshida , n. , & hernquist , l. 2003 , , 596 , l135 bullock , j. s. , kolatt , t. s. , sigad , y. , somerville , r. s. , kravtsov , a. v. , klypin , a. a. , primack , j. r. , & dekel , a. 2001 , , 321 , 559 chiu , w. a. , fan , x. , & ostriker , j. p. 2003 , , 599 , 759 cuby , j .- , le f ` evre , o. , mccracken , h. , cuillandre , j .- c . , magnier , e. , & meneux , b. 2003 , , 405 , l19 dijkstra , m. , haiman , z. , & loeb , a. 2004a , , 613 , 646 dijkstra , m. , haiman , z. , rees , m. j. , & weinberg , d. h. 2004b , , 601 , 666 draine , b. t. & bertoldi , f. 1996 , , 468 , 269 eke , v. r. , navarro , j. f. , & steinmetz , m. 2001 , , 554 , 114 fan , x. , narayanan , v. k. , strauss , m. a. , white , r. l. , becker , r. h. , pentericci , l. , & rix , h. 2002 , , 123 , 1247 galli , d. & palla , f. 1998 , , 335 , 403 gnedin , n. y. 2000 , , 535 , 530 haiman , z. , abel , t. , & rees , m. j. 2000 , , 534 , 11 haiman , z. , & holder , g. p. 2003 , , 595 , 1 haiman , z. , rees , m. j. , & loeb , a. 1996 , , 467 , 522 hu , e. m. , cowie , l. l. , mcmahon , r. g. , capak , p. , iwamuro , f. , kneib , j .- p . , maihara , t. , & motohara , k. 2002 , , 568 , l75 hutchins , j. b. 1976 , , 205 , 103 kashlinsky , a. , arendt , r. g. , mather , j. , & moseley , s. h. 2005 , , 438 , 45 kitayama , t. , yoshida , n. , susa , h. , & umemura , m. 2004 , , 613 , 631 kneib , j. , ellis , r. s. , santos , m. r. , & richard , j. 2004 , , 607 , 697 kodaira , k. , et al . 2003 , , 55 , l17 kogut , a. , et al . 2003 , , 148 , 161 kuhlen , m. , & madau , p. 2005 , , 860 lacey , c. & cole , s. 1993 , , 262 , 627 lacey , c. & cole , s. 1993 , , 271 , 676 madau , p. , & rees , m. j. 2000 , , 542 , l69 miralda - escud ' e , j. 2003 , , 597 , 66 miralda - escude , j. , & rees , m. j. 1998 , , 497 , 21 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 oh , s. p. & haiman , z. 2003 , , 346 , 456 onken , c. a. & miralda - escud ' e , j. 2004 , , 610 , 1 oh , s. p. , nollett , k. m. , madau , p. , & wasserburg , g. j. 2001 , apj , 562 , l10 raig , a. , gonzlez - casado , g. , & salvador - sol , e. 2001 , , 327 , 939 rhoads , j. e. , et al . 2004 , , 611 , 59 ricotti , m. , & ostriker , j. p. 2004 , , 350 , 539 santos , m. r. , bromm , v. , & kamionkowski , m. 2002 , , 336 , 1082 schaerer , d. 2002 , , 382 , 28 shapiro , p. r. , iliev , i. t. , & raga , a. c. 2004 , , 348 , 753 sheth , r. k. & tormen , g. 2002 , , 329 , 61 shull , j. m. , & van steenberg , m. e. 1985 , , 298 , 268 sokasian , a. , yoshida , n. , abel , t. , hernquist , l. , & springel , v. 2004 , , 350 , 47 spergel , d. n. , et al . 2003 , , 148 , 175 tegmark , m. , silk , j. , rees , m. j. , blanchard , a. , abel , t. , & palla , f. 1997 , , 474 , 1 whalen , d. , abel , t. , & norman , m. l. 2004 , apj , 610 , 14 white , r. l. , becker , r. h. , fan , x. , & strauss , m. a. 2003 , , 126 , 1 wise , j. h. , & abel , t. 2005 , , 629 , 615 yoshida , n. , abel , t. , hernquist , l. , & sugiyama , n. 2003 , , 592 , 645 consider a halo of mass @xmath54 at time @xmath61 . the average number of smaller halos with mass between @xmath70 and @xmath71 at @xmath187 that are located within the halo of mass @xmath54 at @xmath61 is given by the extended ps formalism , @xmath188 where @xmath189 is the conditional probability for a mass element to be part of a halo of mass @xmath70 at @xmath72 , given that it is part of a larger halo of mass @xmath190 at a later time @xmath191 ( eq . [ 2.16 ] in lacey & cole 1994 ) . the rate at which stars form at @xmath72 in halos of mass between @xmath70 and @xmath71 , which will be found within the halo of mass @xmath54 at @xmath61 , is @xmath192 where @xmath193 is the probability per unit of time that a halo with mass @xmath70 forms a star at @xmath72 . this latter probability is obtained by noting that the probability for a star to form in a halo with mass @xmath70 between @xmath72 and @xmath194 is equal to the probability for the corresponding halo to form between @xmath195 and @xmath196 , being @xmath197 the required cooling time in this halo to form a star at @xmath72 , times the probability @xmath198 that no halo ancestor has previoulsy formed any star , @xmath199 in equation ( [ i ] ) , @xmath200 is the probability distribution function of halo formation times @xmath55 , and the derivative @xmath201 is to be evaluated by requiring the condition @xmath202 . although a better behaved definition of halo formation time with known analytical expressions of the corresponding probability distribution function can be found in the literature ( raig et al . 2001 ) , for simplicity we use here that provided by lacey & cole ( 1993 ) . these authors define the formation time of a halo with mass @xmath54 at @xmath61 as the earliest time @xmath203 at which some progenitor reaches a mass @xmath204 . at the same time , the average number of stars that have previously formed in ancestors of the halo with mass @xmath54 at @xmath61 , @xmath205 , is obtained by integration of equation ( [ nstar ] ) , @xmath206\ , . \label{nave}\ ] ] assuming a poisson distribution of the number of metal - free stars formed in ancestors of a halo of mass @xmath54 at @xmath61 , with mean expectation value provided by equation ( [ nave ] ) , the probability that no ancestor has previously formed any metal - free star is @xmath207 notice that in order to calculate the previous probability we must proceed in an iterative way since the term @xmath208 in equation ( [ nave ] ) includes the factor @xmath198 affecting its ancestors , implying that we need to save at each step the information of all previous halos along the merger tree . cccccc 1 & & 0.46 & 0.33 & 0.18 & 0.07 + 2 & & 0.80 & 0.61 & 0.40 & 0.19 + 3 & & 0.39 & 0.29 & 0.16 & 0.06 + 4 & & 0.73 & 0.56 & 0.38 & 0.18 + 5 & & 0.31 & 0.24 & 0.14 & 0.06 + 6 & & 0.62 & 0.49 & 0.34 & 0.16 +

results of theoretical calculations of ionization rates of rb and na rydberg atoms by blackbody radiation ( bbr ) are presented . calculations have been performed for _ _ n__s , _ _ n__p and _ _ n__d states of na and rb , which are commonly used in a variety of experiments , at principal quantum numbers _ _ n__=8 - 65 and at three ambient temperatures of 77 , 300 and 600 k. a peculiarity of our calculations is that we take into account the contributions of bbr - induced redistribution of population between rydberg states prior to photoionization and field ionization by extraction electric field pulses . the obtained results show that these phenomena affect both the magnitude of measured ionization rates and shapes of their dependences on _ n_. the calculated ionization rates are compared with the results of our earlier measurements of bbr - induced ionization rates of na _ _ n__s and _ _ n__d rydberg states with _ _ n__=8 - 20 at 300 k. a good agreement for all states except _ _ n__s with is observed . we also present the useful analytical formulae for quick estimation of bbr ionization rates of rydberg atoms .

physics0702192

blackbody radiation ( bbr ) is known to strongly affect the populations of atoms in highly excited rydberg states @xcite . it has also been shown that at the ambient temperature of 300 k bbr can photoionize rydberg atoms with @xmath1 at astonishingly high rates ( @xmath2 ) @xcite . strong effect of bbr on rydberg atoms is related to large matrix elements of bound - bound and bound - free transitions between rydberg states in the microwave and far infrared spectral range @xcite . interaction of rydberg atoms with bbr has been studied earlier in various contexts . farley and wing @xcite calculated the dynamic stark shifts and depopulation rates of rydberg levels of alkali atoms with @xmath3 at 300 k. the temperature dependence of bbr - induced transitions rates from the 19s state of sodium was calculated and measured by spencer et al @xcite . galvez et al @xcite have studied bbr - induced cascade transitions from the initially populated @xmath4=24 - 29 states of na , both theoretically and experimentally . although interaction of rydberg atoms with blackbody radiation has been studied for years , both theoretically and experimentally , only a few works were devoted to bbr - induced ionization of rydberg atoms itself . the temperature dependence of the bbr - induced ionization rate of na 17d state was numerically calculated using the quantum defect method in a coulomb approximation and experimentally measured by spencer et al @xcite . a simple scaling law for bbr ionization rates was also introduced in that work . more recently , the interest to bbr - induced ionization of rydberg atoms has been related to the spontaneous formation of ultracold plasma in dense samples of cold rydberg atoms @xcite , and to the prospects of its use as a convenient reference signal in absolute measurements of collisional ionization rates @xcite . even nowadays , however , the studies of spontaneous evolution of ultracold rydberg atoms to a plasma caused by bbr @xcite use the simple estimates of bbr ionization rates taken from the well - known work @xcite . the numerical calculations of direct bbr photoionization rates were performed by lehman @xcite for h and na rydberg states with @xmath4=10 - 40 using a herman - skillman potential with an additional term to account for core polarization . however , the theoretical data for rb atoms , which are widely investigated in experiments on ultracold plasma , are lacking . no systematic experimental studies of the @xmath4-dependences of bbr ionization rates of alkali - metal rydberg atoms in a wide range of principal quantum numbers are known . a major problem in the interpretation of measured ionization rates of rydberg atoms is related to the fact that populations of rydberg states are affected by bbr - induced processes to an unknown ( or , at least , not straightforwardly predictable ) extent , which depends on the combination of specific experimental conditions ( principal quantum numbers _ n _ , ambient temperature , duration of measurements , extracting electric field strengths , etc . ) . therefore , a more detailed study of bbr - induced ionization of rydberg atoms under typical experimental conditions is required , especially for higher rydberg states that are often explored in the experiments with cold atoms . in this article we present the results of numerical calculations of bbr ionization rates of rb and na rydberg atoms for the most commonly used _ _ n__s , _ _ n__p and _ _ n__d states with _ _ n__=8 - 65 at the ambient temperatures of 77 , 300 and 600 k. in these calculations we take into account two phenomena that may affect the observed ionization rates : the time - dependent bbr - induced population redistribution between rydberg states prior to photoionization and the selective field ionization ( sfi ) of high - lying rydberg states by the electric field applied for extraction of ions from the excitation zone . although both phenomena were mentioned in some of the earlier studies @xcite , their effects have not yet been studied in sufficient detail neither theoretically , nor experimentally . in particular , in the measurements of direct bbr photoionization rate of the sodium 17d state @xcite with effective lifetime 4 @xmath5s at the ambient temperature of 300 k , the effect of population redistribution was diminished due to the short measurement time of 500 ns and weak electric field used for extraction of ions . however , for rydberg states with @xmath6 and effective lifetimes less than 500 ns even this very short time interval is insufficient to avoid mixing with neighboring states , so that mixing processes must be necessarily accounted in calculations . in the present study we address to both phenomena and provide the useful analytical formulae , which can be applied to any rydberg atom for estimates of the direct bbr - induced ionization rates and of the contribution of the sfi . the simplest ( but often insufficient ) way of considering bbr - induced ionization after the excitation of an atom _ a _ to a given _ nl _ rydberg state is to limit the problem to the direct photoionization of the initial _ nl _ state in one step by absorption of bbr photons : @xmath7 where @xmath8 is the energy of absorbed bbr photon , @xmath9 is the atomic ion , and @xmath10 is the free electron emitted in the ionization . in the reality , however , ionization of rydberg atoms exposed to bbr is a complex process , in which the following main components can be identified [ see fig . 1(a ) ] : ( i ) direct photoionization of atoms from the initial rydberg state via absorption of bbr photons , ( ii ) field ionization by extraction electric field pulses of high rydberg states , which are populated from the initial rydberg state by absorption of bbr photons , ( iii ) direct bbr - induced photoionization of atoms in the neighboring rydberg states , which are populated due to absorption and emission of bbr photons prior to photoionization , and ( iv ) field ionization of other high - lying states , which are populated via population redistribution involving two or more steps of bbr photon absorption and/or emission events . our calculations show that all these processes can contribute to the total ionization rate to a comparable extent , and , therefore , none of them can be disregarded . in what follows we will consider the above processes separately and calculate the total bbr ionization rates , both analytically and numerically . ionization mechanisms of rydberg atoms exposed to bbr are illustrated in fig . 1 . the total brr - induced ionization rate can be written as consisting of four separable contributions : @xmath11 [ fig1 ] rydberg states due to spontaneous and bbr induced transitions from the initial state . ( b ) calculated spontaneous and bbr induced transition rates from the initial 16s state to other @xmath12p states . ( c ) calculated spontaneous and bbr induced transition rates from the initial 16d state to other @xmath12p states.,title="fig:",width=264 ] the first contribution , @xmath13 , is the direct bbr photoionization rate of the initially excited _ nl _ state , which will be discussed in section ii a. the second term , @xmath14 , is the rate of the selective field ionization ( sfi ) of high @xmath15 rydberg states , which are populated from the initial rydberg state _ nl _ via absorption of bbr photons . this field ionization will be discussed in section ii c , while redistribution of population between rydberg states will be described in section ii b. the third term , @xmath16 , is the total rate of bbr - induced photoionization of neighboring @xmath17 rydberg states , which are populated via spontaneous and bbr - induced transitions from the initial state . the last term , @xmath18 , is the rate of sfi of high - lying rydberg @xmath15 states that are populated in a two - step process via absorption of bbr photons by atoms in @xmath17 states ( note , that here , in contrast to @xmath14 , we consider lower @xmath17 , states which can not be directly field ionized ) . these latter two ionization rates , which are related to population redistribution between rydberg states , will be considered in section ii d . the atomic units are used below , unless specified otherwise . direct bbr - induced photoionization rate @xmath13 of a given _ nl _ state is calculated from the general formula @xcite : @xmath19 where _ c _ is the speed of light , @xmath20 is the photoionization threshold frequency for the _ nl _ rydberg state with the effective principal quantum number @xmath21 , where @xmath22 is a quantum defect , and @xmath23 is the photoionization cross - section at the frequency @xmath24 . the volume density @xmath25 of bbr photons at the temperature _ t _ is given by the plank distribution : @xmath26}},\ ] ] where _ kt _ is thermal energy in atomic units . for isotropic and non - polarized thermal radiation field , the value of @xmath23 is determined by the radial matrix elements @xmath27 of dipole transitions from discrete _ nl _ rydberg states to the continuum states with _ _ l@xmath28__1 and photoelectron energy _ e _ : @xmath29 where _ _ l__@xmath30 is the largest of @xmath31 and @xmath32 . the main problem in the calculation of _ _ w__@xmath33 for an arbitrary rydberg state is thus to find the values of @xmath34 and their frequency dependence . in order to achieve high accuracy of the matrix elements , numerical calculations should be used . in this work we used the semi - classical formulae derived by dyachkov and pankratov @xcite . in comparison with other semi - classical methods @xcite , this method is advantageous as it gives orthogonal and normalized continuum wavefunctions , which allow for the calculation of photoionization cross - sections with high accuracy . we have verified that photoionization cross - sections of the lower sodium s states calculated using the approach of @xcite are in good agreement with the sophisticated quantum - mechanical calculations by aymar @xcite . approximate analytical expressions for _ _ w__@xmath33 would also be useful , since they illustrate how the ionization rate depends on parameters _ n _ , _ l _ , and _ t_. such expressions can be obtained using the analytical formulae for bound - bound and bound - free matrix elements deduced by goreslavsky , delone and krainov ( gdk ) @xcite in the quasiclassical approximation . for the direct bbr - induced photoionization of an _ nl _ rydberg state the cross - section is given by : @xmath35,&&\end{aligned}\ ] ] where @xmath36 is the modified bessel function of the second kind . this formula was initially derived to describe the photoionization of hydrogen atom . it was assumed that the formula can be extended to alkali atoms simply by replacing _ n _ by @xmath37 , where @xmath38 is the quantum defect of the rydberg state . in the reality , however , its accuracy in absolute values is acceptable only for truly hydrogen - like states with small quantum defects . a disadvantage of the gdk model is that it disregards non - hydrogenic phase factors in the overlap integrals of dipole matrix elements . the main contribution to _ _ w__@xmath33 in eq . ( [ eq3 ] ) is due to bbr frequencies near the ionization threshold frequency @xmath39 , because the bound - free dipole moments rapidly decrease with increasing @xmath40 . for rydberg states with _ _ n@xmath41__1 and low _ _ one has ( @xmath42__l__@xmath43/3)@xmath441 . in this case ( [ eq6 ] ) can be simplified to the form : @xmath45 . & & \end{aligned}\ ] ] the combination of eqs . ( [ eq3 ] ) , ( [ eq4 ] ) and ( [ eq7 ] ) yields : @xmath46}\times & & \nonumber\\ \times \frac{{d\omega } } { { \mathrm{e}^{\omega /\left ( { kt } \right ) } - 1}}.&&\end{aligned}\ ] ] the expression in square brackets is a slowly varying function of @xmath40 . taking into account that the main contribution to _ _ w__@xmath33 is due to frequencies near the ionization threshold , one can replace @xmath40 by 1/(2__n__@xmath47 ) . after such replacement the integral in eq . ( [ eq8 ] ) can be calculated analytically , and the final result is : [ fig2 ] @xmath48 \mathrm{ln}\left ( { \frac{{1}}{{1 - \mathrm{exp}\left ( { - \frac{{\omega _ { nl } } } { { kt } } } \right ) } } } \right).\ ] ] equation ( [ eq9 ] ) gives the approximate direct bbr photoionization rate in atomic units for _ t _ measured in kelvins . alternatively , it can be rewritten to yield _ _ w__@xmath33 in the units of s@xmath49 for temperature _ t _ taken in kelvins as follows : @xmath50 \times \qquad \qquad \qquad & & \nonumber \\ \times \mathrm{ln}\left ( { \frac{{1}}{{1 - \mathrm{exp}\left ( { - \frac{{157890}}{{tn^{2 } } } } \right ) } } } \right ) \,\,\left [ { \mathrm{s}^ { - 1 } } \right ] . \quad & & \end{aligned}\ ] ] here _ _ c__@xmath51 is an _ l_-dependent scaling coefficient , which will be discussed later . by replacing _ n _ with the effective principal quantum number , eq . ( [ eq10 ] ) can be used for quick estimations of direct bbr ionization rates for various rydberg atoms and states . the precise values , however , should be calculated numerically . the results of our numerical and analytical calculations of the direct bbr - induced photoionization rates for sodium and rubidium _ _ n__s , _ _ n__p and _ _ n__d states with _ _ n__=5 - 80 at the ambient temperatures _ _ t__=77 , 300 and 600 k are shown in fig . 2 . a good agreement of our numerical results obtained using the dyachkov and pankratov model with the theoretical data obtained by lehman @xcite is observed . for the case of rubidium such comparison is not possible because no other published data are available , to the best of our knowledge . it is also interesting and instructive to compare the results obtained by numeric calculation with those obtained using the analytical formula ( [ eq10 ] ) . figure 2(a ) shows a remarkable disagreement between the numerical and analytical results ( assuming _ _ c__@xmath51=1 ) for sodium _ _ n__s states , which are known to possess a large quantum defect ( @xmath52=1.348 ) . at the same time , 2(b ) and 2(c ) show that in the case of na _ _ n__p and _ _ n__d states , which have smaller quantum defects ( @xmath53=0.855 and @xmath54=0.015 ) , the agreement between the numerical and analytical results is much better . this is not unexpected , since eq . ( [ eq6 ] ) is valid only for states with small quantum defects . formally , the disagreement for the non - hydrogenic _ _ n__s states stems from peculiarities of the asymptotic behavior of bessel functions in eq . ( [ eq6 ] ) for states with @xmath55 : the analytical expression of gdk model yields close photoionization cross - section values for _ _ n__s , _ _ n__p and _ _ n__d states , while the accurate numerical calculations yield significantly smaller cross - sections for sodium _ _ n__s states [ see fig . at the same time , one can see from fig . 2(a ) that shapes of the analytical curves are quite similar to the numerical ones . therefore , one can simply introduce a scaling coefficient in eq . ( [ eq10 ] ) in order to make it valid also for _ _ n__s states . in order to illustrate that , in figs . 2(d)-(f ) we show the rescaled results of eq . ( [ eq10 ] ) for the case of rb rydberg atoms _ _ n__p , and _ _ n__d states , which all have large quantum defects ( @xmath38 = 3.13 , 2.66 , and 1.34 , respectively ) . a good agreement between the analytical and numerical results was obtained when the rate obtained from eq . ( [ eq10 ] ) was scaled by a factor of _ _ c__@xmath56=0.2 , _ _ c__@xmath57=0.2 and _ _ c__@xmath58=0.4 for the case of _ _ n__p , and _ _ n__d states , respectively . our precise numerical data on _ _ w__@xmath33 are summarized in tables ( [ table1])-([table6 ] ) of appendix . bbr causes not only direct photoionization of the initially populated levels . it also induces transitions between neighboring rydberg states , thus leading to a population redistribution @xcite . for example , after laser excitation of the na 16s state , the bbr - induced transitions populate the neighboring @xmath12p states [ fig . the calculations show that these states have significantly higher direct photoionization rates _ _ w__@xmath33 than the 16s state itself . hence , bbr - induced population transfer to @xmath12p state can noticeably affect the measured effective bbr ionization rate . the rates of spontaneous and bbr - induced transitions from the initial 16s and 16d states to a number of @xmath12p states have been calculated by us in @xcite and are shown in figs . 1(b ) and 1(c ) . importantly , absorption of bbr induces also transitions to higher rydberg states , which are denoted as @xmath59 in fig . these states can be field ionized by the electric field pulses usually applied in experiments in order to extract ions into channeltron or microchannel plate detectors . extraction electric field pulses , which are commonly used to extract ions from the ionization zone to the ionization detector , ionize rydberg states with principal quantum numbers _ n _ exceeding some critical value @xmath60 . this critical value @xmath60 depends on the amplitude of the applied electric field and it can be found from the approximate formula @xcite @xmath61 were _ _ e__@xmath62 is the critical electric field for @xmath63 . hence , if a bbr mediated process populates a state with @xmath64 , this state will be ionized and thus will contribute to the detected ionization signal @xcite . we calculated the radial matrix elements @xmath65 of all dipole - allowed transitions to other @xmath17 states with @xmath66 using the semi - classical formulae of @xcite . the rate of a bbr - induced transition between the states @xmath67 and @xmath17 is expressed through the rate of spontaneous transitions given by the einstein coefficient @xmath68 : @xmath69 where @xmath70 is the transition frequency . note that spontaneous transitions are possible only to lower levels , while bbr leads to transitions both upwards and downwards . the total rate @xmath71 of bbr transitions to all rydberg states with @xmath72 was calculated by summing the individual contributions of @xmath73 transitions given by eq . ( [ eq12 ] ) : @xmath74 the values of @xmath71 were numerically calculated for various amplitudes _ e _ of the electric field pulses . we also compared the numerical values with those obtained from the approximate analytical formulae , which has been derived with the bound - bound matrix elements of the gdk model : @xmath75}\times & & \nonumber\\ \times \frac{{d\omega } } { { \mathrm{e}^{\omega /\left ( { kt } \right ) } - 1}}.\qquad & & \end{aligned}\ ] ] the integration limits are chosen such that the integral accounts for transitions to those rydberg states , for which @xmath76<\omega<\omega _ { nl}$ ] ( i.e. , states above the field ionization threshold ) . integration of eq . ( [ eq14 ] ) gives another useful analytical formula that is similar to eq . ( [ eq10 ] ) : @xmath77\times \qquad \qquad \qquad \qquad & & \nonumber \\ \times \left(\mathrm{ln}\frac{{1}}{{1 - \mathrm{exp}\left ( { \frac{{157890}}{{tn_{c}^{2 } } } - \frac{{157890}}{{tn^{2 } } } } \right)}}\,- \right . \qquad \qquad & & \nonumber \\- \left . \mathrm{ln}\frac{{1}}{{1 - \mathrm{exp}\left ( { - \frac{{157890}}{{tn^{2 } } } } \right ) } } \right ) \left [ \mathrm{s}^ { - 1}\right ] , \qquad & & \end{aligned}\ ] ] where _ t _ is in kelvins . [ fig3 ] the obtained numerical and analytical data on _ _ w__@xmath78 are presented in fig . 3 . for na atoms , the calculations were made for the _ _ n__s , _ _ n__p and _ _ n__d states with _ _ n__=5 - 35 at the ambient temperature _ _ t__=300 k [ figs . 3(a ) , 3(b ) and 3(c ) ] . the amplitudes _ e _ of the extracting electric field pulse are chosen as 100 and 200 v / cm ( corresponding to @xmath79 and 36 , respectively ) . these values are close to those used in our recent atomic - beam experiment on ionization of na atoms @xcite . alternatively , figs . 3(d ) , 3(e ) and 3(f ) show the calculated rates @xmath71 for _ _ n__s , _ _ n__p and _ _ n__d states of rb for the lower _ e _ values ( 5 v / cm and 10 v / cm corresponding to @xmath63=91 and @xmath63 = 76 , respectively ) , which are more adequate to experiments with rydberg atoms in cold gases . in order to achieve a better agreement between the analytical formula ( [ eq15 ] ) and the results of numerical calculations , scaling coefficients @xmath80 in eq . ( [ eq15 ] ) were also introduced . the best agreement was obtained at @xmath81 , @xmath82 and @xmath83 for the rb _ _ n__s , _ _ n__p and _ _ n__d states , respectively . in the case of _ _ n__p and _ _ n__d states of na data such scaling was not necessary , while in the case of _ _ n__s states of na the best agreement [ not shown in fig . 3(a ) ] was found at @xmath84 . the possibility to achieve a satisfactory agreement between the numerical and analytical data of fig . 3 suggests that , using appropriate scaling coefficients , the analytical formula ( [ eq15 ] ) is also suitable to quick estimates of bbr - induced sfi rates for various rydberg atoms and states . our precise numerical data on @xmath71 are summarized in tables ( [ table1])-([table6 ] ) of appendix . in this section we shall analyze the influence of time evolution of populations of rydberg states upon interaction with ambient bbr photons . the typical timing diagram of laser excitation of rydberg states and detection of ions is shown in fig . such scheme was used in our recent experiment on collisional ionization of na rydberg atoms @xcite . the first electric - field pulse is applied immediately after the laser excitation pulse in order to remove the atomic @xmath9 and molecular @xmath85 photoions produced by the laser pulse . then atoms are allowed to interact with ambient bbr during the time interval ( _ _ t__@xmath86 , _ _ t__@xmath87 ) . the second electric - field pulse extracts the ions , which have been produced by collisional and bbr - induced ionization , to the ion detector . the atomic and molecular ions were distinguished using a time - of - flight method . at the ground - state density of @xmath88@xmath89 the bbr ionization is the main source of atomic ions @xcite and the contribution from collisonal ionization of penning type is negligible . let us consider first the simplest case of laser excitation of a single _ _ n__s state . the evolution of the number @xmath90 of atomic ions produced via absorption of bbr photons by atoms in the initial _ _ n__s state is given by @xmath91 where @xmath92 is the total number of rydberg atoms remaining in the ns state as a function of time , and @xmath93 is an effective lifetime of the @xmath4s state at a given ambient temperature . the registered photoions are produced during the time interval ( @xmath94 ) . the total number of ions can then be found by integrating eq . ( [ eq16 ] ) from @xmath95 to @xmath96 . @xmath97,&&\end{aligned}\ ] ] this result can be rewritten by introducing an effective interaction time @xcite : @xmath98.&&\end{aligned}\ ] ] blackbody radiation also induces transitions to other rydberg states @xmath12p , as was discussed in section ii b. evolution of populations of these states is described by the rate equation @xmath99\,n_{ns } \left ( { t } \right)-&&\nonumber\\ -{n_{{n}'p } \left ( { t } \right)}/{\tau_{eff}^{n'p}},\qquad & & \end{aligned}\ ] ] where @xmath100 and @xmath101 are the rates of population of the @xmath12p state due to spontaneous transitions to lower levels and bbr induced transitions upwards and downwards from the initial _ _ n__s state , and @xmath102 is the effective lifetime of the @xmath12p state . [ fig4 ] and molecular @xmath103 ion signals ; ( d ) detector gates . , title="fig:",width=226 ] a combination of eq . ( [ eq19 ] ) with eqs . ( [ eq16 ] ) and ( [ eq17 ] ) yields @xmath104 } } { \left(\tau _ { eff}^{n'p}\right)^{-1}- \left(\tau_{eff}^{ns}\right)^{-1 } } } \times & & \nonumber\\ \times w_{bbr } \left ( { { n}'p } \right)\left ( { 1 - \frac{{t_{eff}^{ns } } } { { t_{eff}^{{n}'p } } } } \right).\quad & & \end{aligned}\ ] ] the main contribution to the sum in eq . ( [ eq20 ] ) is from @xmath12p states with @xmath105 [ see fig . the effective bbr ionization rates for _ _ n__p and _ _ n__d states were determined in the same way as for _ _ n__s states , taking into account the population transfer to both @xmath106 and @xmath107 states . the rate @xmath18 describes the second - order process of bbr - induced transitions from the neighboring @xmath17 states to highly excited states @xmath108 with @xmath109 [ see fig . 5(a ) ] , followed by ionization of these states by extracting electric field pulses . this rate can be calculated using the same eq . ( [ eq20 ] ) , in which @xmath110 is replaced by @xmath71 and the summation is done over the states with @xmath111 . our precise numerical data on @xmath112 and @xmath18 are summarized in tables ( [ table1])-([table6 ] ) of appendix . [ fig5 ] for ( a ) _ _ n__s and ( b ) _ _ n__d states of na with _ _ n__=8 - 35 for the extracting electric field pulses of 100 v / cm and 200 v / cm at the ambient temperatures of _ _ t__=77 , 300 and 600 k. experimental points are taken from ref . , title="fig:",width=226 ] [ fig6 ] calculated total bbr - induced ionization rates @xmath113 for na _ _ n__s , _ _ n__p and _ _ n__d states at the temperatures _ t _ = 77 , 300 , 600 k and for the extracting electric field pulses of _ _ e__=100 v / cm and 200 v / cm , as well as their comparison with our experimental data from ref . @xcite , are shown in fig . 5 . since the values of @xmath114 depend on the timing of the experiment ( see fig . 4 ) , the calculations were performed for @xmath115s and @xmath116s , which were used in our experiment @xcite . at the 300 k temperature a good agreement is observed for _ _ n__d states with _ _ n__=8 - 20 and for _ _ n__s states with _ _ n__=8 - 15 . at the same time , for _ _ n__s states with @xmath0 the calculated ionization rates start to deviate from the experimental values ; the measured values exceed the theoretical ones by a factor of 2.1 for _ _ n__=20 , and the shape of the experimental _ n_-dependence differs from the theoretical one . one possible explanation of such anomaly for _ _ n__s states is related to their specific orbit that penetrates into the atomic core . the penetration causes a strong interaction between the rydberg electron and the core , e.g. , due to core polarization @xcite . this results in large quantum defect and a cooper minimum in the photoionization cross - sections . this assumption is supported by the good agreement of theory and experiment for the hydrogen - like _ _ n__d states , which have a small quantum defect and almost non - penetrating orbits . total bbr - induced ionization rates were also calculated for na and rb @xmath4s , @xmath4p and @xmath4d states in a broader range of @xmath4 and for lower amplitudes of the electric field pulses ( 5 v / cm and 10 v / cm ) . such fields are more relevant to the experiments with cold rydberg atoms , e.g. , to ultracold plasma formation from cold rydberg atoms clouds @xcite , since these experiments explore the atoms excited to relatively high rydberg states that are ionized in weaker electric fields ( 35 v / cm for @xmath117 and 10 v / cm for @xmath118 ) . the calculated ionization rates are presented in fig . it can be seen that sfi and bbr - induced level - mixing processes alter the shapes of _ n_-dependences of total ionization rates @xmath113 , which effect is more pronounced at lower temperatures and larger _ n_. as fig . 6 has a logarithmic scale , the visible effect of additional processes looks small . however , from the tables of appendix it can be seen that , e.g. , for the rb 65s state at @xmath119=300 k the total bbr ionization rate is twice larger than the direct photoionization rate ( see table 5 ) . precise results of our numerical calculations for a five - step set of principal quantum numbers _ n _ are summarized in tables ( [ table1])-([table6 ] ) of appendix . the values of ionization rates at the other _ n _ values can be obtained either by extrapolation or upon request from the authors of this paper . we have calculated the total bbr - induced ionization rates of na and rb _ _ n__s , _ _ n__p and _ _ n__d rydberg states for principal quantum numbers _ _ n__=8 - 65 at the ambient temperatures of 77 , 300 and 600 k. our calculations take into account the effect of bbr - induced mixing of rydberg states and their field ionization by extracting electric field pulses . useful analytical formulae have been derived , which allow for quick estimation of ionization rates and their dependences on the principal quantum number _ n_. the numerical results are in a good agreement with our recent experiment data on na _ _ n__s and _ _ n__d states , except for _ _ n__s states with @xmath0 , which is most probably associated with the cooper minimum in the photoionization cross - section . the obtained results show that bbr - induced redistribution of population over rydberg states and their field ionization by extracting electric fields affect both the magnitudes of total ionization rates and shapes of their dependencies on the principal quantum number . this suggests that these processes are important and can not be ignored in the calculations and measurements of bbr ionization rates . equations ( [ eq16])-([eq20 ] ) , as well as the analytical formulae ( [ eq9 ] ) and ( [ eq15 ] ) , can be used to calculate total ionization rates @xmath113 under particular experimental conditions . the obtained numerical results may be helpful to the analysis of ionization signals measured in experiments on collisional ionization and spontaneous formation of ultracold plasma , since bbr - induced ionization is the main channel of delivering atomic ions . at the same time , as we have revealed that theoretical data for na @xmath4s - states noticeably disagree with experiment at @xmath120 , new experimental data for na and rb in a broader range of principal quantum numbers would be of interest for the further improvement of theory , especially for the non - hydrogen - like states . this work was supported by intas grant no . 04 - 83 - 3692 , russian foundation for basic research ( grants no . 05 - 02 - 16181 , 05 - 03 - 33252 ) , siberian branch of ras , eu fp6 tok project lamol , european social fund , latvian science council , and nato grant eap.rig.981387 . @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + & & & & & & & & + 10 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125&@xmath126 & @xmath127&@xmath128 + 15 & 0.009 & 0.019 & 0.042 & 0.002 & 0.003 & 0.006 & 0.033&0.059 + 20 & 0.805 & 1.655 & 3.83 & 0.059 & 0.118 & 0.240 & 2.64&4.94 + 25 & 4.04 & 10.23 & 25.29 & 0.185 & 0.420 & 0.891 & 14.9&30.4 + 30 & 8.45 & 27.12 & 81.09 & 0.257 & 0.706 & 1.689 & 36.5&91.5 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 11.8 & 2.47 & 5.20 & 0.267 & 0.0488 & 0.107 & 14.6&17.4 + 45 & 14.6 & 4.12 & 9.09 & 0.195 & 0.0466 & 0.106 & 19.0&24.0 + 55 & 14.1 & 5.76 & 13.9 & 0.124 & 0.0412 & 0.102 & 20.1&28.3 + 65 & 12.6 & 8.01 & 23.6 & 0.0773 & 0.0377 & 0.111 & 20.7&36.4 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + & & & & & & & & + 10 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 & @xmath134 & @xmath135&@xmath136 + 15 & 0.0996 & 0.195 & 0.389 & 0.00111 & 0.00200 & 0.00409 & 0.298 & 0.493 + 20 & 5.22 & 10.2 & 20.5 & 0.0217 & 0.0455 & 0.0977 & 15.4&25.8 + 25 & 22.1 & 50.8 & 107 & 0.0545 & 0.132 & 0.308 & 73.1&130 + 30 & 42.4 & 120 & 290 & 0.0662 & 0.197 & 0.554 & 163&333 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 57.0 & 10.5 & 22.9 & 0.0726 & 0.0144 & 0.0318 & 67.6 & 80.0 + 45 & 66.2 & 16.2 & 37.1 & 0.0514 & 0.0135 & 0.0313 & 82.5&103 + 55 & 61.9 & 21.6 & 53.7 & 0.0320 & 0.0117 & 0.0297 & 83.5&116 + 65 & 53.8 & 28.6 & 86.5 & 0.0186 & 0.00997 & 0.0304 & 82.4&140 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + & & & & & & & & + 10 & @xmath137 & @xmath138 & @xmath139 & @xmath140 & @xmath141 & @xmath142&@xmath143 & @xmath144 + 15 & 0.251 & 0.483 & 0.947 & 0.00270 & 0.00472 & 0.00953&0.742&1.21 + 20 & 6.59 & 12.7 & 25.2 & 0.0371 & 0.0762 & 0.158&19.4 & 32.0 + 25 & 22.6 & 50.0 & 104 & 0.0926 & 0.215 & 0.467&72.9 & 127 + 30 & 39.0 & 107 & 256 & 0.120 & 0.331 & 0.812&146 & 296 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 49.3 & 8.88 & 19.4 & 0.131 & 0.0239 & 0.0527&58.3 & 68.8 + 45 & 54.6 & 13.0 & 29.6 & 0.0995 & 0.0230 & 0.0528&67.7&84.3 + 55 & 49.8 & 16.7 & 41.5 & 0.0663 & 0.0204 & 0.0503&66.7 & 91.5 + 65 & 42.7 & 21.8 & 65.9 & 0.0424 & 0.0180 & 0.0519&64.6&109 + @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + & & & & & & & & + 10 & 7.50 & 1.33 & 2.34 & 13.6 & 3.02 & 4.88&25.5&28.3 + 15 & 114 & 39.4 & 70.2 & 78.6 & 24.2 & 39.6 & 256 & 302 + 20 & 223 & 99.2 & 185 & 77.6 & 29.2 & 49.2 & 429 & 534 + 25 & 254 & 155 & 314 & 53.6 & 26.2 & 46.9 & 489 & 669 + 30 & 242 & 218 & 537 & 34.2 & 23.2 & 46.9 & 518 & 861 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 217 & 19.3 & 39.5 & 22.6 & 1.55 & 3.29 & 261 & 283 + 45&166&22.8 & 49.1 & 10.2 & 1.04 & 2.30 & 200 & 227 + 55 & 126&27.3 & 64.2 & 5.15 & 0.780 & 1.87 & 159 & 197 + 65 & 97.4&35.0 & 100 & 2.80 & 0.652 & 1.86 & 136 & 202 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + 10 & 79.92 & 15.99 & 25.84 & 2.426 & 0.8949 & 1.456 & 99.23 & 109.6 + 15 & 778.2 & 240.9 & 393.6 & 17.82 & 5.476 & 9.174&1042 & 1199 + 20 & 1295 & 501.8 & 845.4 & 19.05 & 6.384 & 11.24 & 1823 & 2171 + 25 & 1374 & 709.0 & 1272 & 13.50 & 5.610 & 10.76 & 2102 & 2670 + 30 & 1253 & 932.0 & 1919 & 8.444 & 4.660 & 10.56 & 2198 & 3191 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 1090 & 79.97 & 169.6 & 6.145 & 0.3428 & 0.7388 & 1177 & 1267 + 45 & 795.2 & 89.28 & 198.3 & 2.830 & 0.2290 & 0.5203 & 887.5&996.9 + 55 & 586.0 & 102.1 & 247.0 & 1.419 & 0.1679&0.4172 & 689.7 & 834.8 + 65 & 441.8 & 124.9 & 366.0 & 0.7254 & 0.1267 & 0.3768&567.5 & 808.9 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + 10 & 157.5 & 37.01 & 59.22 & 13.71 & 2.596 & 4.237 & 210.8 & 234.7 + 15 & 890.1 & 268.2 & 434.4 & 39.52 & 8.003 & 13.30&1206 & 1377 + 20 & 1253 & 467.7 & 782.0 & 36.76 & 8.176 & 14.05 & 1765 & 2086 + 25 & 1244 & 613.5 & 1094 & 26.17 & 6.907 & 12.61 & 1891 & 2377 + 30 & 1099 & 774.0 & 1592 & 17.38 & 5.679 & 11.70 & 1896 & 2720 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 936.2 & 65.00 & 137.7 & 12.31 & 0.4207 & 0.8980 & 1014 & 1087 + 45 & 666.5 & 70.18 & 155.6 & 6.040 & 0.2783 & 0.6197&743.0 & 828.8 + 55 & 483.1 & 78.52 & 189.6 & 3.240 & 0.2025 & 0.4856&565.1 & 676.4 + 65 & 360.0 & 94.60 & 277.3 & 1.849 & 0.1557 & 0.4332&456.6 & 639.6 + @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + 10 & 284.4 & 41.90 & 71.08 & 1396 & 187.6 & 294.9 & 1910 & 2046 + 15 & 910.1 & 185.6 & 321.5 & 1396 & 237.4 & 378.9 & 2729 & 3007 + 20 & 991.5 & 290.5 & 526.8 & 770.7 & 175.8 & 289.9 & 2229 & 2579 + 25 & 874.7 & 378.8 & 748.8 & 411.1 & 130.9 & 229.0 & 1796 & 2264 + 30 & 728.5 & 489.6 & 1172 & 229.0 & 105.5 & 208.4 & 1553 & 2338 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 601 & 42.9 & 87.6 & 139 & 7.04 & 14.8 & 790 & 843 + 45&416&48.3 & 104 & 57.1 & 4.46 & 9.82 & 526 & 586 + 55&299&56.5 & 132 & 27.5 & 3.25 & 7.76 & 386 & 467 + 65&222&71.3 & 203 & 14.6 & 2.67 & 7.57 & 311 & 448 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + 10 & 2580 & 353.0 & 554.1 & 190.1 & 31.47 & 49.95 & 3155 & 3374 + 15 & 6054 & 1054 & 1681 & 279.2 & 45.01 & 73.60 & 7433 & 8088 + 20&5900 & 1433 & 2364 & 173.8 & 34.97 & 60.14 & 7541 & 8497 + 25&4884 & 1715 & 3015 & 96.59 & 26.01 & 48.63 & 6722 & 8044 + 30&3898 & 2077 & 4183 & 52.98 & 19.78 & 43.40 & 6048 & 8178 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 3115 & 177.6 & 375.1 & 35.25 & 1.458 & 3.135 & 3330 & 3529 + 45 & 2049 & 188.9 & 417.8 & 14.72 & 0.9238 & 2.094 & 2254 & 2484 + 55 & 1426 & 210.8 & 508.0 & 7.008 & 0.6582 & 1.632 & 1644 & 1942 + 65 & 1029 & 254.5 & 742.4 & 3.470 & 0.4856 & 1.439 & 1288 & 1777 + 35 & 3115 & 177.6 & 375.1 & 35.25 & 1.458 & 3.135 & 3330 & 3529 + + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 100v / cm & 200v / cm & & 100v / cm & 200v / cm & 100v / cm & 200v / cm + 10 & 3679 & 506.3 & 787.9 & 750.7 & 68.35 & 108.5 & 5004 & 5326 + 15 & 6226 & 1052 & 1666 & 631.2 & 58.47 & 94.75 & 7968 & 8618 + 20 & 5547 & 1286 & 2106 & 368.8 & 40.12 & 67.39 & 7242 & 8090 + 25 & 4405 & 1458 & 2549 & 211.4 & 28.43 & 50.80 & 6103 & 7216 + 30 & 3434 & 1709 & 3438 & 124.9 & 21.11 & 42.54 & 5288 & 7039 + & & & & & & & & + & @xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 35 & 2702 & 143.5 & 302.9 & 81.66 & 1.594 & 3.390 & 2929 & 3090 + 45 & 1733 & 148.1 & 327.0 & 36.94 & 0.9921 & 2.202 & 1919 & 2099 + 55 & 1185 & 162.0 & 389.4 & 19.08 & 0.6971 & 1.667 & 1367 & 1595 + 65 & 844.1 & 192.7 & 562.0 & 10.72 & 0.5188 & 1.438 & 1048 & 1418 + @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + & & & & & & & & + 10 & @xmath145 & @xmath146 & @xmath147 & @xmath148 & @xmath149 & @xmath150 & @xmath151 & @xmath152 + 20 & 0.2826 & 0.04174 & 0.08730 & 0.005807 & @xmath153 & 0.001622 & 0.3309 & 0.3773 + 30 & 6.541 & 1.108 & 2.343 & 0.04932 & 0.006986 & 0.01519&7.706 & 8.949 + 40 & 12.61 & 2.740 & 5.967 & 0.04750 & 0.008213 & 0.01840&15.41 & 18.64 + 50 & 13.91 & 4.201 & 9.678 & 0.03120 & 0.007135 & 0.01687&18.15 & 23.64 + 60 & 12.97 & 5.813 & 14.96 & 0.01917 & 0.006156 & 0.01615&18.81 & 27.97 + 65 & 12.21 & 6.863 & 19.53 & 0.01502 & 0.005856 & 0.01688&19.10 & 31.78 + + & & & & & & & & + 10 & @xmath154 & @xmath155 & @xmath156 & @xmath157 & @xmath158 & @xmath159 & @xmath160 & @xmath161 + 20 & 0.489 & 0.0636 & 0.136 & 0.00443 & @xmath162 & 0.00132 & 0.558 & 0.631 + 30 & 7.21 & 1.02 & 2.22 & 0.0412 & 0.00656 & 0.0139&8.28 & 9.49 + 40 & 11.8 & 2.08 & 4.65 & 0.0447 & 0.00887 & 0.0193 & 14.0 & 16.5 + 50 & 11.9 & 2.83 & 6.70 & 0.0321 & 0.00854 & 0.0195&14.8 & 18.7 + 60&10.5 & 3.60 & 9.49 & 0.0212 & 0.00797 & 0.0201&14.1 & 20.0 + 65 & 9.65 & 4.09 & 11.9 & 0.0171 & 0.00785 & 0.0217&13.8 & 21.6 + + & & & & & & & & + 10 & @xmath163 & @xmath164 & @xmath165 & @xmath166 & @xmath167 & @xmath168 & @xmath169 & @xmath170 + 20 & 1.66 & 0.227 & 0.475 & 0.0149 & 0.00207 & 0.00442 & 1.90 & 2.15 + 30 & 15.7 & 2.46 & 5.19 & 0.0710 & 0.0110&0.0237 & 18.2 & 21.0 + 40 & 24.7 & 4.99 & 10.8 & 0.0662 & 0.0127&0.0280 & 29.7 & 35.6 + 50 & 25.4 & 7.15 & 16.5 & 0.0450 & 0.0113 & 0.0264 & 32.6 & 41.9 + 60 & 22.9 & 9.60 & 24.9 & 0.0287 & 0.0101&0.0260 & 32.5 & 47.8 + 65 & 21.3 & 11.3 & 32.5 & 0.0229 & 0.00969&0.0277 & 32.6 & 53.9 + @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 10 & 0.604 & 0.0195 & 0.0390 & 0.408 & 0.0113 & 0.0228&1.04&1.07 + 20 & 183 & 7.58 & 15.3 & 20.8 & 0.633 & 1.30 & 212 & 220 + 30&235&14.8 & 30.3 & 9.25 & 0.390 & 0.819 & 260 & 276 + 40&189&18.5 & 39.2 & 3.70 & 0.227 & 0.492 & 211 & 232 + 50&143&21.9 & 49.1 & 1.67 & 0.150 & 0.344 & 167 & 194 + 60&109&26.7 & 67.0 & 0.842 & 0.113 & 0.288 & 137 & 177 + 65&96.1&30.3 & 83.9 & 0.619 & 0.103 & 0.287 & 127 & 181 + + 10 & 3.69 & 0.102 & 0.206 & 0.0598 & 0.00432 & 0.00873 & 3.85 & 3.96 + 20 & 300 & 9.25 & 19.0 & 8.14 & 0.311 & 0.626 & 317 & 327 + 30 & 294 & 13.0 & 27.3 & 4.87 & 0.255 & 0.520 & 312 & 327 + 40&207 & 13.8 & 30.0 & 2.20 & 0.173 & 0.363 & 223 & 240 + 50&144&14.7 & 33.8 & 1.07 & 0.126 & 0.278 & 160 & 179 + 60&103&16.5 & 42.3 & 0.568 & 0.103 & 0.246 & 120 & 146 + 65&87.8&18.0 & 51.0 & 0.426 & 0.0965 & 0.251 & 106 & 140 + + 10 & 27.8 & 0.782 & 1.58 & 2.73 & 0.0532 & 0.108 & 31.4 & 32.2 + 20 & 540 & 18.7 & 37.6 & 21.4 & 0.626 & 1.28 & 581 & 600 + 30&517&27.3 & 55.9 & 10.4 & 0.439 & 0.908 & 555 & 584 + 40&381&31.6 & 66.9 & 4.60 & 0.282 & 0.600 & 418 & 453 + 50&277&36.2 & 81.2 & 2.25 & 0.199 & 0.444 & 315 & 360 + 60&205&43.5 & 110 & 1.20 & 0.157 & 0.385 & 250 & 316 + 65&179&49.1 & 138 & 0.904 & 0.146 & 0.391 & 229 & 318 + @|c|c|c|c|c|c|c|c|c| + & & & & & & & & + @xmath4&@xmath110&@xmath71&@xmath71&@xmath112&@xmath18&@xmath18&@xmath114&@xmath114 + & direct & 5v / cm & 10v / cm & & 5v / cm & 10v / cm & 5v / cm & 10v / cm + 10 & 81.20 & 1.429 & 2.838 & 177.4 & 2.395 & 4.806 & 262.4&266.2 + 20 & 941.6 & 26.22 & 52.51 & 259.5 & 4.569 & 9.340 & 1232 & 1263 + 30 & 734.1 & 35.55 & 72.58 & 68.99 & 1.925 & 4.022 & 840.6 & 879.7 + 40 & 500.8 & 40.35 & 85.19 & 23.30 & 1.009 & 2.181 & 565.4 & 611.4 + 50 & 350.5 & 45.92 & 102.7 & 9.740&0.6394 & 1.458 & 406.8&464.4 + 60 & 255.3 & 54.93 & 137.1 & 4.738&0.4709 & 1.188 & 315.4&398.3 + 65 & 220.8 & 61.92 & 170.6 & 3.445&0.4253 & 1.176 & 286.6&396.0 + 10 & 373.9 & 5.011 & 10.06 & 20.19 & 0.6466 & 1.299 & 399.7 & 405.5 + 20 & 1690 & 30.95 & 63.33 & 100.7 & 2.073&4.156 & 1824&1858 + 30 & 1009 & 30.95 & 64.80 & 35.35 & 1.170&2.375 & 1077&1112 + 40 & 592.4 & 30.00 & 64.98&13.44 & 0.7148&1.488 & 636.5&672.3 + 50 & 374.1 & 30.71 & 70.43 & 6.029 & 0.4986 & 1.086 & 411.3 & 451.6 + 60 & 252.3 & 33.85 & 86.50&3.076 & 0.3940&0.9362 & 289.6 & 342.8 + 65 & 211.3 & 36.78 & 103.7&2.279 & 0.3675&0.9453 & 250.7&318.2 + + 10 & 1263 & 17.45 & 34.98 & 279.7 & 2.919 & 5.900&1563 & 1584 + 20 & 2696 & 57.56 & 115.5 & 220.0 & 3.948 & 8.007&2978 & 3040 + 30 & 1681 & 63.97 & 130.5&71.48&2.059 & 4.235&1819 & 1887 + 40 & 1051 & 68.29 & 144.0 & 27.59 & 1.211 & 2.556&1148 & 1225 + 50 & 699.6 & 75.56 & 168.9&12.59 & 0.8201 & 1.814 & 788.6 & 883.0 + 60 & 492.7 & 89.13 & 223.8 & 6.510 & 0.6320 & 1.534 & 589.0 & 724.5 + 65 & 420.5 & 100.1 & 280.1 & 4.847 & 0.5827 & 1.544&526.1 & 706.9 + r. f. stebbings , c. j. latimer , w. p. west , f. b. dunning , and t. b. cook , phys . a * 12 * , 1453 ( 1975 ) ; t. ducas , m. g. littman , r. r. freeman , and d. kleppner , phys . lett . * 35 * , 366 ( 1975 ) ; t. f. gallagher , l. m. humphrey , r. m. hill , and s. a. edelstein , phys . rev . lett . * 37 * , 1465 ( 1976 ) .

this paper considers the distributed sampled - data control problem of a group of mobile robots connected via distance - induced proximity networks . a dwell time is assumed in order to avoid chattering in the neighbor relations that may be caused by abrupt changes of positions when updating information from neighbors . distributed sampled - data control laws are designed based on nearest neighbor rules , which in conjunction with continuous - time dynamics results in hybrid closed - loop systems . for uniformly and independently initial states , a sufficient condition is provided to guarantee synchronization for the system without leaders . in order to steer all robots to move with the desired orientation and speed , we then introduce a number of leaders into the system , and quantitatively establish the proportion of leaders needed to track either constant or time - varying signals . all these conditions depend only on the neighborhood radius , the maximum initial moving speed and the dwell time , without assuming _ a prior _ properties of the neighbor graphs as are used in most of the existing literature . , , , , distributed control , unicycle , synchronization , sampled - data , hybrid system , leader - follower model

1609.02174

cooperative control of multi - robot / agent systems ( mrs / mas ) has generated wide interest for researchers in control and robotics communities . compared with a single robot , multiple robots can cooperatively accomplish complicated tasks with the advantages of high efficiency and robustness to the link failures . over the last decade , mrs have wide applications in implementing a large number of tasks ranging from coverage , deployment , rescue , to surveillance and reconnaissance . among these tasks , a basic one is to reach synchronization , i.e. , all robots reach the same state , which actually has close connection with many important engineering applications , such as rendezvous problem @xcite , agreement problem @xcite , distributed optimization @xcite and formation control @xcite . recently , the synchronization problem of mas has been extensively studied in the literature where the neighbor relations are typically modeled as graphs or networks . for example , and , respectively , studied the first - order discrete - time mas with undirected graphs and directed graphs . studied the mas with first - order continuous - time dynamics . the nonholonomic unicycle mrs are investigated by and . mas with nonlinear dynamics , time delays , and measurement noises are also considered @xcite . in almost all existing results , the neighbor graphs are required to satisfy certain connectivity assumptions for synchronization . how to verify or guarantee such conditions has been a challenging issue . in order to maintain connectivity of dynamical communication graphs , potential function methods are commonly used when designing the distributed control laws @xcite . for a real world mrs , it is more practical that the dynamics of the system are modeled in a continuous - time manner whereas the control laws are designed based on the sampled - data information . the sampled - data technique is of interest in many situations , such as unreliable information channels , limited bandwidth , transport delay . the synchronization of mas with sampled - data control laws has been studied @xcite , where the neighbor graphs are also required to satisfy certain connectivity assumptions . it is clear that the potential function techniques are not applicable for the analysis of mas with continuous - time dynamics and sampled - data control , because connectivity of the networks might be lost between sampling instants . how to analyze the synchronization behavior of such kind of systems becomes more challenging . in this paper , we first present a distributed sampled - data algorithm for a group of nonholonomic unicycle robots with continuous - time dynamics , and provide a comprehensive analysis for the synchronization of the closed - loop hybrid system . in our model , each robot has limited sensing and communication range , and the neighbor relations are described by proximity networks . a dwell time is assumed when updating information from neighbors , implying that the control signals are kept constant between the sampled instants and only updated at discrete - time instants . with such sampled - data information , our design of distributed control laws based on nearest - neighbor rules will clearly result in a hybrid closed - loop system , which is different from the case of discrete - time mas studied by and , and is also different from the previous results given by where the control law for the rotational speed is designed using the continuous - time information . for a multi - agent system , we may design a distributed algorithm to guarantee synchronization of the system , but the synchronization state is inherently determined by the initial states and model parameters . in many practical applications , we expect that the system achieves a desired synchronization state and we can treat that state as a reference signal . the agents that have access to the reference signal are referred to as leaders , and they can help steer the mrs to the desired state . although a large number of theoretical analysis and results for the leader - follower model have been provided , further theoretical investigation is still needed due to some limitations in the existing theory : i ) similar to the leaderless case , the neighbor graphs are required to be connected or contain spanning trees to guarantee that the followers track the reference signal @xcite , but there are few results to address how to verify such conditions . ii ) in order to guide all agents to accomplish complicated tasks , such as tracking time - varying signals and the containment control problem , a number of ( not only one ) leaders need to be introduced into the system @xcite . however , quantitative theoretical results for the number of leaders needed are still lack . hence , this paper considers also a multi - unicycle system with multiple leaders and presents some new quantitative results . the sampled - data information is used to design the control laws for the followers and leaders . for the case of the constant reference signal , we analyze the mrs with heterogeneous agents where the leaders and followers have different dynamics since the reference signal is only obtained by the leaders , and quantitatively provide the proportion of leaders needed to track the reference signal . in addition , we investigate the case where the reference signal is dynamic but piecewise constant , and provide quantitative results for the proportion of leaders needed to track a slowly time - varying signal by analyzing the hybrid dynamics at each stage . the main contributions of this paper are summarized into the following three aspects . ( i ) for the leaderless case , we establish a sufficient condition , imposed on the neighborhood radius , the dwell time and the maximum moving speed , to guarantee synchronization of the nonholonomic unicycles , which overcomes the difficulty of requiring _ a prior _ connectivity assumption on neighbor graphs used in most of the existing results . ( ii ) for the leader - follower model , we provide the proportion of leaders needed to guide all robots to track a reference signal which can be constant or slowly time - varying . these quantitative results illustrate that adding leaders is a feasible approach to guide mrs to accomplish some complicated tasks . ( iii ) for both the leaderless case and leader - following case , we provide comprehensive analysis for nonlinear hybrid closed - loop systems . different from the work of and , we need to estimate the synchronization rate of the continuous - time variables ( i.e. , speed and orientation ) . here the speed and orientation are determined by the corresponding values at sampling time instants and they are updated according to the states of relevant neighbors , and the neighbors are defined via the positions of all robots . hence , the positions , orientations and moving speeds of all robots are coupled . we deal with the coupled relationships by combining the dynamical trajectories of the robots at discrete - time instants with the analysis of continuous - time dynamics in sampling intervals . the rest of this paper is organized as follows . in section [ leaderless ] , we present the problem formulation for a leaderless model and provide the main result for synchronization . in section [ leader ] , we first investigate the leader - following problem where the leaders have constant reference signal , and quantitatively provide the ratio of the number of leaders to the number of followers needed to track the signal . we then extend our result to the dynamical tracking where the leaders have time - varying reference signal , and present some simulations to illustrate our theoretical results . concluding remarks are presented in section [ concluding ] . notations : for a vector @xmath0 , @xmath1 denotes the transpose of @xmath2 , and @xmath3 denotes the 2-norm , i.e. , @xmath4 . for a square matrix @xmath5 , @xmath6 denotes the 2-norm of @xmath7 , i.e. , @xmath8 . for any two positive sequences @xmath9 and @xmath10 , @xmath11 means that there exists a positive constant @xmath12 independent of @xmath13 , such that @xmath14 for any @xmath15 ; @xmath16 ( or ( @xmath17 ) ) means that @xmath18 ; @xmath19 , if there exist two positive constants @xmath20 and @xmath21 , such that @xmath22 . consider a group of @xmath13 unicycle robots ( or agents ) moving in a plane . for a robot @xmath23 @xmath24 , the position of its center at time @xmath25 @xmath26 is denoted by @xmath27 . the orientation and moving speed of each robot are affected by the states of its local neighbors . the pair of two robots is said to be neighbors if their euclidean distance is less than a pre - defined radius @xmath28 . we use @xmath29 to denote the set of the robot @xmath23 s neighbors at time @xmath25 , i.e. , [ 10 ] _ i(t)=\{j : _ ij(t ) < r_n},where @xmath30 is the euclidean distance between robots @xmath23 and @xmath31 . the cardinality of the set @xmath29 , i.e. , the degree of the agent @xmath23 , is denoted as @xmath32 . when the robots move in the plane , the neighbor relations dynamically change over time . we use graph @xmath33 to describe the relationship between neighbors at time @xmath25 , where the vertex set @xmath34 is composed of all robots , and the edge set is defined as @xmath35 . the neighbor graphs are distance - induced , and also called geometric graphs or proximity networks . let @xmath36 and @xmath37 denote the moving orientation and translational speed of the @xmath23th robot at time @xmath25 . the dynamics of the robots with nonholonomic constraint for pure rolling and nonslipping is described by the following differential equations ( for @xmath38 ) , [ 1.3 ] \ { ll _ i(t)=v_i(t)_i(t ) , & + _ i(t)=v_i(t)_i(t ) , & + _ i(t)=_i(t ) , & + _ i(t)=u_i(t ) , & . where @xmath39 and @xmath40 denote the acceleration and rotational speed of the robot @xmath23 at time @xmath25 . for robot @xmath23 , what we can control is its rotational speed @xmath40 and its acceleration @xmath39 , which is an extension from the standard unicycle model where one controls the translational speed directly . we however need to point out that with this simplified extension physical forces affecting the angular motion , such as side - slip forces and friction forces , are ignored . for the feasibility of information processing by the robots , we assume that the robots can only receive information and design the control law at discrete - time instants @xmath41 . to simplify the analysis , we suppose that the dwell time is the same and denoted by @xmath42 , i.e. , @xmath43 . at discrete - time instant @xmath44 , each agent is assumed to sense the relative speed and the relative orientation of its neighbors . that is , for robot @xmath23 , it receives the following sampled - data information at time @xmath44 , @xmath45 from , we see that the neighbor set @xmath29 is determined by the positions of all agents , so it is a continuous - time variable . the neighbor relations might change abruptly when all robots are in motion . the introduction of the dwell time avoids introducing chattering in the abrupt changes caused by the evolution of positions . for an agent , the relative positions of its neighbors can be measured by e.g. , the geolocation and positioning technologies . using the position information and the orientation information of the agents , the relative speed at the sampling instants can be calculated . moreover , the relative speed can also be estimated through observer - based methods by introducing reference robots , which is a different framework and falls into our future research . for the sake of simplicity , we assume that each agent can receive the relative speed of its neighbors in this paper . the objective of this section is to design the distributed control law for the nonholonomic multi - robot system based on the sampled - data information , such that the closed - loop system becomes synchronized in both orientation and moving speed . here by synchronization we mean that there exists a common vector @xmath46 , such that for all @xmath38 , we have @xmath47 and @xmath48 for robot @xmath23 , we design the distributed control law according to the widely used nearest - neighbor rule for @xmath49 ( @xmath50),[1.7 ] \ { ll u_i(t)=_j_i(t_k)(v_j(t_k)-v_i(t_k ) ) , & + _ i(t)=_j_i(t_k)(_j(t_k)-_i(t_k ) ) , & .where @xmath51 is the degree of robot @xmath23 at discrete - time @xmath44 . substituting into , we obtain the following hybrid closed - loop dynamical system : & & [ 2.3 ] \ { ll _ i(t)=v_i(t)_i(t ) , & + _ i(t)=v_i(t)_i(t ) , & + _ i(t)=_j_i(t_k)(_j(t_k)-_i(t_k ) ) , & + _ i(t)= _ j_i(t_k)(v_j(t_k)-v_i(t_k ) ) . & thus , we have for @xmath52 $ ] , [ 1.4 ] & & _ i(t)=(1-)_i(t_k)+_j_i(t_k)_j(t_k ) , + & & [ 2.4.1]v_i(t)=(1-)v_i(t_k)+_j_i(t_k)v_j(t_k).in particular , at discrete - time instant @xmath53 , the orientation and moving speed evolve according to the following equations : & & [ 4.5 ] _ i(t_k+1)=_j_i(t_k)_j(t_k ) , + & & [ 4.6 ] v_i(t_k+1)=_j_i(t_k)v_j(t_k).in order to investigate the synchronization behavior of the hybrid closed - loop system , we need to analyze the discrete - time system at sampling time instants and the continuous - time system in the sampling intervals simultaneously . for the discrete - time system and , the algebraic properties of neighbor graphs play a key role . denote the laplacian matrix of the graph @xmath54 as @xmath55 . the normalized laplacian matrix is defined as @xmath56 , where the degree matrix is defined as @xmath57 . the matrix @xmath58 is non - negative definite , and 0 is one of the eigenvalues . thus , we can arrange the eigenvalues of @xmath58 according to such a non - decreasing manner @xmath59 . the spectral gap of graph @xmath54 is defined as @xmath60 1 ) the positions , orientations and speeds of all robots at the initial time instant are mutually independent ; 2 ) for all robots , the initial positions are uniformly and independently distributed ( u.i.d . ) in the ( normalized ) unit square @xmath61 ^ 2 $ ] ; the initial headings are u.i.d . in @xmath62 ; the initial speeds are u.i.d . in the interval @xmath63 $ ] . [ initialremarl]under assumption [ initial ] , the initial neighbor graph is called a random geometric graph ( rgg ) , whose properties are well investigated by . however , from , the independency between positions of all agents will be destroyed for @xmath64 . thus , the properties concerning the connectivity of static rgg can not be used . denote the sample spaces of the position , orientation and speed as @xmath65 ^ 2 $ ] , @xmath66 $ ] and @xmath67 $ ] . by assumption [ initial ] , our synchronization problem is discussed on the sample space @xmath68 , where @xmath69 denotes the cartesian product . divide the unit square @xmath61 ^ 2 $ ] into @xmath70 equally small squares labeled as @xmath71 , where @xmath72 with @xmath73 satisfying @xmath74 . denote @xmath75 @xmath76 as the number of agents in the corresponding small square . introduce the sets & & [ square]b_n=\{:_1jm_nn_j = na_n^2(1+o(1 ) ) } , + & & [ theta0]_n=\ { : _ iv|_j_i(0)_j(0)|=o(f_n ) } , + & & [ v ] _ n=\ { : _ iv|_j_i(0)(v_j(0)-)|=o(v_n f_n ) } , where @xmath77 . using lemma 7 given by , the event @xmath78 happens with the probability @xmath79 for large @xmath13 , where @xmath80 denotes the probability of an event . by assumption [ initial ] and the methods used in lemma 9 given by , for large @xmath13 , @xmath81 and @xmath82 hold true . furthermore , by the independency of the orientations and moving speeds for the given agents , it is clear that for large @xmath13 , @xmath83 . in the following , we will investigate the dynamical behavior of all agents on the set @xmath84 , and we omit the words almost surely " ( a.s . ) for simplicity . under the condition that the neighborhood radius satisfies @xmath85{\log n / n}\ll r_n\ll 1 $ ] , the maximum and minimum initial degrees satisfy the following equalities [ degree ] d_=nr_n^2(1+o(1 ) ) ; d_min(0)=(1+o(1)).in fact , for the case where the neighborhood radius independent of @xmath13 , the degrees of the agents can also be estimated by similar methods . the detailed analysis for the estimation of the initial degrees can be found in theorem 2 given by for details . we rewrite the orientation and speed update equations and into the following matrix form for @xmath86 $ ] , & & [ 10.1](t)=(1-)(t_k ) + p(t_k)(t_k ) , + & & [ 10.2]v(t)=(1-)v(t_k ) + p(t_k)v(t_k ) . correspondingly , the equations and are rewritten into the following form , & & [ 10.3](t_k+1)=p(t_k)(t_k ) , + & & [ 10.4]v(t_k+1)=p(t_k)v(t_k),where the average matrix @xmath87 is defined as : @xmath88 if @xmath89 , and @xmath90 otherwise . a known result for synchronization of the system and can be stated as follows . if the neighbor graphs @xmath91 are connected , then all agents move with the same orientation and with the same speed eventually ( cf . , ) . the synchronization condition of the system and is imposed on the dynamical properties of neighbor graphs . however , for the system under consideration , the neighbor graphs are defined via the positions of all agents . a comprehensive analysis for the system should include how the change of the positions of all robots affects the dynamical properties of neighbor graphs , which brings challenges for our investigation . intuitively , for uniformly distributed agents , if the two agents are not neighbors at the initial time instant , then they are still not neighbors with high probability as the system evolves . that is , if the initial neighbor graph is disconnected , then it is hard to obtain the connectivity of graph @xmath92 for @xmath93 . in order to reach synchronization , the connectivity of the initial neighbor graphs is needed . when the population size of the agents increases , the number of neighbors of each agent will also increase . thus , the interaction radius can be allowed to decay with the number of agents to guarantee connectivity of the initial neighbor graph . the properties of such graphs have been widely investigated in the fields of wireless sensor networks and random geometric graphs ( cf . . moreover , the change of neighbor graphs at the time instants @xmath94 is positively correlated with the moving speed and the dwell time . hence , the connectivity of the dynamical neighbor graphs at all discrete - time instants may be preserved by assuming that the moving speed and the dwell time are suitably small . we now establish synchronization conditions for the hybrid system depending only on the neighborhood radius , moving speed and the dwell time . the states of the agents including positions , orientations , and speeds of the system are continuous - time variables . in order to investigate the synchronization behavior of the system , we need to estimate the synchronization rate of the continuous - time variables @xmath36 and @xmath37 . moreover , by and @xmath36 and @xmath37 are affected by the discrete - time variables @xmath95 and @xmath96 that are updated by the orientations and speeds of relevant neighbors at @xmath97 . while the neighbors are defined via the positions of all robots . the positions , orientations and moving speeds of all robots are coupled , which will be handled by resorting to the mathematical induction and analyzing the dynamics of the system . thus , the comprehensive analysis combines the dynamical trajectories of all agents at discrete sampling time instants with the continuous time dynamics in sampling intervals . the main theorem is stated as follows . [ propo ] if the neighborhood radius @xmath28 , the maximum initial speed @xmath98 , and the dwell time @xmath42 satisfy the conditions @xmath99{\frac{\log n}{n}}\ll r_n \ll 1 $ ] and @xmath100 , where @xmath101 is a positive constant satisfying @xmath102 and @xmath103 is taken as @xmath104 with @xmath105 satisfying @xmath106 . under assumption [ initial ] , the system reaches synchronization in orientation and speed almost surely for large @xmath13 . * proof : * it is clear that under the condition for the neighborhood radius , the initial neighbor graph is connected with probability one ( cf . , ) . if the connectivity of the neighbor graphs can be preserved , then synchronization can be reached . hence , a key step is to show that the distance between any two robots @xmath23 and @xmath31 at any sampling instant @xmath44 ( @xmath107 ) satisfies the following inequality , [ 1.5 ] & & |_ij(t_k)-_ij(0)|_n r_n . we use mathematical induction to prove . it is obvious that the inequality holds for @xmath108 . we assume that holds for all @xmath109 , and prove that it is true for @xmath110 . by the position update equation in , it is clear that the the distance between any two robots @xmath23 and @xmath31 at contiguous discrete - time instants satisfies [ distancerecur ] & & |_ij(t_k+1)-_ij(t_k)|2_t_k^t_k+1v(t)dt + & & 1.6 cm + 2_1in|v_i(t_k)|_t_k^t_k+1(t)dt , k1,where @xmath111 and @xmath112 denote the dissimilarity of the orientation and the speed between the two agents . in particular , the distance at time @xmath113 satisfies [ initialdistance ] |_ij(t_1)-_ij(0)| & & 2(_0^t_1v(t ) dt+2_1inv_i(0)_n ) + & & 8v_n_n.the proof of is put in [ prooflemma ] , and the inequality can be obtained by following the proof of . by , we see that the analysis of the distance depends on the convergence properties of @xmath114 and @xmath115 . we first estimate the convergence rate of @xmath116 . by the induction assumption , we have for @xmath117[average ] p(t_l)-p(0)80_n(1+o(1)),whose proof is presented in [ prooflemma ] . using and lemma 2 given by , we have for @xmath118 [ deltav ] & & v(t_l+1 ) + & & ( ( 0)+_1slp(s)-p(0))^lv(t_1 ) + & & 2(1-(1+o(1))+160_n)^lv(t_1 ) ( 1+o(1 ) ) + & & 2(1-(1+o(1)))^lv(t_1)(1+o(1 ) ) + & & 2(_n)^lv(t_1)(1+o(1 ) ) , where @xmath119 by using , and @xmath120 is the spectral gap of the initial neighbor graph @xmath121 . by lemma 16 given by , @xmath120 can be estimated as @xmath122 . hence , by the speed update equation , we have for @xmath118 [ deltav2 ] & & _ t_l^t_l+1v(t)dt_t_l^t_l+1dt 2(_n)^l-1 + & & v(t_1 ) _ t_l^t_l+1(1-)dt(1+o(1 ) ) + & & = ( 1+_n)(_n)^l-1v(t_1)_n(1+o(1)).on the other hand , using and , we have @xmath123 @xmath124 . thus , we obtain that for @xmath125 & & _ t_l^t_l+1v(t)dt2_n_iv v_i(t_l)_n v_n(1+o(1)).set @xmath126 then we have @xmath127 @xmath128 thus , [ 2.0 ] & & _ l=1^k_0_t_l^t_l+1v(t)dt_l=1^l_0 - 1_nv(t_l)dt + & & + _ l = l_0^k_0(1+_n)(_n)^l-1v(t_1)_n(1+o(1 ) ) + & & v_n_n(l_0+_l = l_0^k_0(_n)^l - l_0)(1+o(1 ) ) + & & v_n_n(+2+)(1+o(1 ) ) + & & ( + 3 ) + & = & [ 1.2],where the inequality @xmath129 for @xmath130 is used in . by a similar analysis as that of , we have for @xmath131 [ 3.1 ] & & _ t_l^t_l+1(t)dt_t_l^t_l+1dt2(_n)^l-1 + & & ( t_1)_t_l^t_l+1 ( 1-)dt(1+o(1 ) ) + & & = ( 1+_n)(_n)^l-1(t_1)_n(1+o(1)).meanwhile , we have for @xmath125 [ 3.1.1 ] & & _ t_l^t_l+1(t ) dt2_n_iv|_i(t_1)|(1+o(1)).by and , we have @xmath132 . similar to the analysis of , it is clear that [ 3.0]&&_l=1^k_0 _ 1inv_i(t_l)_t_l^t_l+1(t)dt + & & _ 1in|_i(t_1)|(1+o(1 ) ) + & = & o().substituting and into , we can derive that the distance between agents @xmath23 and @xmath31 at time @xmath133 satisfies [ 5.2 ] & & |_ij(t_k_0 + 1)-_ij(0)|_l=0^k_0|_ij(t_l+1)-_l(0)| + & & |_ij(t_1)-_ij(0)|+ 2_l=1^k_0_t_l^t_l+1v(t)dt + & & + 2_l=1^k_0_1in|v_i(t_l)|_t_l^t_l+1(t)dt . + & & _ nr_n , where the conditions on the speed and the dwell time are used in the last inequality . using the mathematical induction , we see that the inequality holds for all @xmath134 . as a consequence , we have for @xmath107,p(t_k)-p(0)=80_n(1+o(1)).thus , the inequality holds for all @xmath135 , i.e. , [ 1.1.3 ] & & v(t)v(t_k ) + & & v_n(1-)^k-1(1+o(1 ) ) + & & 0 , 1.5 cm k. by , it is clear that @xmath136 ( resp . @xmath96 ) are non - increasing ( resp . non - decreasing ) sequences . thus , both the sequence @xmath138 and @xmath139 have bounded limits as @xmath140 . moreover , by , @xmath138 and @xmath139 have the same limit , and the translational speed @xmath37 tends to the same value for all @xmath141 . by a similar analysis , we can prove that the orientations of all agents tend to the same value as @xmath142 . @xmath143 generally speaking , the neighborhood radius has some physical meanings in practical systems , e.g. , reflecting the sensing ability of sensors . the smaller the neighborhood radius is , the less the energy consumption will be . from practical point of view , the neighborhood radius should be as small as possible . however , for the system under consideration , the larger the neighborhood radius , the easier the system reaches synchronization . the neighborhood radius for synchronization in theorem [ propo ] describes a tradeoff between these two factors . theorem [ propo ] establishes scaling rates for the neighborhood radius and moving speed for synchronization under assumption [ initial ] . the conditions on these parameters can be adjusted according to the practical demands . assume that the agents are u. i. d. in the square @xmath144 ^ 2 $ ] . let @xmath145 and @xmath146 . it is clear that @xmath147 is u. i. d. in the unit square @xmath61 ^ 2 $ ] . by , @xmath148 and @xmath149 are updated according to the equations @xmath150 , and @xmath151 , respectively . the neighbor relations can also be rewritten as @xmath152 , where @xmath153 . following the proof line of this paper , similar results for synchronization of the mrs can be obtained just by replacing @xmath98 by @xmath154 and replacing @xmath28 by @xmath155 . by this and theorem [ propo ] , we can easily obtain the following result , let the neighborhood radius @xmath156 and the initial maximum moving speed @xmath157 be two positive constants . if the dwell time satisfies @xmath158 with @xmath159 being a positive constant depending on @xmath157 and @xmath156 , then under assumption [ initial ] the mas reaches synchronization almost surely for large @xmath13 . in section [ leaderless ] , we designed the distributed control law for each robot using the sampled - data information , and provided parameter conditions to guarantee synchronization of all robots . it is clear that the resulting orientation and speed of unicycles are determined by the initial states of all robots and model parameters . for many practical applications , e.g. , avoiding collision with obstacles , following a given path , we may expect to guide all robots towards a desired orientation and speed . to achieve this goal , a cost efficient way is to introduce some special robots that have the reference signal about the desired behavior of the whole system . these special robots with the reference signal are called leaders , and other ordinary robots are called followers . in this section , we study the system composed of heterogeneous agents . the sampled - data control laws of both leaders and followers are designed , and some quantitative results for the proportion of leaders needed to track the constant and time - varying signals are established . we consider the system composed of @xmath13 followers and @xmath160 leaders , where @xmath161)$ ] is the ratio of the number of leaders to the number of followers . we denote the follower set and leader set as @xmath162 and @xmath163 , respectively , and @xmath164 . the dynamics of both leaders and followers is described by . for the followers , they receive the sampled - data information @xmath165 at discrete - time instant @xmath44 , and the closed - loop dynamics is described by , i.e. , for @xmath166 and @xmath167 , & & [ 3.3 ] \ { ll _ i(t)=v_i(t)_i(t ) , & + _ i(t)=v_i(t)_i(t ) , & + _ i(t)=_j_i(t_k)(_j(t_k)-_i(t_k ) ) , & + _ i(t)= _ j_i(t_k)(v_j(t_k)-v_i(t_k ) ) . & different from followers , the leaders can receive the relative information of the desired orientation and the desired speed , in addition to the relative translational speed and relative orientation of their neighbors at discrete - time @xmath44 @xmath168 with @xmath169 . thus , for a leader robot @xmath170 , it has the following information at time instant @xmath44 , @xmath171 , where @xmath172 and @xmath98 are the desired orientation and the desired speed , respectively . we adopt the control law for the rotational speed @xmath40 and the acceleration @xmath39 of leaders with the following form for @xmath49 [ 3.6 ] & & _ i(t)=\{(_0-_i(t_k ) ) + & & + _ j_i(t_k)(_j(t_k)-_i(t_k ) ) } , + [ 3.6.1 ] & & u_i(t)=\{(v_n - v_i(t_k ) ) + & & + _ j_i(t_k)(v_j(t_k)-v_i(t_k ) ) } , where the positive constant @xmath173 reflects the balance between the expected behavior and local interactions with neighbors . substituting and into , we can obtain the closed - loop dynamics for the headings and speeds of leaders for @xmath49 , [ 4.4 ] & & _ i(t)=(1-)_i(t_k ) + & & 1.8cm+\{_0+_j_i(t_k)_j(t_k ) } , + & & [ 2.4]v_i(t)=(1-)v_i(t_k)+ + & & 1.8cm\{v_n+_j_i(t_k)v_j(t_k)}.particularly , for @xmath53 , & & [ 4.2 ] _ i(t_k+1)=_0+_j_i(t_k)_j(t_k ) , + & & [ 3.9 ] v_i(t_k+1)= v_n+_j_i(t_k)v_j(t_k).note that after the leaders are added , the neighbor set of both leaders and followers is composed of two parts : leader neighbors and follower neighbors . for a robot @xmath174 , we use @xmath175 and @xmath176 to denote its follower neighbor set and leader neighbor set at discrete - time instant @xmath44 , respectively . that is , @xmath177 , and @xmath178 . denote the cardinality of the sets @xmath175 and @xmath176 as @xmath179 and @xmath180 , respectively . thus , we have @xmath181 , and @xmath182 . for the leader - follower model , if the union of neighbor graphs in bounded time intervals contains a spanning tree rooted at the leaders , then all robots will move with the desired orientation and with the desired speed eventually . how to guarantee the existence of the spanning tree is unresolved . we aim at establishing the quantitative relationship between the proportion of leaders and the neighborhood radius , moving speed and dwell time such that all agents move with the expected orientation @xmath172 and the expected speed @xmath98 eventually . it is clear that the initial distribution of leaders is crucial for the proportion of leaders needed . for example , assume that the leaders and the followers at the initial instant are distributed in two disjoint areas , and the distance between these two areas are large . at the initial time , the followers are not affected by the reference signals of the leaders since the followers do not have leader neighbors , and they evolve in a self - organized manner . by and , the leaders converge to the desired orientation and speed in a certain rate . when the system evolves , the distance between the subgroups of leaders and followers becomes larger and larger . as a result , the followers can not be guided to the desired behavior no matter what the proportion of leaders is . in this part , we proceed with our analysis under assumption [ initial ] in which the uniform distribution of leaders makes it possible to investigate the proportion of leaders for synchronization . for @xmath38 , we denote @xmath183 and @xmath184 what we concern is the ratio of leaders @xmath185 needed such that for all @xmath38 , we have @xmath186 and @xmath187 as @xmath142 . different from the leaderless case where all agents have the same closed - loop dynamics , the leaders and followers in the leader - following model have different closed - loop dynamics . thus , we need to analyze the synchronization of the system with heterogeneous agents . moreover , the orientation of each follower is affected by the orientations of its neighbors including leader neighbors and follower neighbors , and the orientation of each leader is affected by the orientations of its leader neighbors and follower neighbors . the coupled relationship makes the analysis of the leader - following model more challenging . we first present a preliminary result for the convergence of orientations of leaders and followers.owers . [ delta]if there exist positive constants @xmath7 and @xmath188 , such that @xmath189 , @xmath190 , and for @xmath191 and @xmath38 , @xmath192 , where @xmath193 , then , we have for @xmath93 , @xmath194 , and @xmath195 , where @xmath196 * proof . * we prove the lemma by virtue of the mathematical induction . first , the lemma holds for @xmath197 . we assume that the lemma holds at discrete - time instant @xmath44 with @xmath198 , i.e. , the inequalities @xmath199 and @xmath195 hold . thus , for a follower agent @xmath200 , we have by & & _ iv_1_i(t_k+1)=_iv_1| -_0| + & = & _ iv_1| + & & + | + & & _ iv_1\{(1-_i(t_k))_iv_1_i(t_k)+_i(t_k)_iv_2_i(t_k ) } + & & _ iv_1 \{(1-_i(t_k))^k-1 a+_i(t_k)(1-)^k-1 a } + & & _ iv_1(1-(_i(0)- ) ) ^k-1a^ka.while for a leader agent @xmath201 , using we have & & _ iv_2_i(t_k+1)=(1-)_iv_2| + & & + | ( 1- ) + & & _ iv_2\ { ( 1-_i(t_k))_iv_1_i(t_k)+_i(t_k)_iv_2_i(t ) } + & & ( 1-)^ka.thus , the results of the lemma holds at discrete - time instant @xmath202 . this completes the proof of the lemma . @xmath143 for the moving speed , we have a similar result . [ lambda]if there exist positive constants @xmath203 and @xmath188 , such that @xmath204 , @xmath205 and for @xmath206 , @xmath192 , then for @xmath93 , we have @xmath207 and @xmath208 the proof of lemma [ lambda ] is similar to that of lemma [ delta ] , and we omit the proof details . similar to the analysis in section [ leaderless ] , we divide the unit square @xmath61 ^ 2 $ ] into @xmath70 equally small squares labeled as @xmath71 , with @xmath209 where @xmath73 satisfies @xmath210 . denote @xmath211 and @xmath212 with @xmath213 as the number of followers and leaders in the corresponding small square , respectively . introduce the following sets [ square0 ] & & b^_n=\{:_1jm_nn_j1=na_n^2(1+o(1 ) ) , + & & _ 1jm_nn_j2=_n a_n^2(1+o(1 ) } , + & & [ theta00]^_n=\ { : _ iv|_j_i(0)_j(0)|=o(f_n ) } , + & & [ v0 ] ^_n=\ { : _ iv|_j_i(0)(v_j(0)-)|=o(v_n f_n ) } , where @xmath214 . by multi - array martingale lemma ( lemma 7 given by ) , we have @xmath215 for large @xmath13 if @xmath216 . similar to the synchronization analysis for the leaderless model , and using the independency of the orientation and heading at the initial time instant , we have @xmath217 . all of the following analysis is proceeded on the set @xmath218 without further explanations . based on this , we can give some estimations on the characteristics concerning the initial states , which are presented in [ estiamtion ] . we characterize the change of the follower neighbors and leader neighbors of a robot @xmath219 by the following two sets : [ eq6]&&_i1=\{jv_1 : ( 1-)r_n_ij(0)(1+)r_n } , + [ eq7 ] & & _ i2=\{jv_2 : ( 1-)r_n_ij(0)(1+)r_n},where the positive constant @xmath220 satisfies @xmath221 . we denote the cardinality of the sets @xmath222 and @xmath223 as @xmath224 and @xmath225 , respectively . here we briefly address why a certain number of leaders is needed in order to guarantee that the followers track the reference signal of leaders . suppose that a very small number of leaders are added into the system , for example , only one leader . then the influence of the leader is very weak , resulting in a low tracking rate . while the leaders converge to the desired state with a certain rate . as a consequence , all agents may form two disjoint clusters before their orientations and speeds are synchronized to the desired states : leader cluster and follower cluster , and it is impossible for the followers to track the behavior of leaders . intuitively , the larger the neighborhood radius , the easier the initial neighbor graph has a spanning tree ; the smaller the moving speed and dwell time , the easier the spanning tree is kept during the evolution . for such a situation , the smaller the ratio of the number of leaders will be needed to track the reference signal . in the following theorem , we illustrate this intuition from a theoretical point of view , and establish a quantitative result for the ratio of the number of leaders needed . [ largenumber1]assume that the neighborhood radius satisfies @xmath226 . if the ratio @xmath185 of the number of leaders to the number of followers satisfies one of the following two conditions : 1 . @xmath227 , provided that @xmath228 ; 2 . @xmath229 , provided that @xmath230 or @xmath231 , then all robots move with @xmath172 and @xmath98 eventually . * * lemmas [ delta ] and [ lambda ] show that the estimation of @xmath232 is a key step for the convergence of @xmath233 and @xmath234 . @xmath235 is defined via the number of leader neighbors and the number of follower neighbors of the agent @xmath23 at time @xmath25 . in order to estimate @xmath232 , we show that the distance between any pair of robots @xmath23 and @xmath31 at any discrete - time instant @xmath44 satisfies the following inequality , [ eq8 ] |_ij(t_k)-_ij(t_0)|r_n , k0,where @xmath220 is a positive constant taking the same value as that in and . if holds , then for a robot @xmath23 , the change of its follower neighbors and leader neighbors at time @xmath44 in comparison with those at the initial time is included in the sets @xmath222 and @xmath223 defined by and , respectively . for @xmath38 , we have @xmath236 and @xmath237 by the estimation of the number of follower neighbors and leader neighbors at the initial time instant given in [ estiamtion ] , we have [ 6.4 ] & & _ iv|_i(t_k)-_i(0)| + & = & _ iv|- | + & & + & & 256 _ n.using lemmas [ delta ] and [ lambda ] , as @xmath140 , @xmath238 and @xmath239 . moreover , by and , as @xmath142 , @xmath240 and @xmath241 , which mean that both the followers and leaders will move with the same desired orientation @xmath172 eventually . by a similar analysis , we can prove that all robots move with the same desired speed @xmath98 eventually . now , we use the mathematical induction to prove . it is clear that holds for @xmath108 . we assume that it holds for @xmath242 . by the analysis of , we have @xmath243 . using lemmas [ delta ] and [ lambda ] , for @xmath242 , we have @xmath244 , @xmath245 , and @xmath246 , @xmath247 @xmath248 , where @xmath249 is defined in lemma [ thetaleader ] in [ estiamtion ] , and @xmath250 . for followers , using and , we have for @xmath251 , l=0,1,2,\cdots , k_0 $ ] , & & _ iv_1|_i(t)|(1-)_iv_1|_i(t_l)| + & & + _ iv_1|_j_i(t_l)_j(t_l)| + & & ( 1-)_iv_1|_i(t_l)|+ + & & _ iv_1((1-_i(t_l))_iv_1|_i(t_l)|+_i(t_l)_iv_2|_i(t_l)| + & & ( 1-)(_n)^l-1(|_0|+l_n).similarly , @xmath252 @xmath253 while for leaders , using and , we have @xmath254 , and @xmath255 using lemma [ thetaleader ] in [ estiamtion ] , the following result can be obtained , [ 3.2 ] & & _ l=1^k_0_ivv_i(t_l)_t_l^t_l+1_iv|_i(t)|dtv_n_l=1^k_0_t_l^t_l+1 + & & \{(1-)(_n)^l-1(|_0|+l_n)}dt + & = & _ l=1^k_0v_n_n_n^l-1(|_0|+l_n)(1+o(1 ) ) + & = & ( 1+o(1)).moreover , we have [ 3.2.1 ] & & _ l=1^k_0_t_l^t_l+1_iv & & \{(1- ) ( _ n)^l-1(1+o(1))}dt + & & _ l=1^k_0 + & = & . thus , using and , we obtain [ 5.3 ] & & |_ij(t_k_0 + 1)-_ij(0)| + & & _ l=0^k_0x_i(l+1)-x_j(l+1)-x_i(l)+x_j(l)_2 + & & 4v_n_n+4_l=1^k_0\{_t_l^t_l+1_iv|_i(t)|dt + & & + _ ivv_i(t_l)_t_l^t_l+1_iv|_i(t)|dt } + & & r_n , where the condition on the ratio of the number of leaders is used in the last inequality . this completes the proof of .@xmath256 for some complicated tasks , such as path following , avoiding collision with obstacles , the reference signal of leaders may vary with time . in this part , we consider the dynamic tracking of unicycle robots . in order to present the problem and the result clearly , we consider the case where the desired orientations of leaders may change over time , but the desired speed keeps unchanged . for a leader robot @xmath170 , it has the following information at discrete - time instant @xmath44 , @xmath257 , where @xmath98 is the desired speed and @xmath258 is the desired orientation at time @xmath44 . the notations , including the set of follower neighbors @xmath259 , the set of leader neighbors @xmath260 , @xmath222 and @xmath223 have the same meanings as those in subsection [ problem ] . the closed - loop dynamics of followers is still described by . for the leaders , we adopt the similar distributed control law as that of for @xmath261 , just replacing @xmath262 by @xmath258 . thus , we obtain the closed - loop dynamics of the orientation and speed of leaders for @xmath167 , & & _ i(t)= \{(_k-_i(t_k ) ) + & & + _ j_i(t_k)(_j(t_k)-_i(t_k ) ) } , + & & _ i(t)= \{(v_n - v_i(t_k ) ) + & & + _ j_i(t_k)(v_j(t_k)-v_i(t_k ) ) } , where @xmath263 is a positive constant . for a large crowd , it is apparent both mathematically and intuitively that if the desired orientation of leaders changes too fast , then it is impossible for the followers to track . in this paper , we consider the situation where the desired orientation of the leaders is piecewise constant in such a way that it keeps constant until the followers track the reference signal in the sense that the maximum dissimilarity for the orientations is less than a pre - defined tracking error @xmath264 , as shown in fig . [ fgg1 ] where the time instants @xmath265 depend on the tracking error . although this assumption seems to be rather restrictive , it is nevertheless reasonable for applications such as crowd control by active intervention . denote the difference between the desired orientations at two contiguous time instants as @xmath266 . we assume @xmath267 . since what we concern is the tracking effect of the robots at each stage @xmath268 , l=1 , 2 , \cdots$ ] , we need to analyze the dynamical behavior of the leaders and followers stage by stage , and the ending states of the robots at the latest stage is just the starting states at the current stage . we give the main result for the system with dynamic leaders as follows . [ dynamictrack ] assume that the neighborhood radius satisfies @xmath226 . then for any given tracking error @xmath264 , we have for @xmath269 , @xmath270 and [ or ] _ iv|_i(t_k_l+1)-_l| , l=0 , 1 , 2 , , if the ratio of the number of leaders satisfies one of the following conditions \1 ) @xmath271 provided that @xmath228 . \2 ) @xmath229 provided that @xmath230 or @xmath231 . * we first prove that for any pair of robots @xmath23 and @xmath31 , we have [ 6.6.1]&&|_ij(t_k)-_ij(0)|r_n , k0 . using theorem [ largenumber1 ] , we see that holds for @xmath272 $ ] by taking @xmath273 . for @xmath274 , we have @xmath275 . thus , @xmath276 , and @xmath277 , where @xmath278 . we now analyze the dynamical behavior of all agents for @xmath279 $ ] @xmath280 with @xmath281 we have at the time instant @xmath282 , @xmath283 , and @xmath284 at the time instant @xmath285 , the desired orientation of leaders changes to @xmath286 , and the orientations of the followers and leaders at time @xmath285 satisfy @xmath287 , and @xmath288 , where @xmath289 . assume that holds for @xmath290 , p=2 , 3 , \cdots $ ] . then we have @xmath275 for @xmath291 . using lemmas [ delta ] and [ lambda ] , we have for @xmath292 , p=2,\cdots , p-1 $ ] , @xmath293 , and @xmath294 thus , the distance between agents @xmath23 and @xmath31 satisfies [ 5.3.3 ] & & & & \{|_t_l^t_l+1dt| + & & + |_t_l^t_l+1dt| } + & & 2(1+_n)v_n_n(1+o(1))(_l=0^k_1_n^l-1(|_0|+l_n ) + & & + _ s=1^p-1 _ k = k_s+1^k_s+1_n^l - k_s-1(+d_s-1 ) + & & + _ l = k_p+1^k_0_n^l - k_p-1(+d_p-1 ) ) + & & + _ l=0^k_0 ( 1+_n)v_n_n_n^l-1(1+o(1 ) ) + & & 8v_n_n(1+o(1))(|_0|+_s=0^p-1d_s+1)r_n . by the above analysis , we see that for @xmath295 , p=0 , 1 , \cdots,$ ] the assertion holds , and we have for @xmath296 , @xmath297 moreover , using lemma [ lambda ] , we obtain & & _ iv_1_i(t_k)(_n)^k-10 , k , + & & _ iv_2_i(t_k)(1-)(_n)^k-10 , k.@xmath256 a direct consequence is that the ratio of the number of leaders to the number of followers depends on the times and the amplitude that the desired orientations change during evolution . we illustrate the feasibility of guiding a group of ordinary robots to accomplish a complicated task by introducing dynamic leaders whose desired orientation may change with the requirement of the task . the system is composed of @xmath298 ordinary agents labeled @xmath299 , and we aim to guide these robots to move to the right along the bottom line , go across an oval - shaped obstacle and not to collide with it as shown in fig . the desired orientations can be taken as @xmath300 , @xmath301 , @xmath302 @xmath303 , @xmath304 . to complete such a task , we introduce 3 leaders into the system which are labeled @xmath305 and @xmath306 . the initial states of all agents are taken to satisfy assumption [ initial ] with the maximum initial speed @xmath307 , and the neighborhood radius @xmath308 . the dwell time is taken as @xmath309 . the initial orientations , velocities and positions of the leaders and followers are listed in fig . [ fgg3 ] shows the trajectories of all robots , and we see that the leaders can guide the followers to achieve the pre - defined task . in this paper , we proposed a sampled - data distributed control law for a group of nonholonomic unicycle robots , and established sufficient conditions for synchronization for the leaderless case without relying on dynamical properties of the neighbor graphs . in order to steer the system to a desired state , we then introduced leaders with constant or time - varying signals into the system , and provided the proportion of the leaders needed to track the static or dynamic signals . in our model , the robots are connected via distance - induced graphs , and a dwell time is assumed for the feasibility of information sensing and processing which as a consequence avoids issues such as chattering caused by abrupt changes of the neighbor relations . some interesting problems deserve to be further investigated , for example , how to design the control law based on the information of relative orientations and relative positions , how to design the sampled - data control law to avoid collisions , and how to design the observer - based control laws for the case where the relative speed and relative orientation can not be directly measured . * proof of . * by , we have @xmath310 and @xmath311 @xmath312 . denote @xmath313 . the distance between any two agents @xmath23 and @xmath31 satisfies the following inequality : [ distant ] & & |_ij(t_k+1)-_ij(t_k)| + & = & |x_i(t_k+1)-x_j(t_k+1)-x_i(t_k)-x_j(t_k)| + & & x_i(t_k+1)-x_i(t_k)-(x_j(t_k+1)-x_j(t_k ) ) + & = & ( c _ t_k^t_k+1[v_i(t)_i(t)-v_j(t)_j(t)]dt + _ t_k^t_k+1[v_i(t)_i(t)-v_j(t)_j(t)]dt + ) + & & _ t_k^t_k+1\{|v_i(t)_i(t)-v_j(t)_j(t)| + & & + |v_i(t)_i(t)-v_j(t)_j(t)|}dt.using , the first term of satisfies [ b.2 ] & & _ t_k^t_k+1|v_i(t)_i(t)-v_j(t)_j(t)|dt + & & _ t_k^t_k+1|(v_i(t)-v_j(t))_i(t)| + & & + |v_j(t)(_i(t)-_j(t))|dt_t_k^t_k+1v(t)dt + & & + 2_1inv_i(t_k)_t_k^t_k+1||dt + & & _ t_k^t_k+1v(t)dt+_1inv_i(t_k)_t_k^t_k+1(t)dt , and the second term of satisifies [ b.3 ] & & _ t_k^t_k+1|v_i(t)_i(t)-v_j(t)_j(t)|dt_t_k^t_k+1v(t)dt + & & + 2_1inv_i(t_k)_t_k^t_k+1||dt + & & _ t_k^t_k+1v(t)dt+_1inv_i(t_k)_t_k^t_k+1(t)dt.substituting and into yields the inequality .@xmath256 * proof of . * we see that if at the initial time instant , the distance between @xmath23 and @xmath31 satisfies @xmath314 , then by we have @xmath315 ; otherwise , if @xmath316 , then by we have @xmath317 . compared with the initial time instants , the change of the agent @xmath23 s neighbors at time instant @xmath44 is characterized by the following set , [ 1.1 ] _ i=\{j : ( 1-_n ) r_n_ij(0)(1+_n ) r_n}. denote the maximum number of agents in the set @xmath318 defined by as @xmath319 . by the fact @xmath320 with @xmath78 defined in , we have for large @xmath13 , @xmath321 . since the inequality holds for @xmath109 , the number of each agent s neighbors changed at time @xmath322 in comparison with its initial neighbors is bounded by @xmath319 . using lemma 3 in @xcite , we have for large @xmath13 & & p(t_l+1)-p(0 ) ) + & & ( 1+o(1)=80_n(1+o(1)).@xmath256 by the fact @xmath323 with @xmath324 defined in , we can directly obtain the following results . [ thetaleader ] under the assumptions in lemma [ dddd ] , the following assertions hold almost surely for large @xmath13 & & _ iv_1|_i(t_1)||_0|+l_n , + & & _ iv_2|_i(t_1)|(1-w)(|_0|+l_n ) , + & & _ iv_1|_i(t_1)|=(1+o(1 ) ) , + & & _ iv_2|_i(t_1)|=(1-w)(1+o(1 ) ) , where @xmath329 . olfati - saber r. , & murray r. 2004 . consensus problems in networks of agents with switching topology and time - delays , _ ieee trans . , 49 ( 9 ) , 1520 - 1533 . moshtagh n. , michael n. , jadbabaie a. , and daniilidis k. 2009 . vision - based , distributed control laws for motion coordination of nonholonomic robots . _ ieee trans . robotics _ , 25(4 ) , 851 - 860 . li t. , & zhang j. f. 2009 . mean square average consensus under measurement noises and fixed topologies : necessary and sufficient conditions . _ automatica _ , 45(8 ) , 1929 - 1936 . shi g. d. , & johansson k. h. 2013 . robust consensus for continuous - time multi - agent dynamics . _ siam journal on control and optimization _ , 51(5 ) , 3673 - 3691 . gupta p. , & kumar p. critical power for asymptotic connectivity in wireless networks . in _ stochastic analysis , control , optimization and applications _ , birkhauser boston , boston , ma , 547 - 566 . ji m. , & egerstedt m. 2007 . distributed coordination control of multiagent systems while preserving connectedness . _ ieee trans . robotics _ , 23(4 ) , 693 - 703 . dimarogonas d. v. , & kyriakopoulos k. j. 2007 . on the rendezvous problem for multiple nonholonomic agents . _ ieee trans . , 52(5 ) , 916 - 922 . dimarogonasa d. v. , tsiotras p. , & kyriakopoulos k. j. 2009leader - follower cooperative attitude control of multiple rigid bodies . _ systems and control letters _ , 58 , 429 - 435 . ajorlou a. , & aghdam a. g. 2013connectivity preservation in nonholonomic multi - agent aystems : a bounded distributed control strategy , _ ieee trans . autom . control _ , 58(9 ) , 2366 - 2371 . liu s. , li t. , xie l. , fu m. , & zhang j. f. 2013continuous - time and sampled - data based average consensus with logarithmic quantizers . _ automatica _ , 49(11 ) , 3329 - 3336 . xiao f. , & chen t. 2012sampled - data consensus for multiple double integrators with arbitrary sampling . _ ieee trans . control _ , 57(12 ) , 3230 - 3235 . tang g. g. , & guo l. 2007convergence of a class of multi - agent systems in probabilistic framework . _ journal of systems science and complexity _ , 20(2 ) , 173 - 197 . liu z. x. , & guo l. 2009synchronization of multi - agent systems without connectivity assumption . _ automatica _ , 45(12 ) , 2744 - 2753 . tove g. , dimarogonas d. v. , egerstedt m. , & hu x. m. 2010sufficient conditions for connectivity maintenance and rendezvous in leader - follower networks . _ automatica _ , 46(1 ) , 133 - 139 . cao y. , ren w. , & li y. 2009distributed discrete - time coordinated tracking with a time - varying reference state and limited communication . _ automatica _ , 45(5 ) , 1299 - 1305 . couzin i. d. , krause j. , franks n. r. , & levy s. 2005effective leadership and decision - making in animal groups on the move . _ nature _ , 433 , 513 - 516 . cao y. c. , ren w. , & egerstedt m. 2012distributed containment control with multiple stationary or dynamic leaders in fixed and switching directed networks . _ automatica _ , 48(8 ) , 1586 - 1597 .

this paper reports calculations for compressed ce ( ) , pr ( ) , and nd ( ) using a combination of the local - density approximation ( lda ) and dynamical mean field theory ( dmft ) , or lda+dmft . the moment , spectra , and the total energy among other properties are examined as functions of volume and atomic number for an assumed face - centered cubic ( fcc ) structure . these materials are seen to be strongly localized at ambient pressure and for compressions up through the experimentally observed fcc phases ( phase for ce ) , in the sense of having fully formed hund s rules moments and little spectral weight at the fermi level . subsequent compression for all three lanthanides brings about significant deviation of the moments from their hund s rules values , a growing kondo resonance at the fermi level , an associated softening in the total energy , and quenching of the spin orbit since the kondo resonance is of mixed spin - orbit character while the lower hubbard band is predominantly . while the most dramatic changes for ce occur within the two - phase region of the volume collapse transition , as found in earlier work , those for pr and nd occur within the volume range of the experimentally observed distorted fcc ( dfcc ) phase , which is therefore seen here as transitional and not part of the localized trivalent lanthanide sequence . the experimentally observed collapse to the-u structure in pr occurs only on further compression , and no such collapse is found in nd . these lanthanides start closer to the localized limit for increasing atomic number , and so the theoretical signatures noted above are also offset to smaller volume as well , which is possibly related to the measured systematics of the size of the volume collapse being 15% , 9% , and none for ce , pr , and nd , respectively . + pacs numbers : 71.27.+a , 71.20.eh , 75.20.hr

cond-mat0504380

a number of the trivalent lanthanides undergo first order phase transitions under pressure that are characterized by unusually large volume changes , ce ( 15%),@xcite pr ( 9%),@xcite gd ( 5%),@xcite and dy ( 6%).@xcite these `` volume collapse '' transitions are believed to be caused by changes in the degree of @xmath3 electron correlation from strongly correlated ( localized ) at pressures below the transitions to more weakly correlated ( itinerant ) above.@xcite associated with these two regimes are differences in physical properties , with high - symmetry crystal structures below the collapse transitions characteristic of metals without @xmath7 electrons , and low - symmetry early - actinide - like structures at pressures above , suggesting a greater participation of @xmath7 electrons in the bonding . the magnetic susceptibility is of curie - weiss character in the localized regime characteristic of hund s rules @xmath3 moments , and is believed to become temperature independent similar to pauli paramagnetism in the itinerant regime . the latter behavior has actually been observed only for the collapsed @xmath6 phase of ce , but is inferred for the other lanthanides based on analogies to the actinides . in contrast to these apparently general trends , it has been recently shown that nd reaches the characteristic itinerant @xmath6-u structure in a relatively continuous fashion without any phase transitions of unusually large volume change,@xcite while in an analogous actinide system , am is noted to undergo two collapse transitions.@xcite it would therefore appear that the evolution from localized to itinerant character in these @xmath7 electron metals occurs more generally in a continuous manner which may or may not be accelerated by one or more large - volume - change collapse transitions . it was demonstrated many years ago that the volume collapse transitions in the lanthanides and actinides do not involve the promotion of electrons from @xmath7 to other states , and it was argued that these metals were instead undergoing mott transitions in their respective @xmath3 and @xmath8 electron systems.@xcite local density approximate ( lda ) calculations modified to enable orbital polarization,@xcite to include the self - interaction correction,@xcite and to explicitly incorporate the hubbard repulsion @xmath9,@xcite appear to provide such transitions for ce and pr . they describe large - volume magnetically - ordered solutions with hubbard split @xmath3 bands which become unstable under compression to paramagnetic solutions where these bands are grouped near the fermi level . it has been suggested , however , that the rather abrupt delocalization of these modified - lda transitions is an artifact of the static - mean - field nature of these techniques , and that a correlated solution would allow a continuous transfer of @xmath7 spectral weight between the hubbard side bands and the fermi level over an extended volume range.@xcite this is the case for the kondo - volume collapse model of the ce transition , based on a many - body solution of the anderson impurity model , which identifies kondo - like screening of the local moment as the driving force for the transition.@xcite more realistic calculations which combine lda with correlated dynamical mean field theory ( dmft),@xcite so called lda+dmft,@xcite generally corroborate these characteristics for ce,@xcite and the same technique has been applied to the actinides.@xcite beyond the specific case of ce , a more general understanding of the lanthanide volume - collapse transitions or their absence may come from putting them in context of the extended evolution from localized to itinerant character which occurs as these materials are compressed . as a step in this direction , the present paper reports lda+dmft calculations over a wide range of volume for ce ( @xmath0 ) , pr ( @xmath1 ) , and nd ( @xmath2 ) , the last as mentioned being notable for the absence of a volume collapse transition . these calculations have been carried out as in earlier work on ce,@xcite except that the spin - orbit interaction is now incorporated . they also assume a face centered cubic ( fcc ) structure throughout , although lda estimates of the structural energy differences suggest these are significantly smaller than the relevant contributions to the correlation energy , which may then be discussed in terms of their volume dependence as more important leading order effects . the trivalent lanthanides follow a general structural sequence under pressure , namely hcp @xmath10 sm - type @xmath10 dhcp @xmath10 fcc @xmath10 dfcc , where these abbreviations are for hexagonal close packed ( hcp ) , double hcp ( dhcp ) , and distorted fcc ( dfcc).@xcite these are all just different stacking variants within the family of close packed structures , except for dfcc , a soft l - point phonon distortion of fcc,@xcite a phase which only ce does not assume . the heaviest lanthanides traverse the full sequence , whereas the lighter ones begin part way , e.g. , ce , pr , and nd are all dhcp at ambient conditions , although fcc ce is also metastable there . since the same sequence is seen in y which has no nearby @xmath7 states,@xcite and is theoretically understood to depend on the @xmath11 occupancy,@xcite it would appear to have no relation whatsoever with the @xmath7 electrons , which are then presumed to be fully localized . indeed , in the present work we find evidence for fully formed hund s rules moments and little @xmath3 spectral weight at the fermi level throughout the experimentally observed stability field of the fcc phases ( @xmath4 for ce ) , and for lower pressures . further compression takes ce through a 15% volume collapse from the @xmath4 ( fcc ) to the @xmath6 ( also fcc ) phase , while pr undergoes a 9% collapse from the dfcc to an @xmath6-u structure , and nd passes from the dfcc through two other low - symmetry structures before also reaching @xmath6-u , however , without any large volume changes . the present work finds a variety of signatures to accompany this compression regime : substantial deviation of the moments away from their hund s rules values , rapid growth in @xmath3 spectral weight at the fermi level ( the kondo resonance ) at the expense of the hubbard side bands , an associated softness in the total energy , and quenching of the spin orbit due to the mixed-@xmath12 character of the kondo resonance versus the predominant @xmath5 lower hubbard band . the most dramatic changes in these signatures for the case of ce _ coincide _ with the two - phase region in the @xmath4@xmath6 transition . this behavior was observed in earlier work on ce,@xcite and is consistent with the kondo volume collapse scenario.@xcite in regard to signatures in the total energy , a unique feature of the kondo volume collapse model is the assertion that the @xmath4@xmath6 transition is driven by a rapid drop in entropy from the @xmath4 to the @xmath6 side , reflecting screening of the @xmath3 moments . the transition therefore need not occur at @xmath13,@xcite as has been documented for ce alloys,@xcite implying a featureless @xmath13 total energy . nevertheless , due to the interrelationships between the different thermodynamic functions , and even though it is the free energy which actually determines the phase transition , the _ finite-_@xmath14 total energy should show an associated softening in the vicinity of the phase transition , and is therefore a useful diagnostic.@xcite the present work finds the same signatures for compressed pr and nd , however , their most dramatic change takes place over the volume range where the dfcc phase is experimentally observed , suggesting this phase is of transitional character and not the end member of the localized trivalent lanthanide sequence as has been assumed . only on further compression is pr observed to collapse into the @xmath6-u structure , and as noted nd has no collapse . an important difference among the lanthanides is that they become more localized with increasing atomic number , due to the increased but incompletely screened nuclear charge . this is evident in the present work by the offset of the concurrent correlation - related signatures to smaller volumes from ce to pr and then to nd . one consequence is that the softening effect in the energy associated with the growing kondo resonance must compete with the remaining , dominant part of the energy which has a curvature that grows ever larger with decreasing volume . while a region of negative bulk modulus is required for an isostructural transition as in ce , it may be more generally that a low bulk modulus favors larger volume changes in structural phase transitions , which would then be consistent with the decreasing size of the collapse from 15% ( ce ) , to 9% ( pr ) , to none ( nd ) . the observed collapse transitions in gd and dy would then appear inconsistent ; however , these involve filling of the @xmath15 subshell which complicates matters . in the remainder of this paper , the theoretical methods are reviewed in sec . [ theorysec ] , numerical results for compressed ce , pr , and nd are presented in sec . [ resultsec ] , and then a summary and discussion is given in sec . [ summarysec ] . the results in this paper unless otherwise indicated have been obtained by the lda+dmft(qmc ) method , which refers to the merger of the local density approximation ( lda ) with dynamical mean field theory ( dmft),@xcite in order to create a composite technique@xcite which rigorously treats onsite electron correlations and yet retains the material realism and specificity of the lda . an essential component of the dmft method is solution of an auxiliary impurity problem which is achieved here by a quantum monte carlo ( qmc ) algorithm.@xcite the present calculations have been carried out as described in previous work for ce,@xcite except that the spin - orbit interaction has now been added and a broader set of materials considered . in the remainder of this section we give a brief review of the method , various computational details , and calculation of the moment . the lda contribution to the lda+dmft method is to provide an effective hamitonian which includes all valence electron degrees of freedom , @xmath16_{ljm , l^\prime j^\prime m^\prime } \,\hat{c}^\dagger_{{\bf k}\,ljm } \hat{c}^ { } _ { { \bf k}\,l^\prime j^\prime m^\prime } \nonumber \\ & & + \ ; \frac12 \ , u_f \!\!\ ! \sum_{{\bf i},j m , j^\prime m^\prime } \!\!\!\!\!\!\ ! ^ { ' } \ , \hat{n}_{{\bf i}fjm}\ , \hat{n}_{{\bf i}fj^\prime m^\prime}. \label{hameqn}\end{aligned}\ ] ] here , * k * are brillouin zone vectors , @xmath17 are lattice sites , @xmath18 is the orbital angular momentum , @xmath12 is the total angular momentum ( @xmath19 except just @xmath20 for @xmath21 ) , @xmath22 , @xmath23 , and the prime signifies @xmath24 . the relevant valence states for the present lanthanide case are @xmath25 , @xmath26 , @xmath11 , and @xmath3 , and so the matrices @xmath27 are @xmath28 . as described elsewhere,@xcite the @xmath29 are orthogonalized one - electron hamiltonian matrices obtained from converged linear muffin - tin orbital lda calculations,@xcite in which the @xmath3 site energies are shifted so as to avoid double counting the @xmath7-@xmath7 coulomb interaction @xmath9 which is explicitly incorporated into eq . ( [ hameqn ] ) . the spin - orbit interaction was included as a perturbation,@xcite with the spin - orbit coupling parameters @xmath30 kept fixed across the bands , although evaluated at the most important energies , namely at the respective centers of gravity of the occupied state density for each orbital type . the resultant @xmath27 matrices were calculated over a grid of volumes for ce , pr , and nd in an assumed face centered cubic ( fcc ) structure , and companion lda constrained - occupation calculations used to get the volume and material dependent coulomb interactions @xmath9.@xcite dynamical mean field theory assumes a local or * k*-independent self - energy @xmath31 , which together with the one - body part of eq . ( [ hameqn ] ) provides the lattice green function , @xmath32^{-1 } . \label{latgrneqn}\ ] ] this is solved in tandem with an auxiliary impurity problem defined by a similar dyson - like equation with the same local self energy . @xmath33^{-1 } \ , . \label{impgrneqn}\ ] ] the idea is to guess an initial @xmath34 , calculate the lattice green function @xmath35 from eq . ( [ latgrneqn ] ) , identify its * k * average with @xmath36 , and then find the noninteracting or bath green function @xmath37 for the impurity problem from eq . ( [ impgrneqn ] ) . the impurity problem is completely defined by @xmath38 , which describes its one - body part , together with the coulomb interaction for its single site as in eq . ( [ hameqn ] ) without the * i * index . it may be solved exactly to within statistical uncertainties by qmc techniques.@xcite this gives a new @xmath36 which together with the input @xmath39 defines a new self - energy @xmath40 again via eq . ( [ impgrneqn ] ) . the cycle is then repeated by starting anew with @xmath40 in the lattice green function eq . ( [ latgrneqn ] ) , and so on until self consistency . quantities of interest such as the number @xmath41 of @xmath5 or @xmath42 @xmath3 electrons may be obtained from the lattice green function @xmath43 e^{i\omega_n 0^+ } \ , , \label{njeqn}\ ] ] where the sums are over * k * vector and matsubara frequency @xmath44 , and the trace is over those @xmath45 @xmath3 states of type @xmath12 . the total number of @xmath3 electrons is then @xmath46 . similarly the energy per site for the effective hamiltonian in eq . ( [ hameqn ] ) is @xmath47 e^{i\omega_n 0^+ } + u_f \ , d \ , , \label{edmfteqn}\ ] ] which involves the double occupation @xmath48 as discussed elsewhere.@xcite in principle this is an average over the sites * i * , although in practice we obtain this quantity form the qmc impurity problem . to evaluate the total lda+dmft energy @xmath49 including all core and outer electrons , we add a correction to the paramagnetic all - electron lda energy @xmath50 @xmath51 which consists of the dmft energy @xmath52 from eq . ( [ edmfteqn ] ) less an lda - like solution of the model hamiltonian eq . ( [ hameqn ] ) , thus `` model lda '' or @xmath53 . to be more accurate , our lda total energy @xmath50 is scalar relativisitc and so omits spin orbit , and therefore the correction @xmath53 should be and is obtained from a modification of eq . ( [ hameqn ] ) with spin orbit removed . using this modified effective hamiltonian , @xmath53 is determined by a self - consistent solution of eqs . ( [ latgrneqn ] ) and ( [ njeqn ] ) for @xmath54 taking a self - energy @xmath55 . from this , the kinetic energy is calculated by the first term of eq . ( [ edmfteqn ] ) and the potential energy by @xmath56 . the simple form of eq . ( [ etoteqn ] ) is a reminder of the still fairly new and evolving nature of the lda+dmft method . there are recent discussions of the formal context of the method,@xcite as well as some sense that a gw platform in place of the lda may allow for more natural integration with dmft.@xcite mutual self - consistency between the lda and dmft parts has also been stressed , i.e. , feeding a different dmft value of @xmath57 back into a new lda calculation.@xcite this is impractical in the present case given the expense of the qmc solutions , although we do provide a perturbative change in @xmath57 in going from the lda to the dmft , and this may well be sufficient . a step towards answering many of these questions is simply to provide more quantitative tests against experiment of lda+dmft implementations with clearly stated approximations , and this is the spirit of the present work . while the matrices @xmath35 , @xmath27 , and @xmath58 in eq . ( [ latgrneqn ] ) are all @xmath28 , it is customary to take @xmath58 non - zero only within the interacting or @xmath59 @xmath7-@xmath7 block , which permits the same reduction in the impurity problem itself . even so , one must still determine seven distinct functions of @xmath60 to fill out @xmath31 even for the full cubic point group,@xcite and more otherwise . on the other hand , if we ignore crystal field , then the @xmath7-@xmath7 self - energy becomes @xmath61 and we need only find functions for the two spin - orbit states @xmath5 and @xmath42 . this is an excellent approximation over most of the volume range studied here , including the volume - collapse regions , where the crystal field splitting ranges from about 1 to 10% of the spin - orbit splitting . at the very smallest volumes considered here , this ratio approaches 50% , however , it is to be emphasized this omission is only in the self - energy , and all such effects are reflected in @xmath27 where hybridization dominates anyway . in practice we input a bath green function of similar structure , @xmath62 , into the qmc and then improve statistical uncertainties by averaging over the six @xmath5 and eight @xmath42 states in determining the output @xmath63 . it was noted previously that the dmft(qmc ) iterations could be greatly accelerated by separately converging the leading , constant , hartree - fock - like part of the self - energy outside of the qmc part of the cycle.@xcite for the present case this constant part is @xmath64u_f \ , . \label{sig0eqn}\ ] ] therefore after each qmc cycle we extract just the frequency dependent part of the self - energy @xmath65 using here the qmc @xmath41 values @xmath66 \ , , \label{nfqmcgeqn}\ ] ] where @xmath67 is the imaginary time fourier transform of @xmath68 . then with @xmath69 held fixed , we construct a new self - energy @xmath70 _ but this time _ taking the @xmath41 defining @xmath71 as obtained from the lattice green function via eq . ( [ njeqn ] ) . inserting this self - energy back into the lattice green function eq . ( [ latgrneqn ] ) leads to a self - consistent condition on both @xmath72 and the @xmath41 which may be completely converged with negligible cost , again with the @xmath73 held fixed . only after this convergence would we take the resultant @xmath72 and begin the process of mapping back onto the auxiliary impurity problem in order to update @xmath69 . we found the procedure of converging @xmath74 in between each qmc iteration for @xmath75 to work best when the current qmc ratio @xmath76:@xmath77 was retained during the cheap iterations which were then used only to fix the total @xmath7 charge @xmath57 . also note this approach requires the strict equivalence of the @xmath41 calculation in both the lattice and impurity problems , which dictates using @xmath78 in conjunction with eq . ( [ slfeeqn ] ) in solving eq . ( [ impgrneqn ] ) for the bath green function @xmath79 . the output of each qmc cycle is the imaginary time green function @xmath67 which contains statistical uncertainties , and must be fourier transformed in order to define the new self - energy @xmath80 . following earlier work,@xcite our approach is to use multipole fits @xmath81 using basis functions @xmath82 , which have fourier transforms @xmath83 , thus trivially giving the overall fourier transforms . these fits were carried out requiring positive weights @xmath84 , and constrained to give the qmc values at @xmath85 @xmath86 as well as a number of correct moments @xmath87 @xmath88 here @xmath89 , @xmath90 , and the high - frequency moments @xmath91 are defined by @xmath92 . in our previous work on ce we imposed two ( @xmath93 ) of the constraints in eq . ( [ csn2eqn ] ) , however , we find in the present effort for pr and nd at the largest volumes that it was necessary to add a third moment ( @xmath94 ) . the reason is that for such strongly localized functions where the @xmath67 fall rapidly away from their @xmath85 and @xmath95 values , one may get large and unphysical values of @xmath96 and @xmath97 of opposite signs . this can lead to unphysical structure in the spectra far from the chemical potential , although we could detect no impact on the total energy or @xmath41 from this problem . we found that adding the @xmath94 or @xmath96 constraint was sufficient to fix this difficulty . the moments @xmath91 may be obtained from * k*-averages and appropriate traces of powers of @xmath98 with in addition @xmath99 from eq . ( [ sig0eqn ] ) for @xmath100 , and both @xmath99 and @xmath101 for @xmath96 . @xmath101 may be obtained from the double occupation matrix,@xcite or approximated using a hubbard - i type of self - energy tuned to give the qmc @xmath41 values . the results reported in this paper are for a temperature of 632 k ( 0.004 ry ) , which given the @xmath102 cost of the qmc calculations is about as low as is practicable . here the imaginary time interval @xmath103 for the qmc was discretized by @xmath104 , @xmath105 , and @xmath106 divisions , carrying out @xmath107 , @xmath108 , and @xmath109 qmc sweeps , respectively , for each dmft(qmc ) iteration . the smaller number of sweeps for larger @xmath110 was dictated purely by the @xmath111 expense of the calculations , and the statistical uncertainties necessarily increased . nonetheless systematic and generally @xmath112 behavior was seen allowing extrapolation to eliminate the trotter errors.@xcite starting self - energies were already quite good , taken from converged dmft calculations with the hubbard i self - energy,@xcite or from nearby volumes , or temperatures ( 0.01 ry ) . at least thirty iterations were performed at each volume , following five discarded warmup iterations . at a number of volumes this process was repeated taking the final self - energy as the initial guess and starting the process from the beginning . the results were unchanged to well within the statistical uncertainties . all matsubara sums for an argument @xmath113 were carried out using an asymptotic two - pole approximation @xmath114 with the parameters fixed by the first four high - frequency moments of @xmath115 . the infinite sum was then given by a finite sum over the difference @xmath116 plus the analytic result for the infinite sum over @xmath117 , with @xmath118 positive matsubara frequencies used in the former . we found a noticeable increase in the qmc statistical uncertainties for the total energy in going from ce to pr to nd , likely reflecting the larger role played by the potential energy @xmath119 and the qmc determined value of the double occupation , eq . ( [ dbleqn ] ) , with the increasing number of @xmath3 electrons . at large volumes , for example , @xmath120 should be approximately @xmath121 , @xmath122 , and @xmath123 for ce , pr , and nd , respectively , at low temperatures . similarly we found differences in the nature of the trotter corrections , where again at larger volumes we found the coefficient of the @xmath112 dependence to be quite small for ce , larger for pr , and significantly larger for nd . we were able to get decent spectra from the multipole fits eq . ( [ gtfiteqn ] ) by increasing the number of poles by well over two orders of magnitude . as described earlier,@xcite we took equally spaced @xmath124 grids of @xmath125 points , and systematically eliminated poles with negative weights , examining o(@xmath126 ) such grids of varying centroid and width to find the best fit . for the spectra we combined the @xmath127 best fits so long as the worst of these had a root - mean - square agreement with the qmc data no more than 20% larger than that of the best . more important , we also averaged over the fits for the last half of the dmft(qmc ) iterations . the resultant collections of o(@xmath128 ) poles were broadened by gaussians of @xmath129 ev full - width at half maximum . we found systematic evolution of these spectra with volume providing one measure of their validity . since we approximate the @xmath7-@xmath7 self - energy by the form @xmath130 , a comparable treatment of the double occupation matrix is @xmath131 & \mbox{if $ j\!=\!j^\prime$ , $ m\!\neq\!m^\prime$ } \\ d_{5/2,7/2}/48 & \mbox{if $ j\!\neq\!j^\prime$ } \\ \end{array } \right . \ , , \label{n1n2eqn}\ ] ] noting that @xmath132 , and @xmath5 and @xmath42 . we obtain @xmath41 and @xmath133 from the qmc auxiliary impurity problem by summing over the appropriate blocks in the @xmath59 matrix @xmath134 . as these are block totals , the total number of @xmath7 electrons per site is @xmath135 , and the double occupation @xmath120 appearing in the expression for the total energy is @xmath136 . we find the present @xmath137 for @xmath138 to be bounded above by their uncorrelated values from @xmath139 for @xmath140 , or @xmath141 similarly , we find the @xmath137 to be bounded below by a typical strongly correlated expression @xmath142 \ , , \label{dmineqn}\ ] ] where @xmath143 is an integer such that @xmath144 and eq . ( [ dmineqn ] ) is a piecewise linear function of @xmath41 with values @xmath145 at integer @xmath41 . the expectation of the onsite squared @xmath7 moment is given by @xmath146 since our hamiltonian is rotationally invariant , and thence using eq . ( [ n1n2eqn ] ) by @xmath147 in the local - moment regime at large volume and low temperature where @xmath148 , and @xmath123 for ce , pr , and nd , respectively , @xmath149 , and @xmath150 , eq . ( [ jsqeqn ] ) yields @xmath151 , @xmath152 , and @xmath42 , respectively , via @xmath153 . this compares to the hund s rules ground state values of @xmath154 , @xmath155 , and @xmath156 , respectively , and reflects the fact that including the spin - orbit interaction in the absence of intraatomic exchange will give the correct moment only when the @xmath5 subshell has a single electron or hole , or is trivially empty or full . from the thermodynamic perspective , it is the degeneracy of the hund s rules multiplet that matters , and this changes from @xmath157= 14 , 91 $ ] , and @xmath158 to @xmath159= 6 , 15 $ ] , and @xmath160 for ce , pr , and nd , respectively , on adding spin orbit , which is quite an improvement given the correct degeneracy @xmath161 , and @xmath162 , respectively . although not described here , we do find dmft calculations with the hubbard - i self - energy to yield the expected low - temperature entropy plateaus of @xmath163 , @xmath164 , and @xmath165 in units of @xmath166 for ce , pr , and nd , respectively , at large volume , similar to earlier work for ce without the spin - orbit interaction.@xcite one may use eqs . ( [ dmaxeqn ] ) and ( [ dmineqn ] ) to find lower and upper bounds on @xmath167 in eq . ( [ jsqeqn ] ) . if one assumes that the spin orbit is also quenched , @xmath168 , in addition to the uncorrelated eq . ( [ dmaxeqn ] ) , then @xmath169 with this additional assumption , eq . ( [ jsqunceqn ] ) no longer provides a lower bound on @xmath167 except at the smallest volumes where the spin orbit is quenched , but will nevertheless prove useful . moment @xmath170 for ce , pr , and nd . the dmft(qmc ) results are given by the solid curves with data points ; local - moment values , by the horizontal long - dashed lines ; and an uncorrelated estimate with quenched spin orbit , by dash - dot lines . the vertical short - dashed lines denote transitions with significant volume collapse by shading . both @xmath6 and @xmath4 ce phases are fcc . [ jsqfig],width=307 ] this section reports calculations for compressed ce , pr , and nd obtained by the lda+dmft method with a qmc determination of the self - energy , to be referred to more simply as just dmft . all theoretical results have been carried out for an assumed fcc structure and at a temperature of 632 k ( 0.004 ry ) unless otherwise noted . this temperature is about as low as is practical given the @xmath102 expense of the qmc . nevertheless , it is cold enough , since previous work for ce found that the total energy and entropy at @xmath171 k are relatively close to the low - temperature limit.@xcite thus , e.g. , the slope of the @xmath4@xmath6 phase line in ce comes primarily from the explicit @xmath14 in the @xmath172 term of the free energy @xmath173 , and not from @xmath14-dependence in either the energy or entropy.@xcite the extended aspect of the evolution from localized to itinerant character in the three lanthanides is illustrated by the dmft results ( data points with error bars ) in fig . [ jsqfig ] for the square of the onsite @xmath3 electron moment @xmath170 calculated using eq . ( [ jsqeqn ] ) . the vertical dashed lines mark volumes at which transitions are observed experimentally . the horizontal long - dashed lines at large volume give the strongly - correlated local - moment values @xmath174 with @xmath175 the proper hund s rules @xmath176 value for ce , although somewhat smaller than the proper values for pr and nd due to our omission of intraatomic exchange ( see sec . [ dblmomsec ] ) . it is apparent for each material that the local - moment regime persists under compression up to and through most of the stability field of the fcc phase ( large volume @xmath4 fcc phase for ce ) . the dash - dot curves are uncorrelated approximations to @xmath170 from eq . ( [ jsqunceqn ] ) which also presume that the spin orbit is quenched , i.e. , that @xmath177 , which is largely responsible for the significant offset of these curves from the local - moment lines at large volume . the dmft results approach these uncorrelated and spin - orbit quenched curves under compression , reaching reasonably close agreement by the @xmath6 phase of ce and @xmath6-u phases of pr and nd . as will be discussed subsequently , the significance of the quenched spin - orbit lies in the fact that the lower hubbard band is of predominant @xmath5 character , while the @xmath3 kondo resonance which grows at the fermi level at the expense of the hubbard side bands is of mixed @xmath5 , @xmath42 character . therefore quenched spin orbit reflects dominance of the kondo peak which may be viewed as a signature of itineracy . a quantitative measure of the degree of correlation is provided by the lower ( strongly correlated ) and upper ( uncorrelated ) bounds on the double occupation given by eqs . ( [ dmaxeqn ] ) and ( [ dmineqn ] ) , respectively , which in turn provide bounds on @xmath170 . the resultant curves have roughly the same shape as the dmft results in fig . [ jsqfig ] , and bracket these results in each case , with the dmft curves switching from more or less perfect agreement with the strongly - correlated limit at large volume to much closer to the uncorrelated limit at small volume . even so , we find in the vicinity of @xmath178@xmath179/atom that for each of these materials the dmft results have switched only about 60% of the way from the strongly correlated to the totally uncorrelated limit , consistent with the fact that we still find residual hubbard side bands at this volume . on the other hand , standard paramagnetic lda certainly does well enough in predicting the @xmath180 ratio in bct ce and structural characteristics of bct and @xmath6-u phases among the early actinides,@xcite so it may well be that for all practical purposes such phases are weakly correlated enough . electrons @xmath57 less an integer @xmath181 , @xmath182 , and @xmath123 for ce , pr , and nd , respectively . [ nffig],width=307 ] electrons of @xmath5 character as a function of volume . [ rn1fig],width=307 ] the moment @xmath170 in fig . [ jsqfig ] is a bare quantity and does not reflect any screening effects by the other electrons . its large values at the smaller volumes are due via eq . ( [ jsqeqn ] ) to an increase in the number of @xmath3 electrons @xmath57 under compression as seen in fig . [ nffig ] , where @xmath183 is plotted with @xmath181 , @xmath182 , and @xmath123 for ce , pr , and nd , respectively . it is well known that the lanthanides undergo electronic @xmath184-@xmath120 transition under pressure , during which the @xmath25 and @xmath26 states begin to pass above the fermi level thereby increasing the @xmath11 and @xmath3 occupations.@xcite figure [ rn1fig ] shows the fraction @xmath185 of @xmath3 electrons which are of @xmath5 character , and it is quite clear that there is quenching of the spin orbit in the region of interest in this work . density of states @xmath186 for @xmath187 ( dashed line , shaded ) and @xmath15 ( solid line ) as obtained from multi - pole fits to the qmc @xmath188 with @xmath129 ev fwhm gaussian broadening . areas are normalized @xmath189:@xmath190 for the appropriate @xmath12 s , and energy is relative to the chemical potential @xmath191 . [ prdosfig],width=307 ] the interplay between spin orbit and compression induced changes in @xmath3 spectra may be seen in fig . [ prdosfig ] , where the total ( brillouin zone summed ) but @xmath12-resolved @xmath3 spectra or density of states @xmath186 is plotted for pr at a number of volumes . these dmft results were obtained by multi - pole fits to the qmc @xmath188 as described in sec . [ cmpdtlsec ] , and were broadened by a gaussian of @xmath129 ev full width at half maximum . at the largest volume @xmath192 @xmath179/atom , one sees the pure @xmath5 lower hubbard band near @xmath193 ev , and the mixed-@xmath12 upper hubbard band near @xmath182 ev , with the splitting consistent with @xmath194 ev at this volume.@xcite under compression both hubbard bands lose spectral weight at the expense of the growing kondo peak near the chemical potential @xmath191 , which becomes dominant at the smallest volumes . since this fermi - level structure is also of mixed-@xmath12 character , the population of these states at the expense of the initially pure @xmath5 lower hubbard band relates the quenching of spin orbit in these materials to their growing itinerant character under compression . note , however , as is especially evident at @xmath195 @xmath179/atom , that the @xmath5 and @xmath42 contributions to the kondo peak are split by about the spin - orbit energy,@xcite with the former more centered at @xmath191 . consequently , the occupied part of the kondo peak is initially mostly @xmath5 character , although the @xmath15 part catches up as is also evident from fig . [ rn1fig ] . the ce and nd spectra are quite similar with the primary visual difference the increasingly more prominent lower hubbard band in the progression from ce to pr and nd , as is to be expected . the ce spectra are available elsewhere.@xcite a cross check on the spectra in fig . [ prdosfig ] is provided by @xmath196 where this expression becomes exact in the low - temperature limit.@xcite equation ( [ dosefeqn ] ) tracks the values at @xmath197 in fig . [ prdosfig ] to within better than 20% at all volumes . the volume dependence of @xmath198 from eq . ( [ dosefeqn ] ) looks very much like that plotted for ce in the case without spin orbit in fig . 3 of ref . . for decreasing volume , these values are first quite small then rise to a maximum signifying growth of the kondo peak , and then begin to decrease as the hybridization induced broadening of the @xmath3 bands begins to dominate . ce , pr , and nd all behave similarly with the onset of growth in the kondo peak starting near the low - volume side of the fcc ( @xmath4-ce ) phases and then rising to successively higher maxima occurring at smaller volumes , roughly at 29 , 23 , and 20 @xmath179/atom , respectively . comparing these volumes with the transitions indicated in fig . [ jsqfig ] , it may be seen that the ce value reaffirms earlier observations that the initial rapid rise of the kondo peak in that material coincides with the volume collapse.@xcite for pr and nd , on the other hand , the growth of the kondo peak coincides with the observed stability field of the dfcc phase , suggesting a transitional role for this phase rather than being the end member of the localized trivalent lanthanide series as has been assumed . in contrast to the @xmath5 function , @xmath199 starts to increase for decreasing volume at about the same place as @xmath198 , but does so more gradually , steadily growing until it becomes comparable to @xmath198 at the smaller volumes . in @xmath200 refers to the number of @xmath3 bands split off below the fermi level in the hf calculations , with @xmath201 the customary paramagnetic solution , @xmath202 indicating spin- and orbitally - polarized solutions , and @xmath203 , @xmath182 , and @xmath123 for ce , pr , and nd , respectively , being the fully polarized solutions . [ ecorfig],width=307 ] figure [ ecorfig](a ) shows the pr correlation energy , namely the total energy less the result @xmath204 for a paramagnetic hartree fock solution of the effective hamiltonian eq . ( [ hameqn ] ) . one may obtain a variety of metastable hartree fock ( hf ) solutions in which different numbers @xmath205 of `` localized '' @xmath3 bands are split off below the fermi level , ranging from the paramagnetic solution @xmath201 to the fully spin- and orbitally - polarized solution ( maximum @xmath205 ) , which in the case of @xmath1 pr is @xmath206 . the @xmath4 phase of @xmath0 ce has been described by analogous fully - polarized lda+u solutions ( @xmath203),@xcite and the @xmath4@xmath6 volume collapse by the @xmath207 transition.@xcite fig . [ ecorfig](a ) suggests a two step process in these hf solutions for pr , as first one split - off band pops back up to the fermi level under compression ( the @xmath206 and @xmath122 curves cross ) , and then later the second does the same ( the @xmath208 and @xmath121 curves cross ) , a @xmath209 scenario for which there is no experimental evidence . on the other hand there is some truth here , as the intermediate @xmath203 solution is really just a crude attempt to add @xmath3 spectral weight at the fermi level in the continued presence of some hubbard splitting . for a truly correlated calculation , in contrast , there is an entirely continuous transfer of spectral weight from the hubbard sidebands to the kondo resonance at the fermi level , consistent with the smooth behavior seen in the dmft curve in fig . [ ecorfig](a ) . it agrees with the fully polarized hf result at large volumes ( where the hf correctly captures the hubbard splitting ) , but then bends smoothly away from this hf solution under compression ( where the hf fails to describe the kondo resonance ) . the same behavior has been seen in earlier dmft results for ce,@xcite as well as in more rigorous qmc solutions for the anderson lattice hamiltonian.@xcite it seems intuitively clear that such deviation from the fully polarized hf solution as volume is reduced should be associated with the growth of @xmath3 spectral weight at the fermi level , i.e. , the kondo resonance . the total energy differences between dmft and fully polarized hf solutions of eq . ( [ hameqn ] ) are shown in fig . [ ecorfig](b ) for fcc ce , pr , and nd . ferromagnetic order was assumed for the hf . the three insets ( shifted vertical axes ) give a clearer view of the large volume behavior , and the systematically larger qmc uncertainties from ce to pr to nd have been discussed in sec . [ cmpdtlsec ] . our calculations of the @xmath3 spectra indicate growth of the kondo resonance for each material first begins on the small - volume side of the experimentally observed fcc or @xmath4-ce stability range ( see fig.[jsqfig ] ) , and then @xmath198 reaches a maximum at @xmath210 , @xmath211 , and @xmath212@xmath179/atom for ce , pr , and nd , respectively . these volumes are close to where the curves in fig . [ ecorfig](b ) begin to bend away from the fully polarized hf ( 0 baseline ) , consistent with the association of this energy with the kondo resonance . the fact that these deviations occur at successively smaller volumes from ce to pr to nd is consistent with the ratio of the coulomb interaction to band width @xmath213 for the @xmath3 states , which near equilibrium volume ( @xmath214@xmath179/atom ) is 3.7 , 4.4 , and 4.9 for ce , pr , and nd , respectively.@xcite thus pr and nd are successively more localized than ce at ambient conditions , and greater compression is needed to bring about similar changes in electron correlation . a region of negative bulk modulus is of course essential to obtain a first - order isostructural transition as in ce . if more generally a region of low bulk modulus were to favor larger volume changes in structural phase transitions , then the delayed response of the softening effects in fig . [ ecorfig](b ) for pr and nd means competing against the ever more dominant positive bulk moduli of the remaining contributions to the total energy as they continue to grow with decreasing volume . this is at least intuitively consistent with a 15% volume collapse in ce , 9% in pr , and apparently none in nd . it is difficult to quantify this point as the qmc uncertainties in fig . [ ecorfig](b ) do nt admit very reliable calculation of the curvature of these results . there are also some important differences between the ce and pr volume collapse transitions , beyond the fact that the former is isostructural and the latter involves a structural change . as already noted , the experimentally observed two - phase region ( 27.8@xmath215@xmath179/atom ) for the @xmath4@xmath6 collapse in ce coincides with the region of rapid grown in the kondo resonance . it also overlaps a rather sharp breakaway of the dmft curve from the hf result , as seen near @xmath216@xmath179/atom in fig . [ ecorfig](b ) , which is consistent with a large negative contribution to the bulk modulus . here , as for earlier results on ce without the spin orbit,@xcite as well as analyses of the anderson lattice hamiltonian,@xcite these features corroborate the kondo - volume collapse model of the ce @xmath4@xmath6 transition.@xcite for pr , on the other hand , the volume collapse appears to occur at pressures above the region of rapid growth of the kondo resonance . the maximum in the pr @xmath217 spectral weight at the fermi level is near @xmath218@xmath179/atom ( see also fig . [ prdosfig ] ) , and the pr curve in fig . [ ecorfig](b ) begins to bend away from the fully polarized hf at slightly larger volume , both of which lie outside and to the large - volume side of the observed two - phase region for pr ( 19.8@xmath219@xmath179/atom ) . if the ce collapse serves to initiate the evolution from localized to itinerant in this material , such evolution is already well underway by the pr collapse , a point which is also quite evident from behavior of the local moment in fig . [ jsqfig ] . local - density bct fcc energy difference,@xcite and added this to the present fcc dmft result . the experimentally observed 300-k equilibrium volume @xmath220 and stability fields for various phases are marked . [ etotcefig],width=307 ] -u estimate took a @xmath13 local - density @xmath6-u fcc energy difference,@xcite and added this to the present fcc dmft result . the experimentally observed 300-k equilibrium volume @xmath220 and stability fields for various phases are marked ( cp denotes close packed ) . [ etotprfig],width=307 ] and stability fields for various phases are marked ( cp denotes close packed ) . [ etotndfig],width=307 ] comparison of the theoretical and experimental energies are given for ce , pr , and nd in figs . [ etotcefig ] , [ etotprfig ] , and [ etotndfig ] , respectively . the hf and dmft total energies are all at 632 k and for an assumed fcc structure as noted . experimental pressure data @xmath221 exists for pr over a wide range of volume at @xmath222 k,@xcite from which one may obtain the free energy @xmath223 and total energy @xmath224 to within arbitrary constants , which we fix by setting @xmath225 at the experimental 300-k equilibrium volume @xmath220 . raising the temperature generally makes the free energy a steeper function of volume . however , since @xmath226 , the total energy is generally less steep than the free energy . it is a fortunate coincidence that these effects roughly cancel so that the 300-k free energy is within 0.07 ev / atom of the 632-k total energy for pr throughout the range plotted for these experimental quantities in fig . [ etotprfig ] , noting again that both are zeroed at @xmath220 . experimental measurements of @xmath221 for ce exist up 208 gpa at 300 k,@xcite and over the temperature range 299573 k within the fcc ( @xmath4 and @xmath6 ) regime.@xcite assuming @xmath227 is independent of volume , we may use the former results to extend the latter up to higher pressures . extrapolating these results also to somewhat higher temperatures , one may obtain the @xmath228k@xmath229 and @xmath230k@xmath229 curves seen in fig . [ etotcefig ] which like pr are also fairly close . we presume that it is also legitimate to compare room - temperature free energies for nd to our 632-k calculated total energies for that material . the debye temperatures for ce , pr , and nd are all below 150 k at ambient conditions,@xcite and with reasonable grneisen parameters all three materials should still be in the high - temperature limit at 632 k throughout the volume range studied here . the phonon contribution to the total energy is then @xmath231 and has no impact on the volume dependence examined in figs . [ etotcefig][etotndfig ] . the total energies in figs . [ etotcefig][etotndfig ] provide another perspective on the correlation issues discussed earlier . the fully - polarized hf results ( dashed ) curves are analogous to lda+u and should do well at large volumes in the strongly localized limit as is the case here . the dmft results ( open circles ) then show how such static mean field theories begin to break down as volume is reduced due to their inability to account for the growing kondo resonance and its contribution to the correlation energy . this effect is most pronounced for ce in fig . [ etotcefig ] , less so for pr in fig . [ etotprfig ] , and smallest for nd in fig . [ etotndfig ] , reflecting the fact that pr and nd start out successively further in the localized limit than ce , therefore requiring greater compression to achieve comparable changes in the correlation energy . although the agreement between the dmft results and the experimental isotherms is not ideal , and will be discussed further , it is clear that experiment does confirm this systematic progression . a comparison of theoretical and experimental values for the bulk properties is given in table [ bulktab ] . the theory results were obtained from 6-term fits @xcite to the energies over the range @xmath232@xmath233@xmath179/atom in order to better average over scatter in the qmc results . table [ bulktab ] shows quite decent agreement with experiment,@xcite and is a partial validation of the present effective hamiltonians eq . ( [ hameqn ] ) and the manner of total energy calculation eq . ( [ etoteqn ] ) . of particular note is the fact that the dmft results for @xmath220 and @xmath234 are in better agreement with experiment and 3% and 36% smaller , respectively , than the fully polarized hf values in the case of ce , as compared to @xmath235% and @xmath189@xmath236% smaller , respectively , for pr and nd . this suggests a small but not unimportant effect of the kondo resonance even at @xmath237 in ce , effects which are shifted to smaller volume and have less impact for pr and nd . . comparison of theoretical and experimental values for the equilibrium volumes @xmath220(@xmath179/atom ) and bulk moduli @xmath234(gpa ) . the theory used fits to the 632-k results in figs . [ etotcefig][etotndfig ] , while the experiment is at 300 k. however , the ce hf results change by only @xmath238% and @xmath239% for @xmath220 and @xmath234 , respectively , on reducing the temperature from 632 to 316 k. [ cols="^,^,^,^,^,^,^",options="header " , ] [ bulktab ] an estimate of the structural contribution to the energy at smaller volumes is provided by the `` bct '' and `` @xmath6-u '' results ( @xmath240 symbols ) in figs . [ etotcefig ] and [ etotprfig ] for ce and pr , respectively . these were obtained by adding paramagnetic local - density functional values for the bct fcc energy difference in ce,@xcite and the @xmath6-u fcc difference in pr,@xcite to the dmft fcc values . throughout the ranges shown in figs . [ etotcefig ] and [ etotprfig ] these structural energy differences are 10% or less for ce and 25% or less for pr of the more dominant correlation energy contribution represented by the separation between the dmft and the polarized hf curves . these structural corrections improve the agreement with experiment somewhat for ce and pr , however , a similar correction for nd would move the dmft results to lower energies further away from the experiment . lda results,@xcite while the hf , dmft , and structural estimates ( `` bct '' ce , `` @xmath6-u '' pr ) are negative volume derivatives of the total energies in figs . [ etotcefig][etotndfig ] . [ prsfig],width=307 ] figure [ prsfig ] gives the pressure - volume results corresponding to figs . [ etotcefig][etotndfig ] , with experimental 300-k data shown for ce,@xcite pr,@xcite and nd.@xcite except for the @xmath13 lda results ( thin solid lines),@xcite the theoretical results are negative volume derivatives of the 632-k total energies in these figures , an approximation justified by the @xmath241 comparisons in figs . [ etotcefig ] and [ etotprfig ] . multi - term fits to the energies were used in obtaining the derivatives.@xcite the dmft points ( open circles ) were fit ignoring the error bars , with separate fits made for the regions above and below the observed collapse transitions for ce and pr . it is evident that the present paramagnetic dmft results do well in both low - pressure localized and high - pressure itinerant extremes , especially considering the structural corrections [ dashed lines in figs . [ prsfig](a ) and ( b ) ] in the later regime , although there are evident discrepancies in between as will be discussed . standard paramagnetic lda does well in the itinerant regime as seen by the thin black lines , but must be combined with one of the magnetically - ordered modified - lda techniques ( e.g. , lda+u ) to capture the larger volume behavior . this then leads to a magnetic order - disorder transition which is not observed at room temperature , and , e.g. , an incorrect prediction of a volume collapse ( 16% at 39 gpa ) in nd.@xcite turning to the collapse transitions , the kink in the dmft total energy for ce near @xmath216@xmath179/atom in fig . [ etotcefig ] lies within the experimental two - phase region , is within the qmc error bars , and is similar to earlier results without spin orbit.@xcite the addition of the spin - orbit interaction has lowered the @xmath4-phase energy more than that of the @xmath6 phase , so that a common tangent construction to the dmft energy curve in fig . [ etotcefig ] gives a pressure of @xmath242 gpa as compared to a slightly negative value in the earlier work . while the experimental transition ranges from ( an extrapolated ) @xmath243 gpa at @xmath13 to 1.8 gpa at 485 k,@xcite an error of 4 gpa is not huge on the scale of fig . [ prsfig](a ) . the volume change at 4 gpa in the dmft results ( thick solid lines ) in fig . [ prsfig](a ) is in decent agreement with that observed at room temperature . in the case of pr , the estimated @xmath6-u curve ( dashed line ) in fig . [ prsfig](b ) is in good agreement with experiment over all but the lowest observed range of this phase . this raises the question of whether the structural energy difference may be critical to the pr collapse . indeed , while we find a softening in the pr equation of state for the fcc structure due to the contribution in fig . [ ecorfig](b ) , there is no convincing evidence that this drives the bulk modulus negative in the vicinity of the collapse . finally , we turn to the discrepancies between theory and experiment in figs . [ etotcefig][prsfig ] . the fact that the dmft results for ce in fig . [ etotcefig ] are too high at smaller volumes in comparison with experiment , those for pr in fig . [ etotprfig ] are reasonable , while those for nd in fig . [ etotndfig ] are somewhat low is suggestive of the intraatomic exchange interaction , @xmath244 , which has been omitted in the present work . in the strongly localized regime including @xmath245 , this exchange interaction should shift the energy by @xmath121 , @xmath246 , and @xmath247 for ce , pr , and nd , respectively . at small volumes , the itinerant expectation would be @xmath248 taking @xmath249 . if the total energies are shifted to @xmath121 at @xmath220 as done in these figures , this would suggest shifts of @xmath250 , @xmath251 , and @xmath252 ev at @xmath253@xmath179/atom for ce , pr , and nd , respectively , taking values of @xmath57 from fig . [ nffig ] and @xmath254ev . this would imply completely itinerant states at @xmath255@xmath179/atom which is unlikely especially for nd . indeed , polarized hf calculations suggest exchange corrections at this volume of @xmath256 , @xmath257 , and @xmath258ev,@xcite respectively , although this also is only an estimate . nevertheless , these estimates do suggest that intraatomic exchange would significantly improve the present comparisons between theory and experiment . unfortunately , a rigorous inclusion of the exchange interaction , one that would also yield the the correct hund s rules values for the pr and nd local moments , would require dmft calculations for the full @xmath7-@xmath7 coulomb interaction including all four slater integrals and the non - density - density exchange and pair hopping terms . this has been a challenge for qmc , although there is recent progress.@xcite similarly , implementing mutual self - consistency between the lda and dmft parts of the calculations , as has been advocated,@xcite would be prohibitively expense with the present qmc implementation of the self - energy . both improvements as well as dmft calculations of the structural energy differences could be considered with faster although less rigorous approaches to the self energy.@xcite a better understanding of the electron - correlation driven volume collapse transitions in the compressed lanthanides may come from putting this behavior in context of their extended evolution from localized to itinerant character . to this end , the present paper has reported calculations for compressed ce ( @xmath0 ) , pr ( @xmath1 ) , and nd ( @xmath2 ) using a combination of the local density approximation ( lda ) and dynamical mean field theory ( dmft ) , so called lda+dmft . results for the @xmath3 moment @xmath259 , spectra , correlation energy , and equation of state among other quantities have been presented over a wide range of volume at a temperature of 632 k. this temperature is the lowest feasible with the present quantum monte carlo ( qmc ) implementation of the self energy , yet is still reasonably close to the low - temperature limit.@xcite while a face - centered cubic ( fcc ) structure was assumed , lda estimates of the important structural energy differences are significantly smaller than the relevant contributions to the correlation energy which may then be described as leading order volume - dependent effects . we find the three lanthanides to remain rather strongly localized under compression from ambient conditions up through the observed stability fields of the fcc ( @xmath4 ce ) phases , in the sense that the @xmath3 moments are close to the hund s rules values , there are fully formed hubbard sidebands which are themselves a signature of the local moments , and little @xmath3 spectral weight lying in between at the fermi level . subsequent compression brings about significant deviation of the moments from the hund s rules values , growth of @xmath3 spectral weight at the fermi level ( the kondo resonance ) at the expense of the hubbard sidebands , an associated softening in the total energy , and quenching of the spin orbit given that the kondo peak is of mixed-@xmath12 character in contrast to the predominantly @xmath5 lower hubbard band . the most dramatic evolution in these signatures is seen to coincide with the two - phase region of the @xmath4@xmath6 phase transition in the case of ce , consistent with earlier work,@xcite and in agreement with the kondo volume - collapse scenario.@xcite for pr and nd , on the other hand , these signatures change most rapidly over the volume range where the distorted fcc ( dfcc ) structure is experimentally observed to be stable , suggesting that this phase is transitional and not part of the localized trivalent lanthanide sequence . only on subsequent compression is pr experimentally observed to undergo a collapse from the dfcc phase to an @xmath6-u structure , while nd passes from dfcc to @xmath6-u through two other low symmetry phases without any substantial volume changes . due to the increasing but incompletely screened nuclear charge , the lanthanides shift towards the localized limit for larger atomic number , and we see a similar off - set to smaller volume of the above mentioned signatures from ce to pr , and then nd . in particular , a softening contribution to the total energy associated with growth of the kondo resonance must compete with the remaining contributions which become ever more dominant due to a steadily increasing bulk modulus as volume is reduced . if a region of low bulk modulus were to favor larger volume changes in structural transitions , then this would be qualitatively consistent with the observed sequence 15% ( ce ) , 9% ( pr ) , and none ( nd ) . this speculation is apparently contradicted by the 5% and 6% collapse transitions in gd ( @xmath260 ) and dy ( @xmath261 ) , respectively . however , these lanthanides correspond to filling a different spin - orbit subshell , which may significantly complicate matters given the profound manner in which spin - orbit is involved . there are still some correlation effects evident at even the smallest volumes considered here , such as residual hubbard side bands in the @xmath3 spectra and moments whose values have evolved only @xmath262% of the way from the strongly localized hund s rules values to those characteristic of totally uncorrelated electrons . nevertheless , standard paramagnetic lda does quite well within the bct phase of ce and the @xmath6-u phases of pr and nd , and so for all practical purposes these phases appear weakly correlated enough . the success of lda calculations for the structural parameters in these phases for ce as well as the light actinides is also well known.@xcite this work has also included a detailed quantitative comparison between the present lda+dmft results and experiment for the total energies and pressures of ce , pr , and nd . the comparison is encouraging , and serves to corroborate such theoretical observations as the systematic offset of kondo - like correlation signatures to smaller volume for increasing atomic number in these three lanthanides . there are also clear deficiencies , most notably the need to include intraatomic exchange , ideally in its full rotationally invariant form to enable rigorous calculation of the hund s rules moments in general cases . the need for accurate lda+dmft structural energy differences is also apparent . these improvements as well as making the lda and dmft parts mutually self - consistent will require a self - energy treatment that is both more precise and considerably less expensive , which unfortunately is likely to require giving up some of the rigor of the present qmc approach . this work was performed under the auspices of the u.s . department of energy by the university of california , lawrence livermore national laboratory under contract no . w-7405-eng-48 . the author gratefully acknowledges extended interactions with stewardship science academic alliances collaborators at u.c . davis ( doe / nnsa grant na00071 ) , especially r.t scalettar . he has benefited from conversations with k. held , and with members of the doe / bes funded computational materials science network cooperative research team on `` predictive capability for strongly correlated systems . '' a modification of the qmc code of ref . d ) was used in the present work . s. heathman , r. g. haire , t. le bihan , a. lindbaum , k. litfin , y. mresse , and h. libotte , phys . lett . * 85 * , 2961 ( 2000 ) ; a. lindbaum , s. heathman , k. litfin , y. mresse , r. g. haire , t. le bihan , and h. libotte , phys . b * 63 * , 214101 ( 2001 ) . for a tutorial see , k. held , i.a . nekrasov , g. keller , v. eyert , n. blmer , a.k . mcmahan , r.t . scalettar , t. pruschke , v.i . anisimov , and d. vollhardt , in _ quantum simulations of complex many - body systems : from theory to algorithms _ , eds . j. grotendorst , d. marx and a. muramatsu , nic series volume 10 ( nic directors , forschungszentrum jlich , 2002 ) , p. 175 - 209 ; cond - mat/0112079 . a rapid drop in entropy across the @xmath4@xmath6 collapse in ce may be viewed via @xmath263 as arising from a specific heat @xmath264 peak which lies below , e.g. , @xmath265k on the larger - volume @xmath4 side , but moves above @xmath266k on the smaller - volume @xmath6 side . the finite-@xmath14 total energy , @xmath267 , should then show a concomitant softening near the collapse , possibly even a region of negative curvature , even though it is the free energy which dictates the transition , and regardless of whether the @xmath13 energy @xmath268 is featureless . in the full cubic point group , @xmath31 would be characterized by five unique diagonal matrix elements corresponding to the irreducible representations @xmath269 ( @xmath15 ) , @xmath270 ( @xmath176 & @xmath42 ) , @xmath271 ( @xmath176 & @xmath42 ) , with unique off - diagonal couplings between the two @xmath270 and between the two @xmath271 . m. s. s. brooks , b. johansson , and h. l. skriver , in _ handbook on the physics and chemistry of the actinides _ , edited by a. j. freeman and g. h. lander ( north - holland , amsterdam , 1984 ) , vol . 1 , p. 153 . the hubbard @xmath9 values for ce , pr , and nd are plotted in ref . , and , e.g. , for pr are @xmath272 , @xmath273 , and @xmath274 ev for @xmath275 @xmath276 , @xmath277 , and @xmath278 @xmath179/atom , respectively . to test exchange , we performed fully polarized hf calculations with and without the slater integrals @xmath280 for @xmath281 , using @xmath282 , and @xmath283 [ g. h. dieke , _ spectra and energy levels of rare earth ions in crystals _ , edited by h. m. crosswhite and h. crosswhite ( interscience , new york , 1968 ) , p. 63 . ] . we identified the monopole integral @xmath284 with the lda constrained occupation result @xmath9 , approximated dieke s result for @xmath285 by ( @xmath286 for integer @xmath57 , and used the volume - dependent @xmath9 to scale all parameters .

it has been proposed that the first , intermediate - mass ( ) black holes might form through direct collapse of unpolluted gas in atomic - cooling halos exposed to a strong lyman - werner ( lw ) or near - infrared ( nir ) radiation . as these systems are expected to be compton - thick , photons above 13.6 ev are largely absorbed and re - processed into lower energy bands . it follows that direct collapse black holes ( dcbhs ) are very bright in the lw / nir bands , typically outshining small high - redshift galaxies by more than 10 times . once the first dcbhs form , they then trigger a runaway process of further dcbh formation , producing a sudden rise in their cosmic mass density . the universe enters the dcbh era " at when a large fraction of atomic - cooling halos are experiencing dcbh formation . by combining the clustering properties of the radiation sources with monte carlo simulations we show that in this scenario the dcbh mass density rises from mpc at to the peak value mpc at in our fiducial model . however , the abundance of _ active _ ( accreting ) dcbhs drops after , as gas in the potential formation sites ( unpolluted halos with virial temperature slightly above k ) is photoevaporated . this effect almost completely suppresses dcbh formation after . the dcbh formation era lasts only myr , but it might crucially provide the seeds of the supermassive black holes ( smbhs ) powering quasars .

1402.5675

the formation process of the intermediate - mass black holes ( bhs ) through direct collapse has been investigated by numerical and theoretical studies @xcite . such bhs , due to their much larger mass than the bhs form after the death of pop iii stars ( typically several tens @xmath3 , see @xcite ) , are good candidates of the seeds of the supermassive black holes ( smbhs ) which were observed at @xmath11 @xcite . in this direct collapse black hole ( dcbh ) scenario , the formation of a dcbh requires a host halo : ( a ) with virial temperature above @xmath8 k ; ( b ) exposed to strong lyman - werner ( lw , photons with energy between 11.2 ev and 13.6 ev ) radiation or near - infrared ( nir ) radiation with energy above @xmath12 ev @xcite . under these conditions , the @xmath13 can no longer form and the gas in this halo can only cool through ly@xmath14 emission , leading to a quasi - isothermal contraction without fragmentation and finally allowing a bh seed with mass @xmath15 to form at the center of the halo . if such dcbhs largely formed at @xmath16 and are all compton - thick , with a typical mass @xmath17 and typical accretion time scale @xmath18myr , they might be excellent candidate sources of the fluctuations of the source - subtracted near infrared background ( nirb ) @xcite which is hard attributed to normal galaxies @xcite . as proposed in @xcite , dcbhs also naturally generate a cosmic x - ray background ( cxb ) that is correlated with the nirb , which is observed by @xcite . the minimum lw intensity necessary to sustain the formation of the dcbh has been somewhat debated in recent literatures . early studies favored values as high as @xmath19 ( in units of @xmath20erg s@xmath21@xmath22hz@xmath21sr@xmath21 ) for a @xmath8 k blackbody spectrum , which represents a typical spectrum of pop ii stars ; and @xmath23 for a @xmath24 k blackbody spectrum , which characterizes metal - free stars @xcite . the critical intensity for the @xmath8 k blackbody spectrum is lower because in this case , the main mechanism that quenches the @xmath13 formation is not to directly dissociate the @xmath13 themselves , but to dissociate the @xmath25 @xcite the necessary intermediary for the @xmath13 formation through the nir radiation above @xmath12 ev . however , a more detailed work by @xcite revised the value downward : e.g. for a @xmath8 k blackbody spectrum , it is @xmath26 . the critical lw intensity could be even lower if a more refined treatment of self - shielding factor is implemented ( @xcite , z. haiman , personal communication ) . usually the probability for a halo to be embedded in a radiation field with super - critical lw intensity cooling is suppressed due to the direct dissociation of @xmath13 , or super - critical nir intensity which can be characterized by the intensity at lw band when the spectrum is already known in case the @xmath27 dissociation plays the main role . ] is very small if normal galaxies and/or pop iii stars are the only sources of lw photons @xcite ; in this case dcbhs are very rare . however , @xcite pointed out that a compton - thick bh could be very luminous below 13.6 ev due to the reprocessed nebular emission . this means that , once the first dcbhs form , they could ignite a runaway process favoring the formation of more dcbhs and leading to a steep rise in the dcbh abundance ; we dub this well - defined period during cosmic evolution as the `` dcbh era '' , lasting for about 150 myr . the probability that a halo sees a lw intensity above the critical value can be calculated by monte carlo simulations @xcite , in which the lw radiation from previously formed dcbhs is explicitly taken into account . with this probability , and the probability of a halo being unpolluted , we get the formation rate of the dcbh and build the picture of the rise process of its abundance . the process comes to an end when the host atomic - cooling halos ( @xmath28 k ) are rapidly photoevaporated by ionizing photons . photoevaporation drastically reduces the dcbh formation rate and almost completely quenches their formation for @xmath29 . this is the basic idea of this work . we aim at building a solid picture of the rapid rise ( and fall ) of the dcbh formation rate during the dcbh era . the layout of this paper , @xmath30 and @xmath31 to denote the mass of halos with @xmath32 k , @xmath8 k and @xmath33 k respectively at the corresponding redshift . the cosmological parameters are the same as used in @xcite : @xmath34=0.26 , @xmath35=0.74 , @xmath36=0.73 , @xmath37=0.041 , @xmath38 and @xmath39=0.8 . we use the the transfer function in @xcite . ] is as follows : in sec . [ methods ] we introduce the clustering of the spatial distribution of halos at high redshifts , the lw luminosity of these sources and the various mechanisms that influence the formation of dcbhs . we give results in sec . [ results ] , the conclusions are presented in sec . [ conclusions ] . a list of the mostly used symbols is given in tab . [ symbols ] . the calculation of the probability that a halo sees a super - critical lw intensity is introduced in appendix [ appa ] . ll symbol & definition + & + @xmath40 & + @xmath41 & + @xmath42 & + @xmath43 & + @xmath44 & + @xmath45 & + @xmath46 & + @xmath47 & + @xmath48 & + @xmath49 & + @xmath50 & + @xmath51 & + @xmath52 & + @xmath53 & + @xmath54 & + @xmath55 & + @xmath56 & + @xmath57 & + @xmath58 & + + @xmath59 & + + @xmath60 & + + [ symbols ] according to the definition of the two - point correlation function " , for a given halo of mass @xmath61 the number of surrounding halos , @xmath62 , with mass between [ @xmath63 , @xmath64 and at a proper distance in [ @xmath65 , @xmath66 is @xmath67 where @xmath68 is the halo mass function @xcite , @xmath69 is the two - point correlation function for two halos with mass @xmath61 and @xmath63 respectively and separated by a proper distance @xmath65 at redshift @xmath70 . to compute @xmath69 we use the formulae obtained in @xcite ( see also @xcite ) ; these formulae , based on the excursion set formalism , include the non - linear clustering of halos . we set @xmath71 when @xmath65 is smaller than the sum of the virial radii of the two halos . , averaged over surrounding halos in different mass ranges at @xmath72 for a central halo with @xmath73 , as a function of physical distance @xmath65 in units of the central halo virial radius . _ bottom _ : fractional contribution of surrounding sources at distance @xmath74 to the radiation intensity as seen by the central halo ( eq . [ fj ] ) . in this figure we adopt @xmath75 , where @xmath76 is the virial radius of the central halo at the corresponding redshift . ] the @xmath69 averaged over all surrounding halos in the mass range [ @xmath77 , @xmath78 is @xmath79 where @xmath80 is the mean number density of halos between [ @xmath77 , @xmath78 . for a reference halo of mass @xmath73 , we show @xmath81 as a function of @xmath65 at @xmath82 and @xmath83 in the upper panel of fig . we consider separately the mass range [ @xmath84 , @xmath30 ] , representing minihalos harboring pop iii stars , and @xmath86 , corresponding to the hosts of normal galaxies or dcbhs . at high redshifts , halos with mass above @xmath30 are still rare ( @xmath87 mpc@xmath4 at @xmath82 ) . however , such halos are strongly clustered , i.e. they are likely to be surrounded by close neighbors . for example , as shown in the upper panel of fig . [ cfl ] , at @xmath82 at a distance @xmath88 the central halo virial radius , @xmath76 , the number density of surrounding halos with mass between [ @xmath84 , @xmath30 ] is already @xmath89 the mean cosmic density . for more massive surrounding halos with @xmath90 the excess is further enhanced by two orders of magnitude . it is natural then to expect that , as a result of the local halo density enhancement , the radiation intensity as seen by the central halo overwhelmingly exceeds the mean cosmic background intensity . as an estimate , we can consider that the contribution of surrounding halos to the specific intensity of the radiation as seen by the central halo is @xmath91 the fractional contribution from all surrounding sources in the mass range [ @xmath77 , @xmath78 and with distance @xmath74 is @xmath92 for @xmath73 , we plot @xmath93 as a function of @xmath65 at @xmath82 and @xmath83 in the bottom panel of fig . the results shown in the bottom panel are for the same mass ranges as for the upper one with the same line styles.the transition from the `` close neighbors '' dominating regime ( note however that the real number of close neighbors fluctuates considerably around the mean number given by @xmath94 ) to the one dominated by `` distant sources '' as redshift decreases is evident . considering the radiation received by a central @xmath73 halo at @xmath82 from halos above @xmath30 at a distance @xmath95 ( roughly corresponding to the mean free path of lw photons set by ly@xmath96 absorption ) , more than 90% of it comes from close neighbors within @xmath97 , as at such high redshift the background sources are still rare . even considering only sources in the lower mass range [ @xmath84 , @xmath30 ] , this fraction still reaches about 60% . however , at @xmath83 the fraction decreases to about 10% and 1% if surrounding sources are in above two mass ranges respectively . the highly clustered spatial distribution of halos helps to build a very inhomogeneous radiation field at high redshifts . the radiation intensity as seen by some halos is much higher than the mean cosmic background intensity . this fact creates extremely favorable ( although short lasting ) conditions for triggering the formation of the dcbh by the strong lw radiation from the close neighbors . when calculating the lw radiation as seen by atomic - cooling halos with mass @xmath98 and assessing whether dcbhs could form therein , we consider two kinds of sources contributing to the lw radiation : dcbhs and normal galaxies ( stars in halos with mass @xmath98 ) . their lw luminosities are denoted by @xmath99 and @xmath100 respectively . in principle it would be necessary to include also the contribution from pop iii stars in surrounding minihalos ( @xmath101 ) when computing the impinging radiation spectrum / intensity . however , for the following reason it is unnecessary here . the radiation of pop iii stars is characterized by a hard spectrum3 and 10 ev , i.e. , the range probed by near - infrared observations on earth , see @xcite . ] that easily prevents @xmath13 cooling by dissociating these molecules directly and eventually suppresses the formation of pop iii stars in minihalos ; while we include radiation from pop iii stars to calculate the lw feedback on the formation of pop iii stars themselves ( see sec . [ genetic ] ) , its role is irrelevant in triggering the formation of dcbhs due to requiring a very high lw critical intensity , as found by @xcite who explored a wide range of conditions . therefore we simply ignore the radiation from pop iii stars here . the spectrum of both normal galaxies and dcbhs is softer than the pop iii stars spectrum , the dissociation of @xmath27 by nir photons plays a significant role in this case , even we still use the radiation intensity at the lw band as a criterion , it should be understood as the effects of nir photons characterized by the lw intensity , as the spectrum is already known . regarding the lw luminosity of the dcbh , we use the spectral energy distribution ( sed ) of compton - thick bhs derived in @xcite ( see the bottom panel of fig . 1 in that paper ) . the luminosity depends on the bh mass , @xmath102 , which grows following the law @xmath103 , where @xmath104 is the mass of the bh seed , @xmath105 myr is the eddington time scale @xcite . at any given time , the _ active _ dcbh population is composed by bhs formed at a time @xmath106 before ; then the mass distribution follows @xmath107 and @xmath108 if @xmath109 ; here @xmath110 , @xmath111 is the dcbh formation rate @xmath112 before the considered time . throughout this paper we use @xmath113 and @xmath114 myr @xcite . for star - forming , normal galaxies , their lw luminosity is computed using starburst99 @xcite . in the instantaneous burst " mode ( which is more appropriate for star formation in dwarf galaxies ) , for a salpeter stellar initial mass function ( imf ) @xcite in the mass range @xmath115 and metallicity @xmath116 , more than 60% of the lw photons are emitted within 10 myr from the burst ; the mean emission rate is @xmath117 s@xmath118 , and 3600 lw photons are produced per baryon into stars during this time interval . therefore the lw luminosity of a halo with mass @xmath63 is @xmath119 , \label{lwgal}\end{aligned}\ ] ] where we adopt a mean energy @xmath120 ev and bandwidth @xmath121 hz for lw photons . as most of lw photons are produced in 10 myr , only a fraction @xmath122 of galaxies are actively producing lw photons , where @xmath123 is the collapse fraction in halos above @xmath30 , and @xmath124 the increment of this quantity in the past 10 myr . similarly to eq . ( [ lwgal ] ) , from the starburst99 spectrum template we also get the mapping between the galaxy luminosity at @xmath125 and the halo mass . combining these formulae with the halo mass function we predict the galaxy luminosity function ( lf ) at any redshift . we calibrate the star formation efficiency of normal galaxies , @xmath47 , by comparing the predicted lf to the observed one at @xmath126 and observed upper limit at @xmath127 @xcite , finally fixing the star formation efficiency to @xmath128 . some of these normal galaxies might harbor pop iii stars @xcite , and in principle they should have a slightly different spectrum . however , we do not make such distinction and use the mean lw luminosity eq . ( [ lwgal ] ) for all star - forming normal galaxies . this appears a reasonable assumption as pop iii stars have short lifetimes , rapidly pollute their hosts , inducing a transition to pop ii star formation ; the galaxy spectrum becomes practically indistinguishable from standard pop ii galaxies @xcite . to gain a quantitative estimate of the relative importance of normal galaxies and dcbhs , we provide some reference numbers in the following . at @xmath129 , the mean lw luminosity of a star - forming normal galaxy in a host with mass @xmath30 is @xmath130erg s@xmath21hz@xmath21 . for comparison , the lw luminosity of a @xmath131 dcbh is @xmath132erg s@xmath21hz@xmath21 . hence , at this epoch , a typical compton - thick dcbh is more than 50 times brighter than the smallest normal galaxies ( these small galaxies , however , provide most of the integrated light as they are much more numerous ) . dcbhs then play a crucial role in prompting further formation of dcbhs in other halos . ( [ dn ] ) gives the mean number of surrounding halos . to get the probability distribution of the lw specific intensity as seen by the central halo , we follow @xcite and use monte carlo simulations . the details of the method are given in appendix [ appa ] . by generating a large set of monte carlo realizations , we directly get the probability that a central halo sees a super - critical lw intensity . we denote this probability by @xmath56 . as already mentioned , for an atomic - cooling halo , being immersed in an intense lw radiation field represents a necessary but not sufficient condition to allow the formation of a dcbh . a second requirement is that the gas of the candidate dcbh host does not contain metals , which - similar to @xmath13 cooling - would provide the required radiative energy loss to trigger gas fragmentation and star formation . halos can be enriched by heavy elements in two independent manners : genetically ( i.e. inheriting the heavy elements present in their progenitors ) or through winds from neighboring galaxies . in this section , we study the probability for a halo with @xmath133 k to be enriched by metals from either its progenitors or neighbors . to study this enrichment mode , we need to understand in detail how metals are propagated through different hierarchical halo generations , starting from the very first pop iii stars formed in minihalos . in the presence of lw radiation , pop iii stars can only form in metal - free minihalos where the gas has enough time to cool and collapse before @xmath13 is dissociated by this radiation . as the cooling and collapse timescales decrease with halo mass , for a given @xmath134 pop iii stars can only form in minihalos above a critical mass @xmath50 . this critical mass , as a function of lw intensity , has been obtained via numerical simulations by @xcite and more recently confirmed by @xcite and @xcite . a widely used fitting formula is @xmath135,\ ] ] where @xmath134 is in units of @xmath20erg s@xmath21@xmath22hz@xmath21sr@xmath21 . as there is no redshift dependence in the above formula , following @xcite , we rewrite it as @xmath136 , \label{mcritpop3}\ ] ] where @xmath137 is the critical mass in the absence of lw feedback . we use @xmath138 , where @xmath139 is the minimum virial temperature above which @xmath13 cooling becomes efficient . usually @xmath140 k @xcite , so we adopt a value of @xmath141 k in this work . for simplicity , when computing the @xmath50 by eq . ( [ mcritpop3 ] ) , we only use the uniform background lw intensity . this is a conservative assumption as spatial fluctuations from neighboring sources might be effective in suppressing pop iii star formation in even more massive minihalos , and further increase the formation probability of dcbhs @xcite . the lw background is the sum of contributions from pop iii stars , star - forming normal galaxies and active dcbhs , @xmath142 we assume that pop iii stars follow a salpeter imf in the mass range 1 - 100 @xmath3 and compute the emission rate of lw photons in a pop iii `` instantaneous burst '' mode ( more appropriate for the very tiny minihalos ) according to . we find that in the instantaneous burst " mode most of lw photons are produced in a time interval @xmath143 yr after the burst ; the mean photon emission rate is @xmath144 photons s@xmath145 . repeating the calculation in eq . ( [ lwgal ] ) , the lw luminosity ( per stellar mass ) of pop iii stars is then @xmath146erg s@xmath21hz@xmath118 . after the virialization of a minihalo , it will take some more time , @xmath147 , for the gas to cool and collapse ; then pop iii stars form and emit most of their lw photons within @xmath148 . the mass density of pop iii stars emitting photons at redshift @xmath70 is @xmath149 where @xmath48 is the star formation efficiency of pop iii stars , @xmath150 for @xmath151 and equals zero otherwise . the timescale for gas to cool and collapse is the maximum between the cooling and free - fall timescales of a halo , @xmath152 . \label{tdelay}\ ] ] for a halo with mass @xmath63 at redshift @xmath153 , the @xmath13 cooling timescale for gas enclosed in a radius @xmath154 is @xmath155 where @xmath156 is the boltzmann constant and @xmath157 is the mean number density of the enclosed gas . we use the cooling function of @xmath13 , @xmath158 , given by , and always use a maximum @xmath13 fraction of a halo @xmath159 in @xcite , which is independent of @xmath154 . then the mean cooling timescale weighted by the mass of gas enclosed within @xmath154 , @xmath160 is used as the cooling timescale of this halo , where @xmath161 is the enclosed gas mass derived from the gas profile @xcite . analogously , the average of the free - fall timescale weighted by the enclosed dark matter mass ( eq . ( 21 ) in @xcite ) is used as the free - fall time scale of this halo . we checked that for minihalos with virial temperature @xmath162 k , the @xmath147 determined above agrees well with the time interval from when the halo reaches virial temperature @xmath162 k to the runaway core collapse in simulations of @xcite ( panel _ c _ in their fig . 1 and fig . 2 ) . with the mass density of pop iii stars expressed by eq . ( [ rho_pop3 ] ) we therefore have their contribution to the lw background radiation @xmath163 where @xmath164 is taken to be equal to the mean free path of lw photons set by ly@xmath96 absorption , i.e. the proper distance from @xmath70 to @xmath165 . the lw radiation contributed by star - forming normal galaxies is @xmath166 \frac{\delta f_{\rm coll}}{f_{\rm coll}}l_{\rm max } ; \end{aligned}\ ] ] and the radiation from active dcbhs is @xmath167l_{\rm max},\end{aligned}\ ] ] where @xmath168 is the mean lw luminosity of active bhs whose mass distribution is described by eq . ( [ dnbh ] ) . the fraction of active dcbhs , @xmath169 , is always defined as the number density of active dcbhs divided by the number density of halos with mass between @xmath30 and @xmath31 . substituting @xmath170 for @xmath134 in eq . ( [ mcritpop3 ] ) , one can determine the critical mass of minihalos for the formation of pop iii stars , @xmath50 , at redshift @xmath70 . as @xmath50 at @xmath70 depends on its value at previous epochs , in practice we need to solve such equation step by step , starting from the redshift at which the lw feedback is negligible , i.e. , the critical mass is @xmath171 . having obtained @xmath50 , for each halo we can calculate the probability that its most massive progenitors ( the earliest progenitors with a certain mass ) ever hosted pop iii stars . the maximum of these probabilities for all possible masses of the most massive progenitors is the probability of genetic enrichment @xcite . to get this probability , the distribution of the formation redshift of the most massive progenitors is needed . for a halo with mass @xmath172 at redshift @xmath173 , the probability that its most massive progenitor with @xmath63 formed before redshift @xmath70 is ( fitted by @xcite according to their simulations ) , @xmath174 where @xmath175 @xmath176 and @xmath177 finally we get the genetic probability via the following steps . for a progenitor of mass @xmath63 , using eq . ( [ tdelay ] ) , we determine a redshift @xmath178 defined as the maximum of @xmath179 and the redshift when @xmath180 , where @xmath181 is the redshift interval corresponding to a time @xmath182 before @xmath70 . progenitor with mass @xmath63 must have formed before @xmath178 , otherwise pop iii stars would fail to form before @xmath173 in halos with mass @xmath172 ( no enough time for the gas to cool and collapse ) . the formation probability of pop iii stars in progenitor of @xmath63 before @xmath178 is therefore @xmath183 . the final genetic enrichment probability of a host halo @xmath172 at redshift @xmath173 , @xmath184 , is the maximum probability obtained after scanning all possible @xmath185 values . the potential birth place of a dcbh might also be polluted by metals coming from neighboring galaxies . metals can be carried to large distances from their origin halos by galactic winds expanding at a speed @xmath186 . hence , the enrichment probability in the wind mode , @xmath57 , is just the probability for a halo that has neighboring galaxies within the radius @xmath187 ( due to the finite energy of supernovae , there is an upper limit @xmath188 comoving mpc on the propagation distance of the wind ; @xmath189 km / s for normal galaxies , see fig . 1 and fig . 2 in @xcite ) , where @xmath190 is the redshift at which the wind is first ignited . taking @xmath190 as the redshift at which the most massive progenitor of a halo reaches a mass , as long as they do not contain _ active _ dcbhs . even if we have assumed that only a fraction @xmath122 of them are star - forming , the remaining halos , including those that previously had active dcbhs , must have experienced star formation after the bh stop accreting gas , see @xcite . ] of @xmath30 , the probability distribution of @xmath190 is again calculated using eq . ( [ pf ] ) . the computation therefore is very similar to the @xmath56 and we carry it on together with that using the same monte carlo simulations . in addition to normal galaxies , minihalos may also eject metals into the environment if they experienced pop iii star formation . however , in this case the propagation distance of metals is small and if the minihalos are massive metals even can not escape the host halos @xcite . in addition , @xcite pointed out that the mixture of metals carried by galactic winds with gas in halos is not effective enough , while @xcite found that the metals are re - accreted by the host halos and can not significantly enrich the environment . we therefore ignore this contribution to the metal enrichment . we only consider the influence of metal enrichment on central halos with @xmath133 k , as the major mechanism quenching pop iii star formation in minihalos is lw feedback , fully captured by @xmath50 . we believe that this is a good approximation at @xmath70 25 , considering that the formation of pop iii stars is quenched quickly by dcbhs shortly thereafter according to our results . photoevaporation plays a significant role in suppressing galaxy / dcbh formation , particularly in smaller halos that could be fully photoevaporated by ionizing radiation . for halos with mass @xmath172 at @xmath173 , at any @xmath191 , the median mass of their most massive progenitors is obtained by taking @xmath192 in eq . ( [ pf ] ) @xcite . we use the evolution of the median of the most massive progenitors as a typical growth history of a halo with mass @xmath172 at @xmath173 . the evolution of the lw intensity as seen by this typical most massive progenitor is expressed by eq . ( [ jlw ] ) using @xmath193 in each cell , and replacing @xmath61 with this typical progenitor mass at each redshift . the ionization intensity ( at 13.6 ev ) as seen by a halo located in an ionized bubble is derived from the lw intensity of normal galaxies alone , as dcbhs are highly obscured at energies @xmath194 13.6 ev , and photoevaporation becomes significant only at later redshifts ( @xmath195 ) when the formation of pop iii stars is almost totally quenched in our models . if a halo resides outside ionized bubbles , the ionizing radiation intensity vanishes , since the optical depth of neutral gas to ionizing photons is very high . therefore , the ionizing intensity can be expressed as @xmath196 where @xmath197 is a normalization factor considering the combination of escape fraction and the absorption due to neutral absorbers in the igm . for normal galaxies , the spectral ratio @xmath198 . @xmath199 is the lw intensity corresponding to an ionizing field intensity set by ionization equilibrium : seeing an ionizing intensity larger than @xmath199 indicates that the halo must reside in an ionized bubble . if the ionizing radiation has a @xmath200 power - law shape @xcite , considering that the ionization cross - section has instead a @xmath201 form , the ionization intensity at ionization equilibrium is @xmath202 $ ] , where @xmath203 is the comoving hydrogen number density , @xmath204 is the case b recombination rate , @xmath205 is the planck constant , @xmath206 is the h - ionization cross - section at 13.6 ev , @xmath207 is the neutral hydrogen fraction at the edge of ionized bubbles for which we adopt a value @xmath208 . let us define the redshift at which a given halo is first engulfed by the expanding ionization front and embedded into the bubble as @xmath209 : this is the moment at which photoevaporation of the halo gas starts . we compute the fraction of the original baryons left in @xmath172 at @xmath173 using eq . ( 6 ) in @xcite . formulae in @xcite are fitted from results of simulations assuming constant ionization intensity ; however , in our calculation , from @xmath209 to @xmath173 , we find that the ionization intensity seen by the typical most massive progenitor increases by about two orders of magnitude , see the upper panel of fig . therefore we always use the @xmath210 seen by this progenitor at a median redshift @xmath211 as an effective intensity . we assume the dcbh could only form in halos whose mass is above @xmath30 , and have at least 20% leftover gas content . the net result of photoevaporation feedback is to boost the minimum mass , @xmath51 , of galaxy or dcbh host halos . at redshift @xmath70 , the number density of halos with mass between @xmath63 and @xmath212 is @xmath213 ; among these halos , a fraction @xmath214 ( @xmath215 is given by eq . ( [ pf ] ) ) have most massive progenitors with mass @xmath216 - @xmath30 at redshift @xmath153 - between @xmath153 and @xmath217 . the time interval between @xmath153 and @xmath70 is that required for the gas to collapse into a bh . such timescale has been estimated to be of several myr @xcite , therefore we use 5 myr here . some of these progenitors have already been polluted or were not exposed to a strong enough lw radiation , therefore bhs only form in a fraction @xmath218p_{\rm j - w}(m^\prime_{\rm t4},z^\prime)$ ] of them , where @xmath58 is the probability for a halo exposed to a super - critical lw radiation and not polluted by metals from neighboring galaxies , a quantity directly obtained from our monte carlo simulations . we solve for @xmath219 analytically as described in sec . [ genetic ] . the formation rate of dcbhs at @xmath70 is then @xmath220p_{\rm j - w}(m^\prime_{\rm t4},z^\prime ) \nonumber \\ & \times \int_{m_{\rm t4}}^{m_{\rm 2t4 } } dm\frac{dn}{dm}(z)\frac{dp}{dz^\prime}(z^\prime , m^\prime_{\rm t4}|m , z)\frac{dz^\prime}{dz}. \label{dnbhdz}\end{aligned}\ ] ] with eq . ( [ dnbhdz ] ) , the number density of _ active _ dcbhs at @xmath70 is then @xmath221 where @xmath222 is the redshift corresponding to a time @xmath223 before @xmath70 ( dcbhs formed earlier than @xmath222 have already exhausted / ejected their gaseous fuel and therefore faded away at @xmath70 ) . the combination of eqs . ( [ dnbhdz ] ) and ( [ nbh ] ) is an integro - differential equation and is integrated step by step . we start from a high redshift , e.g. 50 , when the number of dcbhs is negligible , so the initial @xmath224 is set to be zero . after each step of the integration , we update @xmath224 . note both @xmath50 and @xmath51 depend on the evolution history of @xmath224 before this redshift , so we also update them accordingly . we then update @xmath55 by using the new @xmath50 , and @xmath58 by using the new @xmath224 in the monte carlo realizations . we then finally get the evolution of the number density of dcbhs . the formation rate in eq . ( [ dnbhdz ] ) also allows to derive the mass density of active dcbhs : @xmath225 where @xmath112 is the time interval between @xmath222 and @xmath153 , and the cumulative ( i.e. active + dead ) bh mass density @xmath226 } \frac{dn_{\rm bh}}{dz^\prime}dz^\prime.\ ] ] .parameters in different models , @xmath46 is in units of @xmath20erg s@xmath21@xmath22hz@xmath21sr@xmath21 , while @xmath227 is the minimum distance between two halos in units of the sum of the virial radii of them . parameters not listed in this table are the same for all models , e.g. , @xmath228 k , @xmath229 , @xmath230 km / s and @xmath231 comoving mpc . model @xmath232 is our fiducial model . [ cols="<,^,^,^,^,^,^,^",options="header " , ] . ] in fig . [ rho_bhz ] , we plot the evolution of the mass density ( left panel ) of active dcbhs , and the black hole cumulative mass density ( right ) for our models , whose parameters are listed in tab . [ table : parameters ] . in each panel , as a reference we add the best - fitting mass density evolution required to fit nirb observations @xcite , and the mass density range corresponding to @xmath233 uncertainties of the model free parameters in that paper . in all models , the adopted @xmath234 is consistent with the critical intensity given by @xcite for a @xmath8 k black - body spectrum , which is between 30 and 300 . to gain further insight into the prediction of the model , we plot in fig . [ probabilities ] the evolution of @xmath58 , @xmath56 , @xmath57 , @xmath55 and @xmath235 for a halo with mass @xmath30 in the fiducial model . we recall that the @xmath235 describes the formation probability of the dcbh in recently formed @xmath30 halos . let us start from the analysis of the fiducial model results . initially , all halos with mass @xmath30 are exposed to a lw intensity above @xmath46 , as at high redshift they represent rare , highly biased peaks of the density field ( dash - dotted line in fig . [ probabilities ] ) . however , dcbhs can hardly form in such systems as the neighboring galaxies providing the necessary lw illumination also eject metals , yielding @xmath236 and @xmath237 ( long and short - dashed lines , respectively ) . during cosmic evolution , these halos become less clustered and the metal enrichment probability by winds decreases considerably . as a result , at @xmath238 @xmath57 starts to drop and @xmath58 starts to rise . at this time some dcbhs start to form in halos containing pristine gas ; once formed , they provide extra lw photons in addition to normal galaxies . as the universe keeps expanding , eventually @xmath56 also starts to drop ( @xmath239 ) at a high rate , because at this time the previously formed dcbhs are still too few to provide a sufficiently large number of lw photons . as a result , the @xmath58 reaches a peak of @xmath240 at @xmath239 , then it drops again , reaching the trough @xmath241 at @xmath242 . the combined probability @xmath235 closely traces the @xmath58 trend albeit with a much smaller value as the majority of halos ( 90% ) are polluted by metals from their progenitors ( solid line ) . note that since @xmath243 and until @xmath244 , the genetic enrichment is the dominant metal enrichment ( and dcbh suppression ) mode . a simple analogy will help understanding the physical evolution at this point . a dcbh can be visioned as a flame , with the newly formed pristine gas halos representing dry wood . if enough dry wood is provided the flame will grow stronger and dry more other wood by itself , reaching a kind of self - supported regime ; alternatively , the flame will rapidly extinguish due to the lack of fuel supply . the period @xmath245 represents a dangerous stage for the formation of dcbhs , as initially a large fraction of @xmath90 halos is already polluted , leaving only a minority of pristine ones as candidate hosts of new dcbhs . if the drop of @xmath56 would continue beyond that epoch , the initial dcbh `` weak flame '' would be extinguished or remain very weak . this would be the result if the lw radiation from dcbhs themselves is not considered . dcbhs are about two orders of magnitude brighter than most of normal galaxies in the lw band . thus , even if they are rare at high redshifts , they dramatically impact cosmic structure formation in several ways . the first obvious effect is that the formation probability of pop iii stars in minihalos is reduced significantly and even quenched after @xmath246 . therefore the genetic enrichment probability is reduced ( i.e. minihalos are sterilized ) and the fraction of @xmath30 halos with pristine gas increases . secondly , lw radiation from previously formed dcbhs can trigger the formation of further dcbhs much more efficiently than normal galaxies . for example , at @xmath82 , the lw intensity of a normal galaxy in a host halo with mass @xmath30 drops to 30 at a distance @xmath247 kpc ( physical ) . the distance is only slightly larger than @xmath248 of the host halo , so it is only marginally possible for a single normal galaxy to trigger the formation of dcbhs at this redshift . on the other hand , for a bh with mass of @xmath131 , the corresponding distance is about 7 kpc . at the same redshift , in the volume within this distance , during the accretion timescale of a bh , about 50 new halos with mass between @xmath30 and @xmath31 form . it means that a single bh can ideally prompt the formation of about 50 new dcbhs ! in practice , though , this number is likely smaller because at such high redshift many of these halos are metal - enriched when they form . whatever its precise strength is , this positive feedback exerted by dcbhs on their own formation can hardly be overlooked . the final results depend on the competition between the negative factors ( the expansion of the universe and metal pollution ) and the above positive feedback . if the latter dominates , the formation probability of the dcbh will rise again . at @xmath249 , the abundance of dcbhs increases to a level that the trend of decreasing @xmath56 set by the negative factors can be reversed , @xmath56 and @xmath58 then start to rise , while around this time @xmath55 drops very fast . the rise of the dcbh population and the sharp decrease of the genetic enrichment are interlinked , of course . the universe then at @xmath250 enters the dcbh era , in which @xmath56 grows so rapidly due to positive feedback that essentially all newly formed , unpolluted @xmath30 halos become populated with active dcbhs , whose density reaches a peak of about @xmath251mpc@xmath4 at @xmath252 ( model @xmath232 in fig . [ rho_bhz ] , left panel ) . the dcbh era extends until @xmath9 , or about 150 myr . what causes the end of the dcbh era ? the answer is _ photoevaporation _ , which represents a third negative feedback - in addition to cosmic expansion and metal enrichment - that sets in at later stages . to understand this key point let us inspect the upper panel of fig . [ jion ] , where we show the ionizing background intensity , @xmath210 , as seen by the most massive progenitors of two halos of mass @xmath253 and @xmath254 at @xmath255 , respectively . we mark @xmath209 on each curve . we also plot ( bottom panel ) the minimum mass of halos in which dcbhs could still form even under the effects of photoevaporation . photoevaporation feedback does not affect the formation of dcbhs before @xmath256 , when even halos with mass @xmath30 are massive enough to hold more than 20% of their gas ( long - dashed curve in the bottom panel of fig . [ jion ] ) . however , after then dcbhs could only form in halos with @xmath257 , which increases steadily since @xmath256 . for example , at @xmath9 @xmath51 is about 1.8 times @xmath30 . the number of newly formed halos with @xmath258 which can hold at least 20% of their gas is reduced by about 70% . as photoevaporation feedback becomes effective , the abundance of dcbhs decreases ( after a time delay @xmath259 ) steadily until a sharp drop occurs at @xmath9 , when even halos as massive as @xmath31 can not prevent their gas from being almost completely photoevaporated . we stress that the above scenario is largely independent of the upper mass limit of dcbh host halos . adopting a higher mass cut would make the dcbh disappearance phase slightly more gentle as a few among the most massive host halos are able to retain their gas . considering also the uncertainties in the details of our photoevaporation model , it is clear that the detailed late evolution of the dcbh abundance deserves additional future work . ( solid ) and @xmath31 ( dashed ) at redshift 13 . we mark @xmath209 by crosses . _ bottom : _ critical mass of halos that can retain at least 20% of their gas against photoevaporation ( long - dashed ) , @xmath30 ( solid ) and @xmath31 ( dashed ) . ] besides the fiducial one , we have explored different models ( see tab . [ table : parameters ] ) to check the robustness of our conclusions . the evolution of mass density in these models is also plotted in fig . [ rho_bhz ] . the first variation is model @xmath260 without genetic enrichment . this is motivated by the possible suppression of pop iii star formation in progenitor minihalos exposed to a sufficiently large lw intensity @xmath261 erg s@xmath21@xmath22hz@xmath21sr@xmath21 , as proposed by @xcite . this is combined with a larger @xmath262 value . we confirm the rise of the dcbh population , which now occurs earlier and reaches a higher peak @xmath263 @xmath3mpc@xmath4 ( model @xmath260 in fig . [ rho_bhz ] ) . with model @xmath264 in fig . [ rho_bhz ] we check the sensitivity of the results to the assumed minimum interhalo distance . as shown by fig . [ cfl ] , at redshift 30 , if the sources of the radiation received by a central @xmath30 halo are halos with mass @xmath98 , @xmath265% of the radiation comes from neighbors within 10@xmath76 . however , close neighbors may merge with the central halo on a short timescale . to assess the impact of possible mergers on our conclusions , model @xmath264 ignores the contribution from close neighbors within a certain distance by adopting a larger @xmath227 , and uses the minimum @xmath46 given in @xcite . we posit @xmath266 equal to 2 times the sum of the virial radii of two halos . the basic change with respect to the fiducial model is that both @xmath56 and @xmath57 are reduced by the larger @xmath227 . however , as metal enrichment by galactic winds is more sensitive to the distance between halos , at early stages @xmath58 is somewhat higher than in the fiducial model . an even larger @xmath227 value would fail to produce the dcbh population rise after @xmath267 ; if genetic enrichment is turned off , though , the rise can occur even for @xmath268 . we have studied the formation and evolution of dcbhs at high ( @xmath70 13 ) redshifts . the most striking result is that we found that a well - defined dcbh formation era might occur in the early universe extending approximately for 150 myr from @xmath250 to @xmath9 . the intermediate masses ( @xmath269 ) of dcbhs make them excellent seeds for the subsequent rapid growth of the smbh observed at @xmath70 6 . in addition , such bhs could be very abundant in this era and might leave a clear signature in the observed fluctuations of the source - subtracted nirb @xcite . during such era the abundance of active dcbh grows , reaches a peak and eventually comes to a sudden halt when the required conditions for their formation disappear . these conditions include the availability of pristine gas in atomic - cooling halo ( @xmath28 k ) embedded in a sufficiently strong lw or nir radiation field . whether and for how long these conditions can be met during cosmic evolution depends on a complex network of negative and positive feedbacks that we have described in detail . a key result is that the ( so far neglected ) positive feedback of dcbh on their own formation fosters a rapid multiplication of the number of dcbhs . this is possible because dcbhs are likely compton - thick systems ( see @xcite ) , they are very bright in the rest - frame uv / nir band , typically outshining small high - redshift galaxies by more than 10 times ; in their surroundings the lw flux is therefore largely enhanced with respect to the level provided by galaxies , thus enabling additional dcbhs to form . dcbh formation spreads very rapidly , similarly to a flame in a haystack . their mass density raises from @xmath2 @xmath3mpc@xmath4 at @xmath270 to the peak value @xmath271 @xmath3mpc@xmath4 ( corresponding to a density parameter @xmath272 ) at @xmath7 in our fiducial model . the dcbh formation era however does not last very long . this is due to a combination of several negative feedback effects that we have discussed in sec . [ methods ] . in addition to the effect of a decreasing bias of dcbh host halos with time , metal pollution , either due to heavy elements inherited by the progenitor halos or carried by supernova winds coming from neighbors , is perhaps an obvious way to quench dcbh formation . however , we find that collectively these effects are not able to quench the dcbh widespread formation before @xmath273 . instead , the rapid dcbh birthrate drop is caused by photoevaporation of the gas from the gravitational well of the smallest atomic - cooling halos . this effect reduces the formation rate of the dcbh , almost completely suppressing their formation for @xmath29 . this is the reason why dcbh formation era lasts for a relatively short time interval during cosmic evolution . however , as they are compton - thick , their ionizing photons are fully reprocessed into lower energy radiation before escaping into the igm and , as such , they play no active role in driving reionization . the above scenario is robust against variations of the free parameters of the model , for which we have explored ( see sec . [ modvar ] ) , including the critical lw intensity threshold for dcbh formation , the strength of the genetic enrichment , and the influence of the assumed minimum distance between two halos . in spite of these , we do not observe substantial deviations from the evolutionary trend described by the fiducial case , witnessing a dcbh formation history firmly controlled by feedback(s ) action . our work is based on the most popular assumption that dcbh can only form under specific conditions , i.e. , the gas is metal - free and the @xmath13 formation is suppressed by strong external radiation . it is possible that even in the presence of metals or @xmath13 , if the accretion timescale is smaller than the fragmentation timescale , dcbh may still form @xcite ; particularly because the supersonic turbulence suppresses the fragmentation @xcite . albeit interesting , this scenario still needs to overcome a number of open questions that have been discussed in a recently published paper by two of the present authors @xcite . we can not exclude different dcbh formation channels with respect to the canonical one explored here , however , the photoevaporation of haloes with circular velocity 30 km / s , which is discussed in this paper , might in practice make these channels only have modest influence on our conclusions . we also note that purely forming dcbh in metal free , uv illuminated halos already produces a bh cosmic mass density comparable to the present - day one . we finally comment on a possibly intriguing implication of our model . the end of the dcbh era leaves behind a large number of `` fossil '' dcbh seeds ( @xmath274 ) . therefore to have sufficient number of seeds for smbhs is not a problem any more in this scenario ( see also @xcite ) . some of dcbhs are ready to be embedded in other galaxies and in smbhs . while only a small fraction of these bhs will continue to grow following merging events and possibly becoming smbhs , the majority of them will evolve passively . hence a large number of intermediate - mass black holes could be present in local universe , which has escaped so far detection . in the future it will be interesting to consider suitable strategies to find them . it is also tempting to speculate that dcbh gas devoid , low - mass dark matter halo hosts might be related with the large missing halo population predicted by @xmath275cdm models at @xmath276 . it is a pleasure to acknowledge m. dijkstra , s. salvadori and e. sobacchi for discussions and comments . we acknowledge financial support from prin miur 2010 - 2011 , project `` the chemical and dynamical evolution of the milky way and local group galaxies '' , prot . 2010ly5n2 t . by and xc also acknowledges the support of the nsfc grant 11073024 and the most project 863 grant 2012aa121701 . yx is supported by china postdoctoral science foundation and by the young researcher grant of national astronomical observatories , chinese academy of sciences . 75 natexlab#1#1 b. , davis a. j. , khochfar s. , natarajan p. , dunlop j. s. , 2013 , , 432 , 3438 b. , khochfar s. , johnson j. l. , neistein e. , dalla vecchia c. , livio m. , 2012 , , 425 , 2854 m. c. , 2010 , , 402 , 673 m. c. , rossi e. m. , armitage p. j. , 2008 , , 387 , 1649 m. c. , shlosman i. , 2009 , , 702 , l5 m. c. , volonteri m. , rees m. j. , 2006 , , 370 , 289 s. , mayer l. , callegari s. , 2014 , , 437 , 1576 r. j. et al . , 2011 , , 737 , 90 v. , loeb a. , 2003 , , 596 , 34 n. et al . , 2013 , , 769 , 68 r. , riquelme m. a. , 2008 , , 674 , 644 j .- h . , shlosman i. , begelman m. c. , 2013 , , 774 , 149 l. , kuhlen m. , shapiro p. r. , 2007 , , 665 , l85 a. , gong y. , smidt j. , santos m. g. , 2012 , , 756 , 92 a. et al . , 2012 , , 490 , 514 m. , haiman z. , mesinger a. , wyithe j. s. b. , 2008 , , 391 , 1961 d. j. , hu w. , 1998 , , 496 , 605 x. et al . , 2001 , , 122 , 2833 a. , haardt f. , salvaterra r. , 2013 , , 434 , 2600 a. , barkana r. , visbal e. , tseliakhovich d. , hirata c. m. , 2013 , , 432 , 2909 s. r. , loeb a. , 2003 , , 588 , 18 d. , palla f. , 1998 , , 335 , 403 l. , yoshida n. , abel t. , frenk c. s. , jenkins a. , springel v. , 2007 , , 378 , 449 c. , tormen g. , sheth r. k. , 2012 , , 422 , 185 k. , ricotti m. , kashlinsky a. , 2012 , , 752 , 113 t. , omukai k. , yoshida n. , yorke h. w. , 2011 , science , 334 , 1250 i. t. , scannapieco e. , martel h. , shapiro p. r. , 2003 , , 341 , 81 j. l. , whalen d. j. , fryer c. l. , li h. , 2012 , , 750 , 66 j. l. , whalen d. j. , li h. , holz d. e. , 2012 , arxiv e - prints arxiv:1211.0548 a. , arendt r. g. , ashby m. l. n. , fazio g. g. , mather j. , moseley s. h. , 2012 , , 753 , 63 a. , arendt r. g. , mather j. , moseley s. h. , 2005 , , 438 , 45 a. , arendt r. g. , mather j. , moseley s. h. , 2007 , , 654 , l5 s. m. , bullock j. s. , dekel a. , 2004 , , 354 , 292 m. r. , klein r. i. , mckee c. f. , 2007 , , 656 , 959 m. r. , mckee c. f. , 2005 , , 630 , 250 m. a. , schleicher d. r. g. , schmidt w. , niemeyer j. , 2013 , , 433 , 1607 m. a. , schleicher d. r. g. , schmidt w. , niemeyer j. c. , 2013 , , 436 , 2989 c. , ortiz otlvaro p. a. , bresolin f. , kudritzki r .- , lo faro b. , pauldrach a. w. a. , pettini m. , rix s. a. , 2010 , , 189 , 309 c. et al . , 1999 , , 123 , 3 g. , natarajan p. , 2006 , , 371 , 1813 m. e. , bryan g. l. , abel t. , 2001 , , 548 , 509 p. , ferrara a. , rees m. j. , 2001 , , 555 , 92 n. , sasaki s. , suto y. , 1998 , , 497 , 555 t. et al . , 2011 , , 742 , 124 l. , kazantzidis s. , escala a. , callegari s. , 2010 , , 466 , 1082 a. l. , gnedin o. y. , gnedin n. y. , zemp m. , 2013 , , 773 , 19 p. a. et al . , 2012 , , 745 , 110 s. p. , haiman z. , 2002 , , 569 , 558 k. , 2001 , , 546 , 635 b. w. , norman m. l. , 2008 , , 673 , 14 f. i. , di matteo t. , ciardi b. , 2007 , , 665 , 107 a. , ferrara a. , salvaterra r. , 2012 , , 422 , 1690 j. a. , haehnelt m. g. , 2009 , , 396 , 343 e. e. , 1955 , , 121 , 161 e. e. , 1964 , , 140 , 796 r. , ferrara a. , dayal p. , 2011 , , 414 , 847 e. , barkana r. , 2002 , , 571 , 585 d. , 2003 , , 397 , 527 c. , bryan g. l. , haiman z. , 2010 , , 402 , 1249 r. k. , mo h. j. , tormen g. , 2001 , , 323 , 1 r. k. , tormen g. , 1999 , , 308 , 119 e. , mesinger a. , 2013 , , 432 , l51 h. , kitayama t. , 2000 , , 317 , 175 m. , silk j. , rees m. j. , blanchard a. , abel t. , palla f. , 1997 , , 474 , 1 r. i. , eisenstein d. , fan x. , rieke m. , kennicutt r. c. , 2007 , , 657 , 669 m. , stiavelli m. , 2007 , , 667 , 38 m. , stiavelli m. , 2009 , , 694 , 879 g. a. , leitherer c. , 2005 , , 621 , 695 d. , van veelen b. , oshea b. w. , norman m. l. , 2008 , , 682 , 49 j. h. , abel t. , 2007 , , 671 , 1559 j. , haiman z. , 2012 , , 425 , l51 j. , haiman z. , bryan g. l. , 2011 , , 418 , 838 b. , ferrara a. , salvaterra r. , chen x. , 2013 , , 431 , 383 b. , ferrara a. , salvaterra r. , xu y. , chen x. , 2013 , , 433 , 1556 e. , rydberg c .- e . , schaerer d. , stlin g. , tuli m. , 2011 , , 740 , 13 . ( [ dn ] ) gives the mean number of surrounding halos . to get the probability distribution of the lw specific intensity as seen by the central halo , we follow @xcite , we generate a large number of monte carlo realizations by dividing the mass and distance ranges of the surrounding halos into @xmath277 bins . in the @xmath278-th bin with central mass and central distance ( @xmath279 ) , there are @xmath280 halos on average . the actual number in each bin for each realization , @xmath281 , follows the poisson distribution with this mean . regarding to the minimum distance between two halos , @xmath227 , a natural choice is the sum of their virial radii , but we also discuss models with higher minimum distance . the maximum distance is set by ly@xmath96 absorption , as described in sec . [ genetic ] . in each realization , the lw specific intensity as seen by the central halo is in the igm should be included , however , this effect is rather uncertain and complicated , we therefore have to ignore it in current work . ] @xmath282 where @xmath283 , the lw luminosity of the @xmath284-th source in the @xmath278-th bin , can take the value of one from @xmath285 , depending on @xmath286 . when photoevaporation feedback is allowed , both dcbhs and galaxies could only form in halos that are massive enough to retain enough gas ( we assume at least 20% in this paper ) . we use the same critical mass , @xmath51 , for both dcbh and galaxy formation , as discussed in sec . [ photoevaporation ] . @xmath283 is therefore determined as follows : * if @xmath287 , @xmath288 . * if @xmath289 and @xmath51 is smaller . using a halo mass corresponding to @xmath290 k as an upper limit gives similar results . in this case even when most of gas in halos below @xmath31 has already been evaporated by ionizing photons , a few dcbhs can still form in more massive halos . however , the actual difference is small compared with the uncertainties of photoevaporation models . ] than @xmath31 , for @xmath291 , we suppose that a fraction @xmath169 of the halos in this mass range contains active dcbhs , and for each @xmath284 we decide whether @xmath292 , or @xmath100 , or zero by the value of a randomly generated number , @xmath293 , following an uniform distribution : if @xmath294 , then @xmath292 ; if instead @xmath295 , we then generate a new and independent uniformly distributed random number @xmath296 . if @xmath297 , @xmath298 , otherwise @xmath299 . if @xmath300 , the hosts only hold normal galaxies , again we decide whether @xmath301 equals @xmath100 or zero checking if a third random number @xmath302 is smaller or larger than @xmath122 . * if @xmath289 and @xmath51 is larger than @xmath31 , we generate a random number @xmath303 ; if it is smaller than @xmath304 , @xmath298 , otherwise @xmath288 .

a sample of 123 radio sources that exhibit significant variations at 1.4 ghz on a seven year base - line has been created using first vla b - array data from 1995 and 2002 on a strip at near the south galactic cap . this sample spans the range of radio flux densities from to 1000 mjy . it presents both in size and radio flux density range a unique starting point for variability studies of galaxies and quasars harboring lower luminosity active galactic nuclei ( agn ) . we find , by comparing our variable and non - variable control samples to the sloan digital sky survey the following : 1 ) the quasar fraction of both the variable and non - variable samples declines as a function of declining radio flux density levels ; 2 ) our variable sample contains a consistently higher fraction of quasars than the non - variable control sample , irrespective of radio flux ; 3 ) the variable sources are almost twice as likely to be retrieved from the optical sdss data than the non - variable ones ; 4 ) based on relative numbers , we estimate that quasars are about five times more likely to harbor a variable radio source than are galaxies ; and 5 ) there does not appear to be any significant optical color offset between the two samples , even though the suggestive trend for sources to be bluer when variable has been detected before and may be real . this leads us to conclude that both radio variability and radio flux density levels , in combination with accurate optical information , are important discriminators in the study of ( radio ) variability of galaxies . the latter start to dominate the source counts belowmjy . in any case , variability appears to be an intrinsic property of radio sources , and is not limited to quasars . radio variability at low flux density levels may offer a unique tool in agn unification studies .

astro-ph0402117

all is flux , nothing stays still ; nothing endures but change . this seems to apply particularly well to extragalactic radio sources . important advances in our understanding of radio source intrinsic and extrinsic variability mechanisms have been made by studying their amplitudes , characteristic time - scales , and cross - correlations with observations at other wavelengths . radio variability can be classified in three broad classes ( padrielli et al . 1987 ) , based on their frequency dependence : 1 ) variability only occurs at low frequencies ( @xmath3ghz ) , 2 ) variability occurs both at low and high ( @xmath4ghz ) frequencies , but is not correlated in time , and 3 ) low and high frequency variability , but the variations are correlated ( although they need not occur simultaneously ) . this classification also conveniently separates the underlying variability mechanisms into `` most likely extrinsic '' ( low frequency variations only ) , and `` most likely intrinsic '' ( correlated broadband variability ) . the accepted model for the extrinsic variability of radio sources is refractive interstellar scintillation ( iss ) , which predominantly induces variations at low frequencies ( @xmath5ghz ) , with low amplitudes ( @xmath6 ) , and with typical timescales of a few days @xcite . variations of this type have been detected in multi - frequency studies ( e.g. , mitchell et al . 1994 ; lazio et al . perhaps the strongest arguments for the external origin of these variations are correlations with galactic latitude ( e.g. , gregorini et al . 1986 ; gaensler et al . 2000 ) , and an annual modulation due to earth s orbital motion ( e.g. , lainela & valtaoja 1993 ; bondi et al . 1994 ; 1996 ) . low - frequency variations with much longer timescales ( @xmath7 years ) , and/or larger amplitudes are thought to be source intrinsic , however ( e.g. , rs & machalski 1990 or , more recently , gaensler & hunstead 2000 ) . broadband radio variability , in which the variations usually appear first and more strongly at higher frequencies , are thought to be intrinsic to the radio source . a model describing this behavior , in which shocks propagate along the radio jets , has been proposed by @xcite . the model further implies that sources viewed close to the line of sight ( i.e. , flat - spectrum quasars and bl lacs ) should be more variable and have shorter variability timescales . for objects very close to the line of sight , relativistic beaming will become important , amplifying the variations and shortening the timescales to @xmath5 day ( e.g. , lister 2001 ) . indeed , intraday variability has been readily detected in samples of compact , flat - spectrum sources . some of the variability occurs on such short timescales that the apparent source brightness temperature ( e.g. , gopal - krishna et al . 1984 ) can be as high as @xmath8k ( e.g. , quirrenbach et al . 2000 ) , well beyond the inverse compton limit of @xmath9k ( kellermann & pauliny - toth 1969 ) . doppler boosting factors of up to a few hundred are needed to reconcile these numbers ( the apparent t is proportional to the doppler factor to the third power ) . these doppler factors are actually at the high end of the observed range , and it has been suggested that these velocities are pattern speeds that do not reflect the lorentz factors of the jet itself ( e.g. , lister et al . 2001 ) . nonetheless , intraday variability has been uniquely identified with the blazar class of objects ( e.g. , quirrenbach et al . 1992 ; lister et al . 2001 ) . comparatively little is known about ( intrinsic ) radio variability in non - blazar sources at low flux density levels ( @xmath7 mjy ) . most of the variability research has focused on either bright radio samples ( e.g. , gregorini et al . 1986 , @xmath10 jy ; lister et al . 2001 , @xmath11 jy ; aller et al . 2003 , @xmath12 jy ) , or small samples at much lower flux levels ( a few 10 s of mjy e.g. , rs & machalski 1990 ; bondi et al . 1996 ; riley et al . recently , carilli et al . ( 2003 ) used deep multi - epoch vla observations of the lockman hole area to study variability at the 0.1 mjy level , resulting in just 9 variable sources . in this paper , we will use two epoch observations of the first zero - dec strip ( sect . [ observations ] ) of about 9000 radio sources to expand significantly the number of variable sources down to flux density levels of @xmath13 mjy . unlike much brighter radio samples , the typical radio source population at these levels is dominated by star - forming galaxies and steep - spectrum systems , with only a small percentage of flat - spectrum agn ( e.g. , windhorst et al . 1999 ; richards et al . 1999 ; de vries et al . 2002 ) . with our sample , therefore , we are in an excellent position to investigate the radio variability properties of lower radio luminosity galaxies . interstellar scintillation is not expected to contribute significantly to the variability in our sources , since our observations were carried out at 1.4 ghz , near the minimum in expected iss ( e.g. , padrielli et al . 1987 ; mitchell et al . furthermore , given our variability criterion ( see sect . [ sample ] ) , only sources that vary by significant amounts , especially toward the lower flux levels , are included . these variation levels exceed the typical iss variations of a few percent by a large factor . the paper is outlined as follows . in the next two sections , the radio observations and sample selection are described . in sect . [ optid ] we discuss the optical properties of the variable sample , and compare these against a non - variable control sample . our radio imaging data have been taken as part of the faint images of the radio sky at twenty centimeters survey ( first , see becker et al . 1995 for a detailed description ) conducted with the vla in b - array between 1993 and 2002 . the observing strategy of the first survey provides a limited capability to search for variability , although since adjacent fields have significant overlap , most sources were observed more than once . for the most part , data were collected in strips of constant declination so adjacent fields in the east - west direction are sensitive to variations on a time scale of 3 minutes . first observing runs were usually separated by multiples of 24 hours so adjacent fields in the north - south direction are sensitive to variations on timescales of 1 to 5 days . this led , for instance , to the discovery of a number of radio - variable stars @xcite . a search for variability between the first survey and the nrao vla sky survey ( nvss , condon et al . 1998 ) is also possible . this is , however , complicated by the difference in angular resolution : the first survey can resolve out flux seen by the nvss . this results in so many nvss sources appearing brighter than their first counterparts that it is necessary to restrict the analysis to sources which appear brighter in first ( see sect . [ nvsscomp ] ) , making a straightforward flux comparison to test for variability difficult . the best opportunity to search for long term variability has been afford by our decision to re - observe the south galactic cap ( sgc ) first zero dec strip ( hereafter fzdb ) during the summer of 2002 ( the initial set of observation were made in 1995 , which will be referred to as fzda ) . the observations were made for the dual purpose of a quality control test of the first survey , as well as a search for variability . the zero - dec strip was chosen for this purpose because it has been observed by the sloan digital sky survey as well ( sdss , see stoughton et al . 2002 for the early data release , and abazajian et al . 2003 for data release 1 updates ) . a total of 720 fields were re - observed covering 120.2 square degrees and encompassing 9086 radio objects in the fzdb catalog . the fzdb survey area is shown in fig . [ area ] . the first catalog has a uniform sensitivity across the surveyed zero - dec strip area , whereas the fzdb has not ( fig . [ area ] ) . this makes a direct comparison of their relative completeness limits and source surface densities difficult . with an average surface density of @xmath14 radio sources per square degree @xcite , the first catalog should contain about 10800 objects within the nominal zero - dec strip survey area . however , this number can not be directly compared to the fzdb count of 9086 , due to the latter s non - uniform survey coverage . the best way to compare the two catalogs is to create an fzda catalog ( derived from first ) which has its source surface density tapering off toward the survey edges , just like the fzdb catalog . the simplest way of obtaining such a catalog is to list the unique sources contained in first around the fzdb catalog positions . this makes sense , since we are interested in objects present in both catalogs in the first place . the total number of sources in such an fzda catalog vary with the radius up to which the fzdb positions are matched : 6605 sources within 3 , 7343 within 1 , and 11778 within 10 . we will use the first two lists to compare the relative completeness of the fzdb catalog . the 6605 entry list represents our matched catalog from which we will select our variable sample ( sect . [ varcrit ] ) , and the 7343 constitutes our comparison fzda sample . the combined catalog uses a matching radius of 3 , which is about half of the fwhm of the b - array beam . this automatically excludes the fraction of extended sources for which the source detection program assigned different components and/or different positions though . as a consequence , the number of 6605 is considerably lower than the original first surface density ( even lower than expected based on the non - uniform coverage ) . to be consistent , we exclude _ all _ extended components . ] from our combined sample . there are good scientific reasons for this as well . extended radio emission is most likely due to lobe ( radio hotspot ) emission which can not vary on the short timescales we are sampling here ( @xmath15 years ) . since we are interested in shorter timescales , we will have to study those sources in which the ( feeding ) processes close to the agn itself vary . these sources are consequently very small , and unresolved by our imaging data ( with its resolution of @xmath16 arcseconds ) . the resulting reduced sample sizes have been listed in table [ samples ] . figure [ matchfrac ] illustrates how well the fzda and fzdb compare . at the higher flux density levels we expect to find almost all of the point sources in the combined catalog this is the case for sources with flux densities exceeding 2 mjy . for both the fzda and fzdb catalogs , over 98% of those sources end up in the combined catalog . none of the remaining catalog - unique entries turned out to be real on close inspection : they were either resolved sources for which a particular component was assigned a differing position between fzda and fdzb ( separated by more than 3 ) , or the catalog entry turned out to be an imaging artifact that made it into one but not both catalogs ( e.g. , ringing around very bright objects ) . the situation is quite different for fainter sources , however . an increasing fraction of faint objects are only present in either catalog , which indicates a systematic effect rather than genuine source variability . disentangling the truly variable sources from the random noise at these flux density levels can only be achieved by obtaining deeper surveys . indeed , none of these faint sources end up meeting our variability criterion ( sect . [ varcrit ] ) , which is based on the survey sensitivity level . all the variable sources that _ do _ meet the criterion have flux densities @xmath17mjy ( see fig . [ vp2med ] ) , and are in the well - matched part of fig . [ matchfrac ] . so , in summary , transient phenomena ( e.g. , gamma ray burst afterglows ) have not been found for flux density limits exceeding 2 mjy . fainter events may be present , but we do not have any way of assessing their reality . since we are interested in radio source variability , we begin by defining our variability ratio ( hereafter vr ) as : @xmath18 for the @xmath19 source in our catalog . the indices a and b denote the measurement epochs 1995 and 2002 . because the source brightness does not enter into this equation , we can not use this ratio at face value : toward the lower flux density levels the ( sky background ) noise will account for a significant fraction of the `` variability '' in such sources . we therefore have to weight the vr value of each source individually based on its signal - to - noise ratio ( s / n ) and local noise values ( see next section ) . since we are only considering unresolved sources , ideally their ( fitted ) peak flux density and integrated flux density should be the same . this is , however , not the case . in fact , the integrated flux density values are found to have larger intrinsic scatter than the fitted peak flux values . both the peak and integrated flux density are based on gaussian fits to the actual data , but unlike the peak fit , the integrated value is also based on fits to the major and minor axes of the object . noise in the measurement of the axes therefore increases the noise in the integrated flux density compared to the peak value . in fig . [ relvar ] we have plotted the relative flux density difference ratio ( defined as the larger flux density divided by the smaller flux density ) for both the source integrated and peak flux density values . it is clear from this figure that the scatter in the peak flux measurements ( the solid blue median line ) is considerably less than that of the integrated flux density ratios ( as indicated by the solid purple median line ) . to minimize the number of variable false positives , we only consider the source peak flux density values . following @xcite , the relative flux density errors can be written as : @xmath20 this equation reflects the two components of the measurement error , with @xmath21 due to a constant systematic error , and @xmath22 dependent on the s / n ratio . this expression can be used to estimate the significance of radio source variation , as defined by the quotient of the new and old flux densities . if one defines the flux density and noise values in the first epoch as @xmath23,@xmath24 , and in the second as @xmath25,@xmath26 , the 4@xmath27 variability envelopes are then given by : @xmath28 which can be further reduced by substituting @xmath29 , @xmath30 mjy , @xmath31 , and @xmath32 . the @xmath27 values have been measured to be on average 0.150 and 0.148 mjy for the 1995 ( a ) and 2002 ( b ) epochs , respectively . the @xmath22 value has been chosen to match the value of @xcite and @xcite , and seems fairly robust across different instruments and survey setups . the @xmath31 value ( which is lower than the @xmath33 of the latter two papers ) , is set by the high s / n end of the source distribution . its adopted value is a good match between including / excluding high s / n variable sources : the fraction of variable sources as function of s / n is kept acceptably flat ( see figs . [ varplot2 ] and [ vp2med ] , with s / n defined as @xmath34 ) . the upper and lower @xmath35 envelopes can now be expressed ( as a function of s / n ) as : @xmath36 equation [ fluxerror3 ] forms the basis of our variability criterion . any source having a vr exceeding either @xmath35 envelope is considered genuinely variable . this yields 146 variable sources , of which 58 are brighter in the 2002 epoch , and 88 are fainter ( fig . [ varplot2 ] ) . the solid red curve in fig . [ varplot2 ] represents the median flux density ratio as function of s / n , plotted on a linear scale to improve clarity ( as opposed to the logarithmic equivalent in green ) . as can be gathered from this curve , the medians quickly fall below 1 as the s / n decreases , with a maximum deviation of 8% at s / n @xmath37 . most of the offsets are within @xmath1% , however . even though the offsets are small , there is a systematic trend which is affecting our variability sample selection , something we wish to avoid . if uncorrected , it implies that fainter sources are more likely to fade between 1995 and 2002 than brighter sources , something which would be hard to explain in any variability scenario . we have two parameters to play with : a small ( additive ) zero - point correction which will straighten out the median curve , and a multiplicative sensitivity correction which will raise or lower the curve along the y - axis . the best correction ( with the resulting median curve having the smallest variations along the @xmath38 line ) is obtained by adjusting the fzdb flux densities by : @xmath39 which amounts to a 90 @xmath40jy zero - point offset , and a 1.16% sensitivity correction . the 90 @xmath40jy offset is small compared to both the 150 @xmath40jy rms noise value and the 250 @xmath40jy clean bias ( a flux density correction applied per beam for the first survey , see white et al . 1997 ) . and the 1.16% factor is well within the nominal @xmath41 systematic uncertainty in the flux density scale . after applying these corrections ( fig . [ vp2med ] ) , we end up with 128 candidate variable sources / components ( corresponding to 2.0% of the sample ) , listed in table [ varsample ] , of which 70 are brighter in the 2002 epoch , and 58 fainter . as a reality check , we compared the fzdb catalog against the nvss catalog to see whether the same sources are deemed variable , using basically an identical variability criterion . there are , however , a few differences . the nvss survey has a typical background noise of 0.45 mjy ( condon et al . 1998 ) , instead of the 0.15 mjy for the first survey . another complicating factor is the resolution difference between the nvss and fzdb . sources that are not resolved by the @xmath42 nvss beam are often resolved by the fzdb , with each component only a fraction of the nvss total flux density . we tried to minimize this effect by first excluding resolved nvss sources outright from the correlation , and secondly using the fzdb integrated ( instead of peak ) flux density measurements . furthermore , we used a very restrictive @xmath43 matching radius in another attempt to just match unresolved sources both in the fzdb and nvss . the resulting matched sample contains 2201 sources . of these 2201 sources , 10 are brighter at the fzdb epoch , and 91 are fainter . this is after applying a similar fzdb flux density correction as in the previous section ( albeit with slightly different constants ) . clearly , there is still some flux density being resolved out by the longer base - line fzdb observations compared to the nvss , which results in quite a few bogus variable sources . this implies we can not really compare the fzdb - nvss faint variable sources to table [ varsample ] , since an unknown fraction of them will be spurious . the 10 fzdb - nvss bright variable sources on the other hand , are not be affected by this . indeed , 9 out of the 10 sources are recovered in table [ varsample ] ( separately footnoted ) , with the sole `` newly '' identified variable source first j222646@xmath44005210 . upon closer inspection , this source appears to be stable between fzda and fzdb ( @xmath45 mjy for both epochs ) . its nvss flux density of 615 mjy is just within the quoted @xmath41 systematic flux density uncertainty between the two surveys . based on this , we therefore conclude that our variability criterion is robust , and results in a sample of truly variable sources . this does tacitly assume that the cataloged radio flux densities are always correct . however , as the next section will show , this is actually not valid for some cases . our variable sample actually contains 128 variable , unresolved , _ components_. while most of these component entries turn out to be single ( unresolved ) radio pointsources , some of them are part of a multi - component radio source . the varying part is then presumably associated with the agn / radio - core position . however , there may still be some cases where the fzda and fzdb cataloging software divided up radio flux differently between adjacent components , introducing artificial variability . it is therefore important to check all the multi - component radio sources in our sample ( 28 ) individually . in order to characterize the _ overall _ radio properties of our sample , we start by defining 6 morphological categories : 1 ) single pointsource ( ps in column 11 of table [ samples ] ) ; 2 ) pointsource with an elongation , or linear feature ( cj , `` core - jet '' ) ; 3 ) variable core with two non - variable lobes ( cl ) ; 4 ) core embedded in diffuse halo ( ch ) ; 5 ) complex , multi - component structure ( cx ) ; and 6 ) possible hotspot variability ( hs ) . each of the sources in table [ samples ] have been categorized accordingly . the relative number break - down of the sample is : 1 ) 100 ( 78.1% ) ; 2 ) 14 ( 10.9% ) ; 3 ) 9 ( 7.0% ) ; 4 ) 1 ( 0.8% ) ; 5 ) 1 ( 0.8% ) ; and 6 ) 3 ( 2.3% ) . except for the last two classes ( containing 4 sources , or 3.1% of the total sample ) , all of the variability is consistent with it being due to the agn / core component . we will discuss a few selected sources in more detail below . radio maps of all the extended variable radio sources can be found in fig . [ radiomaps ] . * j000257@xmath46002447 ( ch ) * this source contains a variable pointsource component embedded in an extended radio halo . the radio surface brightness of this halo remains constant between the fzda and fzdb epochs , whereas the peak flux density of the central pointsource varies by more than 10% . * j001800@xmath44000313 ( hs ) * the north - eastern lobe of this double radio source is slightly resolved , just enough so that it did not make it into our variable component list . as it turns out , both lobes / hotspots appear to be variable . the total flux density for the radio source increased by 13.6% between fzda and fzdb , from 62.14 to 70.56 mjy . unfortunately , two unrelated sources close to this source also `` varied '' by @xmath47 , casting serious doubt on its reality . the source is very close to the edge of its particular survey field , which may be the reason for the flux density discrepancy . it is taken out of further consideration , and has been placed separately in table [ samples ] . * j003246@xmath46001917 ( hs ) * upon close inspection of both the fzda and fzdb catalogs , we have to come to the conclusion that this source is not variable . the problems stems from the fact that fzda lists 5 components for this source , and fzdb only 4 . this leads to quite different flux density assignments to the north - western lobe . the source is taken out of further consideration . * j012213@xmath46001801 ( cl ) * both the eastern and western lobes are constant to within @xmath48 between fzda and fzdb . flux density values for the bright eastern lobe are 123.84 and 122.61 mjy , a variation of 1.0% . the core , on the other hand , brightened over the same time period from 348.28 to 404.79 mjy . * j020234@xmath44000301 ( cl ) * the lobes do not vary beyond the 1% level . the core , however , fades from its fzda flux of 44.01 mjy to 35.92 ( an 18.4% decrease ) . * j021840@xmath46001515 ( cx ) * this is the only source in our sample with a hard to categorize radio morphology . the marked component position ( fig . [ radiomaps ] ) coincides with a @xmath49 quasar . the two components immediately to the north do not have optical counterparts , though there appears to be a faint object in the sdss image that is situated in between the lobes . all of the radio components fade between fzda and fzdb by on average 16% . a nearby , unrelated , radio source observed in the same field remains constant , however . the component associated with the quasar varies the most : from 14.12 mjy down to 10.92 mjy ( a decline of 22.7% ) . at the moment it is not clear to us how the various components relate to each other , if at all . * j021202@xmath46002750 ( cl ) * the core component faded by at least 4 mjy ( @xmath50 ) relative to the lobes . * j212058@xmath44000612 ( hs ) * as is the case for j003246@xmath46001917 , this source is not variable . the south - eastern lobe is constant , and the perceived variability in the north - western lobe is due to differing fzda and fzdb component flux density assignments . this source is also taken out of consideration . * j212955@xmath44000758 ( cl ) * the `` variability '' is due to the fzda having 3 components , whereas fzdb lists 4 . the summed fluxes do not vary . * j220017@xmath46000133 ( cl ) * the core component in this source brightened ( by 20% ) , so we are inclined to regard this source as variable even though the number of fitted components is different between the fzda and fzdb . this faint extra component in the fzda catalog ( just to the north of the core ) does not appear to contribute anything to the total flux of the core in the fzdb catalog ( where this component is absent ) , based on the fitting parameters . * j222729@xmath44000522 ( cl ) * the core variability is 17.2% ( relative to the lobes ) , with the flux declining from 97.51 mjy to 80.69 mjy . * j235050@xmath46002848 ( cl * the core flux increased , whereas the lobes remained more or less constant . * j235828@xmath44003934 ( cl * all of the components faded by about the same amount , casting serious doubt on the reality of core variability . we have taken this source out of further consideration . so , after close inspection , we found all 3 potential hotspot variable sources ( hs ) and 2 core - lobe ( cl ) sources to be non - variable . they have been excluded from our subsequent studies , but are left in table [ samples ] and fig . [ radiomaps ] as reminders of how careful one has to be using automated selection criteria . we next investigate the possible connection between optical properties and the presence of radio variability . variability correlations in the optical and radio have been proposed before ( e.g. , tornikoski et al . 1994 ; hanski et al . 2002 ) , with the highly beamed blazars as the clearest examples of sources which exhibit variability across the electromagnetic spectrum ( e.g. , bregman et al . 1990 ) . the first step would be to correlate our variable sample with known optical catalogs , and compare the results against a sample of non - variable radio sources . we therefore created 12 sets of 123 non - variable sources ( i.e. , each control sample has the same number of elements as the variable sample ) . to ensure non - variability , we only selected sources inside @xmath51 variation curves ( as opposed to sources _ outside _ @xmath35 variations , fig . [ vp2med ] ) , while matching the radio flux density distribution function of the variable sample . this was done by matching the number of non - variable sources to the amount of variable sources per s / n bin ( binned logarithmically from 1.0 to 3.4 , with binsize 0.4 dex ) . given that there are only 1922 sources ( out of our initial sample of 6605 ) that are less variable than these @xmath51 limits , 12 sets of control samples are about as many as one can create without starting to have a significant fraction of sources shared between them . all of our fzda / fzdb survey area is covered by the automatic plate measuring ( apm ) machine catalog . the apm facility ( in cambridge , uk ) lists identifications and positions based on scanned uk and poss ii schmidt plates , covering currently more than 15 , 000 deg@xmath52 of sky . our sample of 123 variable sources has been matched against the apm catalogs with a @xmath53 matching radius . while this does discriminate against possible off - nuclear radio sources , it is necessary to limit the number of chance matches , given the density of objects in the apm catalog . the matching results are listed in table [ apmmatch ] . what is immediately striking is the difference in detection rates between the variable and non - variable samples . about half of the variable sources are retrieved from the apm catalogs , but only 1 in 6 of the non - variable sources have optical counterparts . the rms variation between the 12 control samples is 2.8% , resulting in a @xmath54 detection rate difference . clearly , variable sources are easier to detect not only because on average they are brighter , but also due to a higher fraction of quasars ( see sect . [ sdssmorph ] ) . the brightness difference is not large ; actually a cumulative distribution plot against mean magnitude ( fig . [ apmids ] ) shows how comparable their distributions are . based on a ks - test , these two distributions do not differ to any level of significance . fortunately , a significant fraction ( @xmath55 ) of the survey is covered by the sdss data release 1 ( dr1 ) , so we can use these deeper and more uniform imaging data to investigate this further . the sdss data also show a higher fraction of optical identifications for variable sources . since the latter data are deeper than the photographic plates the apm catalogs are based on , we identify 82% of the variable sources , and on average 38% of the non - variable ones ( table [ sdssmatch ] ) . these values are up from the apm values of 54% and 18% , respectively . as mentioned earlier , one way of explaining the higher match rate for the variable sample is that it may contain a higher fraction of quasars . since these objects are typically unresolved , they are easier to detect at any given magnitude than resolved galaxies ( of the same integrated magnitude ) . table [ sdssmatch ] lists the morphological breakdown of each sample , based on the sdss classification . with a relative stellar fraction of 53% , the variable sample contains significantly more quasar candidates than the non - variable control sample ( on average 19% ) . the results of table [ sdssmatch ] can also be used to calculate how much more quasars are likely to be variable compared to galaxies . this is a function of radio flux density ( fig . [ morphrf ] ) , but the low number of sources ( per bin ) only allow for an overall ratio . we assume the non - variable sample numbers ( 19% of the radio population are quasars , 81% are galaxies ) are representative of the general population . these numbers compare very well to the ones quoted by @xcite for a much larger sample of sdss sources detected with first ( 83% of those are galaxies ) . now , if we introduce the terms quasar variability rate qvr and galaxy variability rate gvr ( both defined as the fraction of quasars / galaxies that are variable ) , and an initial sample size of @xmath56 , we get : @xmath57 this means that the variable sample will contain in total @xmath58 variable sources . we know the relative quasar and galaxy fractions to be 53% and 47% . this leads to ( after dividing out @xmath56 ) : @xmath59 if one assumes a @xmath60 uncertainty in the relative fractions ( table [ sdssmatch ] ) , this relative value ranges between 3.4 and 7.5 . in other words , not surprisingly , quasars are on average 5 times as variable ( in the radio ) than are galaxies . the radio source population changes as a function of radio flux density ( e.g. , windhorst et al . 1999 ; richards et al . 1999 ; de vries et al . 2002 ) , from mainly agn - dominated sources at higher flux density levels down to starburst dominated sources at the sub - mjy level . this will affect our variable sample as well , and we expect to see a decline of the quasar fraction as function of lower flux density limits . this is nicely illustrated in fig . [ morphrf ] , which plots the relative quasar fraction with flux density . the solid green line represents our variable sample , and the dashed red line is the mean of our 12 control samples . the bins have a constant number of members ( except for the last bin in the control sample ) , hence the variable spacing . the binning of the control samples has been adjusted to approximately match the binning of the actual sample ( 4 bins of 16 each for the variable sample , and 5 bins of 60 plus 1 bin of 41 members for the control samples ) . while it is not known how much uncertainty is present in the sdss classification ( and hence quasar fraction ) toward the faintest magnitudes ( see sect . [ clasrel ] below ) , it is apparent from fig . [ morphrf ] that the quasar fractions for the variable and non - variable samples differ significantly ( the formal fraction errors are 5% for the control sample , and 12% for the variable sample ) . this leads us to two robust conclusions based on this figure : 1 ) a radio variable sample will contain more quasars than a non - variable sample , irrespective of radio flux density , and 2 ) the relative contribution of quasars to the sample declines as a function of declining flux density : below @xmath2mjy , galaxies account for the majority of the variable sources . the last section critically hinges on the reliability of the sdss galaxy versus quasar classification scheme . if for one reason , most of the excess quasar classifications in our variable sample were made toward the very faint end of the magnitude range , one might become suspicious about the reality of the effect . a direct test is plotting the quasar fraction as function of optical ( r - band in this case ) magnitude , instead of radio flux density as in fig . [ morphrf ] . this has been done in fig . [ morphmag ] , with again our variable sample in green and the control sample in red . it is reassuring to see that most of the excess quasar identifications in the variable sample are made for magnitudes brighter than @xmath61 . stoughton et al . ( 2002 ) estimate a 95% confidence level for classifications up to this magnitude . it becomes much less certain for fainter sources . without doubt some of the current classifications are wrong , but there is no indication that our consistently higher quasar fraction in the variable sample , both as function of radio flux density and optical magnitude , is significantly affected by this . we have established that there is a significant difference between the variable and non - variable radio samples in terms of their optical counterparts . the former have a higher matching rate with optical identifications ( given a fixed detection limit ) , which is at least in part ascribed to the higher fraction of quasars . we have already seen that the magnitude distributions are not significantly different ( fig . [ apmids ] ) , however , this does not rule out that significant optical color differences may be present between the samples . the clearest difference between galaxies and quasars ( at least in the restframe ) is provided by the ( u@xmath46r ) vs ( r@xmath46z ) color plot ( see fig 13 in stoughton et al . 2002 for intermediate color - color plots ) . this allows for sampling of the 4000 break in galaxies up to a redshift of about unity . this break will offset the galaxies from the quasars , which usually exhibit fairly flat , powerlaw continua ( see fig . 3 in de vries et al . our data are presented in fig . [ sdsscol ] . the sample means and their 1@xmath27 errors ( indicated by the crosses in fig . [ sdsscol ] ) are presented in table [ sdsscoltable ] . both the variable quasar and galaxy distributions in the color - color plane are ( statistically ) distinct from the non - variable control sample . this trend for both the galaxies and quasars to become bluer when they vary has been seen before ( e.g. , giveon et al . 1999 ; trevese & vagnetti 2002 ; vagnetti et al . 2003 ; de vries et al . still , the color spread among the sources in the sample is much larger than the _ individual _ color change as a source goes through an outburst . indeed , the studies of giveon , trevese , vagnetti and their respective collaborators all rely on following individual objects through their outbursts , instead of using a statistical sample . the presence of the mean optical color offsets in our radio variable sample implies two things : 1 ) on average , radio variable sources also exhibit optical ( color ) variations , and 2 ) the optical variation time - scales are shorter than our 7 year base - line , since the sdss data were not taken at the epoch of radio outburst . however , optical colors for an individual source , while useful in differentiating between quasars and galaxies , do not provide an effective selection mechanism for ( radio ) variability . nonetheless , as we have seen in the previous sections , the presence of radio variability has a clear impact on the sample s morphological makeup . we have created a sample of 123 variable radio sources using two epoch observations of a zero - dec strip toward the south galactic cap . this sample spans the range of radio flux densities from @xmath1 to 1000 mjy . it presents both in size and radio flux density coverage a unique starting point for variability studies of more normal , less agn - dominated galaxies , especially toward the lower flux density limits . we compared both our variable and non - variable samples to the sloan digital sky survey optical data . we found that the quasar fraction of the sample sharply declines as a function of declining radio flux density levels . this is consistent with earlier findings that the radio source population demographics change as one samples at progressively lower flux density levels : agn - dominated systems tend to be found at the brighter radio flux density levels ( @xmath62 mjy ) , whereas star - forming and normal galaxies dominate the counts at sub - mjy flux density levels ( at 1.4 ghz , e.g. , windhorst 2003 ) . our variable sample contains a consistently higher fraction of quasars than the non - variable control sample , independent of radio flux density . while this explains part of our almost @xmath63 higher optical matching rate of the variable sample compared to the non - variable one ( quasars are easier to detect at a given brightness limit than galaxies ) , it does imply that our variable sample contains on average slightly brighter sources ( though not significantly so , see sect . [ apmid ] ) . based on relative number statistics , we estimate that quasars are about 5 times more likely to harbor a variable radio source than galaxies . however , at flux density levels @xmath64 mjy , the majority of ( radio ) variable sources are identified as galaxies . and finally , galaxies and quasars that harbor a variable radio source exhibit , on average , bluer optical colors than hosts of non - variable sources . all of this underlines the fact that both galaxies and quasars can harbor variable radio sources , albeit at different occurrence rates . some of this is obviously due to the beamed nature of the ( variable ) quasars , enhancing the variability both by boosting their brightnesses and shortening the variability timescales . nonetheless , especially toward lower radio flux density limits , a statistically significant study of variability is possible provided one starts with large enough samples . large scale optical surveys like the sdss provide the crucial radio source optical `` environmental '' information . as more sdss data become available on the zero - dec strip , we will investigate this further . the authors like to thank steve croft for useful discussions and careful reading of this paper . wdvs work was performed under the auspices of the u.s . department of energy , national nuclear security administration by the university of california , lawrence livermore national laboratory under contract no . w-7405-eng-48 . the authors also acknowledge support from the national radio astronomy observatory , the national science foundations ( grants ast 00 - 98259 and ast 00 - 98355 ) , the space telescope science institute , and microsoft . this research has made use of the nasa / ipac extragalactic database ( ned ) which is operated by the jet propulsion laboratory , caltech , under contract with the national aeronautics and space administration . lrrrrcrrrrr j000111@xmath46002011 & 00 01 11.19 & @xmath4600 20 11.6 & 33.57 & 23.28 & 1.46 & 15.6 & 0.518 & 18.56 & q & cl + j000257@xmath46002447 & 00 02 57.19 & @xmath4600 24 47.4 & 112.93 & 124.90 & 0.92 & 4.2 & & 19.52 & g & ch + j000549@xmath44005048 & 00 05 49.92 & 00 50 48.1 & 8.87 & 5.13 & 1.77 & 10.1 & & 21.59 & g & ps + j001158@xmath46000208 & 00 11 58.81 & @xmath4600 02 08.2 & 5.95 & 27.34 & 0.22 & 43.2 & & 20.98 & q & ps + j001507@xmath46000801 & 00 15 07.01 & @xmath4600 08 01.8 & 10.54 & 12.67 & 0.85 & 4.3 & 1.704 & 18.23 & q & ps + j001611@xmath46001512 & 00 16 11.07 & @xmath4600 15 12.3 & 827.61 & 1050.26 & 0.80 & 11.2 & 1.575 & 19.70 & q & ps + j002354@xmath44001919 & 00 23 54.58 & 00 19 19.5 & 4.75 & 7.46 & 0.66 & 5.2 & & 22.69 & g & cj + j002738@xmath44000627 & 00 27 38.25 & 00 06 27.6 & 4.56 & 6.91 & 0.68 & 5.4 & & 22.02 & q & ps + j003007@xmath46000007 & 00 30 07.90 & @xmath4600 00 07.4 & 50.52 & 39.32 & 1.30 & 12.1 & & 19.55 & q & ps + j003127@xmath44003959 & 00 31 27.92 & 00 39 59.4 & 9.58 & 12.31 & 0.79 & 6.3 & & 21.80 & q & ps + j003536@xmath46000627 & 00 35 36.30 & @xmath4600 06 27.2 & 17.28 & 14.52 & 1.21 & 6.1 & & 18.94 & g & ps + j003540@xmath46002529 & 00 35 40.22 & @xmath4600 25 29.2 & 4.01 & 6.57 & 0.63 & 6.4 & & 21.82 & g & ps + j003703@xmath44003537 & 00 37 03.36 & 00 35 37.4 & 12.84 & 15.28 & 0.86 & 4.5 & & 21.51 & q & ps + j004332@xmath44002459 & 00 43 32.71 & 00 24 59.8 & 120.94 & 108.53 & 1.13 & 6.1 & 1.127 & 19.20 & q & ps + j004819@xmath44001457 & 00 48 19.11 & 00 14 57.1 & 98.76 & 89.40 & 1.12 & 5.6 & 1.536 & 19.92 & q & ps + j005205@xmath44003538 & 00 52 05.58 & 00 35 38.2 & 34.58 & 81.46 & 0.43 & 38.1 & 0.399 & 16.09 & q & cj + j005212@xmath44000945 & 00 52 12.47 & 00 09 45.2 & 11.92 & 9.77 & 1.24 & 5.5 & & & & ps + j005225@xmath44002627 & 00 52 25.67 & 00 26 28.0 & 12.13 & 9.98 & 1.24 & 5.2 & & 20.75 & g & ps + j005717@xmath46002433 & 00 57 17.01 & @xmath4600 24 33.2 & 89.47 & 114.74 & 0.79 & 11.6 & 2.790 & 19.19 & q & cj + j010525@xmath44001121 & 01 05 25.52 & 00 11 21.7 & 33.97 & 41.73 & 0.83 & 8.3 & & 21.45 & g & ps + j010745@xmath44003952 & 01 07 45.21 & 00 39 52.8 & 10.50 & 8.88 & 1.21 & 4.5 & & & & ps + j010838@xmath44002814 & 01 08 38.56 & 00 28 14.6 & 5.33 & 3.91 & 1.40 & 4.2 & & & & ps + j011106@xmath44000846 & 01 11 06.79 & 00 08 46.1 & 5.25 & 3.43 & 1.57 & 4.9 & & 22.70 & q & ps + j011515@xmath44001248 & 01 15 15.78 & 00 12 48.5 & 46.84 & 43.13 & 1.10 & 4.4 & 0.045 & 14.39 & g & ps + j012213@xmath46001801 & 01 22 13.92 & @xmath4600 18 01.0 & 386.60 & 331.60 & 1.18 & 8.3 & & 20.23 & q & cl + j012528@xmath46000555 & 01 25 28.85 & @xmath4600 05 55.8 & 1333.76 & 1481.35 & 0.91 & 4.8 & 1.076 & 16.47 & q & ps + j012753@xmath44002516 & 01 27 53.70 & 00 25 16.5 & 131.32 & 90.08 & 1.48 & 19.1 & & 20.76 & q & ps + j013457@xmath44003942 & 01 34 57.42 & 00 39 43.0 & 6.09 & 2.87 & 2.18 & 8.7 & & 22.04 & q & ps + j013815@xmath44001446 & 01 38 15.02 & 00 14 46.5 & 42.98 & 51.76 & 0.84 & 8.1 & & 21.64 & g & cj + j015329@xmath46002214 & 01 53 29.75 & @xmath4600 22 14.3 & 17.46 & 14.98 & 1.19 & 5.8 & & 19.07 & q & ps + j015528@xmath44001204 & 01 55 28.47 & 00 12 04.6 & 17.23 & 19.82 & 0.88 & 4.8 & & 22.08 & g & ps + j015950@xmath46002407 & 01 59 50.09 & @xmath4600 24 07.2 & 10.68 & 12.63 & 0.86 & 4.3 & & 21.50 & g & ps + j020141@xmath44003825 & 02 01 41.04 & 00 38 25.5 & 4.50 & 6.35 & 0.73 & 4.7 & & & & ps + j020214@xmath46001748 & 02 02 14.30 & @xmath4600 17 48.3 & 75.12 & 60.00 & 1.27 & 11.5 & & 21.36 & g & ps + j020234@xmath44000301 & 02 02 34.32 & 00 03 01.7 & 30.54 & 39.41 & 0.79 & 10.5 & & 18.42 & g & cl + j020928@xmath46001224 & 02 09 28.85 & @xmath4600 12 24.9 & 4.10 & 2.51 & 1.69 & 4.2 & 0.152 & 16.02 & g & ps + j021202@xmath46002750 & 02 12 02.13 & @xmath4600 27 50.1 & 45.14 & 52.84 & 0.87 & 6.6 & & & & cl + j021301@xmath46001815 & 02 13 01.13 & @xmath4600 18 15.0 & 41.55 & 48.44 & 0.87 & 6.5 & & 21.98 & g & ps + j021553@xmath44001826 & 02 15 53.65 & 00 18 26.9 & 30.90 & 35.82 & 0.88 & 5.6 & & 19.51 & g & ps + j021755@xmath46000935 & 02 17 55.99 & @xmath4600 09 35.9 & 3.96 & 5.85 & 0.70 & 4.4 & & 19.86 & g & cj + j021840@xmath46001515 & 02 18 40.55 & @xmath4600 15 15.9 & 10.06 & 13.17 & 0.78 & 6.4 & 1.171 & 18.81 & q & cx + j022624@xmath44000746 & 02 26 24.61 & 00 07 46.2 & 43.84 & 48.74 & 0.91 & 4.4 & & 22.56 & g & ps + j023105@xmath44000843 & 02 31 05.59 & 00 08 43.5 & 47.60 & 53.86 & 0.90 & 5.0 & 1.338 & 20.00 & q & ps + j025321@xmath44000559 & 02 53 21.04 & 00 05 59.9 & 151.89 & 97.64 & 1.57 & 22.0 & & 22.26 & g & ps + j025333@xmath44002431 & 02 53 33.66 & 00 24 31.7 & 16.86 & 14.86 & 1.15 & 4.6 & & & & ps + j025404@xmath46002628 & 02 54 04.60 & @xmath4600 26 28.7 & 4.39 & 2.71 & 1.67 & 5.2 & & 21.97 & g & ps + j025859@xmath44003618 & 02 58 59.65 & 00 36 18.3 & 29.45 & 23.75 & 1.26 & 9.4 & & 22.28 & q & ps + j025928@xmath46002000 & 02 59 28.52 & @xmath4600 20 00.1 & 243.47 & 225.85 & 1.09 & 4.3 & 2.001 & 17.34 & q & ps + j030137@xmath44001746 & 03 01 37.58 & 00 17 46.1 & 6.86 & 4.45 & 1.58 & 7.0 & & 21.47 & g & cj + j030317@xmath46002644 & 03 03 17.01 & @xmath4600 26 45.0 & 7.41 & 3.68 & 2.06 & 4.9 & & & & cj + j030702@xmath44000651 & 03 07 02.04 & 00 06 51.9 & 5.84 & 8.08 & 0.74 & 5.4 & & & & ps + j030834@xmath44003303 & 03 08 34.31 & 00 33 03.7 & 4.16 & 1.83 & 2.35 & 6.4 & 0.031 & 14.91 & g & ps + j030933@xmath46001901 & 03 09 33.27 & @xmath4600 19 01.2 & 6.31 & 4.58 & 1.41 & 4.6 & & & & ps + j031006@xmath44001549 & 03 10 06.69 & 00 15 49.9 & 1.36 & 5.55 & 0.26 & 11.1 & 0.109 & 18.54 & g & ps + j031118@xmath44000816 & 03 11 18.51 & 00 08 16.4 & 29.30 & 36.54 & 0.81 & 9.1 & & & & ps + j031202@xmath46000442 & 03 12 02.50 & @xmath4600 04 42.5 & 6.42 & 8.59 & 0.77 & 5.2 & 0.038 & 13.50 & g & ps + j031345@xmath46000720 & 03 13 45.04 & @xmath4600 07 20.3 & 3.63 & 6.06 & 0.62 & 6.1 & 2.519 & 20.04 & q & ps + j031353@xmath46000004 & 03 13 53.47 & @xmath4600 00 04.1 & 10.13 & 7.03 & 1.47 & 7.7 & & 21.16 & g & ps + j031357@xmath44003506 & 03 13 57.10 & 00 35 06.9 & 16.54 & 20.23 & 0.83 & 6.7 & & & & ps + j031452@xmath44001346 & 03 14 52.08 & 00 13 46.5 & 6.77 & 10.05 & 0.69 & 7.3 & 3.202 & 19.54 & q & ps + j031609@xmath44000107 & 03 16 09.55 & 00 01 07.9 & 4.02 & 6.03 & 0.69 & 4.6 & & & & ps + j031634@xmath46002039 & 03 16 34.96 & @xmath4600 20 39.6 & 18.17 & 15.63 & 1.18 & 5.5 & & & & ps + j031814@xmath46002948 & 03 18 14.43 & @xmath4600 29 48.9 & 113.20 & 93.91 & 1.22 & 9.7 & & 21.77 & g & ps + j032007@xmath44000354 & 03 20 07.01 & 00 03 54.0 & 12.17 & 10.20 & 1.22 & 5.0 & & 22.10 & g & ps + j212000@xmath46001159 & 21 20 00.72 & @xmath4600 11 59.5 & 42.94 & 48.33 & 0.90 & 4.9 & & & & ps + j212447@xmath44000747 & 21 24 47.34 & 00 07 47.6 & 5.79 & 3.82 & 1.56 & 5.3 & & & & ps + j213206@xmath44003520 & 21 32 06.15 & 00 35 20.1 & 13.65 & 11.95 & 1.16 & 4.1 & & & & ps + j213638@xmath44004154 & 21 36 38.57 & 00 41 54.3 & 4136.68 & 3546.71 & 1.18 & 8.3 & 1.932 & 16.79 & q & ps + j213748@xmath44001219 & 21 37 48.43 & 00 12 19.9 & 41.69 & 36.02 & 1.17 & 7.0 & 1.666 & 17.92 & q & ps + j213804@xmath44000714 & 21 38 04.06 & 00 07 14.7 & 10.31 & 16.33 & 0.64 & 11.6 & & & & cj + j214138@xmath44000319 & 21 41 38.55 & 00 03 19.8 & 11.08 & 14.37 & 0.79 & 6.8 & & & & ps + j214324@xmath44003502 & 21 43 24.37 & 00 35 02.7 & 36.03 & 45.22 & 0.81 & 9.7 & 2.030 & 19.37 & q & ps + j214419@xmath44002055 & 21 44 19.88 & 00 20 55.8 & 10.34 & 7.31 & 1.44 & 7.5 & & & & ps + j214613@xmath44000930 & 21 46 13.31 & 00 09 30.8 & 9.94 & 7.88 & 1.29 & 5.6 & & & & ps + j214807@xmath46000630 & 21 48 07.73 & @xmath4600 06 30.6 & 68.57 & 90.18 & 0.77 & 12.7 & & & & cj + j214811@xmath46001230 & 21 48 11.48 & @xmath4600 12 30.6 & 7.76 & 9.95 & 0.80 & 4.7 & & & & cj + j215349@xmath44003119 & 21 53 49.75 & 00 31 19.6 & 82.50 & 103.65 & 0.81 & 10.4 & & & & ps + j215353@xmath46001339 & 21 53 53.89 & @xmath4600 13 39.5 & 13.48 & 17.27 & 0.79 & 7.3 & & & & ps + j215359@xmath44004412 & 21 53 59.83 & 00 44 12.6 & 9.76 & 6.73 & 1.48 & 7.8 & 1.030 & 19.07 & q & ps + j215733@xmath46000340 & 21 57 33.66 & @xmath4600 03 40.5 & 5.36 & 3.60 & 1.53 & 4.9 & & & & ps + j215949@xmath44005146 & 21 59 49.91 & 00 51 46.7 & 12.29 & 9.21 & 1.36 & 7.8 & & & & ps + j215954@xmath46002150 & 21 59 54.46 & @xmath4600 21 50.1 & 4.07 & 2.46 & 1.71 & 4.7 & 1.960 & 16.98 & q & ps + j220017@xmath46000133 & 22 00 17.37 & @xmath4600 01 33.6 & 7.84 & 6.15 & 1.30 & 4.9 & & & & cl + j220755@xmath46000215 & 22 07 55.25 & @xmath4600 02 15.0 & 78.40 & 61.93 & 1.28 & 11.9 & & & & cj + j220822@xmath44002352 & 22 08 22.88 & 00 23 52.7 & 4.62 & 2.42 & 1.97 & 6.1 & & & & ps + j221001@xmath46001309 & 22 10 01.82 & @xmath4600 13 09.9 & 125.83 & 115.31 & 1.10 & 4.8 & & & & ps + j221031@xmath46001356 & 22 10 31.46 & @xmath4600 13 56.1 & 14.06 & 12.00 & 1.19 & 5.2 & & & & ps + j221909@xmath44003113 & 22 19 09.40 & 00 31 13.4 & 9.13 & 11.70 & 0.80 & 5.3 & & & & cj + j222036@xmath44003334 & 22 20 36.33 & 00 33 34.2 & 15.27 & 12.76 & 1.22 & 5.8 & & & & ps + j222135@xmath46001100 & 22 21 35.00 & @xmath4600 11 00.1 & 14.11 & 8.64 & 1.66 & 12.6 & & & & ps + j222235@xmath44001536 & 22 22 35.87 & 00 15 36.7 & 46.04 & 51.13 & 0.91 & 4.5 & & & & cj + j222704@xmath44004517 & 22 27 04.24 & 00 45 17.5 & 8.13 & 5.57 & 1.49 & 7.0 & & & & ps + j222726@xmath44001059 & 22 27 26.53 & 00 10 59.3 & 6.05 & 4.52 & 1.37 & 4.2 & & & & ps + j222729@xmath44000522 & 22 27 29.08 & 00 05 22.2 & 77.84 & 91.64 & 0.86 & 7.4 & 1.510 & 18.86 & q & cl + j222744@xmath44003450 & 22 27 44.58 & 00 34 50.6 & 22.13 & 30.53 & 0.74 & 12.1 & 1.540 & 19.14 & q & ps + j222758@xmath44003705 & 22 27 58.13 & 00 37 05.2 & 68.69 & 99.12 & 0.70 & 17.3 & & & & ps + j223047@xmath44002756 & 22 30 47.46 & 00 27 56.3 & 8.54 & 13.82 & 0.63 & 11.4 & & & & ps + j224224@xmath44005512 & 22 42 24.14 & 00 55 13.0 & 17.71 & 13.44 & 1.34 & 9.5 & & & & cj + j224331@xmath46001233 & 22 43 31.94 & @xmath4600 12 33.1 & 13.85 & 21.40 & 0.66 & 14.8 & 2.040 & 18.45 & q & ps + j224448@xmath46000619 & 22 44 48.11 & @xmath4600 06 19.8 & 8.24 & 5.58 & 1.51 & 7.3 & & & & ps + j224627@xmath46001214 & 22 46 27.68 & @xmath4600 12 14.2 & 85.55 & 56.00 & 1.55 & 20.9 & & & & ps + j224730@xmath44000006 & 22 47 30.19 & 00 00 06.1 & 464.52 & 183.71 & 2.56 & 43.8 & 0.094 & 18.50 & q & ps + j224922@xmath44001804 & 22 49 22.28 & 00 18 04.6 & 10.41 & 8.51 & 1.25 & 5.3 & & & & ps + j225852@xmath46001857 & 22 58 52.94 & @xmath4600 18 57.3 & 3.45 & 1.28 & 2.80 & 6.0 & & & & ps + j230157@xmath44000351 & 23 01 57.85 & 00 03 52.0 & 3.75 & 7.04 & 0.55 & 7.8 & & 23.23 & q & ps + j230314@xmath44000052 & 23 03 14.85 & 00 00 52.2 & 3.72 & 1.72 & 2.24 & 5.8 & & & & ps + j230655@xmath44003638 & 23 06 55.16 & 00 36 38.1 & 12.99 & 15.19 & 0.87 & 4.3 & & & & ps + j231541@xmath44002936 & 23 15 41.67 & 00 29 36.6 & 21.93 & 13.94 & 1.60 & 15.7 & & 20.81 & q & ps + j231558@xmath46001205 & 23 15 58.64 & @xmath4600 12 05.5 & 3.52 & 6.27 & 0.58 & 7.0 & & 22.85 & g & ps + j231845@xmath46000754 & 23 18 45.81 & @xmath4600 07 54.7 & 3.03 & 5.34 & 0.59 & 5.5 & 0.867 & 19.44 & q & ps + j231856@xmath44001437 & 23 18 56.66 & 00 14 37.7 & 20.16 & 18.34 & 1.12 & 4.2 & 0.030 & 12.76 & g & ps + j231910@xmath44001859 & 23 19 10.33 & 00 18 59.0 & 28.90 & 33.36 & 0.88 & 5.6 & & 22.60 & q & ps + j232038@xmath44003139 & 23 20 38.01 & 00 31 39.8 & 72.69 & 82.95 & 0.89 & 5.7 & 1.911 & 19.03 & q & ps + j232323@xmath44003327 & 23 23 23.95 & 00 33 27.5 & 16.73 & 12.52 & 1.36 & 9.2 & & & & ps + j233448@xmath46001400 & 23 34 48.06 & @xmath4600 14 01.0 & 24.85 & 29.64 & 0.85 & 6.8 & & 21.56 & q & ps + j233822@xmath44001146 & 23 38 22.35 & 00 11 46.6 & 18.19 & 15.44 & 1.20 & 6.0 & & 22.24 & q & ps + j233852@xmath44004843 & 23 38 52.46 & 00 48 43.5 & 11.32 & 9.18 & 1.26 & 5.9 & & & & ps + j234623@xmath44004301 & 23 46 23.73 & 00 43 01.1 & 13.49 & 11.79 & 1.17 & 4.6 & 2.861 & 19.09 & q & ps + j234939@xmath46001315 & 23 49 39.90 & @xmath4600 13 15.2 & 8.46 & 6.97 & 1.24 & 4.2 & 1.267 & 20.22 & q & ps + j235022@xmath44001232 & 23 50 22.41 & 00 12 32.4 & 3.92 & 2.03 & 2.00 & 5.1 & & & & ps + j235050@xmath46002848 & 23 50 50.72 & @xmath4600 28 48.7 & 13.12 & 11.57 & 1.15 & 4.1 & & 19.62 & g & cl + j235409@xmath46001948 & 23 54 09.18 & @xmath4600 19 48.1 & 384.58 & 344.79 & 1.13 & 6.1 & 0.462 & 17.93 & q & ps + j235823@xmath44000213 & 23 58 23.91 & 00 02 13.2 & 8.03 & 10.67 & 0.77 & 5.7 & & & & ps + j001800@xmath44000313 & 00 18 00.79 & 00 03 17.9 & 20.76 & 18.82 & 1.12 & 4.2 & & & & hs + j003246@xmath46001917 & 00 32 46.02 & @xmath4600 19 17.8 & 81.29 & 60.80 & 1.35 & 14.5 & & & & hs + j212058@xmath44000612 & 21 20 59.00 & 00 06 12.7 & 31.17 & 28.05 & 1.13 & 5.3 & & & & hs + j212955@xmath44000758 & 21 29 55.68 & 00 07 59.0 & 1.17 & 3.03 & 0.42 & 4.5 & & & & cl + j235828@xmath44003934 & 23 58 28.77 & 00 39 34.1 & 2.48 & 4.16 & 0.62 & 4.3 & & & & cl + lcc variable & 66 & 53.7% + control 1 & 17 & 13.8% + control 2 & 29 & 23.6% + control 3 & 21 & 17.1% + control 4 & 21 & 17.1% + control 5 & 20 & 16.3% + control 6 & 27 & 22.0% + control 7 & 23 & 18.7% + control 8 & 17 & 13.8% + control 9 & 21 & 17.1% + control 10 & 22 & 17.9% + control 11 & 23 & 18.7% + control 12 & 21 & 17.1% + mean control & [email protected] & [email protected]% + lcccc variable & 64 & 82.0% & 53.1% & 46.9% + control 1 & 28 & 37.3%&17.9%&82.1% + control 2 & 30 & 40.0%&13.3%&86.7% + control 3 & 21 & 28.0%&14.3%&85.7% + control 4 & 36 & 48.0%&13.9%&86.1% + control 5 & 26 & 34.7%&26.9%&73.1% + control 6 & 32 & 42.7%&18.8%&81.2% + control 7 & 29 & 38.7%&24.1%&75.9% + control 8 & 33 & 44.0% & 9.1%&90.9% + control 9 & 23 & 30.7%&21.7%&78.3% + control 10 & 30 & 40.0%&16.7%&83.3% + control 11 & 31 & 41.3%&29.0%&71.0% + control 12 & 22 & 29.3%&18.2%&81.8% + mean control & [email protected] & [email protected]% & [email protected]% & [email protected]% + rrr - variable & [email protected] & [email protected] + non - variable & [email protected] & [email protected] + nominal offset & @xmath460.48 & @xmath460.20 + significance & 2.3 & 1.8 + - variable & [email protected] & [email protected] + non - variable & [email protected] & [email protected] + nominal offset & @xmath460.47 & @xmath460.11 + significance & 2.0 & 0.9 +

we present numerical simulations on the propagation of uhe protons with energies of ev in extragalactic magnetic fields over 1 gpc . we use the ors galaxy sample , which allow us to accurately quantify the contribution of nearby sources to the energy spectrum and the arrival distribution , as a source model . the sample is corrected taking the selection effect and absence of galaxies in the zone of avoidance ( ) into account . we calculate three observable quantities , cosmic ray spectrum , harmonic amplitude , and two point correlation function from our data of numerical simulations . with these quantities , we compare the results of our numerical calculations with the observation . we find that the arrival distribution of uhecrs become to be most isotropic as restricting sources to luminous galaxies . however , it is not isotropic enough to be consistent with the agasa observation , even for . in order to obtain sufficiently isotropic arrival distribution , we randomly select sources , which contribute to the observed cosmic ray flux , from the ors sample more luminous than mag , and investigate dependence of the results on their number . we show that the three observable quantities including the gzk cutoff of the energy spectrum can be reproduced in the case that the number fraction of the ors galaxies more luminous than mag is selected as uhecr sources . in terms of the source number density , this constraint corresponds to mpc . however , since mean number of sources within the gzk sphere is only in this case , the agasa 8 events above ev , which do not constitute the clustered events with each other , can not be reproduced . on the other hand , if the cosmic ray flux measured by the hires , which is consistent with the gzk cutoff , is correct and observational features about the arrival distribution of uhecrs are same as the agasa , our source model can explain both the arrival distribution and the flux at the same time . thus , we conclude that large fraction of the agasa 8 events above ev might originate in the topdown scenarios , or that the cosmic ray flux measured by the hires experiment might be better . we also discuss the origin of uhecrs below ev through comparisons between the number density of astrophysical source candidates and our result ( mpc ) .

astro-ph0210132

in spite of a increasing amount of data , the origin of cosmic rays in particular at the highest energy ( @xmath11ev ) is still unknown . it is very difficult to produce such ultra - high energy cosmic rays ( hereafter uhecrs ) in astrophysical shocks , which are thought to be responsible for the galactic cosmic rays . the gyroradii of uhecrs with energies of 10@xmath12ev are larger than our galaxy , and this suggests an extragalactic origin for any astrophysical scenario . using a hillas - plot @xcite , active galactic nuclei ( agn ) , gamma - ray bursts ( grbs ) , dead quasars and colliding galaxies are considered as probable candidates . these conventional acceleration models are called bottom - up scenarios . on the other hand , this situation has in recent years triggered a number of exotic production models based on new physics beyond the standard model of the particle physics ( see * ? ? ? * and references therein ) , which are called top - down scenarios . in such scenarios , uhecrs owe their origin to decay of some supermassive particles , which could be produced from topological defects ( tds ) , or be certain metastable supermassive relic particles ( msrps ) produced in the early universe . in bottom - up scenarios in which protons are accelerated in the potential uhecrs sources , like grbs and agns , the large distances to the earth lead to another problem , called the gzk effect @xcite . uhe protons with energy above 8@xmath13 ev interact with photons of cosmic microwave background ( cmb ) and lose their energy rapidly due to the production of pions over distances of tens of mpc @xcite . this should results in a cut - off of the cosmic ray spectrum at energy around 8@xmath13 ev . the situation for nuclei is considered to be worse due to the photo - disintegration mechanism @xcite . observationally , there seems to be a disagreement between the agasa which revealed the extension of the cosmic ray energy spectrum above @xmath10 ev @xcite and the high resolution fly s eye ( hires ; * ? ? ? * ) which recently reported the cosmic ray flux with the gzk cut - off @xcite . since uhecrs above @xmath10 ev observed by the agasa must originate within the gzk sphere , whose radius is typically about 50 mpc , the arrival directions are expected to point toward their sources within a few degrees , which is typical deflection angle of uhecrs ( e.g. , * ? ? ? however , no plausible source counterparts have been found within the gzk sphere . there is also a problem concerning the arrival directions of uhecrs ( e.g. , * ? ? ? their distribution seems to be isotropic on a large scale with a statistically significant small scale clustering @xcite . the current agasa data set of 57 events above 4 @xmath13 ev contains four doublets and one triplet within a separation angle of 2.5@xmath14 . chance probability to observe such clusters under an isotropic distribution is about 1 @xmath15 @xcite . potential models of uhecr origin are constrained by their ability to reproduce the observed energy spectrum and the isotropic arrival distribution of uhecrs with the small scale clustering . a number of studies has been devoted to the problem of the energy spectrum . yoshida and teshima ( 1993 ) show the gzk cutoff in consideration of evolution of the universe . blanton et al.(2001 ) show that the observed number of events above @xmath10 ev is 2 @xmath16 higher than the expected one , assuming a hard @xmath17 injection spectrum and small local overdensity ( a factor of two ) . berezinsky et al.(2002 ) analyze three source models , grb , agn and local enhancement of uhecr sources . they conclude the agasa events above @xmath10 ev , if confirmed , must be of different origin . although these authors show that it is difficult to reproduce the observed extension of the energy spectrum by the bottom - up scenarios , they do not take into account of the effects of the extragalactic magnetic field ( egmf ) on the energy spectrum . furthermore , since they do not consider angular source distribution as well as the egmf , the problems concerning the arrival distribution of uhecrs can not be discussed . although arrival distributions from a single source are calculated in the presence of magnetic field in some papers @xcite , it appears to be difficult to explain the observational isotropy by a single source . with consideration of both the magnetic field and the source distribution , the energy spectrum and the distribution of the arrival directions are calculated in sigl et al.(1999 ) and lemoine et al.(1999 ) . however the source distribution seems to be too simple , a two dimensional sheet @xcite or a pancake - like profile @xcite , which represent the local super cluster ( lsc ) . smialkowski et al.(2002 ) calculate arrival distribution of uhecrs using the iras pscz catalog of ir galaxies as a realistic source model . however , they restrict their attention to galaxies with far infrared luminosity @xmath18 triggered by collision and merging processes , which are possible sites of uhecr acceleration . it is also suspected that this galaxy sample do not contain some nearby galaxies in the lsc . since nearby sources within the gzk sphere are responsible for uhecrs above @xmath10 ev , galaxy sample with better completeness on nearby galaxies should be used . given these situations , we examined in our previous paper @xcite whether the current observation can be explained by a bottom - up scenario in which source distribution of uhecrs is proportional to that of galaxies . we used the optical redshift survey ( ors ; * ? ? ? * ) as a realistic source model . completeness on nearby galaxies of this sample would be better than that of the iras pscz . by using the ors sample , we can accurately quantify the contribution of nearby sources to the energy spectrum above @xmath10 ev . we also took effects of the egmf on the energy spectrum and the arrival distribution of uhecrs into account . however , there are problems in our previous study . first , we did not correct absence of galaxies in the zone of avoidance ( @xmath1 ) when using the ors sample , so that we could not present accurate discussion of statistics on the arrival directions of uhecrs . second , we considered propagation of uhecrs only over 40 mpc , focusing on uhecrs with energies above @xmath19 ev . in this case , contribution of distant sources ( outside 40 mpc ) to the cosmic ray flux below @xmath19 ev can not be taken into account . finally , we injected uhecrs from all the galaxies including dwarf ones . luminous galaxies are generally larger than dwarf ones , and thus they are expected to be able to accelerate particles to higher energy . thus , we should investigate dependence of the results on the limiting magnitude of galaxies . this paper is an extended study of our earlier work . we calculate the energy spectrum and the arrival distribution of uhecrs using the ors galaxy sample . we correct the sample taking the selection effect and absence of galaxies in the zone of avoidance ( @xmath1 ) into account . this allow us to accurately discuss statistics on the arrival directions of uhecrs . we also extend the propagation distance of uhecrs , from 40 mpc to 1 gpc . as s result , we can calculate the cosmic ray flux below @xmath19 ev , properly including contribution from distant sources . furthermore , we investigate dependence of the results on the limiting magnitude of galaxies . we compare the results of the energy spectra with the current observations , agasa and hires . in order to explore the possibility of reproducing the extension of energy spectrum beyond @xmath10 ev , we adopt several strength of the egmf ( 1,10 and 100 ng ) . we also investigate dependence of the arrival distribution of uhecrs on the limiting magnitude of galaxies and the source number , and estimate the most favorable ones from comparison with the agasa data using the harmonic analysis and the two point correlation function as statistical methods . we show that the three observable quantities including the gzk cutoff of the energy spectrum can be reproduced in the case that the number fraction @xmath5 of the ors galaxies more luminous than @xmath4 mag is selected as uhecr sources . implications of the results for the uhecr source candidate are discussed in detail . in section [ model ] , we show our method of calculation . results are shown in section [ result ] and their implications are discussed in section [ discussion ] . we conclude in section [ conclusion ] . this subsection provides the method of monte carlo simulations for propagating protons in intergalactic space . at first , we assume that the composition of uhecrs is proton , and inject an @xmath20 spectrum within the range of ( @xmath21 - @xmath22)ev . 10000 protons are injected in each of 26 energy bins , that is , 10 bins per decade of energy for the cases of @xmath23 = 40mpc and ( @xmath24 , @xmath23 ) = ( 1,10ng , 10mpc ) , where @xmath24 and @xmath23 are strength and correlation length of the extra galactic magnetic field ( explained below ) . for another cases , we inject 5000 protons in each of the bins . particles below @xmath25 ev lose their energies mainly by pair creations and above it by photopion production @xcite in collisions with photons of the cmb . the pair production can be treated as a continuous loss process considering its small inelasticity ( @xmath26 ) . we adopt the analytical fit functions given by chododowski et al . ( 1992 ) to calculate the energy loss rate for the pair production on isotropic photons . according to them , the energy loss rate of a relativistic nucleus for the pair production on isotropic photons is given by @xmath27 where @xmath28 is the lorenz factor of the particle , @xmath29 ( @xmath30 is the momentum of the particle in units of @xmath31 ) , @xmath32 , @xmath33 , @xmath34 , @xmath35 , and @xmath36 are the photon distribution in the momentum space , the fine - structure constant , the classical electron radius , the charge of the particle , and the rest mass of the particle , respectively . for the energy range @xmath37 ev , @xmath38 can be well represented as @xmath39 @xmath40 since protons lose a large fraction of their energy in the photopion production , its treatment is very important . we use the interaction length and the energy distribution of final protons as a function of initial proton energy which is calculated by simulating the photopion production with the event generator sophia @xcite . we checked that arrival spectrum at earth for mono - energetic injection of energy @xmath41 ev is consistent with that of stanev et al.(2000 ) ( fig . 2 ) , which is also calculated with the event generator sophia , for the case of ( @xmath24 , @xmath23 ) = ( 1ng , 1mpc ) which is the closest parameter set to the one of stanev et al.(2000 ) . the egmf are little known theoretically and observationally . the upper limit for its strength and correlation length @xmath42 , as measured by faraday rotation of radio signals from distant quasars @xcite , is often used @xcite . we adopt in this study not only this widely accepted value , 1ng , but also relatively strong magnetic field ( 10ng , 100ng ) , which is better for explaining the extension of the energy spectrum and the observed isotropic arrival distribution . in this case , deflection angles of uhecrs become so large that the no counterparts problem can be simply solved . we assume that the magnetic field is represented as the gaussian random field with zero mean and a power - law spectrum . thus , @xmath43 can be written as @xmath44 where @xmath23 is the correlation length of the egmf and @xmath45 characterizes the numerical cut - off scale . we use @xmath46 corresponding to the kolmogorov spectrum . according to our previous study @xcite , the correlation length is chosen to be 1 mpc , 10 mpc , and 40 mpc . 1 mpc is roughly equal to mean separation of galaxies and widely used as the typical value of the correlation length . as we show later , the small scale clustering observed by the agasa is difficult to be reproduced for @xmath47 mpc ( see figure [ fig13 ] and [ fig17 ] ) . we do not have to consider the correlation length larger than 40 mpc . physically one expect @xmath48 , but we set @xmath49 in order to save the cpu time . the universe is covered with cubes of side @xmath23 . for each of the cubes , fourier components of the egmf are dialed on a cubic cell in wave number space , whose side is @xmath50 , with random phases according to the kolmogorov spectrum , and then fourier transformed onto the corresponding cubic cell in real space . we create the egmf of 20 @xmath51 20 @xmath51 20 cubes of side @xmath23 , and outside it , adopt the periodic boundary condition in order to reduce storage data for magnetic field components . similar methods for the turbulent magnetic fields have been adopted @xcite . in this study , we assume that the source distribution of uhecrs is proportional to that of the galaxies . we use the realistic data from the ors @xcite galaxy catalog , which is nearly full sky survey , and contains two subcatalogs , one complete to a b magnitude of 14.5 ( ors - m ) and the other complete to a b major axis diameter of @xmath52 ( ors - d ) . another nearly full sky catalog which is often used in this kind of study is the iras pscz survey @xcite . however , it is suspected that some galaxies in the lsc are excluded . since local overdensity of uhecr sources makes the gzk cutoff less sharp or eliminate it , we use the ors as a source model rather than the iras . in any statistical acceleration mechanism , particles must be kept confined within the acceleration site . luminous galaxies are generally larger than dwarf ones , and thus they are expected to be able to accelerate particles to higher energy . furthermore , it is less well known that luminous galaxies in the lsc distribute outward than faint galaxies , contrary to general clusters of galaxies @xcite . accordingly , we inject uhecrs from galaxies with different limiting magnitudes , and explore the source model which reproduces the current observation . as an example , this selection of luminous galaxies corresponds to restricting the sources of uhecrs to host galaxies of agns . it is still unknown how much an ultimate uhecr source contribute to the observed cosmic ray flux . we thus consider the two cases in which all galaxies are the same , and they inject cosmic rays proportional to their absolute luminosity . this allow us to investigate effects of the uncertainty of the cosmic ray luminosity on the results . since we consider the galaxies with different absolute magnitudes , we use the magnitude - limited subcatalog , i.e. , ors - m , which is also composed of two subsamples , eso - m ( 2437 galaxies ) and ugc - m ( 3279 galaxies ) . we further restrict ourselves to galaxies within 8000 km s@xmath53 and with @xmath54log@xmath55log@xmath56 , which corresponds to the apparent magnitude limit of @xmath57 , where @xmath58 is the dimensionless hubble constant @xmath59 km s@xmath53 mpc@xmath60 and @xmath61 is the recession velocity . in the following , we abbreviate @xmath54log@xmath62 to @xmath63 . we exclude very faint nearby galaxies with @xmath64 . galaxies in the zone of avoidance ( @xmath1 ) are not observed . our final ors sample contains 5178 galaxies with 2346 galaxies in the eso - m and 2832 galaxies in the ugc - m . figure [ fig1 ] shows the sky distribution of our ors galaxies in the equatorial coordinate . the strip between the eso and ugc region @xmath65 is covered by the esgc , which is a part of the diameter - limited subcatalog , ors - d . the absolute magnitudes are shown in figure [ fig2 ] as a function of the recession velocity . throughout the paper , we assume that there are no departures from uniform hubble expansion for simplicity , and use @xmath66 in calculating distance between galaxies in the sample and our galaxy from recession velocity . in order to calculate the energy spectrum and the distribution of arrival directions of uhecrs realistically , there are two key elements of the galaxy sample to be corrected . first , galaxies in a given magnitude - limited sample are biased tracers of matter distribution because of the flux limit . for each galaxy in the sample , we compensate galaxies which is not included in the sample using the selection function given in santiago et al.(1996 ) . the positions of compensated galaxies are determined according to gaussian distribution whose mean is the position of the original galaxy and rms is such that @xmath67 , where @xmath68 is the distance from our galaxy and @xmath69 , @xmath70 are number density of galaxies which is also calculated using the selection function , value of the selection function at distance @xmath68 from us respectively . for the reason of the selection effect , the ors is only sampling the universe out to 8000 km s@xmath53 ( see figure [ fig2 ] ) . we assume that contributions from sources outside it is completely isotropic , and calculate its amount from the number of galaxies inside it . second , our ors sample does not include galaxies in the zone of avoidance ( @xmath1 ) and the esgc region @xmath71 . in the same way , we assume that the source distribution in this region is homogeneous , and calculate its number density from the number of galaxies in the observed region , which is corrected for the selection effect . in figure [ fig3 ] , we show the sky distribution of the ors galaxies corrected in the manner explained above . finally , we now explain how the angular probability distribution of uhecrs are calculated . at first , protons with a flat energy spectrum are injected isotropically at a given point in each of the 26 energy bins , and then propagated in the egmf over @xmath72 gpc for @xmath73 gyr . weighted with a factor corresponding to a @xmath20 power law spectrum , this provides distribution of energy , deflection angle , and time delay of uhecrs as a function of the distance from the initial point . with this distribution , we can calculate the flux and the arrival directions of uhecrs injected at a single uhecr source . then , summing contributions from all the sources , we obtain the angular probability distributions of uhecrs . in this paper , we use the distribution of energies and deflection angles integrated over the time delay , assuming that the cosmic ray flux at the earth is stationary . in figure [ fig4 ] , we show number of protons with energies of @xmath74 ev integrated over deflection angle , time delay as a function of propagation distance . it is noted that this number is proportional to the cosmic ray flux in this energy range from sources in concentric shells with an infinitely small width , assuming that a source distribution is homogeneous . in this subsection , we explain the two statistical quantities , the harmonics analysis for large scale anisotropy @xcite and the two point correlation function for small scale anisotropy . the harmonic analysis to the right ascension distribution of events is the conventional method to search for global anisotropy of cosmic ray arrival distribution . for a ground - based detector like the agasa , the almost uniform observation in right ascension is expected . the @xmath75-th harmonic amplitude @xmath68 is determined by fitting the distribution to a sine wave with period @xmath76 . for a sample of @xmath69 measurements of phase , @xmath77 , @xmath78 , @xmath79 , @xmath80 ( 0 @xmath81 ) , it is expressed as @xmath82 where , @xmath83 , @xmath84 . we calculate the first and second amplitude from a set of events generated according to predicted probability density distribution of arrival directions of uhecrs . if events with total number @xmath69 are uniformly distributed in right ascension , the chance probability of observing the amplitude @xmath85 is given by , @xmath86 where @xmath87 the agasa 57 events is consistent with isotropic source distribution within 90 @xmath15 confidence level as shown later . accordingly , we adopt the amplitude such that @xmath88 as observational constraint , rather than the most probable amplitude calculated from the observed 57 events . the two point correlation function @xmath89 contains information on the small scale anisotropy . we start from a set of generated events or the actual agasa data . for each event , we divide the sphere into concentric bins of angular size @xmath90 , and count the number of events falling into each bin . we then divide it by the solid angle of the corresponding bin , that is , @xmath91 , \label{eqn100}\end{aligned}\ ] ] where @xmath92 denotes the separation angle of the two events . @xmath90 is taken to be @xmath93 in this analysis . the agasa data shows strong correlation at small angle @xmath94 with 5 @xmath16 significance of deviation from an isotropic distribution as shown later . in order to quantify the statistical significance of deviations between models and data we introduce @xmath95 , which is defined by , @xmath96 where @xmath97 is the value of the two point correlation function obtained by the agasa observation at angle @xmath98 , and @xmath99 is statistical error of the @xmath100 due to the finite number of simulated events , which is set to be the observed one . in this study , we take @xmath101 to be 10 deg for properly quantifying deviations of models from the sudden increase of correlation at the small angle scale observed by the agasa . we compare the model predictions with the existing agasa data throughout the paper . therefore , we set the number of simulated events to be that of the agasa data above @xmath102 ev ( 57 ) , and restrict the arrival directions of uhecrs in the range of @xmath103 when calculating these statistical quantities . there are also the present working or development of large - aperture new detectors , such as the hires @xcite and south and north auger @xcite . if they report the arrival directions of observed events , we can make comparisons between the model predictions and their data using the same statistical methods . in this subsection , we present the results of the numerical simulations for the angular images of uhecrs in a variety of conditions concerning strength and correlation length of the egmf and the limiting magnitudes of galaxies . we show the angular images for two energy range @xmath104 ev and above @xmath10 ev , about at which the gzk cutoff of the energy spectrum is predicted . for comparison , we also show the agasa events with the angular images . it is noted that the current agasa data set contains 49 events with energies of @xmath104 ev and 8 events above @xmath10 ev . the calculated angular images of uhecrs originating the galaxies more luminous than @xmath105 ( upper ) , @xmath106 ( middle ) , @xmath107 ( bottom ) for number ( left ) and luminosity ( right ) weighted sources in the case of @xmath108 are shown in figure [ fig5 ] for energy range @xmath104 ev . the agasa 49 events in this energy range are also shown as circles of radius proportional to their energies . spatial structure of the lsc is reflected by prominent high intensity region running north and south at @xmath109 for @xmath105 and @xmath106 . since the luminous galaxies which distribute outward than the faint ones make a substantial contribution in the case of luminosity weighted sources , the region of high intensity becomes to be less obvious in the case of @xmath105 . in other cases , there is no significant difference simply due to small variance of absolute magnitudes of the individual galaxy . we can see that the angular image becomes to be isotropic and show good correlation with agasa events as restricting sources to more luminous galaxies . such correlation of the agasa events with angular image of uhecrs is also pointed out in the case of galaxies with huge infrared luminosity @xcite . even for @xmath110 , there are high intensity regions around @xmath111 which do not correlate with the agasa events . however , if we select sources , which contribute to the currently observed cosmic ray flux , from our ors sample , these high intensity regions can be eliminated . an example in the case of @xmath112 is shown in figure [ fig21 ] as we explain later . figure [ fig6 ] shows the predicted angular image of uhecrs for @xmath113(1,1 ) ( upper ) , ( 1,10 ) ( middle ) , ( 10,1 ) ( bottom ) and @xmath105 ( left ) , @xmath107 ( right ) with energies of @xmath104 ev . the angular image is distorted for @xmath113(1,10 ) , ( 10,1 ) , and dependence on the correlation length is relatively weak because the deflection angle of uhecrs is proportional to @xmath114 . as we will show later , this distortion of the angular image is crucial for the small scale clustering . in figure [ fig7 ] , we show the angular image for @xmath108 ( upper ) , @xmath115 ( middle ) , @xmath116 ( bottom ) and @xmath105 ( left ) , @xmath107 ( right ) with the energy range @xmath117 ev . ( highest energy of the events observed by the agasa is @xmath118 ev . ) because protons with energies above @xmath10 ev must originate within about 50 mpc from us due to photopion production , only nearby galaxies are able to make a substantial contribution to the angular image so that the spatial structure of the lsc is clearly visible . there is no correlation for @xmath105 between the agasa events and the angular image of uhecrs . for @xmath110 , arrival distribution is relatively isotropic , but still has no correlation with the agasa events . as we conclude later , the agasa events above @xmath10 ev might be of different origin , if they are confirmed . we present in this subsection the results of the numerical simulations for the energy spectra of uhecrs . figure [ fig8 ] shows the energy spectrum in the case of @xmath119 ( upper ) , @xmath120 ( middle ) , @xmath121 ( bottom ) ng and @xmath110 for number weighted sources . they are normalized arbitrary , but with the same factors for all the values of the correlation length . the substantial deflection in the egmf results in a significant increase of the path length , then leading to a time delay exceeding the age of the universe or a decrease of energy below @xmath122 ev due to pair production for protons injected at sufficiently large distance , as easily seen in figure [ fig4 ] . thus , the stronger egmf make the gzk cutoff around @xmath10 ev less sharp or eliminate it in figure [ fig8 ] . when the correlation length of the egmf is small , protons can not be deflected effectively and thus its path length and equivalently energy loss is smaller than that for longer correlation length . this can also be seen in figure [ fig4 ] and figure [ fig8 ] . especially , effects of changing the correlation length are large at @xmath123 ev for @xmath124 ng and at @xmath125 ev for @xmath126 ng , where gyroradii of uhecrs are about @xmath72 mpc and thus deflection is strong enough for the diffusion approximation to become applicable in the case of @xmath127 and @xmath128 mpc . in order to determine normalization of the energy spectrum and quantify statistical significance of mean deviation between the fitted energy spectrum and the observed one , we introduce @xmath129 in the similar way as the two point correlation function explained in the previous section . this is defined by @xmath130 where @xmath131 is the energy spectrum observed by the agasa at @xmath132 , and @xmath99 is @xmath133 error at this energy bin . the origin of cosmic rays with energies below @xmath21 ev is also one of the major open questions in astro - particle physics . berezinsky et al.(2002 ) showed that predicted uhecrs flux from grbs fall short of the observed flux below @xmath21 ev assuming an injection spectrum @xmath20 , which is advocated in vietri ( 1995 ) and waxman ( 1995 ) , and also used in this study . however , there are another possible extra - galactic and galactic uhecr production sites , such as agns and young neutron stars @xcite . maximum energy of cosmic ray achieved by them may be lower than @xmath21 ev . in this case , these galactic and/or extra - galactic components may substantially contribute to the cosmic ray flux below @xmath21 ev . throughout the paper , we assume that these components bridge the gap between the observed flux and the predicted one of berezinsky et al.(2002 ) , and restrict ourselves to the energy spectrum only above @xmath21 ev . figure [ fig9 ] shows the energy spectrum normalized so as to minimize @xmath129 in the case of @xmath119 ( upper ) , @xmath120 ( middle ) , @xmath121 ( bottom ) ng and @xmath105 ( left ) , @xmath107 ( right ) for number weighted sources . we also show the significance values @xmath129 in this figure . the predicted energy spectrum for @xmath134 provide good fit to the agasa data , including the extension above @xmath10 ev . however , arrival distribution is significantly anisotropic and inconsistent with the agasa ( see the next subsection ) . in the case of @xmath135 , the energy spectrum can be well fitted with the agasa data below @xmath136 ev , but the gzk cutoff is predicted above @xmath136 ev due to the energy loss by photopion production in contrast to the agasa data . this is because luminous galaxies near the earth are less than dwarf galaxies ( see figure [ fig2 ] ) . although the energy spectrum extends beyond @xmath10 ev for the stronger egmf , it does not provide good fit with the agasa data at @xmath137 ev , where the current data have adequate statistics . consequently , @xmath129 for the stronger egmf is larger than that for @xmath119 ng as easily seen in figure [ fig9 ] . in the same way as figure [ fig9 ] , we show the energy spectra for luminosity weighted sources in figure [ fig10 ] . since contributions from the dwarf galaxies are suppressed in this case , the cutoff of the energy spectra for @xmath105 is predicted . for @xmath110 , the results are almost the same as number weighted ones . in this subsection , we show the results of statistics on the arrival directions of uhecrs . figure [ fig11 ] shows the predicted first harmonics for number weighted sources as a function of limiting magnitude . the error bars represent the statistical error due to the finite number of simulated events , whose number is set to be that observed by the agasa in the energy range of @xmath74 ev . arrival directions of uhecrs are restricted in the range @xmath103 in order to compare our results with the agasa data . the solid lines indicate the harmonic amplitude obtained by the agasa data . the shaded region is expected from the statistical fluctuation of isotropic source distribution with the chance probability larger than @xmath138 . clearly visible in figure [ fig11 ] is that the anisotropy is least for @xmath139 , which is caused by broadened distribution of luminous galaxies in the lsc . the amplitude at @xmath3 is consistent with isotropic source distribution within 1 @xmath16 level for all the @xmath140 . compared with the agasa data , a significant anisotropy can be seen for @xmath141 , although the energy spectra in these cases extend above @xmath10 ev as shown in the previous subsection . increase of the amplitude around @xmath142 is mainly due to two giant galaxies existing in the almost same direction @xmath143 in the vicinity of our galaxy ( at @xmath144 30 mpc ) . comparing the results of different strength of the egmf , we find that the anisotropy in the cases of 10 , 100 ng is larger than that of 1 ng below @xmath145 contrary to expectation . we note again that stronger egmf results in shorter cosmic ray horizon because of time delay and energy loss due to the substantial deflection in the egmf ( see figure [ fig4 ] ) . that is , spatial structure of nearby galaxies is strongly reflected in the large scale anisotropy for the strong egmf . above @xmath145 , since luminous galaxies distribute outward , the anisotropy in the cases of the strong egmf is smaller as expected . also , increase of correlation length results in more effective deflections , and then leads to smaller anisotropy in the case of 1 , 10 ng . however , this dependence is opposite for 100 ng , which is caused by a reason somewhat similar to that of slightly large anisotropy for the strong egmf below @xmath145 explained above . gyroradius of proton with energy @xmath9 ev in the egmf of 100 ng is 1 mpc , which is smaller than @xmath146 mpc . in this case , uhecr deflection is strong enough for the diffusion approximation to become applicable . thus , the cosmic ray horizon is extremely shorter than that for @xmath147 mpc , so that the anisotropy arise from nearby galaxies in the lsc . figure [ fig12 ] shows the predicted second harmonics for number weighted sources in the same manner as the first harmonics . dependence on the limiting magnitude is weaker than that of the first harmonics . there is no parameter set @xmath148 which is consistent with isotropic source distribution . we also calculated the first and second harmonics for luminosity weighted sources , and found that the anisotropy is slightly less than that of number weighted sources for @xmath149 because of small contributions from the dwarf galaxies , which concentrate in the lsc . for @xmath150 , the results are almost the same as number weighted sources . the two point correlation functions are shown for number weighted sources in the case of @xmath151 ( left panels ) and @xmath152 ( right panels ) in figure [ fig13 ] . the histograms represent the agasa data , which show the statistically significant correlation at the smallest angle scale @xmath153 . @xmath154 defined in the previous section is also shown . at larger angles @xmath155 , there are relatively strong correlation for @xmath151 , which is not consistent with the observed isotropic distribution , whereas arrival distributions for @xmath152 are sufficiently isotropic . among the cases of @xmath152 , noticeable feature is coexistence of a strong correlation at small angles and sufficiently isotropic distribution at larger scale for @xmath108 . this is a consequence of the sharp peak of the uhecr angular image in the case of @xmath108 ( see figure [ fig5 ] ) . this strong correlation at small angles is reduced or eliminated for longer correlation length or stronger egmf . due to small number of observed events , the statistics of the two point correlation function is limited . for this reason , @xmath154 for @xmath108 is equal to that for @xmath156 . however , if the future experiments like the pierre auger array @xcite confirm the small scale clustering of uhecr arrival directions with more event number , the case of @xmath108 will be favored with better significance . let us summarize the results of the three quantities ( energy spectrum , harmonic amplitude , two point correlation function ) . first , the energy spectrum for @xmath119 ng provide better fit to the agasa data than that for strong egmf in terms of @xmath129 . among this case , the extension of the energy spectrum is also reproduced for @xmath157 , although there is significant large scale anisotropy contrary to the agasa data in this case . in the case of @xmath158 , the gzk cutoff is predicted . second , the amplitude of the first harmonics is consistent with the agasa data at @xmath159 irrespective of @xmath140 . the amplitude of the second harmonics is not consistent with the agasa for any @xmath160 . finally , the two point correlation function for @xmath161 indicate coexistence of a strong correlation at small angles @xmath153 and sufficiently isotropic distribution at larger scale . in sum , the source model of @xmath161 seems to reproduce the observation better than another parameter sets . the amplitude of the second harmonics and the extension of the energy spectrum can not be explained in this source model . up to present , we have not explained the large scale isotropy quantified by the amplitude of the second harmonics and the small scale clustering except for @xmath108 , as well as the extension of the cosmic ray spectrum above @xmath10 ev . since we assumed that all the galaxies more luminous than a given limiting magnitude contribute to the cosmic ray flux , the number of sources is maximal and the fluctuations around the assumed non - isotropic distribution is minimal , which make the anisotropy most visible . furthermore , each source contribute at most one event in this case , it is difficult to obtain clusters which are likely to reflect the point - like sources . given this situation , we select some galaxies from our ors sample , and investigate dependence of the results on the number of selected galaxies . it is also expected that we can know the effect of fluctuation of source number within the gzk sphere on the energy spectrum above @xmath10 ev . as an example , this selection corresponds to restricting the sources of uhecrs to host galaxies of grbs , which contribute to the currently observed cosmic ray flux . we assumed that contributions to arrival distribution from sources outside 8000 km s@xmath53 from us , within which the ors is sampling the universe , is completely isotropic one . however , since we select only some galaxies in the following , point - like nature of sources has to be taken into account . thus , we distribute galaxies homogeneously in this region , and including them , we select some galaxies from our ors sample . we restrict ourselves to the limiting magnitude of @xmath112 and number weighted sources , which produce the most isotropic arrival distribution of uhecrs . it is noted that the number density of galaxies more luminous than @xmath112 in our ors sample is @xmath162 mpc@xmath7 . in figure [ fig14 ] , we show the amplitude of the first harmonics as a function of the number fraction ( nf ) of selected sources to all the ors galaxies more luminous than @xmath112 for all the @xmath140 . for each number fraction , we plot the average over all trial of source selection and realization from the simulated probability distribution with two error bars . the smaller is the statistical error due to the finite number of observed events , while the larger is both the statistical error and the cosmic variance , that is , variation between different selections of sources from our ors sample . in order to obtain the average and variance , we dial the simulated sets of events 100 times from probability distribution predicted by a specific source distribution , and the sets of sources 30 and 10 times from our ors sample for the number fraction of @xmath163 and @xmath164 respectively in the case of @xmath165 ng , but 30 times for all the number fractions in @xmath126 ng . we note that the number of sources , which contributes to the cosmic ray flux , for each set of @xmath140 differ from the others , even for the same number fraction , because of the different range of uhecrs ( see figure [ fig4 ] ) . for example , there is no source within 100 mpc for the number fraction of @xmath166 ( for the number density of @xmath167 mpc@xmath7 ) in the case of @xmath126 ng . as a general trend , anisotropy increases with decreasing the number fraction , because the arrival directions of uhecrs clumps to that of sources which is small in number . for large number fraction @xmath168 , the amplitudes approach to the values in the case that all galaxies more luminous than @xmath112 contribute to the observed cosmic ray flux , as they should . as clearly visible in figure [ fig4 ] , uhecrs can propagate the longest distance for @xmath108 , in which the deflection angle , and then the effective path length , is smallest among for another parameters . accordingly , distant sources make a substantial contribution to the arrival distribution of uhecrs , unlike in the case of stronger egmf . this is reflected in the decrease of the amplitude around the number fraction of @xmath169 for @xmath108 . except for @xmath126 ng , arrival distribution is sufficiently isotropic compared to isotropic source distribution for almost all the number fractions . the amplitude of the second harmonics is also shown in figure [ fig15 ] . dependence on the number fraction is roughly same as that of the first harmonics . for @xmath119 and @xmath120 ng , the predicted amplitude is sufficiently isotropic for the number fraction of @xmath170 . on the other hand , it appears to be inconsistent with the observation for any number fraction in the case of @xmath126 ng . figure [ fig16 ] shows @xmath171 of the two point correlation function as a function of the number fraction . increase of deviations between models and data with decreasing the number fraction is attributed to the clump of uhecrs at directions of their sources , similarly to the case of the harmonic analysis . especially , this dependence is quite noticeable for @xmath108 , where uhecrs hardly deflect so that correlation at small angle is too strong . at larger number fraction ( @xmath172 ) , it can be seen that the model of @xmath108 provide relatively better fit to the observation than another models . in figure [ fig17 ] , we show the two point correlation function predicted by a specific source scenario for @xmath108 , @xmath173 and @xmath115 with the number fraction of @xmath5 . @xmath171 is also shown . each source distribution is selected so that it predicts the smallest value of @xmath171 for respective set of @xmath140 . only for @xmath108 , another condition is also imposed that the harmonic amplitude predicted by the source model is consistent with the agasa observation within @xmath133 level . the correlation of events at small angle scale and sufficient isotropic distribution at larger scale are well reproduced for @xmath108 . there is no significant correlation for @xmath156 and @xmath115 , because of the larger deflection angle of uhecrs . in the cases of strong egmf or longer correlation length , it appears to be difficult to reproduce the strong anisotropy only at the smallest angle scale , even for the case of minimal value of @xmath171 . considering the calculated results of the first and second harmonics and the small scale anisotropy all together , the model of @xmath108 and the number fraction of @xmath5 ( the source number density of @xmath6 mpc@xmath7 ) seems to reproduce the observations better than another parameter sets . however , due to the small number of observed events , the statistics for the two point correlation function is limited . for this reason , @xmath171 does not differ very much from each other . the future experiments like the pierre auger array @xcite will decrease statistical uncertainty and provide more strong constraints to the model predictions . next we discuss dependence of the energy spectrum on the number fraction . in figure [ fig18 ] , we show the predicted energy spectra in the case of @xmath108 for the number fraction @xmath174 , @xmath169 , @xmath175 in descending order . they are normalized so as to minimize @xmath129 . the shaded regions represent 1 @xmath16 error due to the source selection from our ors galaxies more luminous than @xmath3 . as is evident from this figure , decreasing the number fraction increases fluctuation of the predicted energy spectrum , especially above @xmath10 ev . for the number fraction of @xmath175 ( the source number density of @xmath176 mpc@xmath7 ) , since there are no source within the gzk sphere , the gzk cutoff is clearly visible . the predicted energy spectrum is roughly consistent with the agasa observation within 1 @xmath16 level in the case of the number fraction of @xmath169 , where the observed isotropy and clusters are also reproduced well . however , this result should be interpreted with care . since the source number density is @xmath6 mpc@xmath7 in the case of the number fraction of @xmath169 , mean number of sources is @xmath8 within the gzk sphere . therefore , the agasa 8 events above @xmath10 ev , which do not constitute the clustered events with each other , must originate from at most a few sources . we found a source model which explain the extension of the energy spectrum . in this source model , there are two sources within the region covered by the ors ( @xmath177 mpc ) , one at @xmath178 mpc , the other at @xmath179 mpc . in figure [ fig19 ] , we show the energy spectrum predicted by this source model . the dotted line represent contribution from the nearest source . the predicted energy spectrum provide good fit to the observed one including the extension beyond @xmath9 ev . however , almost all the events above @xmath9 are generated by the nearest source as evident from figure [ fig19 ] . these events are strongly concentrated in a single center , which is inconsistent with the agasa 8 events . it seems to be difficult to reproduce the agasa observation above @xmath102 ev ( including above @xmath10 ev ) by a single scenario of uhecr origin , even for @xmath180 . recently , the hires collaboration @xcite reported the cosmic ray flux from @xmath181 ev to over @xmath10 ev including the gzk cutoff @xcite . in our numerical simulations , the sharp gzk cutoff of the energy spectrum is predicted for the number fraction of @xmath182 as shown above . the small scale correlation of events in this case is too strong for @xmath108 as is evident from figure [ fig20 ] , where the two point correlation function is shown in the same way as figure [ fig17 ] , but with the number fraction of @xmath182 . this strong correlation is reduced for @xmath156 and @xmath115 , but still inconsistent with the agasa data . finally we show in figure [ fig21 ] realizations of 57 arrival directions of uhecrs with energies of @xmath74 ev , predicted by the source model of @xmath180 of figure [ fig17 ] . the agasa 57 events in this energy range are also shown . the arrival directions are restricted in the range @xmath103 in order to compare our results with the agasa data . the arrival distribution of uhecrs seems to be isotropic on a large scale with clusters of events meaning the small scale anisotropy . the amplitudes of the first and second harmonics are consistent with the agasa , but the extension of the spectrum is not . ( mean number of simulated events above @xmath10 ev is only 1 . ) in this example , c6 doublet at @xmath183 is well reproduced . also , high intensity regions around @xmath111 mentioned in section 3.1 . are eliminated under the favor of the source selection . we presented the results of numerical simulations on the propagation of uhe protons with energies of @xmath0 ev injected by a discrete distribution of galaxies using the ors data sample , which is properly corrected for the selection effect and absence of galaxies in the zone of avoidance ( @xmath1 ) . we adopted strength of the egmf to be not only 1 ng but also 10,100 ng , in order to explore the possibility of reproducing the extension of the energy spectrum above @xmath10 ev . we calculated the three observable quantities , cosmic ray spectrum , harmonic amplitude , and two point correlation function . with these quantities , we made comparisons between the model predictions and the existing agasa data . at first , we explored the source model consistent with the current observations , as a function of the limiting magnitudes of galaxies . the energy spectrum for the strong egmf shows the extension beyond @xmath10 ev . however , it can not be well fitted with the observed energy spectrum , especially at @xmath137 ev where the current data have adequate statistics . on the other hand , the energy spectrum for @xmath119 ng provide good fit with the observation below @xmath136 ev , and statistical significance of deviation @xmath129 is smaller than that for the strong egmf . among this case , the extension of the energy spectrum is also reproduced for @xmath157 , although there is significant large scale anisotropy contrary to the agasa data in this case . in the case of @xmath158 , the gzk cutoff is predicted . we found that the angular image became to be isotropic and showed good correlation with agasa events as restricting sources to more luminous galaxies . as a consequence , galaxies more luminous than @xmath4 mag produce the angular image of uhecr which is most isotropic roughly irrespective of strength and correlation length of the egmf . however , it is not isotropic enough to be consistent with the observed amplitude of the second harmonics , even for @xmath112 . we further found that the small scale anisotropy is well reproduced in the case of @xmath108 because of the small deflection angle of uhecrs . next , in order to obtain sufficiently isotropic arrival distribution of uhecrs and small scale clustering even for the case of other than @xmath108 , we randomly selected sources from our ors sample , and investigate the dependence of the results on their number . we found that the arrival distribution became to be sufficiently isotropic in the case that the number fraction @xmath170 of the ors galaxies more luminous than @xmath112 is selected as uhecr sources for @xmath119 , @xmath120 ng . among this case , the small scale clustering can also be reproduced for @xmath180 better than the cases of another parameter sets . we furthermore found that the extension of the energy spectrum is explained in this case within roughly 1 @xmath16 error due to source selection from our ors sample because of the large fluctuation above @xmath10 ev . in sum , we showed that the three observable quantities including the gzk cutoff of the energy spectrum can be reproduced in the case that the number fraction @xmath5 of the ors galaxies more luminous than @xmath4 mag is selected as uhecr sources . in terms of the source number density , this constraint corresponds to @xmath6 mpc@xmath7 . as mentioned in the previous subsection , we showed that the three observable quantities including the gzk cutoff of the energy spectrum can be reproduced in the case that the number fraction @xmath5 of the ors galaxies more luminous than @xmath4 mag is selected as uhecr sources . however , we should interpret this result with care . since the agasa 8 events above @xmath10 ev do not constitute the clustered events with each other , the number of their sources within the gzk sphere must be much larger than the event number ( 8) . on the other hand , the number of sources within the gzk sphere is at most only 1 in the case of the number fraction of @xmath169 ( the source number density of @xmath184 mpc@xmath7 ) . it is suspected that cosmic rays above @xmath10 ev ( hereafter ehecrs ; extremely - high energy cosmic rays ) are strongly concentrated in a single center , even if the extension of the energy spectrum is explained . therefore , large fraction of ehecrs observed by the agasa might originate from sources other than ones of uhecr with energies of @xmath185 ev . the number density of ehecr source must be much larger than @xmath184 mpc@xmath7 . we can put forward two candidates as such ehecr production site . it is suggested that iron ions from the surfaces of young strongly magnetized neutron stars may be accelerated to @xmath10 ev through relativistic mhd winds @xcite . however , it seems to be difficult to obtain sufficiently isotropic arrival directions of ehecrs from sources localized in the galactic plane by deflection due to the galactic magnetic field @xcite . the decay of some supermassive particles , which could be produced from tds , or be certain msrps , are also considered as probable ehecrs origin . decay rate of the supermassive particles required to explain the observed ehecr flux are estimated as @xmath186 mpc@xmath7 yr@xmath53 @xcite . multiplying by the current ehecr observing time ( @xmath187 yr ) , we obtain the number density of @xmath188 mpc@xmath7 , which is sufficiently large to reproduce not clustered agasa 8 ehecrs . as for the large scale anisotropy , distribution of tds in the universe would be homogeneous . msrps are expected to be clustered in galactic halos , and thus ehecr flux will be dominated by contribution from our own galactic halo . accordingly , arrival distribution would be isotropic enough to be consistent with the current agasa observation in both cases . furthermore , these scenarios are not constrained by contribution to the cosmic ray flux at lower energy which must be smaller than observed one , since hard injection spectrum is generally predicted . for the reasons stated above , the top - down scenarios may be favored to explain large fraction of the 8 ehecrs observed by the agasa . there is a problem if all the ehecrs observed by the agasa originate from sources other than those of uhecrs . if we consider that sources of ehecrs and uhecrs precisely differ from each other , there would be no clustered event set which includes both ehecrs and uhecrs , because clustered events are expected to reflect a single point source . however , 2 ehecrs out of 8 observed by the agasa constitute the clustered events with uhecrs with energies of @xmath104 ev ( c1 and c3 in hayashida et al.(2000 ) ) . we have to interpret that the two clusters happen to be produced by chance . chance probability of more than two accidental clusterings within @xmath189 between 49 uhecrs and 8 ehecrs is quite low @xmath190 . the agasa observation seems to deviate from our prediction with @xmath191 confidence level . however , about 1 ehecr out of 8 may originate in the bottom - up scenarios . indeed , 57 cosmic rays above @xmath102 ev predicted by the source model of figure [ fig21 ] include an ehecr . in this case , since the source of this single event will be located within the gzk sphere , it is likely that this event constitutes a clustered event set with uhecrs . chance probability of more than one accidental clustering rise up to @xmath192 , which is high enough for us to interpret one of the two doublets to be produced merely by chance . thus , our scenario is consistent with the current agasa observation within 1 @xmath16 level . the event correlation between ehecrs and uhecrs would provide an important test of our conclusion with sufficient amount of data from future experiments . for example , when the event number of uhecrs above @xmath102 increases to 10 times as many as that of the current agasa observation ( 57 ) , the number of ehecrs which lie within @xmath189 from uhecrs by chance becomes @xmath193 . if future experiments observe event number of such ehecrs within this range , our scenario will be supported . here we note that the source number within the gzk sphere is at most 1 in the case of nf=@xmath169 . therefore the number of event clustering between an ehecr and an uhecr which can be explained by bottom - up scenarios is at most 1 . another test of our conclusion is distinguishing the chemical composition of primary cosmic rays which initiate air showers in the earth atmosphere . in the bottom - up scenarios , protons or nuclei are primary particles . on the other hand , photons are considered as primary in the top - down scenarios @xcite . information on the chemical composition is mainly provided by the depth of shower maximum for fluorescence observation of the air shower . there are a number of projects utilizing the fluorescence technique , such as hires @xcite , south and north auger @xcite , owl @xcite and euso @xcite . these experiment would provide information on the chemical composition and a test of our conclusion . the hires collaboration @xcite reported the cosmic ray flux from @xmath181 ev to over @xmath10 ev including the gzk cutoff @xcite . as discussed above , we showed that the three observable quantities except for ehecr events observed by the agasa can be reproduced in the case that the number fraction @xmath5 of the ors galaxies more luminous than @xmath4 mag is selected as uhecr sources . in figure [ fig22 ] , we show the predicted energy spectrum fitted to the data measured by hires - i detector in the case of @xmath180 . as easily seen , the fit is good as compared with the middle panel of figure [ fig18 ] . if the flux measured by the hires collaboration is correct and observational features about the arrival distribution are same as the agasa , our source model can explain both the arrival distribution of uhecrs and the flux at the same time . when the large - area detectors like the pierre auger array @xcite start operating , the statistical error of the observed energy spectrum will decrease . this allows us to draw a conclusion about the origin of ehecrs or whether they exist or not . cccc object & density & ratio & ref + & [ mpc@xmath7 ] & to our result & + & & ( @xmath184 mpc@xmath7 ) & + agn & @xmath195 & 100.0 & @xcite + fr - ii radio galaxy & @xmath196 & 0.03 & @xcite + bl lac & @xmath197 & 0.3 & @xcite + grb & @xmath195 & 100.0 & @xcite + dead quasar & @xmath198 & 500.0 & @xcite + colliding galaxy & @xmath199 & 0.7 & @xcite + in this subsection we discuss implications of our results , obtained by comparison of the model predictions with the existing agasa data , for the origin of uhecrs with energies of @xmath104 ev . the number densities of possible uhecr production sites and ratios to our result ( @xmath6 mpc@xmath7 ) are tabulated in table [ tbl : num ] . blast waves in agn jet could accelerate particle to a few tens of eev @xcite and similarly for agn cores . however , due to the interaction of uhecrs with very high radiation fields in and around the central engine of an agn , these maxima are unlikely to be achieved . this is consistent with our result , because the number density of agn are much larger than our results , and thus , arrival distribution may be inconsistent with the agasa if all the agns contribute to the cosmic ray flux above @xmath102 ev . among agns , the most promising acceleration sites for uhecrs are the so - called hot - spots of fr - ii radio galaxies @xcite . however , their number density seems to be too small , as compared with our result . it is unlikely that fr - ii radio galaxies are responsible for all the agasa 49 events with energies of @xmath104 ev . another possible uhecrs production sites are bl lacertae objects ( bl lacs ) @xcite . their number are larger than fr - ii radio galaxies , but still insufficient to reproduce all the 49 events . grb can be shown to in principle accelerate protons up to @xmath10 ev @xcite . the effective density of grb is determined by the grb rate @xmath200 h@xmath201 mpc@xmath7 yr@xmath53 @xcite ) and the typical time delay @xmath202 of uhecrs , which depends on the propagation distance from sources to the earth . since we are now considering uhecrs with energies of @xmath104 ev , we have to include contribution from sources outside the gzk sphere when calculating the typical time delay . we calculated the mean time delay from our numerical data assuming homogeneous grbs distribution , and obtained @xmath203 yr . this gives the grb effective density @xmath204 mpc@xmath7 . from our result , if about 1 @xmath15 of grbs produce uhecrs , arrival distribution predicted by grbs would be consistent with the agasa . the compact dynamo model has been proposed as a natural mechanism for accelerating cosmic rays in dead quasars @xcite . only 0.2 @xmath15 of dead quasars are required to contribute to the comic ray flux above @xmath102 ev . large fraction of luminous infrared galaxies ( lirgs ) are found to be interacting systems , suggesting that collisions and merging processes are responsible for triggering the huge ir light emission @xcite . in such systems , favorable environments for accelerating cosmic rays to @xmath10 ev are provided by amplified magnetic fields on the scale of tens kpc resulting from gravitational compression , as well as high relative velocities of galaxies and/or superwinds from multiple supernovae explosions @xcite . according to smialkowski et al.(2002 ) , we take the number density of such colliding galaxies as that of galaxies with huge far infrared luminosity @xmath18 . the number density is roughly equal to our result . if all the galaxies with @xmath18 are responsible for uhecrs , predicted arrival distribution would be consistent with the agasa . indeed , smialkowski et al.(2002 ) showed that arrival distribution of uhecrs predicted by lirgs is sufficiently isotropic such that statistical tests they used for large scale anisotropy was not conclusive for distinguishing between isotropic and lirgs source distribution . furthermore , they showed that the probability of occurring the clustered events are more than 10 times higher for lirgs source distribution than that for isotopic distribution . their results well agree with ours . our result is supported by a statistical analysis of clustering of uhecrs developed by dubovsky et al.(2000 ) , which allow an estimate of the minimum number density of sources . they applied their arguments to the 14 events above @xmath10 ev observed by several experiments . they obtained the minimum number density of sources @xmath205 mpc@xmath7 . since we are now considering uhecrs below @xmath10 ev , we have to apply the arguments given in their paper to the cosmic ray events below @xmath10 ev . let us apply the arguments to the 49 events with energies of @xmath185 ev observed by the agasa . according to them , the number density must be larger than @xmath206 where @xmath207 , @xmath208 , @xmath209 are the total number of events , the number of events in clusters , e - folding length of cosmic ray luminosity from a single source respectively . the value of @xmath209 can be obtained as @xmath210 mpc from figure [ fig4 ] . taking @xmath211 and @xmath212 , eq . ( [ minnum ] ) gives @xmath213 mpc@xmath7 which is consistent with our result @xmath6 mpc@xmath7 . throughout the paper , we assumed that the statistical properties of the egmf are uniform . however , simple analytical arguments based on magnetic flux freezing , and large scale structure simulations passively including the magnetic field @xcite demonstrate that the magnetic field is most likely as structured as are the baryons . the egmf as strong as @xmath214 g in sheets and filaments of large scale galaxy distribution , such as in the lsc , are compatible with existing upper limits on faraday rotation @xcite . it is suspected that the arrival distribution of uhecrs depends on the fields in the immediate environment of the observer . however , numerical simulations of uhecr propagation in inhomogeneous egmf over cosmological distances are highly time - consuming . with the assumption of homogeneous egmf , we have performed numerical simulation of spherically symmetric propagation . provided that the egmf is inhomogeneous , the propagation of uhecrs from a single point source has no longer spherical symmetry . as a result , we have to specify the earth position in the universe . we also have to choose the detector ( earth ) size so small enough for us to accurately calculate the arrival directions . in this case , the number fraction of uhecrs arriving at the earth to injected ones is extremely small . this requires the number of particle to be propagated several orders of magnitudes higher than that used in this study , which takes enormous cpu time . although the strong egmf in the lsc is compatible with the faraday ratation measures , the cosmic ray observations do not appear to support such local egmf . indeed , we showed in this study that small scale clustering can be well reproduced in the case that uhecrs propagate along nearly straight lines . if the local strong egmf affects the arrival directions of uhecrs , small scale clustering observed by the agasa will not be obtained . this difficulty can be seen from the studies by another authors . there are several groups who study the propagation of uhecr in such strong egmf in the lsc @xcite . they assume simplified source distributions which represent the lsc . the strong egmf of @xmath214 g leads to substantial deflections of uhecrs , which are better for explaining the observed isotropic arrival distribution of uhecrs . however , consistency of small scale anisotropy and also large scale isotropy predicted by their scenarios with the agasa observation is somewhat worse than that predicted by our scenario . the strong egmf in the lsc seems to be unfavored by the current cosmic ray observation , although it is not ruled out by the faraday rotation measurements . of course , since the event number of uhecrs is currently very small , we can not absolutely rule out such strong egmf in the lsc . fulture experiments would increase the significance level of statement about whether such strong egmf affects the arrival directions of uhecrs or not . in this section , we present our conclusions obtained from comprehensive study on the bottom - up scenarios . for the origin of ehecrs , we conclude that large fraction of the agasa 8 events above @xmath10 ev might originate in the topdown scenarios , or that the cosmic ray flux measured by the hires experiment might be better . we also discussed the origin of uhecrs below @xmath9 ev through comparisons between the number density of astrophysical source candidates and our result ( @xmath6 mpc@xmath7 ) . at present , we can not conclude which source candidates are responsible for uhecrs below @xmath9 ev . however , our result for the source number density ( @xmath6 mpc@xmath7 ) will be useful when searching for the ultimate uhecr source . throughout the paper , we set constraints to the source models so that their predictions reproduce the current agasa observation within 1 @xmath16 error of the statistical fluctuation due to finite number of the observed events . however , if we relax the constraints to within 2 @xmath16 error , the energy spectrum ( figure [ fig9 ] and [ fig10 ] ) and the amplitude of the second harmonics ( figure [ fig12 ] ) can be reproduced in the case of @xmath215 without the source selection . in this case , since the source number density is @xmath216 mpc@xmath7 , there are @xmath217 sources within the gzk sphere . the agasa events above @xmath10 ev , which do not constitute the clustered events with each other , may also be reproduced , contrary to the case of the source number density of @xmath6 mpc@xmath7 . thus , we can not absolutely rule out the bottom - up scenarios as the origin of ehecrs . the statistics for the quantities which we used is limited by the small number of observed events , which make difficult to draw a decisive conclusion on the origin of uhecrs . however , the present working or development of large - aperture new detectors , such as hires @xcite and south and north auger @xcite will considerably decrease the statistical uncertainties . there are furthermore plans for space based air shower detectors such as owl @xcite and euso @xcite which would detect about 1000 events per year above @xmath10 ev @xcite . with the method to constrain the source model and their number density developed in this paper , these experiments will reveal the origin of uhecrs with better significance in the very near future . we would like to thank dr . t. stanev for kindly giving us the data of interaction length and inelasticity of uhe protons due to the photopion production with photons of the cosmic microwave background . this research was supported in part by giants - in - aid for scientific research provided by the ministry of education , science and culture of japan through research grant no.s14102004 and no.s14079202 .

the masses of supermassive black holes ( sbhs ) show correlations with bulge properties in disk and elliptical galaxies . we study the formation of galactic structure within flat - core _ triaxial _ haloes and show that these correlations can be understood within the framework of a baryonic component modifying the orbital structure in the underlying potential . in particular , we find that terminal properties of bulges and their central sbhs are constrained by the destruction of box orbits in the harmonic cores of dark haloes and the emergence of progressively less eccentric loop orbits there . sbh masses , , should exhibit a tighter correlation with bulge velocity dispersions , , than with bulge masses , , in accord with observations , if there is a significant scatter in the relation for the halo . in the context of this model the observed relation implies that haloes should exhibit a faber - jackson type relationship between their masses and velocity dispersions . the most important prediction of our model is that halo properties determine the bulge and sbh parameters . the model also has important implications for galactic morphology and the process of disk formation . 3mark iii # 1#2#3#4=.24 = .24 = .24 = .24

astro-ph0301338

the possibility that supermassive black holes ( sbhs ) inhabit the centers of many if not most galaxies , and the observed correlation between sbh masses and galactic bulge properties , has potentially a fundamental significance for our understanding of galaxy formation and evolution . the relationships between black hole and bulge properties include a loose relationship between sbh and bulge masses , @xmath5 , and an apparently much tighter one between the sbh mass and the velocity dispersion in the corresponding bulge , @xmath6 ( e.g. , ferrarese & merritt 2000 ; gebhardt et al . 2000 ; tremaine et al . 2002 ; cf . reviews by kormendy & gebhardt 2001 ; merritt & ferrarese 2001 ) . in this paper we attempt to provide a physical explanation for these relationships between sbhs and their host galaxies . our model is based on the interaction between the dark haloes of galaxies and the baryonic components settling in their midst . as baryonic matter accumulates to form the bulge and sbh , the orbital structure of the underlying gravitational potential is modified . this , in turn , affects the subsequent accumulation of gas , which is highly dissipative and therefore is sensitive to orbital geometry . consequently , halo properties will determine those of the bulge and the sbh , and this will give rise to correlations between them that are compatible with the observed ones . cuspy , triaxial haloes appear to be a natural outcome of dissipationless collapse in a cold dark matter ( cdm ) dominated universe ( e.g. , cole & lacey 1996 ) . interactions with the baryonic component during the initial stages of collapse may affect the triaxiality , making it milder but still non - negligible ( dubinsky 1994 ) . before it even becomes dominant in the central regions , a clumpy baryonic component can also level off the central cusps of dark haloes , producing harmonic cores ( el - zant , shlosman & hoffman 2001 ) . even though this may affect the equidensity contours , again making them rounder , it need not symmetrize the equipotentials ; these can remain asymmetric if triaxiality is not affected beyond some radius ( for example , a homogeneous bar has a non - axisymmetric force contribution inside its density figure ) . from an observational standpoint it appears that haloes of fully formed galaxies tend to have nearly constant density cores ( de blok & bosma 2002 ) and that residual potential axial ratios of about 0.9 in cdm haloes are plausible , _ even in present day galaxies _ ( e.g. , kuijken & tremaine 1994 ; rix & zaritsky 1995 ; rix 1995 ) . the orbital structure of the inner regions of slowly rotating , non - axisymmetric potentials with harmonic cores is dominated by box orbits ( e.g. , binney & tremaine 1987 ) , which have no particular sense of circulation . they are self - intersecting and , therefore , can not be populated with gas . dissipation causes material to sink quickly toward the only long - lived attractor available the center ( e.g. , pfenniger & norman 1990 ; el - zant 1999 ) . in the process , interaction with the triaxial harmonic core causes the baryonic material to lose most of its angular momentum . the combined system would , therefore , also be slowly rotating . thus , unless star formation terminates the collapse , the final concentration of the first baryonic material could be extremely large . the onset of star formation is expected to occur when the baryonic material becomes self - gravitating roughly speaking , when its density becomes larger than that of the halo core . this also happens to be the criterion for the destruction of the harmonic core and the emergence of loop ( or tube ) orbits , which do have a definite sense of circulation . the role of the sbh is to contribute to the emergence of these orbits in the very central region . it is the collusion between the sbh and the more extended hot baryonic component in creating the loop orbits that leads to the correlations claimed in this paper . whereas , in three dimensions , a box orbit can be represented as a superposition of three ( generally ) incommensurable radial oscillations along mutually perpendicular axes , which are efficiently attenuated by dissipation , loop orbits are best described in terms of modest radial and vertical excursions superposed on rotational motion about the center . while the vertical and radial excursions , like the oscillations characterizing box orbits , are attenuated by dissipation , the rotational motion is not efficiently dissipated among gas clouds populating such orbits ( in the same sense of circulation ) . the minimization of the radial and vertical oscillations results in closed periodic orbits that are confined to a plane . we expect the amount of gas dissipation on loop orbits to depend on their axial ratios for , again , motion on highly eccentric orbits would lead to shocks and the accompanying loss of angular momentum on a short dynamical timescale , as observed in numerical simulations of gas flows in barred galaxies ( e.g. , heller & shlosman 1994 ) . lacking a detailed model for the dissipation rate associated with gaseous motion , we will assume that there is a critical eccentricity _ below _ which the loop orbits can serve as long - lived attractors for dissipative motion . in other words , gas populating these orbits will evolve only secularly and not dynamically . fortunately , this dependence on critical eccentricity will turn out to be weak . in the following we explore , within the above framework , the formation of galactic bulges and central compact objects . we show that it is possible to deduce a well - defined linear correlation between the masses of the central sbhs and corresponding bulges in given dark matter haloes , with mass ratios comparable to those observed . we demonstrate that , if dark matter halo cores follow a faber - jackson type relation ( faber & jackson 1976 ) between their masses and velocity dispersions , then a similar relation also applies to the bulges . an @xmath4 relation between the sbh mass and the bulge velocity dispersion , which under certain assumptions can have smaller scatter than the @xmath7 relation , also follows naturally . within this framework , it is possible to make a number of testable predictions concerning the related structures of bulges , sbhs and their host haloes . we discuss these in the final section . formal aspects of the perturbation analysis applied to bulge - halo systems have been deferred to appendices a and b. preliminary results of this work have been reported by shlosman ( 2002 ) . since , at this stage , we are interested in _ generic _ dynamical phenomena related to orbital structure , the exact form of the halo potential is immaterial as long as it exhibits a harmonic core where no loop orbits can exist . the exact distribution of the baryonic material is also not crucial , except , again , for its central density distribution . if this diverges , say as @xmath8 , loop orbits will be created all the way to the center , since the bulge - halo system no longer possesses a harmonic core . in this case there is no need to form the sbh . if , on the other hand , the ( proto)-bulge has a harmonic core of its own , there will still be a nearly constant density region near the center with only box orbits unless a central point mass is present . we will assume that ` cuspy ' bulges , when they exist , are products of later evolution ( for example , a result of star formation and cold dissipationless collapse : e.g. , lokas & hoffman 2000 ) . we use a logarithmic standard form ( e.g. , binney & tremaine 1987 ) to represent the halo potential : @xmath9 where @xmath10 is the asymptotic ( in the limit @xmath11 and @xmath12 ) circular velocity , @xmath13 is the core radius and @xmath14 , @xmath15 are the potential axis ratios . we will consider the process of bulge formation to be terminated , or at least substantially slowed down , when the combined ( baryonic plus halo ) potential admits sufficiently round non - intersecting loop orbits , which permit the long - term circulation of gas without excessive dissipation . such motions necessarily take place in a symmetry plane determined by the ( orbit - averaged ) angular momentum ( e.g. , frank , king & raine 2002 ) . therefore , for our purposes , it will suffice to consider only orbits in the plane @xmath16 and to ignore the vertical dimension . we fix @xmath14 at @xmath17 ( though the effects of its variation are discussed where relevant ) and adopt a threshold axial ratio @xmath18 for orbits that can be populated with gas . the value of @xmath18 that best describes when this happens remains to be investigated , but its mere existence is what is important here . as we show below , our results are rather insensitive to the exact value of @xmath18 . the halo core mass will be taken to be @xmath19 and we define the halo density to be @xmath20 . this is conveniently close to the value of the density in the region where the potential is effectively harmonic that is , within the region where , in the absence of the bulge component , no loop orbits exist for @xmath21 . we note , however , that a bulge with a larger core radius probes a region of the halo core with smaller average density than the region probed by a bulge with a relatively small core radius . this effect will be discussed in section 4 . since our aforementioned criterion depends on the potential in a chosen symmetry plane , the exact three - dimensional mass distribution of the bulge is unimportant , as the same planar potential may arise from a variety of these . a particularly convenient form for the potential in this situation is that of miyamoto & nagai ( 1975 ) , @xmath22 this potential can approximate a disk - bulge system , with the parameters @xmath23 and @xmath24 determining the scale length and height , respectively . in the symmetry plane , therefore , the potential can arise from a range of density distributions from highly flattened to spherical ones . in general , when we refer to the `` bulge '' we will have a spherical system in mind with @xmath25 . in particular , all densities quoted are calculated under this assumption . we will comment on the effects of assuming a flattened distribution where relevant . in that case @xmath26 in eq . ( [ miyamoto ] ) is replaced by @xmath2 , the spherical bulge mass . the bulge _ core _ is defined by the radius @xmath27 and encloses the core mass @xmath28 . the bulge core density is defined as @xmath29 . the scaling relations that arise do not depend on the absolute masses and lengths , but instead , on the ratios of these quantities among the various contributions to the potential . an important property of the dynamics in our model is that _ any transformation that leaves the relative masses and lengths of the components constant will also not affect the orbital structure once scale transformations are taken into account . _ such transformations only change the time units , hence the dynamics remains invariant on a new characteristic timescale . in addition , _ in regions where the baryonic and halo components have the same mass distributions , i.e. , inside their cores , only the density ratios of these components determine their relative contributions to the potential , _ and hence the orbital structure . for the remainder of this paper , we will describe all components of the potential in terms of units scaled to the dimensions of the halo potential , eq . ( [ logarithm ] ) . we scale all distances to @xmath13 , with @xmath30 , and all masses to @xmath31 so that @xmath32 . the normalized bulge core density is then @xmath33 . the bulge potential in scaled units is @xmath34 the black hole potential , in the same units , is @xmath35 where @xmath36 . note that all orbits in the total potential @xmath37 depend only on three parameters , which may be taken to be @xmath38 , @xmath39 , and @xmath40 . in this section , we demonstrate that the arguments outlined above imply a linear relation between the sbh and bulge masses , provided that the baryonic and cdm density distributions are not completely axisymmetric and exhibit , at some stage , nearly constant density cores . fig . [ homofin ] ( the result of orbital integrations ) exhibits the axis ratios of the closed loop orbits @xmath41 as a function of the longer axis length ( @xmath42 ) . in this figure , we have fixed the scaled bulge core density , @xmath43 , and the sbh - to - bulge mass ratio , @xmath44 , but vary the scaled bulge core radius , @xmath38 . consequently , the bulge masses @xmath45 range over a factor of @xmath46 . as expected , when only the halo contributes to the potential ( dotted line ) , @xmath47 quickly drops to zero inside the halo core , since no loop orbits can exist there . ( for halo potential axis ratios smaller than the adopted value of 0.9 , i.e. , greater triaxiality , @xmath47 reaches 0 at radii that are progressively closer to the halo core radius , @xmath48 . as the baryonic component , both in the bulge and the sbh , becomes progressively more massive , correspondingly rounder loop orbits appear . loop orbits in the very central region are produced by the presence of the sbh . the latter is represented numerically by a plummer sphere with a softening scale of @xmath49 , in scaled units . near the center , where the sbh contribution is dominant , the axis ratios of the loop orbits tend to @xmath50 . to2.5 in ' '' '' the axis ratio curves exhibit two distinct minima . it is apparent that the inner minima of the curves in fig . [ homofin ] have ( nearly ) the same values of @xmath47 , but occur at different radii @xmath51 . this is a consequence of assuming a fixed ratio of sbh - to - bulge mass , @xmath40 , and a constant bulge core density @xmath52 . because the bulge and halo densities are nearly uniform in the inner regions , their relative contributions to the potential are determined by their density ratio . in the absence of the sbh contribution , @xmath47 in fig . [ homofin ] would tend to zero within the effective harmonic core of the bulge - halo system ( see eq . [ [ expo ] ] ) . when the relative density is kept constant ( and @xmath53 ) this is proportional to @xmath54 ( the effect of density variation will be examined in section [ inmin ] ) . the reason for this is that inside the nearly constant - density bulge core , the potential , which is now a superposition of two nearly harmonic potentials , is nearly harmonic ( even if the new core is less triaxial than the halo alone , e.g. , if the bulge is assumed to be spherical ) . the inner minima of the curves in fig . [ homofin ] , therefore , correspond to a transition from the region where the sbh provides the dominant contribution toward the creation of loop orbits to that where the bulge provides this contribution . therefore , the minima occur at radii where the gravitational acceleration due to the sbh is proportional to that due to the bulge+halo , @xmath55 ( since the minimum is located well within the bulge core radius ) . since @xmath40 is taken to be constant , we expect the minima to occur at radii @xmath51 . in section 2 , we have defined the critical value @xmath18 above which gas circulation can be sustained for secular timescales . here we have shown that given a critical value , @xmath18 , for the inner minimum and the relative bulge - to - halo core density ratio , a value of @xmath40 associated with this minimum is determined . however , satisfying the condition @xmath56 at the inner minimum does not imply that the same condition is satisfied at all radii within @xmath13 . until it is satisfied at the outer minimum , as well , further evolution can occur . this is briefly sketched below and outlined in more detail in section 3.4 . until the critical value of @xmath47 is reached at all radii inside the halo core , the bulge will continue to grow , since outside the bulge core @xmath47 declines again . it is clear from fig . 1 that , in contrast to that of the inner minimum , the value of @xmath47 at the outer minimum depends more sensitively on the bulge mass than on its density . the growth of the bulge , therefore , determines a _ minimal _ bulge mass fixed by the condition that @xmath57 at all radii within the halo core . if the sbh did not continue to grow along with the bulge , this growth in the bulge mass would decrease the linear correlation coefficient @xmath40 , but , as we show in section 3.4 , this leads only to near - independence of the final value of @xmath40 on @xmath18 and still keeps the values of @xmath40 within the observed range . as shown in the previous section , the creation of loop orbits with axis ratio above a given value depends , in the central region , solely on the ratio of the bulge - to - halo core density and the mass ratio @xmath58 . the actual value of the bulge mass determines only the position of the minimum ( as a fraction of the halo core radius ) , not the axis ratio at the minimum . but even the radius of the minimum is largely insensitive to the bulge mass , i.e. , @xmath59 , provided that the minimum actually exists and the bulge core density relative to the halo harmonic core density remains constant . we will now examine the bounds on @xmath0 and @xmath40 which ensure the existence of a minimum @xmath57 inside the bulge core and observe the effect of varying the bulge density . we calculate and plot @xmath60 for different choices of @xmath18 ( fig . [ compac ] ) , in the following way . first , without any loss of generality , we choose a bulge mass @xmath45 that is large enough that the outer minimum satisfies @xmath61 , for all @xmath62 . this is done in order to focus on the effects of the inner minima only . ( recall , from the previous section , that the value of the outer minimum depends on the mass of the bulge . ) fig . [ compac ] , therefore , exhibits the effect of @xmath63 and @xmath40 on the inner minima of the @xmath47 curves at @xmath64 . in order to obtain @xmath60 , we vary @xmath0 , which is some fraction @xmath40 of the chosen bulge mass . the density is varied by contracting the bulge ( by decreasing its core radius ) until the condition @xmath61 is satisfied everywhere inside the bulge core . to2.5 in ' '' '' one observes in fig . 2 that at smaller densities @xmath40 tends toward an asymptotic ( and maximal ) value associated with a given @xmath18 . we refer to this hereafter as the `` asymptotic '' regime . in the transition to this regime , along a @xmath65 curve , the value of @xmath40 is monotonically increasing . in other words , the bulge contribution toward the creation of loop orbits of a given eccentricity gradually diminishes , and is compensated by a greater contribution from the sbh component . in the process , the minimum in @xmath47 moves outward . the asymptotic regime corresponds to a situation in which the existence of loop orbits with the required elongation depends on the value of @xmath40 , irrespective of the bulge density ( that is , effectively , only on the sbh mass ) . in this limit the sbh contribution to the potential is sufficient to create loop orbits with @xmath61 at all radii within the halo core ( i.e. , at @xmath66 ) , without additional contributions from the bulge component ( cf . 3 ) . essentially , it corresponds to a rapid collapse to the center and formation of the sbh by a large fraction of the baryonic material , bypassing the formation of a bulge . this regime appears to be of academic interest only , as it implies @xmath67 all the way to the center ; it will not be discussed further . to2.5 in ' '' '' a second regime characterizes the dynamical state of the sbh - bulge - dark halo system with @xmath68 . in this `` scaling regime '' the @xmath69 curves are parallel straight lines with average slopes of about @xmath70 . this can be explained in the following manner . from eq . ( [ expo ] ) we know that the radius of the effective harmonic core of the bulge - halo system , @xmath42 , is proportional to @xmath71 . in the scaling regime , the location of the minimum in @xmath47 will be proportional to @xmath42 . now , in order to maintain the minimum at a specified value of @xmath18 , the gravitational acceleration due to the sbh must be proportional to the acceleration due to the halo ( which contains the nonaxisymmetry ) at @xmath42 . this implies @xmath72 . substituting for @xmath42 and noting that @xmath73 , we obtain @xmath74 . this is approximately what is found from figure 2 and it also holds if we change the mass of the bulge keeping the radius constant and again using eq . ( [ expo ] ) . note , furthermore , that for constant @xmath39 this relation predicts the radius of the effective harmonic core ( and the location of the minimum ) to be proportional to @xmath75 , as expected from the heuristic considerations of the previous section . while the asymptotic values of @xmath40 depend on @xmath2 , the values of @xmath40 in scaling regime do not . the transition between these two regimes can , therefore , be characterized by a sharp change in the behavior of @xmath40 , as seen from fig . 2 . we are mainly interested in the scaling region , because the onset of star formation can be tied to the baryonic component becoming self - gravitating , which corresponds roughly to the bulge core density exceeding that of the background halo , i.e. , @xmath68 . in this case , and for @xmath76 , the sbh contributes significantly to the potential only at radii @xmath77 . the @xmath47 curves , therefore , exhibit a definite inner minimum well inside the halo core . the values of @xmath78 and the initial @xmath40 , in principle , also depend on the critical eccentricity , @xmath18 , of the inner minimum , which is expected to be independent of the bulge mass , but which does depend on complex gas dynamics . if , after the sbh forms , the bulge mass falls short of the value required for creating sufficently round closed loop orbits at all radii inside the halo core , it will continue to grow until the outer minimum of the axis ratio curve also attains @xmath61 . in section 3.4 , we show that the sbh growth , if continued , becomes intermittent . in the next section , we also demonstrate that one can place constraints on the possible range of @xmath40 values by considering this constraint on the outer minimum . to obtain loop orbits rounder than a given @xmath18 _ at all radii _ @xmath79 , and not only in the central region , one in fact needs to take into account the bulge - to - halo core mass ratio and not only the density ratio . this can already be seen in fig . [ homofin ] , where all curves with sbh and bulge contributions exhibit an inner minimum with @xmath80 , while @xmath47 declines significantly as one moves further out . for low bulge masses , an additional outer minimum emerges , within @xmath81 , before @xmath47 rises again outside the halo harmonic core . only one of the @xmath47 curves , corresponding to the most massive bulge , has @xmath82 at _ all _ radii . therefore , for a given halo , there is a minimal bulge mass , @xmath83 , that is necessary to create sufficiently round loop orbits at all radii @xmath84 if we assume that the bulge core density varies at most by a factor of a few ( @xmath85 $ ] ) , @xmath86 , which is only weakly dependent on the bulge mass and density ( @xmath87 ) , varies little . the minimal bulge mass is also nearly constant ( calculations show , for example , that it varies by @xmath88% when @xmath63 changes from 1 to 2.5 ) . there is more sensitivity to the assumed @xmath18 . in fig . [ complots ] ( asterisks ) we show the bulge mass required to create loop orbits with @xmath61 at all radii outside @xmath86 , for @xmath89 . these are the masses necessary , at this density , to produce outer minima with the required values . for bulge densities @xmath90 these minima lie at radii @xmath91 . thus , unless the sbh can contribute significantly to the potential at radii comparable to @xmath13 , which seems implausible , the dominant contributions to the potential at the radii examined here should be only those due to the bulge and halo components . therefore , once the halo parameters are fixed , the creation of loop orbits with given @xmath61 outside the bulge core will depend only on the bulge parameters @xmath86 and @xmath92 . to2.5 in ' '' '' for @xmath89 , a well - defined inner minimum in the @xmath47 curve exists inside the bulge core only if @xmath93 . for larger @xmath18 , the inner minimum moves outward and merges with the outer one . systems with these properties lie outside the scaling region in fig . 2 . in this case , for the condition @xmath61 to be satisfied , the sbh has to contribute significantly at all radii no matter how massive the bulge is . ( for @xmath89 , the curve with @xmath94 in fig . 2 lies in this regime . ) furthermore , the minimal bulge density required to create loop orbits with @xmath61 , without appeal to an overmassive sbh , increases rapidly when @xmath95 ( fig . [ exmags ] ) , suggesting that the bulk of the material may form stars before the condition @xmath61 is reached everywhere within the halo core . if we demand that @xmath96 , then @xmath97 must be smaller than 0.8 in order for the model to be plausible , implying that @xmath40 must lie in the range @xmath98 . to2.5 in ' '' '' on the other hand , star formation can be somewhat delayed to higher densities . a reasonable range lies within @xmath99 . still higher densities in the bulge can be excluded based on the observed rotation curves . in this range of @xmath63 , values of @xmath100 become feasible without invoking unrealistically massive sbhs , as seen from fig . [ exmags ] . one may also reasonably assume that , due to enhanced dissipation , higher densities require larger @xmath18 to maintain long - lived gaseous motion ( detailed modeling , however , will be necessary to determine exactly how @xmath18 depends on the system parameters ) . in this situation , the range in @xmath40 that can be inferred from fig . 2 , is again about @xmath98 and compatible with the observed value of @xmath101 , which exhibits a significant scatter . the values of @xmath40 obtained so far , which should be regarded as _ initial _ values , are fixed by the inner minimum of the axis ratio curves . as argued in section 3.1 , further evolution of the bulge will occur unless sufficiently round loop orbits also exist at the outer minimum . in the next section we examine how this evolution can modify the range of @xmath40 , if at all . there are basic differences between the conditions for satisfying @xmath102 at the inner and outer minima in fig . [ homofin ] . as discussed in section 3.1 , the initial growth of the sbh is terminated after the inner loop orbits reach the critical eccentricity . the growth of the bulge , on the other hand , continues until the required @xmath18 is reached at all radii inside the halo core . since the increase of bulge mass , at roughly constant density , moves the inner minimum outward as described in section 3.1 , the sbh would have to continue growing in pace with the bulge mass if it is to maintain the critical value of @xmath40 ( which is determined only by the values of @xmath39 and @xmath18 ) . does this actually happen , and if so , how ? to2.5 in ' '' '' to see how the parallel growth of the sbh and bulge could come about , consider the solid curve in fig . [ homofin2 ] , which represents the state of the sbh / bulge system at the end of the initial stage of infall . this is equivalent to one of the curves ( say , the dash - dotted line ) in fig . 1 , except that we have chosen @xmath103 for the inner minimum , instead of 0.9 . this line has two maxima : a left - hand one at the origin , determined by the sbh , and a right - hand one , determined by the bulge core . note that the outer part of the solid curve drops below @xmath18 , implying that the bulge will continue to grow . now suppose this growth occurs _ without _ corresponding growth of the sbh ( in contrast to the case in fig . 1 , where @xmath40 is kept constant ) . if the bulge continues to grow while the sbh growth is stopped , the right - hand maximum moves further to the right , while the left - hand maximum stays the same . this opens a widening ` gap ' at the position of the inner minimum which drops below @xmath18 , allowing gas to flow again toward the radius of influence of the sbh , @xmath104 . this is illustrated by the dashed curve in fig . [ homofin2 ] , which corresponds to a factor of 10 increase in @xmath45 accompanied by a similar _ drop _ in @xmath40 . the gas is expected to accumulate in this vicinity . however , once a substantial amount of gas has collected at @xmath104 , it becomes self - gravitating and is prone to global self - gravitating instabilities . the fastest of these instabilities , @xmath105 modes or bar instability ( e.g. , bardeen 1975 ) , have been discussed in a similar context by shlosman , frank & begelman ( 1989 ) . they induce rapid ( dynamical ) gas inflow . hence we expect that the growth of the sbh at this stage will be intermittent , but because the time - averaged conditions for infall at the inner and outer maximum do not change , we expect the value of @xmath40 to stay roughly constant within the range given in fig . 2 , namely @xmath106 , depending on the value of @xmath18 . is there always enough gas to accumulate at @xmath107 in order to trigger a bar instability , due to the opening of the gap between the sbh and the bulge ? one can imagine the opposite extreme to that discussed above , in which insufficient gas enters the widening gap from outside , perhaps because star formation is efficient during the early stages of infall . let us suppose , for the sake of argument , that the sbh does _ not _ grow beyond its initial mass , as determined by the initial value of @xmath40 . how does this affect the range of final @xmath40 values ? figure 4 shows minimal bulge masses , @xmath108 , for a range of @xmath18 values . while we have no estimate for the initial bulge masses which define the initial @xmath40 in fig . 2 , it is possible to rule out very small masses , e.g. , @xmath109 , because the total baryonic mass within the halo core is taken to be about 10% . this means that for lower @xmath110 , the initial bulge mass is equal to or larger than @xmath83 estimated in section 3.3 , and , therefore , will not grow beyond its initial value . it is easy to understand this result , because one needs small baryonic masses to create loop orbits with such large eccentricities . this means that in this regime the initial value of @xmath111 ( in figures 2 and 4 ) is also its final value . alternatively , in the regime of larger @xmath112 , the bulge can grow at most by a factor of @xmath113 , reducing the initial @xmath40 by this amount . luckily , the initial values of @xmath40 for high @xmath18 lie around their high end , @xmath114 . reduction by up to two orders of magnitude brings them again to about @xmath115 . hence , whether the sbh grows after the initial stage or not does not destroy the @xmath7 correlation and does not move the values of @xmath40 outside the observed range . a corollary of this discussion is that the final value of @xmath40 appears insensitive to the value of @xmath18 , if the growth of the sbh is supressed , and depends strongly on @xmath18 , if the sbh grows in tandem with the bulge . note also that the variation of @xmath116 ( represented by the asterisks in fig . 4 ) , follows closely that of @xmath117 ( represented by filled circles ) . the ratio of these two quantities , @xmath118 is , therefore , largely independent of @xmath18 . in other words , the minimal bulge mass is predicted to be about 5 times the geometric mean between the sbh and the halo masses , provided that the sbh grows along with the bulge . to summarize , the initial value of the black hole - to - bulge mass ratio @xmath40 is fixed by the density at which the gaseous material in the inner region becomes self - gravitating and forms stars , and by the value of @xmath18 . however , unless loop orbits of sufficient eccentricity are created in the _ whole _ harmonic halo core region , the mass of the bulge will continue to grow . at the same time , the dynamical infall onto the sbh can choke , if the remaining gas mass is insufficient to cause bar instability and channel it to the sbh . in the latter case @xmath40 will decline below its initial value as the bulge grows , but will tend to approach a final value roughly independent of @xmath18 . in the opposite limit , the sbh will grow in proportion to the bulge , and @xmath40 will stay constant ( as shown by figs . 2 and 4 ) . in either limit , we assume that the process stops when the mass of the baryonic component reaches the minimal mass required to achieve loop orbits with a given @xmath61 , at all radii . for less triaxial haloes , and for a given bulge core density , lower sbh masses are required to produce loop orbits below a critical eccentricity . the minimal mass of the bulge needed to create the required loop orbits in the outer region , however , also decreases especially since , for a mildly triaxial halo , the radius at which no loop orbits exist inside the harmonic core decreases with decreasing triaxiality . thus , in the regime of mild halo triaxiality , the ratio of @xmath40 to ( normalized ) minimal bulge mass should not depend sensitively on the value of @xmath14 . we do ignore the fact that triaxiality can be a function of radius , and choose a fixed value of @xmath14 for simplicity . once orbits of sufficiently large @xmath47 are present , dissipation will reduce radial motion relative to these orbits , as well as vertical motion away from the symmetry plane defined by the angular momentum vector . infalling gas will start to populate the newly - formed , round , non - intersecting periodic loops , leading to the formation of a disk component inside the halo core . at this stage the bulge formation stops . outside the halo core , gas can accumulate at any stage of the formation process on closed loop orbits , which always exist in strongly inhomogeneous density distributions . provided that the halo triaxiality is small , these orbits will be nearly circular ( for potential axis ratio @xmath14 mildly deviating from unity @xmath119 : see , e.g. , rix 1995 ) . if the halo core radius is exceedingly large , however , it is possible that no significant disk component will form at all . for if the core is large relative to the total halo ( virial ) radius , the contracting gaseous component will end up in the core , instead of spinning up and forming an extended disk . losing most of its angular momentum to the halo , it will eventually end up as part of the bulge component . this would lead to the formation of an elliptical rather than a disk - dominated galaxy . in this context , one expects that haloes with larger core radii are host to larger spheroidal components . since it appears that , in general , more massive galaxies are usually of earlier type ( e.g. , persic , salucci & stel 1996 ) , one could deduce that more massive cores have larger core radii . this would be expected if these cores followed a faber - jackson type relation ( for example , if @xmath120 , then @xmath121 ) as tentatively suggested by observations ( burkert 1995 ; salucci & burkert 2000 ; dalcanton & hogan 2001 ) and deduced if halo cores formed via the destruction of the inner ( @xmath122 ) regions of nfw haloes ( navarro , frenk & white 1997 ) . in this case , lower mass haloes form when the universe is denser and are , therefore , more concentrated . as a consequence , the region where @xmath122 is smaller relative to the virial radius for low mass haloes ( el - zant , shlosman & hoffman 2001 ) . the existence of a faber - jackson type relation for halo cores is also required in order to reproduce the @xmath123 relation as discussed in the next section . within the above framework , the formation of the sbh is intimately tied to the bulge component , whereas the formation of the disk component takes place after the processes leading to bulge and sbh formation are essentially complete . the bulk of the disk is also expected to form at scales larger than the halo harmonic core ( but see section 5 ) . thus , the deduced correlation involving the bulge and sbh does not simply generalize to one involving the disk as well , in accordance with observations ( e.g. , kormendy & gebhardt 2001 ) . the fact that all the orbital properties discussed in this paper depend only on the relative magnitudes of the masses and spatial scales of the galactic components involved has important consequences . for the @xmath7 relationship it has the obvious implication that the derived correlation will hold for all halo masses . a more powerful prediction transpires in relation to the @xmath4 relation . for this to hold , it is necessary that the masses and velocity dispersions of halo cores are related in a similar manner . suppose that for a given halo there exist unique values for the sbh and bulge masses , and for the bulge scalelength . if a faber - jackson type relation between the halo core mass and velocity dispersion exists , a corresponding relation will exist between the bulge parameters . it also follows that a similar relation will exist between the sbh mass and the bulge ( and halo ) velocity dispersion . this can be deduced by simple scaling transformations , because the orbital properties we are interested in are all invariant with respect to spatial scale and mass transformations . this means that if we multiply the masses of the sbh , bulge and halo by some constant @xmath124 , thus effectively changing the mass units , the curves in fig . [ homofin ] will remain invariant . in the same manner , if we multiply all lengthscales by some factor @xmath125 , so that @xmath126 and @xmath127 , thus effectively changing the length unit , all curves in fig . [ homofin ] remain the same . this implies that if , for example , @xmath128 , then @xmath129 are equivalent systems in the sense decribed above ( they have the same axis ratio curves for their loop orbits with the same @xmath47 values as a function of rescaled radius ) . these systems will all follow the @xmath3 relation for the halo . in this particular case , @xmath130 implies a similar relationship between @xmath1 and @xmath2 , as well as @xmath0 . it is of course possible that , for a given halo , the masses of the sbh and bulge , as well as the bulge lengthscale , are not unique . in other words , the subset of haloes with a given @xmath31 and @xmath13 may contain bulges and sbhs with a distribution of properties . in section 3.2 we had assumed that the bulge collapse is largely terminated when its density is of the order of the halo core density . this in turn fixes @xmath40 , once @xmath18 is determined . in this section we further assume that the value of @xmath40 is not changed as a result of any subsequent bulge growth . that is , the sbh grows in tandem with the bulge ( cf . section 3.4 ) . this relaxes the assumption that the bulge masses are determined by the minimal mass required to create closed loops with @xmath61 at all radii , allowing for more massive bulges . one would like to infer to what extent variations in the bulge and sbh properies , within a given halo , affect the homology relations discussed above . we now show that when one accounts for variations of the bulge and the sbh parameters , the departure from the aforementioned relation is not dramatic . this comes about basically because the mass and core radius of the bulge are correlated , under the assumptions of our model . first , we note that the average velocity dispersion of the core of the baryonic component can be written as @xmath131 , and the average density @xmath132 . in this case @xmath133 . here @xmath134 depends on the functional form of the density distribution . its exact value is unimportant if all bulges are assumed to have the same functional form for their density distributions . henceforth we set @xmath135 . for a constant bulge - to - halo density ratio @xmath78 a relationship between bulge velocity dispersion and mass @xmath136 results . in a toy model where the halo density does not vary at all with radius a constant bulge density determines a unique @xmath40 , given @xmath18 , and an `` @xmath137 '' relation between bulge properties within a given halo arises , with index equal to 3 . considering more realistic models for the density distribution raises the index somewhat . when the core radius of the baryonic component is not very small compared to that of the halo , the density of the halo will not be strictly constant in the region of interest . as a result , the slope of the @xmath137 relationship will further increase . , where @xmath138 , the slope of @xmath123 is given by @xmath139 $ ] which stays between @xmath140 for @xmath141 , and then rises rapidly . the resulting slope of @xmath123 is some weighted average of @xmath142 , and greater than 3 . ] this occurs because larger bulges see " a smaller mean halo density , and therefore smaller bulge densities suffice to produce the same relative contribution to the potential . since @xmath143 , an inverse correlation between density and @xmath144 steepens the @xmath145 relation . ( a marginal effect is already seen in fig . 1 where the inner maxima produced by bulges with progressively larger @xmath146 also have progressively larger inner minima of @xmath47 . ) if the baryonic component giving rise to the gravitational potential is significantly flattened there will be an increase in the @xmath123 slope to 4 . in this case , the surface , rather than the volume density , will determine the gravitational field , resulting in a situation where @xmath147 . to2.5 in ' '' '' the fact that the index of the relation between @xmath2 and @xmath1 at constant bulge density is close to that of the faber - jackson relationship implies that , even if there is considerable variation in the bulge mass for a given halo , the departure from a faber - jackson relation defined by the halo parameters ( as described above ) is not too large . this is illustrated in fig . [ comsig ] , where we plot the relationship between the average velocity dispersion ( simply defined as @xmath148 ) of the baryonic component and its mass for systems having parameters @xmath40 and @xmath18 corresponding to those in fig . [ compac ] . the plots are obtained by keeping @xmath40 and @xmath18 constant and , for a given bulge mass , decreasing its characteristic radius until @xmath61 at all radii inside the halo core . to obtain physical parameters we set @xmath149 km s@xmath150 and @xmath151 kpc . the solid line in fig . [ comsig ] is actually a superposition of several lines , corresponding to different values of @xmath18 ( and @xmath40 ) , as used in fig . 2 . its average slope is around 3.2 and increases slowly toward higher dispersion velocities . these lines , however , have different end - points , determined by the minimal bulge masses associated with the different @xmath18 ( cf . section 3.3 ) which are denoted by circles in fig [ comsig ] . as the mass decreases . the result is that small hooks " ( not shown here ) appear in place of the circles in fig . [ comsig ] . in principle , this regime may be of interest , but is not considered in this paper because it occurs for a very small range of bulge masses . ] the maximum values of the bulge masses in these curves correspond to 2.5 times the halo core mass . it seems unlikely that bulges would be more massive than this . their characteristic densities or radii would have to be much larger than those of the halo , as would their contribution to the potential . in systems with disks such extreme conditions would violate the shapes of observed rotation curves . to2.5 in ' '' '' in reality , for a given halo , there is a range of possible bulge densities and associated values of the @xmath40 parameter . this , in principle , affects the normalization of the @xmath145 relation . nevertheless , the normalization turns out to be weakly dependent on @xmath78 ( @xmath152 ) . thus , variations in bulge densities within haloes of given mass and size do not significantly affect the relationship between bulge mass and velocity dispersion . in our model , the central ( bulge ) density required for the production of loop orbits with @xmath61 in the scaling regime is also only weakly dependent on @xmath40 ( as @xmath153 , cf . [ compac ] ) . the above leads to an important corollary , _ that large variation in @xmath40 will cause only small changes in _ @xmath145 , as illustrated in fig . [ varkmb ] . this relation , therefore , appears to be robust and is not heavily affected by changes in bulge and sbh parameters . note also that , for less dense bulges the slope of the lines in fig . [ varkmb ] tends to @xmath154 , especially in the limit of large values of the velocity dispersion ( which is also where most of the observations lie ) . this is due to the the effect described above massive bulges with smaller densities are more extended , and , therefore probe larger regions of the halo core . to2.5 in ' '' '' to obtain an @xmath4 relation , one has to multiply the values of the bulge masses in fig . [ comsig ] by the appropriate @xmath40 factors , arriving at an @xmath4 relation _ within a given halo . _ this introduces `` scatter '' by shifting the @xmath155 lines , as can be seen from fig . [ varkmbh ] . for any fixed value of @xmath1 , this figure reveals that the ( vertical ) scatter in @xmath0 is similar to that in @xmath40 . this leads us to an important point . in the present formulation , the @xmath145 relation is tighter than the @xmath4 one . this appears to contradict observations suggesting that @xmath4 is much tighter than the faber - jackson relation . however , this result has been obtained under the assumption that there is _ no scatter in the halo faber - jackson relation . _ a significant scatter in @xmath3 would result in a corresponding scatter in @xmath156 , which follows from the homology scaling discussed in the beginning of this section . moreover , if @xmath18 correlates with @xmath1 in such a way that gaseous sustems embedded in haloes with higher velocity dispersion require larger values of @xmath18 to remain stable , the scatter in the @xmath4 relation due to variations in @xmath40 can be reduced . for example , in fig . [ varkmbh ] , we have assumed a single halo for all the lines with different @xmath18 . if , however , lines with higher values of @xmath18 are associated with haloes with larger @xmath157 ( and thus @xmath1 ) , the constant @xmath18 lines would be shifted in such a way that the vertical distances between them at a given value of @xmath1 decrease . we illustrate this effect in fig [ shift ] , where a linear relationship is assumed between @xmath1 and @xmath18 . to2.5 in ' '' '' we suggest that the relative tightness of @xmath4 could result from a loose relation between @xmath1 and @xmath18 , coupled with scatter in @xmath3 comparable to that of the faber - jackson relation for bulges . such @xmath158 relation can result from a general trend that dissipation is increasing in more concentrated systems . testing this will require detailed modeling of the way in which the dissipation rate depends on @xmath18 , velocity and density , and is outside the scope of this paper . based on the illustrative example of fig . [ shift ] , we only claim here a plausibility of a loose relationship between @xmath18 and the system velocity dispersion . if the postulated correlation persists for haloes of different masses , a steepening of the @xmath4 relation relative to the halo ( and bulge ) mass velocity dispersion relation is expected , implying that the former should have a larger index . in the model presented here , gaseous baryonic material settles inside a mildly non - axisymmetric halo with a nearly constant density core . initially , no orbits with a definite sense of rotation exist . the first infalling baryonic material , therefore , efficiently loses its angular momentum to the core . this initial collapse terminates only when the gas becomes self - gravitating and forms stars . the increased central density concentration which is produced in this first phase , however , destroys the harmonic core , paving the way for the existence of non - intersecting closed loops with a definite sense of rotation . if the loop orbits are too eccentric , the gas will shock and depopulate them . these orbits , therefore , can not represent long - lived attractors of the dissipative motion . thus , if the baryonic component does not possess a central cusp initially , the absence of sufficiently round supporting closed loops will lead to the formation of a central mass concentration including an sbh . at larger radii the potential responsible for creating sufficiently round loop orbits is that of the extended baryonic component , in the form of a bulge . this leads to a linear relationship between the bulge and sbh masses . it is crucial that both the onset of self - gravity and the appearance of increasingly circular loop orbits are subject to the same condition , both depending on the density ratio of the collapsing gas to that of the background halo core . this limits the allowable range in densities . for bulge core densities of order those of the halo core , and for plausible values of the critical eccentricity , the value of @xmath40 lies in the range @xmath98 , which is compatible with present observations . for densities up to an order of magnitude larger , the range of values is the same , provided that the critical values of the closed loop axis ratios are larger . this can be expected if the dissipation rate along loop orbits is dependent on density a plausible assumption . the most important prediction of the model outlined above is that the bulge and sbh parameters are determined by the halo properties . in particular , relationships between sbh and bulge masses and the velocity dispersion of the bulge necessarily arise , and with the right exponents , if the haloes should also exhibit a faber - jackson type relation between their masses and velocity dispersions . moreover , within a given halo , variations in the bulge and sbh properties are not expected to destroy these relations because imposing the condition of critical eccentricity requires that the bulge and sbh masses are related to the bulge velocity dispersion _ via _ a power law , with index also close to that of the faber - jackson relationship . finally , if the faber - jackson relationship for the halo exhibits significant scatter and if , as again seems plausible , the critical eccentricities of the loops anti - correlate with the density of the system , less scatter should be present in the relationships between sbh masses and bulge velocity dispersions than the corresponding relationships between bulges masses and their own velocity dispersions the standard faber - jackson relationship . there is already tentative evidence that halo cores may indeed follow faber - jackson type relationships ( burkert 1995 ; dalcanton & hogan 2001 ) . in addition , a halo core produced by flattening out the inner region ( where @xmath8 ) of the nfw profile would produce such a relation ( el - zant , shlosman & hoffman 2001 ) . the model thus makes testable predictions concerning the relationships among the very inner regions of galaxies , their extended baryonic components , the dark matter haloes they are thought to be embedded in and the cosmology that predicts their existence . in this framework , larger and more massive halo cores produce , on average , larger and more massive bulges . disks form outside the core , or later , when the central concentration produced by the baryons destroys the core . if the core is very large , most of the baryonic material is consumed in the first phase and no significant disk forms . this effect is expected to be prominent in larger mass cores , since if these follow the faber - jackson relation , more massive haloes should have proportionally larger cores . other predictions include the requirement that the average density of the bulge in the central region should be close to that of the halo core . there is also a minimal bulge mass associated with a given core , although this varies significantly with the critical loop orbit eccentricity assumed . a number of additional consequences for galaxy formation and evolution will be discussed elsewhere . to obtain relationships between sbh masses and bulge properties , we have assumed that during the gaseous infall phase the baryonic component did not have a central density cusp . most observed bulges and ellipticals , however , do have such cusps . one then has to assume that once star formation starts , cold dissipational collapse is initiated . this would lead to a central density cusp , as it does in the case of cosmological haloes . memory of the initial state is retained via the total energy , which determines the final velocity dispersion . merritt & ferrarese ( 2001 ) have suggested that a `` self - regulating '' mechanism , related to the threshold mass necessary for the loss of triaxiality in the system , may be behind the close correlation between sbh and bulge properties . as these authors point out , however , the sbh masses required for strong chaotic behavior leading to rapid loss of triaxiality are in fact probably too large being of the order of a few percent of the mass of the system s baryonic component . this is actually of the order of the sbh mass needed to create round loop orbits at _ all _ radii inside the halo core . in our model , however , this is not assumed . instead , an additional baryonic component plays the role of creating these orbits in the outer region . the collusion between this `` bulge '' component and the sbh in destroying the harmonic core and creating a situation whereby stable gaseous motion can exist gives rise to the correlations described in this paper . the correlations obtained here are compatible with the observed ones , and with acceptable scatter , despite our lack of knowledge of the values of such parameters as @xmath18 and its variation with system properties . detailed modeling of the gas dynamics will be required to further constrain this model . it is also possible that our distinction between the dynamical role played by the sbh and that played by the bulge core is too restrictive . we have introduced this to be able to obtain quantitative results , within the model , solely on the basis of the orbital characterisitics . in general , the roles of the two components , i.e. , the sbh and the bulge , may not be too distinct formation of the sbh can take place simultaneously with a cuspy bulge . for this , the even more ambitious task of a self - consistent treatment , including gas and stellar dynamics and star formation , is required . we believe , however , that our results are generic and arise from fundamental dynamical phenomena which will manifest themselves in any formulation of galaxy formation in mildly triaxial haloes with harmonic cores . we thank our colleagues , too numerous to mention , for stimulating discussions . this work was supported in part by nasa grants nag 5 - 10823 , wku-522762 - 98 - 6 and hst go-08123.01 - 97a ( provided by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas5 - 26555 ) to i.s . , nsf grant ast-9876887 to m.c.b . , and nsf grant ast-9720771 to j.f . arnold , v.i . 1989 , mathematical methods of classical mechanics ( new york : springer ) bardeen , j.m . 1975 , proc . a.hayli ( dordrecht : reidel ) , 297 binney , j. , tremaine , s. 1987 , galactic structure , princeton univ . press bogoliubov , n.n . , mitropolsky , y.a . 1961 , asymptotic methods in the theory of non - linear oscillations ( new york : gordon & breach ) burkert , a. 1995 , , 447 , l25 cole , s. , lacey , c. 1996 , mnras , 281 , 716 dalcanton , j.j . , hogan , c.j . 2001 , , 561 , 35 de blok , w.j.g . , bosma , a.c . 2002 , a&a , 385 , 816 de zeeuw , t. , merritt , d. 1983 , , 267 , 571 ( dzm ) dubinsky , j. 1994 , , 431 , 617 el - zant , a. 1999 , phys rep , 311 , 279 el - zant , a. , shlosman , i. , hoffman , y. 2001 , , 560 , 636 faber , s.m . , jackson , r.e . 1976 , , 204 , 668 ferrarese , l. , merritt , d. 2000 , , 539 , l9 frank , j. , king , a. , raine , d. 2002 , accretion power in astrophysics , cambridge univ . press , 3rd edition gebhardt , k. , et al . 2000 , , 539 , l13 heller , c.h . , shlosman , i. 1994 , , 424 , 84 kormendy , j. , gebhardt , k. 2001 , 20th texas symp . on relativistic astrophysics , j.c . wheeler & h. martel , eds . ( aip conf . 586 ) , 363 kuijken , k. , tremaine , s. 1994 , , 471 , 178 locas , e.l . , hoffman , y. 2000 , , 542 , l139 merritt , d. , ferrarese , l. 2001 , the central kpc of starbursts & agn : the la palma connection , j.h . knapen , j.e . beckman , j.h . , i. shlosman & t.j . mahoney , eds . ( asp : san - francisco , vol 249 ) , 335 miyamoto , m. , nagai , r. 1975 , , 27 , 533 navarro , j.f . , frenk , c.s . , white , s.d.m . 1997 , , 490 , 493 persic , m. , salucci , p. , stel , f. , 1996 , , 281 , 27 pfennifer , d. , norman , c.a . 1990 , , 363 , 391 rix , h .- w . , 1995 , iau symp . 169 , unsolved problems of the milky way , astro - ph/9501068 rix , h .- w . , zaritsky , d. 1995 , , 447 , 82 salucci , p. , burkert , a. , 2000 , , 337 , l9 shlosman , i. 2002 , agns : from central engine to the host galaxy , s. collin et al . ( asp conf . series , vol . 290 ) , in press shlosman , i. , frank , j. , begelman , m.c . 1989 , nature , 338 , 45 tremaine , s. , et al . 2002 , astro - ph/0203468 let the potential @xmath159 be a function of the cartesian coordinates in the halo symmetry plane , @xmath160 with @xmath161 . we follow de zeeuw & merritt ( 1983 , hereafter dzm ) and expand the potential in even powers of the coordinates : @xmath162 if no bulge exists , inside the halo s harmonic core only the terms quadratic in the coordinates are important . if , due to an additional component , or at the boundaries of the core , weak nonlinearity is present the series can be truncated , as above , to second order , as higher order terms are unimportant . the coefficients are given by : @xmath163 , @xmath164 , @xmath165 , @xmath166 , @xmath167 , where the derivatives are taken at @xmath168 . in terms of these one defines the auxilliary variables @xmath169 , @xmath170 , @xmath171 . the condition for stability of the loop orbits is ( see dzm ; table 2a third row ) @xmath172 we will be interested in systems that are both mildly nonlinear and mildly nonaxisymmetric : thus @xmath173 and @xmath174 . the above condition reduces to @xmath175 which , under the above conditions , is always satisfied . it is , therefore , the condition for the existence of loop orbits and not their stability that will be of interest to us . in the absence of a central mass ( sbh ) or density cusp these can not be found arbitrarily close to the center , instead there is a _ bifurcation radius _ beyond which these exist . this determines the effective core radius of the system . to second order , the condition for the existence of loop orbits is @xmath176 with @xmath177 and ( to first order ) @xmath178 , where @xmath179 is the hamiltonian . for @xmath180 ( first order potential terms dominate in eq . [ pot ] ) and @xmath181 , the above requires @xmath182 in the unperturbed case , the action variables ( e.g. , binney & tremaine 1987 ) are time - independent and ( exact ) solutions can be written in terms of these in cartesian coordinates as @xmath183 , etc . in the mildly nonlinear case the solutions of the equations of motion , averaged over a dynamical time ( denoted below by a bar " ) , approximate the true solution to first order in the ( relative amplitude of the ) perturbation and for a number of dynamical times inversely proportional to this ( dzm ; see also bogoliubov & mitropolsky 1961 ; arnold 1989 ) . in this case analogous approximate solutions can be given in terms of the corresponding action variables . in particular , for loop orbits in mildly nonlinear potentials one finds @xmath184 and @xmath185 eliminating @xmath186 one gets @xmath187 where @xmath188 , @xmath189 and @xmath190 . at bifurcation , to first order , @xmath191 where @xmath179 is given by eq . ( [ hamiltonian ] ) . these orbits are infinitely thin and represent oscillations along the @xmath192-axis , with amplitude @xmath193 which is the effective core of the bulge - halo system , as mentioned above . as one increases @xmath179 these become thicker with axis ratio @xmath194 the view we have taken in this paper is that when this ratio becomes large enough , @xmath195 , such loop orbits can support gaseous motion , and that stars formed on these orbits can constitute populations of stellar disks , thus ending the bulge formation stage . we note here that , beyond the bifurcation point , the above relation predicts a rather rapid ( as a function of radius ) transition to round loop orbits . this is confirmed by orbital integration , even though if the bulge mass is smaller than a certain minimal mass , in the fully nonlinear treatement , the orbital axis ratio can decrease again . the second minimum in axis ratio curves ( e.g. , fig . 1 ) is thus not reproduced by this perturbation analysis . we now apply the perturbation analysis to the superposition of potentials given by eqs . ( [ logarithm ] ) and ( [ bulgepot ] ) with the goal of calculating the bifurcation radii of the loops . this is the effective harmonic core radius of the bulge - halo system . since we are only interested in orbit shapes , we set @xmath196 and use scaled variables , as defined in section 2 . for the halo , we use @xmath197 where @xmath198 , corresponding to @xmath199 in eq . ( [ logarithm ] ) , parametrizes the nonaxisymmetry . for the bulge , we have @xmath200 with the above definitions for the potential we have @xmath201 @xmath202 @xmath203 and @xmath204 from equations ( [ hamiltonian ] ) , ( [ action ] ) , and ( [ ymax ] ) bifurcation happens when the value of the long axis of the loops satisfies @xmath205 substituting from the expressions above and taking the limits @xmath206 , @xmath207 , and assuming @xmath208 , we obtain @xmath209 this is the effective harmonic core radius of the bulge - halo system .

even though mutually unbiased bases and entropic uncertainty relations play an important role in quantum cryptographic protocols they remain ill understood . here , we construct special sets of up to mutually unbiased bases ( mubs ) in dimension which have particularly beautiful symmetry properties derived from the clifford algebra . more precisely , we show that there exists a unitary transformation that cyclically permutes such bases . this unitary can be understood as a generalization of the fourier transform , which exchanges two mubs , to multiple complementary aspects . we proceed to prove a lower bound for min - entropic entropic uncertainty relations for any set of mubs , and show that symmetry plays a central role in obtaining tight bounds . for example , we obtain for the first time a tight bound for four mubs in dimension , which is attained by an eigenstate of our complementarity transform . finally , we discuss the relation to other symmetries obtained by transformations in discrete phase space , and note that the extrema of discrete wigner functions are directly related to min - entropic uncertainty relations for mubs .

1004.5086

one of the central ideas of quantum mechanics is the uncertainty principle which was first proposed by heisenberg @xcite for two conjugate observables . indeed , it forms one of the most significant examples showing that quantum mechanics does differ fundamentally from the classical world . uncertainty relations today are probably best known in the form given by robertson @xcite , who extended heisenberg s result to two arbitrary observables @xmath3 and @xmath4 . robertson s relation states that if we prepare many copies of the state @xmath5 , and measure each copy individually using either @xmath3 or @xmath4 , we have @xmath6{|\psi\rangle}|\end{aligned}\ ] ] where @xmath7 for @xmath8 is the standard deviation resulting from measuring @xmath5 with observable @xmath9 . the essence of is that quantum mechanics does not allow us to simultaneously specify definite outcomes for two non - commuting observables when measuring the same state . the largest possible lower bound in robertson s inequality ( [ eq : heisenberg ] ) is @xmath10 , which happens if and only if @xmath3 and @xmath4 are related by a fourier transform , that is , they are conjugate observables . a natural measure that captures the relations among the probability distributions over the outcomes for each observable is the entropy of such distributions . this prompted hirschmann to propose the first entropic uncertainty relation for position and momentum observables @xcite . this relation was later improved by @xcite , where @xcite show that heisenberg s uncertainty relation is in fact implied by this entropic uncertainty relation . hence , using entropic quantities provides us with a much more general way of quantifying uncertainty . indeed , it was realized by deutsch @xcite that other means of quantifying `` uncertainty '' are also desirable for another reason : note that the lower bound in is trivial when @xmath5 happens to give zero expectation on @xmath19 $ ] . hence , it would be useful to have a way of measuring `` incompatibility '' which depends only on the measurements @xmath3 and @xmath4 and not on the state . deutsch @xcite himself showed that @xmath20 where @xmath21 , and @xmath22 is the min - entropy arising from measuring the pure state @xmath5 using the basis @xmath14 ( see section [ sec : urquantities ] for more information on the entropic quantities we use ) . if @xmath14 and @xmath15 are related by a fourier transform , then the r.h.s . of becomes @xmath23 , where the minimum is achieved by a state that is invariant under the fourier transform . since the shannon entropy obeys @xmath24 , deutsch s bound also holds for the shannon entropy . better lower bounds have since been obtained for the shannon entropy by maassen and uffink @xcite following a conjecture of kraus @xcite . their uncertainty relations are again strongest ( in the sense that the lower bound is largest ) when the bases @xmath14 and @xmath15 are conjugate , that is , the two bases are related by a fourier transform . apart from their role in understanding the foundations of quantum mechanics , these uncertainty relations play a central role in cryptography in the noisy - storage model @xcite , quantum key distribution @xcite , information locking @xcite , and the question of separability @xcite . in particular , such relations have practical interest in noisy - storage cryptography @xcite where they may enable us to prove security for a larger class quantum memories . here , we are concerned with measurements in multiple bases @xmath25 . entropic uncertainty relations provide a natural way of quantifying `` incompatibility '' of more than two measurements by lower bounding @xmath26 for all states @xmath5 . we call a state @xmath5 that minimizes the average sum of entropies a _ maximally certain state_. when @xmath27 is the shannon entropy , the largest bound we can hope to obtain for any choice of bases is @xmath28 since choosing @xmath5 to be an element of one of the bases yields zero entropy when we subsequently measure in the same basis . if is indeed a lower bound to , we will call the measurements _ maximally incompatible with respect to the shannon entropy_. note that this can only happen if @xmath5 gives us full entropy can never exceed @xmath29 , where the maximum is attained when the distribution over the outcomes is uniform @xmath30 . ] when measured in any other basis , that is , the bases are all mutually unbiased . curiously , however , it was shown that whereas being mutually unbiased is necessary , it is not a sufficient condition to obtain maximally strong uncertainty relations for the shannon entropy @xcite . in particular , there do exist large sets of up to @xmath31 mutually unbiased bases in square dimensions for which we do obtain very weak uncertainty relations @xcite . recently , ambainis @xcite has shown that for any three bases from the `` standard '' mutually unbiased bases construction @xcite in prime dimension , the lower bound can not exceed @xmath32 , for large dimensions . for dimensions of the form @xmath33 and @xmath34 no further assumption is needed , but the proof assumes the generalized riemann hypothesis for dimensions of the form @xmath35 . furthermore , for any @xmath36 , there always exist @xmath37 of these bases such that the lower bound can not be larger than @xmath38 . only if we use the maximal set of @xmath39 mutually unbiased bases that can be found for any given prime power dimension , do we obtain quite strong uncertainty relations @xcite . at present , we merely know that there do exist arbitrarily large sets of two outcome measurements that give us maximally strong uncertainty relations @xcite , and that in larger dimensions selecting a large amount of bases at random does provide us with strong relations @xcite ( for a survey see @xcite ) . indeed , it remains an intriguing open question as to whether there even exist three measurements with three outcomes in dimension @xmath40 that are maximally incompatible with respect to the shannon entropy . in the light of these questions , it is therefore natural to study the structure of mutually unbiased bases(mubs ) to see whether we can identify additional properties which are sufficient for obtaining strong uncertainty relations . in @xcite , wootters and sussman made the interesting observation that for the maximal set of @xmath39 mutually unbiased bases coming from such constructions as @xcite in dimension @xmath1 , the lower bound of the entropic uncertainty relation in terms of the collision entropy given in @xcite is tight , and the minimum is attained by a state that is invariant under a unitary that cyclically permutes the set of all @xmath39 mubs . a similar unitary was noted to exist by chau @xcite . wootters and sussman derive their transformation from phase space arguments . their unitary can in fact easily be generalized to cyclically permute @xmath41 bases , whenever @xmath41 divides @xmath39 ( see section [ sec : urdwggeneral ] ) . the results in @xcite have recently been generalized by appleby @xcite , who shows that in prime power dimensions of the form @xmath42 , there exists a unitary operation that cyclically permutes the first and second halves of the full set of mubs . this raises the pressing question of whether smaller sets of mubs also exhibit such symmetries ? and can we exploit such symmetries to obtain tight uncertainty relations ? in particular , is the minimizing state always an invariant of such a transformation as observed for two bases in ? * main result * we first show by an explicit construction that there exist sets of @xmath43 mutually unbiased bases in dimension @xmath1 with the property that there exists a unitary that cyclically permutes all bases in this set , whenever ( a ) @xmath41 is prime , and ( b ) @xmath41 divides @xmath44 or @xmath45 . more specifically , we provide an explicit construction of mubs @xmath25 with @xmath46 and a unitary @xmath47 such that @xmath48 furthermore , in dimension @xmath2 , we actually find such a unitary for any set of @xmath41 mubs , where @xmath49 . our approach exploits properties of the clifford algebra , which might yield new insights into the structure of these mubs . it is entirely distinct from the phase space approach which was used to construct such a unitary for the full set of @xmath39 mubs @xcite . note that our construction gives at most @xmath50 bases , but shows that there is indeed an additional symmetry which has previously gone unnoticed . for @xmath51 bases , @xmath47 is simply the fourier transform , and it would be interesting to investigate general properties of our transformation and whether it has applications in other areas . we then apply our transformation to the study of uncertainty relations in terms of the _ min - entropy _ ( see ) . since @xmath24 , this also provides us with bounds on uncertainty relations in terms of the shannon entropy . of course , many forms of entropy could be considered when it comes to quantifying uncertainty , and each has its merits . the min - entropy is of particular interest in cryptography , and is also related to the well studied extrema of the discrete wigner function as we will discuss in section [ sec : urdiscretewigner ] . in particular , it will be easy to see that the average min - entropy for the full set of @xmath52 mubs can be bounded as @xmath53\ , \end{aligned}\ ] ] where @xmath54 is the maximum value of the discrete wigner function at the point @xmath55 in discrete phase space . symmetries thereby play an important role in determining @xmath56 . * second result * we prove a simple min - entropic uncertainty relation for an arbitrary set of @xmath41 mutually unbiased bases . for mubs @xmath25 we obtain @xmath57 . \end{aligned}\ ] ] for the case of @xmath58 mubs in dimension @xmath13 , this bound is indeed the same as the bound in . for any small set of @xmath59 mubs , our bound matches the strongest known bound @xcite . we also prove the following alternate lower bound : @xmath60,\end{aligned}\ ] ] which is stronger than for the complete set of @xmath61 mubs in dimension @xmath13 . clearly , when @xmath62 , the two bounds are equivalent . we further show that is in fact tight for @xmath63 mubs in dimension @xmath2 stemming from our construction , where the minimum is attained for an invariant state of the transformation @xmath47 that cyclically permutes all @xmath64 bases . even though this is a somewhat restricted statement , it is the first time that a tight entropic uncertainty relation has been obtained for this case . the minimizing state here has an appealing symmetry property , just as for the case of @xmath51 bases in where the minimum is attained by a state that is invariant under the fourier transform . for the collision entropy @xmath65 , wootters @xcite previously showed that the lower bound from @xcite is attained by an invariant state when considering the full set of @xmath39 mubs . here , however , we exhibit a tight uncertainty relation for these @xmath66 bases in @xmath2 for the collision entropy @xmath65 which has an entirely different structure and the minimum is not attained by an invariant state of our transformation . nevertheless , we have for the first time a _ tight _ entropic uncertainty relation for _ all _ possible mubs in a dimension larger than the trivial case of @xmath67 where the bloch sphere representation makes the problem easily accessible . in @xmath2 , we have a tight relation for @xmath68 for @xmath69 , and tight relations for @xmath65 for @xmath70 . our result indicates that due to the different properties of the minimizing state for different numbers of bases , the problem may be even more daunting than previously imagined . yet , our work shows that in each case the minimizing state is by no means arbitrary . it has a well defined ( albeit different ) structure in each of the cases . * third result * for some sets of mubs we do obtain for the first time , significant insight into the structure of the maximally certain states . in particular , we note in section [ sec : symmetry ] that for @xmath41 mutually unbiased bases that the state that minimizes the min - entropic uncertainty relations is an invariant of a certain unitary whenever @xmath41 divides @xmath39 for @xmath1 . before explaining our construction of mutually unbiased bases for which there exists a unitary that cyclically permutes them , let us define the notions of mubs more formally and recall some known facts . let @xmath71 and @xmath72 be two orthonormal bases in @xmath73 . they are said to be _ mutually unbiased _ if @xmath74 , for all @xmath75 . a set @xmath76 of orthonormal bases in @xmath73 is called a _ set of mutually unbiased bases _ if each pair of bases is mutually unbiased . for example , the well - known computational and hadamard basis are mutually unbiased . we use @xmath77 to denote the maximal number of mubs in dimension @xmath13 . in any dimension @xmath13 , we have that @xmath78 @xcite . if @xmath79 is a prime power , we have @xmath80 and explicit constructions are known @xcite . other constructions are known that give less than @xmath39 mubs in other dimensions @xcite . however , it is still an open problem whether there exists a set of @xmath81 ( or even @xmath64 ! ) mubs in dimension @xmath82 . our construction of mutually unbiased bases makes essential use of the techniques developed in @xcite , together with properties of the clifford algebra . the clifford algebra is the associative algebra generated by operators @xmath83 satisfying @xmath84 and @xmath85 . it has a unique representation by hermitian matrices on @xmath44 qubits ( up to unitary equivalence ) that can be obtained via the famous jordan - wigner transformation @xcite : @xmath86 for @xmath87 , where we use @xmath9 , @xmath88 and @xmath89 to denote the pauli matrices . furthermore , we let @xmath90 note that in dimension @xmath67 these are just the familiar pauli matrices , @xmath91 , @xmath92 and @xmath93 . of particular importance to us will be the fact that we can view the operators @xmath83 , as @xmath94 orthogonal vectors forming a basis for @xmath95 . in particular , for any orthonormal transformation @xmath96 which when applied to the vector @xmath97 gives @xmath98 , there exists a unitary @xmath99 such that @xmath100 the orthonormal transformation that is particularly interesting to us here is the one that cyclically permutes the basis vectors . as described above we can find a corresponding unitary @xmath101 which cyclically permutes the basis vectors @xmath102 . an explicit construction can be found in the appendix . this symmetry can be extended to @xmath103 , see e.g. @xcite . it will also be useful that the set of @xmath104 operators @xmath105 forms an orthogonal basis for @xmath106 hermitian matrices in @xmath1 @xcite . to construct mutually unbiased bases , we follow the procedure outlined in @xcite , but now applied to a subset of the operators in @xmath107 . that is , we will group operators into classes of commuting operators , i.e. , sets @xmath108 of size @xmath109 such that 1 . the elements of @xmath110 commute for all @xmath111 , 2 . @xmath112 for all @xmath113 . it has been shown in @xcite that the common eigenbases of such classes form a set of @xmath41 mubs . first of all , note that no class can contain two generators @xmath114 and @xmath115 since they do not commute . when forming the classes we hence ensure that each one contains exactly one generator @xmath114 , which clearly limits us to constructing at most @xmath0 such classes . the difficulty in obtaining a partitioning that is suitable for our purpose is to ensure that the unitary @xmath47 that cyclically permutes the generators @xmath116 also permutes the corresponding bases by permuting products of operators appropriately . we show in the appendix that our general construction achieves the following : suppose that @xmath43 is prime , and either @xmath41 divides @xmath44 or @xmath45 . then in dimension @xmath117 , there exist @xmath41 mutually unbiased bases @xmath25 for which there exists a unitary @xmath47 that cyclically permutes them @xmath118 let us consider two simple examples of such classes in dimension @xmath2 . these are not obtained from our general construction , but nevertheless provide us with the necessary intuition . for @xmath66 mubs the classes are given by @xmath119 it is easy to see that the unitary @xmath47 that achieves the transformation @xmath120 , but leaves @xmath121 and @xmath122 invariant , cyclically permutes the bases given above . for the collision entropy @xmath65 the minimum is attained for an eigenstate of the _ commuting _ operators @xmath123 , @xmath124 and @xmath125 . this minimizing state also shows that the wellknown bound for @xmath65 ( see e.g. @xcite ) can be _ tight_. for @xmath63 mubs we obtain the classes @xmath126 it is easy to see that the unitary @xmath47 that achieves the transformation @xmath127 , but leaves @xmath128 invariant , cyclically permutes the bases given above . for @xmath63 classes the minimum in the entropic uncertainty relation for @xmath68 is attained for a state that is invariant under the transformation @xmath47 . however , we also know that for @xmath63 or @xmath129 classes in dimension @xmath130 no partitioning of operators is possible that satisfies our requirements . thus , for what values of @xmath41 and @xmath13 such a unitary can be found , remains an interesting open question . we now investigate the relationship between the observed symmetries and entropic uncertainty relations . before comparing different uncertainty relations , we first provide a short introduction to all the entropic quantities we will use . the expert reader may safely skip this section . in general , the _ rnyi entropy _ @xcite of order @xmath55 of the distribution obtained by measuring a state @xmath5 in the basis @xmath12 is given by @xmath131.\end{aligned}\ ] ] indeed , the shannon entropy forms a special case of the rnyi entropy by taking the limit @xmath132 , i.e. , @xmath133 , where we omit the subscript . of particular importance are the _ min - entropy _ , for @xmath134:[def : minentropy ] @xmath135 and the _ collision entropy_[def : collision ] @xmath136 we have @xmath137 and hence uncertainty relations for @xmath138 also provide us with a bound on uncertainty relations for @xmath139 whenever @xmath140 . note that intuitively , the min - entropy is determined by the highest peak in the distribution and most closely captures the notion of `` guessing '' . to see why it is a more useful quantity in cryptography than the shannon entropy , consider the following example distribution @xmath141 : let @xmath142 and let @xmath143 be the all 0 string . suppose that @xmath144 and @xmath145 for @xmath146 , i.e. , with probability @xmath147 we choose @xmath148 and with probability @xmath147 we choose one string uniformly at random . then @xmath149 , whereas @xmath150 ! if @xmath151 would correspond to an encryption key used to encrypt an @xmath44 bit message , we would certainly not talk about security if we can guess the key with probability at least @xmath147 ! yet , the shannon entropy is quite high . apart from its cryptographic applications , min - entropic uncertainty relations are appealing since the problem of determining tight uncertainty relations can be simplified considerably in the presence of symmetries . furthermore , these relations bear an interesting relation to the extrema of the discrete wigner function . first of all , note that for the min - entropy we have by jensen s inequality that @xmath152 where the inequality becomes equality if all terms @xmath153 are the same . for @xmath154 , define @xmath155 note that determining a tight lower bound to is thus equivalent to determining @xmath156 clearly , any @xmath157 such that @xmath158 thus gives us a lower bound for . for any set of bases , this makes the problem of finding a bound more approachable as it reduces the problem to finding the largest eigenvalue for any operator @xmath159 . in particular , it can be phrased as a semidefinite program to minimize @xmath157 such that holds for all @xmath160 . it is now easy to see why symmetries simplify our goal of determining tight uncertainty relations . [ lem : symmetry ] suppose that for every @xmath161 there exists a unitary @xmath162 such that @xmath163 . then there exists a @xmath164 such that the minimum in is attained for a state @xmath165 that is invariant under @xmath166 . first of all , note that @xmath167 and hence for @xmath168 @xmath169 in particular , this holds for the state @xmath170 corresponding to the eigenvector @xmath5 with the largest eigenvalue of @xmath171 . when looking for the minimizing state on the r.h.s of we can thus restrict ourselves to states which are invariant under @xmath172 . note that in this case , we furthermore have that @xmath173 meaning that the inequality is tight in case of such a symmetry which is our claim . the question of course remains , whether such unitaries do exist in general . wootters and sussman @xcite have shown that there exists a unitary @xmath47 that cyclically permutes the set of all @xmath39 mubs for @xmath1 by constructing a unitary that corresponds to a rotation around the origin in phase space . clearly , by considering the unitary @xmath174 one can trivially adapt their construction to obtain a unitary that cyclically permutes @xmath41 mubs whenever @xmath175 . by first translating any point in the phase space to the origin , then applying the transformation @xmath174 and finally translating the origin back to the original point , one can obtain the desired unitaries @xmath162 that enable us to find tight bounds for the min - entropic uncertainty relations . this is the first time we gain significant insight into the structure of the states that minimize . note that our construction only gives unitaries @xmath162 for @xmath176 for any @xmath177 . this means that our complementarity transform @xmath47 , leads to tight bounds only if the largest eigenvalue of any @xmath159 happens to occur for a @xmath160 of this form . this is for example the case for @xmath63 in @xmath2 , where we do not obtain a unitary from the phase space approach of @xcite . [ sec : urdwggeneral ] to see how finding a lower bound for min - entropic uncertainty relations for @xmath39 mubs relates to finding the extrema of the discrete wigner function , let us first recall the properties of the discrete wigner function . the discrete phase space is a two - dimensional vector space over a finite field @xmath178 , where here we focus on the case of @xmath1 . for every state @xmath165 , we can associate a function @xmath179 with every point @xmath55 in the discrete phase space , known as the discrete wigner function . for completeness , we provide a short summary on how to determine @xmath179 ; a detailed account can be found in @xcite . first of all , note that the @xmath180 points of the discrete phase space can be parititioned into @xmath13 parallel lines each of which contains @xmath13 points . any such partition is called a _ striation _ , and it is known that @xmath39 such striations can be found @xcite . one may now define the discrete wigner function by relating each striation to one of the @xmath39 possible mutually unbiased bases @xcite : let @xmath181 denote the @xmath182-th line in the striation @xmath183 . with each such line , we associate a projector @xmath184 onto the @xmath182-th element of the basis @xmath185 , in a specific order so as to satisfy certain symmetry constraints @xcite . defining the phase - space point operator @xmath186 one can now define the discrete wigner function as @xmath187 the _ extrema of the discrete wigner function _ are defined as the minimum and maximum of over quantum states @xmath165 . note that when considering @xmath188 mutually unbiased bases , each point @xmath55 in the discrete phase space can be contained in exactly one line from each basis , as all lines in a striation , i.e. , one basis are parallel . hence , there is a one - to - one correspondence between points @xmath55 in discrete phase space and vectors @xmath189 . in terms of the phase space operator this means that @xmath190 note that the maximum of the discrete wigner function @xmath191 is simply the largest eigenvalue of @xmath192 ( or @xmath193 ) up to a factor of @xmath30 . we thus have that @xmath194\ , \end{aligned}\ ] ] satisfies @xmath195 and the maximum of the discrete wigner function provides a lower bound to the min - entropic uncertainty relations as given in . the extrema @xmath54 were evaluated numerically in @xcite for small @xmath13 . note , however , that as noted in section [ sec : symmetry ] , one may use symmetries to solve the problem of determining @xmath54 directly . as mentioned in section [ sec : urmin - entropy ] , the problem of finding a lower bound for the average min - entropy reduces to the problem of finding the maximum eigenvalue of the operator @xmath159 defined in . in the appendix , we use a result due to schaffner @xcite obtained using the techniques of kittaneh @xcite , to show that for any set of @xmath41 mutually unbiased bases in dimension @xmath13 , the maximum eigenvalue of @xmath159 is bounded by @xmath196 using this , we obtain the following simple bound for the average min - entropy in the appendix [ lem : smalllbound ] let @xmath25 be a set of mutually unbiased bases in dimension @xmath13.then , @xmath197\ .\ ] ] for the case of @xmath51 mubs in dimension @xmath13 , our bound exactly matches the well known result of deutsch ( see ) . for @xmath198 , the only other known lower bound for the average min - entropy is the one obtained in @xcite , where it is shown that for a set of @xmath199 mubs in dimension @xmath200 , the following holds : @xmath201\end{aligned}\ ] ] where @xmath202 are subsets of @xmath44-bit strings . in the case of min - entropic uncertainty relations , these subsets contain only a single string , which corresponds to the peak of the probability distribution induced on the state @xmath5 by the corresponding bases @xmath203 and @xmath204 , so that , @xmath205 thus , in dimension @xmath200 , our bound in is clearly the same as . the reason we obtain a more general bound for @xmath206 mubs in any dimension @xmath13 is that we reduce the problem directly to an eigenvalue problem without going through the representation in terms of bit strings as in @xcite , via a much simpler argument using only the concavity of the @xmath207 . using an alternate approach involving a bloch sphere like representation of the basis vectors @xmath208 , we show that the maximum eigenvalue of @xmath159 can be bound differently , as follows : @xmath209 as we show in the appendix , this implies [ lem : largelbound ] let @xmath25 be a set of mutually unbiased bases in dimension @xmath13.then , @xmath210 \ .\ ] ] notice that this alternate bound on the min - entropy is stronger than when @xmath211 . in particular , for the complete set of @xmath39 mubs in dimension @xmath13 , this alternate bound in is stronger than any previously known bounds . when @xmath62 the two bounds that we derive are indeed equivalent . see figs . [ fig : dim4 ] and [ fig : dim8 ] for a comparison of our bounds in dimensions @xmath2 and @xmath130 respectively . average min - entropy for different sets of mubs in dimension @xmath2 . + the crosses denote numerically computed minima of the average min - entropy for mubs obtained using our construction . the bound in is clearly tight for both @xmath66 and @xmath63 mubs . the second analytical bound in is stronger than for @xmath212 bases . the circle denotes the average min - entropy for the invariant states given in . for @xmath64 mubs in @xmath2 the minimum of the average min - entropy is indeed attained by states invariant under @xmath47.,scaledwidth=52.0% ] finally , we provide an example of a set of mubs where lemma [ lem : smalllbound ] is tight . for the set of @xmath63 mubs in dimension @xmath2 constructed from the classes given in , our bound @xmath213 \approx 0.678\ , \end{aligned}\ ] ] is tight , and the minimum is indeed achieved by a state that is invariant under the unitary transform that cycles through the bases , as defined in . as noted in section [ sec : symmetry ] , in this case , the largest eigenvalue of @xmath159 occurs for a @xmath160 of the form @xmath176 for any @xmath214 . the states that achieve the lower bound are in fact eigenvectors of @xmath47 , which can be expressed in terms of the mub basis vectors as follows , @xmath215 average min - entropy for different sets of mubs in dimension @xmath130 . + the bound in is close to tight for @xmath66 mubs in dimension @xmath130 . the second analytical bound in is stronger than for @xmath216 bases . the circle denotes the average min - entropy for invariant states constructed in dimension @xmath130 , similar to the states described in . for @xmath217 mubs in @xmath130 , the minimum of the average min - entropy is nearly attained by states invariant under @xmath47.,scaledwidth=52.0% ] we have shown that there exist up to @xmath43 mutually unbiased bases in dimension @xmath1 for which we can find a unitary that cyclically permutes these bases , whenever @xmath41 is prime and @xmath41 divides @xmath44 or @xmath45 . this unitary is found by exploiting symmetry properties of the clifford algebra . our approach is quite distinct from the phase space approaches that were previously used to show that there exists such a unitary for the set of all @xmath39 mubs @xcite , or for two halves of the full sets of mubs when @xmath42 @xcite . our unitary can be understood as a generalization of the fourier transform , and it would be interesting to see whether it has other applications in quantum information . it is an interesting open question to generalize our result to other dimensions , or to a different number of bases . in prime dimension , one could consider the generalized clifford algebra @xcite . even though it does not have the full @xmath218 symmetry , it nevertheless exhibits enough symmetries to allow an exchange of generators . this stems from the way the ( generalized ) clifford algebra is obtained @xcite , which permits any transformation that preserves the @xmath219-norm for @xmath220 in dimension @xmath219 . yet , this is only the first step of our construction . as for generalizing our result to any @xmath41 bases in dimension @xmath1 , we note that it _ is _ indeed possible to find such classes even when @xmath41 is not prime , as our example for @xmath63 in dimension @xmath2 shows . however , we also know that for @xmath129 classes in dimension @xmath221 , no partitioning of operators can be found satisfying our requirements . it is an interesting open question as to when such a partitioning can be found in general . finally , we use our complementarity transform to obtain a tight uncertainty relation for the min - entropy for @xmath63 bases in dimension @xmath2 . no tight relations are known for this case before . we also use a slight generalization of the unitary from @xcite to show that when @xmath1 and @xmath41 divides @xmath39 , the minimizing state is an invariant of a certain unitary . this is the first time that significant insight has been obtained on the structure of the minimizing states for min - entropic uncertainty relations for mutually unbiased bases . it is an exciting open question to obtain tight relations in general , and understand the structure of the minimizing states . we are grateful to david gross for pointing us to the relevant literature for the discrete phase space construction . we also thank lukasz fidkowski and john preskill for interesting discussions . pm and sw are supported by nsf grant phy-0803371 . 44 natexlab#1#1bibnamefont # 1#1bibfnamefont # 1#1citenamefont # 1#1url # 1`#1`urlprefix[2]#2 [ 2][]#2 , * * , ( ) . , * * , ( ) . , * * ( ) . , * * , ( ) . , * * ( ) . , * * , ( ) . , * * ( ) . , * * , ( ) . , , , * * , ( ) . , , , * * , ( ) . , , ( ) , . , , , , in _ _ ( ) , pp . . , , , , ( ) , . ( ) , . , * * , ( ) . , , , , , * * ( ) . , * * , ( ) . , * * , ( ) . ( ) , . , * * ( ) . , , , , * * , ( ) . , * * , ( ) . , * * , ( ) . , * * , ( ) . , , , , * * , ( ) . , * * , ( ) , . ( ) , . , * * , ( ) . ( ) , . , ph.d . thesis , ( ) . , * * , ( ) . , ph.d . thesis , ( ) . , in _ _ ( , ) , vol . of _ _ , pp . . , in _ _ ( ) , pp . , . , * * , ( ) . , * * , ( ) . , in _ _ ( ) , pp . . , , , * * , ( ) . , , , * * , ( ) . , * * , ( ) . , _ _ ( , ) . , * * , ( ) . , _ _ ( , ) . , ph.d . thesis , ( ) , . it is well known that for any orthonormal transformation @xmath96 there exists a corresponding unitary transformation @xmath99 @xcite , where we refer to ( * ? ? ? * appendix c ) for instructions on how to obtain explicit constructions . the transformation we wish to construct here , of the form @xmath222 , is thereby particularly simple to obtain . it can be build up from successive rotations in the plane spanned by only two `` vectors '' @xmath114 and @xmath115 . more specifically , we first construct a unitary that corresponds to a rotation around an angle @xmath223 in the plane spanned by @xmath114 and @xmath115 , bringing @xmath114 to @xmath115 . this is simply a reflection around the plane orthogonal to the midvector between @xmath114 and @xmath115 , followed by a reflection around the plane orthogonal to @xmath115 . using the geometric properties of the clifford algebra this corresponds to the unitary @xmath224 to obtain the desired unitary , we now compose a number of such rotations . let @xmath225 if @xmath226 is odd , and @xmath227 if @xmath226 is even . furthermore , let @xmath228 if @xmath41 is odd , and @xmath229 if @xmath41 is even . note that @xmath230 is the unitary that flips the sign of @xmath231 , but leaves all @xmath114 for @xmath232 and @xmath233 invariant . we may then write @xmath234 this unitary hence transforms @xmath235 , but leaves all other generators @xmath114 for @xmath236 invariant . a similar unitary can be found for any transformation @xmath237 @xcite , but is more difficult to construct explicitly . in and we gave examples of constructing @xmath66 and @xmath63 mubs in dimension @xmath2 , such that they are cyclically permuted under the action of a unitary @xmath47 that permutes the clifford generators in @xmath2 . here , we show by a general construction that it is always possible to construct @xmath41 such classes in dimension @xmath200 , whenever @xmath238 and @xmath41 is prime . we also outline a construction for @xmath45 classes , given a unitary @xmath47 that cycles through _ all _ @xmath0 clifford generators , when @xmath0 is prime . given the @xmath94 generators of the clifford algebra in dimension @xmath200 , we consider the set @xmath239 to generate a set of @xmath240 mubs , we seek to group the elements of @xmath241 into @xmath41 classes of commuting operators , ie . sets @xmath242 of size @xmath109 such that 1 . the elements of @xmath110 commute for all @xmath111 , 2 . the classs are all _ mutually disjoint _ , that is , @xmath243 3 . the unitary @xmath47 that cyclically permutes the generators @xmath116 , also permutes the corresponding classes by permuting products of operators appropriately . our approach in obtaining such a set of classes is to first pick @xmath244 elements for the class @xmath245 and then generate the rest of the classes by repeated application of @xmath47 to the elements of @xmath245 . this automatically ensures property ( * p3 * ) . to ensure ( * p1 * ) and ( * p2 * ) , the @xmath244 operators @xmath246 must satisfy the following : 1 . for any pair @xmath247 , @xmath248 = 0 $ ] , and 2 . the operators in @xmath245 cycle through _ mutually disjoint _ sets of operators under the action of @xmath47 . to understand condition ( ii ) better , consider an operator @xmath249 in @xmath245 . then , by construction , @xmath250 for @xmath251 , assuming we construct a total of @xmath41 classes . in addition , property ( ii ) implies @xmath252 , for any @xmath253 . in other words , given any two operators @xmath247 that cycle through the sets @xmath254 respectively under the action of @xmath47 , property(ii ) demands that @xmath255 , for all @xmath256 and @xmath257 . finally , we note that no class can contain two generators @xmath114 and @xmath115 , since they do not commute . when forming the classes we hence ensure that each one contains exactly one generator @xmath114 , which we refer to as the _ singleton _ @xmath258-operator of the class , as opposed to the rest of the elements which will be _ products _ of @xmath258-operators . the fact that each class can contain at most one singleton operator limits us to constructing a maximum of @xmath0 such classes . before proceeding to outline our construction , we establish some useful mathematical facts which will help motivate our algorithm for the construction of mutually disjoint classes . for the rest of the section , we will work with a set of @xmath219 @xmath258-operators @xmath259 that are cycled under the action of @xmath47 , as follows , @xmath260 in other words , we are given a set of @xmath258-operators whose _ cycle - length _ is @xmath219 . first , we consider sets of products of two @xmath258-operators of the form @xmath261 , which we call _ length-2 _ operators . it is convenient to characterize such pairs in terms of the _ spacing _ ( @xmath262 ) between the operators that constitute them . the spacing function @xmath262 , for a given set of @xmath219 operators , is simply defined as : @xmath263 . then , the following holds : [ lem : spacing ] the action of @xmath47 on any length-2 operator @xmath261 leaves its spacing function @xmath264 invariant . thus , length-2 operators that have unique spacings cycle through mutually disjoint sets of operators under the action of @xmath47 . recall , @xmath265 . it clearly follows that @xmath266 similar to defining length-2 operators , we refer to any product of @xmath267 @xmath258-operators as a _ length-@xmath267 _ operator . for operators of length higher than @xmath58 , it becomes convenient to refer to them using their corresponding index sets . for example , the operator @xmath268 will be simply denoted by the index set @xmath269 . in the following lemma , we obtain a condition for any set of length-@xmath267 operators to cycle through mutually disjoint sets under the action of @xmath47 . [ lem : lengthl ] suppose the length-@xmath267 operators ( for @xmath270 ) that belong to the class @xmath271 are such that they correspond to index sets @xmath272 which all sum to the same value @xmath273 then , no given index set of length @xmath267 can belong to more than one class , for prime values of @xmath219 . given the operators @xmath274 , such that the corresponding index sets @xmath275 sum to @xmath276 under the action of @xmath47 , these index sets change to @xmath277 for any index set @xmath278 the sum of the indices corresponding to the new operators @xmath279 becomes @xmath280 proceeding similarly , the corresponding operators in the class @xmath281 have index sets @xmath282 that sum to @xmath283 for all @xmath284 . thus , starting with a constraint on the length-@xmath267 operators in @xmath245 , we have obtained a constraint on the corresponding operators in a generic class @xmath285 . now , to arrive at a contradiction , suppose that an index set @xmath286 whose indices @xmath287 take values from the set @xmath288 , belongs to two different classes , @xmath285 and @xmath289 ( with @xmath290 ) . the constraint imposed by implies @xmath291 without loss of generality , let @xmath292 . since we can form at most @xmath219 classes , the difference @xmath293 can be at most @xmath294 . finally , since @xmath295 , condition can not be satisfied for prime values of @xmath219 . recall that our approach to constructing any @xmath219 classes is to first construct the class @xmath245 , and then obtain the rest by successive application of @xmath47 . therefore , the fact that any index set of a certain length @xmath267 can not belong to more than one class implies that each length-@xmath267 operator in @xmath245 cycles through a unique set of length-@xmath267 operators under @xmath47 . in other words , the length-@xmath267 operators cycle through mutually disjoint sets , as desired . lemma [ lem : lengthl ] thus provides us with a sufficient condition for the set of length-@xmath267 operators in @xmath245 to cycle through mutually disjoint sets under @xmath47 , given a set of @xmath258-operators whose cycle - length is prime - valued . we only need to ensure that the length-@xmath267 operators in the first class that we construct , @xmath245 , correspond to index sets that _ all _ sum to the same value . this condition is of course subject to the constraint that the maximum allowed length for the operators in @xmath245 ( and by extension , in any class ) is @xmath296 . as a warmup , we construct @xmath45 classes in dimension @xmath297 , when @xmath0 is prime . this case is particularly easy , and illustrates how the results of the previous sections will be used in general . [ thm:(2n+1)classes ] let @xmath298 denote the complete set of @xmath299 @xmath258-operators , and let @xmath47 be the unitary that cycles through all of them , that is , @xmath300 if @xmath0 is prime , then there exist @xmath0 classes @xmath301 satisfying properties * ( p1 ) * through * ( p3)*. we prove the existence of @xmath0 classes by construction . we first outline an algorithm to pick @xmath244 operators that constitute the class @xmath271 . the remaining classes are easily obtained by the application of @xmath47 to the elements of @xmath271 . then , we make use of lemmas [ lem : spacing ] and [ lem : lengthl ] to prove that the classes obtained through our construction do satisfy the desired properties . + * * 1 . pick one of the elements of @xmath302 , @xmath303 , as the singleton operator . 2 . pair up the remaining operators in @xmath302 to form @xmath304 length-2 operators which commute with @xmath303 , as follows , @xmath305 where @xmath306 denotes the set of length-2 operators in @xmath307 . since we have left out the pair @xmath308 in the middle , we get , @xmath309 . 3 . form higher length operators that commute with @xmath310 , by combining @xmath303 with appropriate combinations of the length-2 operators . any operator of even length @xmath311 is created by combining @xmath312 pairs in @xmath313 . and any operator of odd length @xmath314 is created by appending @xmath303 to a length-@xmath315 operator . + denoting the sets of length-3 operators as @xmath316 , length-4 operators as @xmath317 , and in general , the set of length-@xmath312 operators as @xmath318 , we have , @xmath319 putting together the operators from steps @xmath320 , @xmath321 , and @xmath322 we get the desired cardinality for the class @xmath271 as follows : @xmath323 the rest of the classes are generated by successive applications of the unitary @xmath47 to the elements of @xmath271 , so that @xmath324 . + it is easy to see that the elements of each class satisfy property ( * p1 * ) above the different length operators have been picked in such a way as to ensure that they all commute with each other . similarly , by construction , they satisfy property * ( p3)*. it only remains to prove property ( * p2 * ) , that the classes are all mutually disjoint . + the elements of @xmath306 correspond to the following set of spacings @xmath325 which are all distinct . so by lemma [ lem : spacing ] , the elements of @xmath306 cycle through mutually disjoint sets of length-@xmath58 operators . for higher length operators , we first show that our construction meets the conditions of lemma [ lem : lengthl ] . for the class @xmath245 , the elements of @xmath306 correspond to index sets that satisfy @xmath326 the length-2 operators of a generic class @xmath281 similarly satisfy @xmath327 since higher length operators are essentially combinations of length-2 operators and the singleton operator , conditions similar to hold for higher length index sets as well . since operators of even length @xmath311 contain @xmath183 pairs from @xmath306 , the corresponding index sets in @xmath245 satisfy @xmath328 similarly , since the odd length operators have @xmath123 appended to the even length operators , the index sets of length @xmath314 in @xmath245 satisfy , @xmath329 to sum up , for any @xmath330 , our construction ensures that index sets of length-@xmath267 belonging to @xmath245 sum to the same value . the conditions of lemma [ lem : lengthl ] are therefore satisfied , with the quantity @xmath331 in taking the value @xmath332 , for all @xmath333 . now , we can simply evoke lemma [ lem : lengthl ] to prove that , when @xmath0 is prime , the higher length operators in @xmath245 cycle through mutually disjoint sets of operators . next , we show that it is possible to obtain an arrangement of operators into @xmath41 classes in dimension @xmath334 , when @xmath41 is prime and @xmath238 , such that the unitary @xmath47 that cyclically permutes @xmath41 @xmath258-operators also permutes the coresponding classes . [ lclasses ] suppose @xmath47 is a unitary that cycles through sets of @xmath41 @xmath258-operators from the set @xmath335 in dimension @xmath334 , where @xmath41 is prime and @xmath238 . then there exist @xmath41 classes @xmath336 that satisfy properties * ( p1 ) * through * ( p3)*. * proof : * note that since @xmath238 we have @xmath337 for some positive integer @xmath338 . the set of @xmath94 clifford generators @xmath339 can then be partitioned into @xmath340 sets as follows : @xmath341 without loss of generality , we can assume the unitary @xmath47 is constructed such that it cyclically permutes the @xmath41 operators within each set , as follows . @xmath342 once again , we begin with an algorithm for picking @xmath244 elements for the class @xmath245 . the algorithm closely follows the one outlined in the previous section , barring some minor modifications . + * * 1 . the `` middle '' element from @xmath343 , @xmath344 , is picked as the singleton element of @xmath245 . the @xmath304 length-2 operators which commute with @xmath344 are picked as follows 3 . @xmath345 pairs are picked from @xmath346 @xmath347 leaving @xmath123 and @xmath348 unused . @xmath349 pairs are picked from each of the sets @xmath343 through @xmath350 , @xmath351 leaving the first operator in each set unused . finally , the unused @xmath258-operators from different sets are put together as specified below , to get the remaining @xmath338 length-@xmath58 operators : @xmath352 the set of length-@xmath58 operators is then given by @xmath353 which gives @xmath354 . pick higher length operators from @xmath241 that commute with @xmath344 and @xmath313 , by combining @xmath344 with appropriate combinations of the length-2 operators . as before , any even - length operator of length @xmath355 is obtained by combining @xmath312 length-2 operators from @xmath306 . any operator of odd - length @xmath356 , is created by appending @xmath344 to a length-@xmath357 operator . putting together all the operators created in steps[1]-[3 ] , we get the desired cardinality for the class ( see ) , that is , @xmath358 . + * proof of properties * ( p1 ) * through * ( p3 ) * : * the different length operators have again been picked in such a way as to ensure that they all commute with each other . since the remaining @xmath359 classes are generated by successive applications of the unitary @xmath47 to the elements of @xmath271 , we have @xmath360 . thus * ( p1 ) * and * ( p3 ) * is satisfied . it remains to prove that the classes constructed here also satisfy property ( * p2 * ) . as in the earlier case of @xmath0 classes , the operators in each of the sets @xmath361 correspond to unique values of the spacing function : @xmath362 , \nonumber\ ] ] which guarantees , by lemma [ lem : spacing ] that these operators cycle through mutually disjoint sets under @xmath47 . since the operators in @xmath363 are formed by combining @xmath258-operators from different sets @xmath364 , each of them cycles through a different set of operators under @xmath47 . thus we see that all the length-@xmath58 operators in @xmath245 cycle through mutually disjoint sets . before we proceed to discuss the higher length operators , we make one further observation about the length-@xmath58 operators . the operators in @xmath306 correspond to index sets which satisfy @xmath365 in particular , the length-2 operators in the set @xmath366 have been picked carefully so as to ensure that the above constraint is satisfied . in fact , this was the rationale behind leaving out the first operator in each of the sets @xmath364 while choosing the corresponding length-@xmath58 elements in @xmath367 . the higher length operators in @xmath245 can be of two types : 1 . those that are comprised of @xmath258-operators from a single set @xmath364 alone , and 2 . operators that comprise @xmath258-operators from more than one set . since a type-(a ) operator can not cycle into a type-(b ) operator under the action of @xmath47 , these two cases can be examined separately . + * type-(a ) : * the maximum length that an operator of type-(a ) can have , as per our construction , is @xmath359 . we have ensured this by leaving at least one operator of each of the sets @xmath364 unused in constructing the length-@xmath58 operators . furthermore , the constraint in implies that the index sets corresponding to such higher length operators in @xmath245 , sum to the same value modulo @xmath41 . more precisely , any even - length index set of length @xmath368 , where the indices are all drawn from a given set @xmath364 , satisfies @xmath369 and any index set of odd length @xmath314 satisfies @xmath370 then , invoking lemma [ lem : lengthl ] with @xmath371 for even values of @xmath267 and @xmath372 for odd values of @xmath267 , we see that no operator of type-(a ) can belong to more than one class , for prime values of @xmath41 . + * type-(b ) : * an operator of type-(b ) is a product of operators from smaller sets @xmath373 . consider a length-@xmath267 operator , @xmath374 which comprises @xmath375 @xmath258-operators from @xmath376 , @xmath377 operators from @xmath343 , and in general , @xmath378 from the set @xmath364 . @xmath379 note that by our construction , the operator @xmath374 exists in more than one class if and only if , for all @xmath380 the product of all operators in @xmath380 also belongs to more than one class . in what follows , we argue that our construction ensures that this is not possible . in particular , given a set of length-@xmath267 operators in @xmath245 which can be broken down into smaller sets as described above , we will argue that there exists at least one set @xmath380 in every such length-@xmath267 operator @xmath374 , such that the products of operators in @xmath380 corresponding to different length-@xmath267 operators cycle through mutually disjoint sets , as defined earlier . note the following two facts about the subsets @xmath381 . first , our construction ensures that any subset @xmath382 of a given size @xmath383 , satisfies either or depending on @xmath383 being even or odd . second , note that the maximum size of these subsets is @xmath384 . however , in order to invoke lemma [ lem : lengthl ] , we still require @xmath383 to be strictly less than @xmath41 . our goal is hence to show that every length-@xmath267 operator must have at least one subset @xmath381 of size @xmath385 . suppose there exists a length-@xmath267 operator such that every subset is of size @xmath41 . then , the operator itself has to be of length @xmath386 however the maximum value of @xmath267 in our construction is @xmath387 , implying that atleast one of the @xmath340 subsets must be of a size strictly smaller than @xmath41 . and , for such a subset of size less than @xmath41 , constraints and ensure that the same subset can not be found in more than one class , provided @xmath41 is prime . the min - entropy of the distribution that an orthonormal basis @xmath389 induces on a state @xmath390 is given by @xmath391\ ] ] we are looking to evaluate a lower bound on the average min - entropy of any @xmath41 mutually unbiased bases ( not necessarily coming from our construction ) in a @xmath13-dimensional hilbert space . the average min - entropy is given by - @xmath392 using jensen s inequality . the problem of finding an optimal uncertainty relation for the min - entropy , thus reduces to the problem of maximizing over all @xmath390 , the quantity @xmath393 . it is easy to see that this maximum is always attained at a pure state , so we can restrict the problem to an optimization over pure states . we can simplify the problem of finding the lower bound of by recasting it as follows . consider states of the form @xmath394 where @xmath395 denotes a string of basis elements , that is , @xmath396 . suppose we can show for all possible strings @xmath160 , @xmath397 then , since @xmath398 $ ] , the bound is simply @xmath399 we have thus reduced the problem to one of finding the maximum eigenvalue of operators of the form @xmath159 , over all possible strings @xmath160 . note that by , it is sufficient to determine @xmath157 in . to solve this eigenvalue problem we recall a result of schaffner @xcite proved using the methods of kittaneh @xcite , that for a set of @xmath41 orthogonal projectors @xmath402 , the following bound holds : @xmath403 where @xmath404 denotes the operator norm , which here is simply the maximum eigenvalue for hermitian operators . applying this result to sums of basis vectors @xmath208 , we have , @xmath405 which implies @xmath406 recall , that for all @xmath407 @xmath408 where @xmath409 denotes some phase factor . further , since the vectors @xmath208 are normalized , the cauchy - schwarz inequality gives @xmath410 combining these with gives the following bound on the maximum eigenvalue of the operator @xmath159 : @xmath411 by , this immediately proves our claim . here , we present an alternate approach to bound the maximum eigenvalue of @xmath159 , using a bloch vector like representation of the mub basis states . the bound that we obtain here , stated in lemma [ lem : largelbound ] , is stronger than the last one when @xmath211 . in particular , when we consider the complete set ( @xmath188 ) of mubs in any dimension @xmath13 , this approach yields the best known bound . first , we switch to working in a basis of hermitian operators , so that every state in @xmath413 has a parametrization in terms of vectors in a real vector space . any state @xmath390 can be written as : @xmath414 where @xmath415 are hermitian , trace - less operators that are orthogonal with respect to the hilbert - schmidt norm : @xmath416 = 2\;\delta_{ij}$ ] , and the scalars @xmath417 . thus we can parameterize any state in our @xmath13-dimensional hilbert space with a vector @xmath418 . when @xmath165 is a pure state ( @xmath419 = 1 $ ] ) , the vector @xmath420 corresponding to this pure state satisfies the following normalization condition @xmath421 & = & 1 \nonumber \\ \rightarrow \frac{1}{d } + \frac{1}{2}\sum_{i=1}^{d^2 - 1}|\alpha^{(i)}|^{2 } & = & 1 \nonumber \\ \rightarrow |\vec{\alpha}| = \sqrt{\sum_{i=1}^{d^2 - 1}|\alpha^{(i)}|^{2 } } & = & \sqrt{\frac{2(d-1)}{d}}\end{aligned}\ ] ] * ( m1 ) _ normalization _ : @xmath424 = 1 $ ] implies that @xmath425 , @xmath426 . ( by an argument similar to the one that leads to . ) * ( m2 ) _ constant inner - product _ : @xmath427 implies that @xmath428 . this is easily seen , as follows : @xmath429 & = & \frac{1}{d } + \frac{1}{2}\sum_{i}\alpha^{(i)}_{(b , j)}\alpha^{(i)}_{(\hat{b},k ) } = \frac{1}{d } \nonumber \\ \rightarrow \vec{\alpha}_{(b , j)}.\vec{\alpha}_{(\hat{b},k ) } & = & 0\end{aligned}\ ] ] now , using this representation of mub states and density operators , we can rewrite the maximization problem of as : @xmath430 = \max_{{|\psi\rangle}}\textrm{tr}\left[\frac{1}{l}\sum_{j}{|b^{(j)}\rangle}{\langleb^{(j)}|}{|\psi\rangle}{\langle\psi|}\right]\nonumber \\ & \leq & \max_{\vec{\alpha}}\frac{1}{l}\sum_{j}\textrm{tr}\left[\left(\frac{\mathbb{i}}{d } + \frac{\sum_{j}\alpha^{j}_{(b^{(j)},j)}\hat{a}_{j}}{2}\right)\left(\frac{\mathbb{i}}{d } + \frac{\sum_{i}\alpha^{(i)}\hat{a}_{i}}{2}\right)\right ] \nonumber \\ & = & \max_{\vec{\alpha}}\frac{1}{l}\sum_{j}\left(\frac{1}{d } + \frac{1}{2}\vec{\alpha}_{(b^{(j ) } , j)}.\vec{\alpha}\right ) \nonumber \\ & = & \frac{1}{d } + \max_{\vec{\alpha}}\frac{1}{2l}\sum_{j}\vec{\alpha}_{(b^{(j ) } , j)}.\vec{\alpha } \end{aligned}\ ] ] now we only need to find the real @xmath431-dimensional vector @xmath420 , that maximizes the sum @xmath432 . if we now define an `` average '' vector corresponding to each string @xmath160 , as follows @xmath433 then , it becomes obvious that the maximum is attained when @xmath420 is parallel to @xmath434 . since it is a vector corresponding to a pure state , its norm is given by , so that @xmath435 note that this maximizing vector has a constant overlap with all vectors @xmath436 , for a given string @xmath160 . in other words , for each string @xmath160 , the maximum is attained by the vector that makes equal angles with all the vectors that constitute the `` average '' vector ( @xmath437 ) corresponding to that string . note however that this vector may not always correspond to a valid state . now that we know the maximizing vector , we can go ahead and compute the value of @xmath157 in . @xmath438 & \leq & \frac{1}{d } + \max_{\vec{\alpha}}\frac{1}{2l}\sum_{j}\vec{\alpha}_{(b^{(j ) } , j)}.\vec{\alpha } \nonumber \\ & = & \frac{1}{d } + \frac{1}{2}\max_{\vec{\alpha}}\vec{\alpha}_{(\textrm{avg})}.\vec{\alpha } \nonumber \\ & = & \frac{1}{d } + \frac{1}{2}\frac{\vec{\alpha}_{(\textrm{avg})}.\vec{\alpha}_{(\textrm{avg})}}{|\vec{\alpha}_{(\textrm{avg})}|}\sqrt{\frac{2(d-1)}{d } } \nonumber \\ & = & \frac{1}{d } + \frac{1}{2}|\vec{\alpha}_{(\textrm{avg})}|\sqrt{\frac{2(d-1)}{d } } \nonumber \\ & = & \frac{1}{d } + \frac{1}{2\sqrt{l}}\frac{2(d-1)}{d } \nonumber \\ & = & \frac{1}{d}\left(1 + \frac{d-1}{\sqrt{l}}\right)\end{aligned}\ ] ] where we have used the fact that the vector @xmath437 have a constant norm which can be computed as follows : @xmath439 \nonumber \\ \rightarrow |\vec{\alpha}_{(\textrm{avg})}| & = & \frac{1}{\sqrt{l}}\sqrt{\frac{2(d-1)}{d}},\end{aligned}\ ] ] thus proving our claim . the second step follows from the fact that vectors corresponding to different mub states have zero inner product ( see property @xmath440 above ) . note that the fact that the bases are mutually unbiased was crucial in giving rise to properties @xmath441 and @xmath440 which in turn enabled us to identify the maximizing vector @xmath442 . indeed the maximizing vector corresponding to a given string @xmath160 might not always correspond to a valid state , in which case the bound we derive can not be achieved . however , there exist strings of basis elements @xmath160 , for which we can explicitly construct a state that has equal trace overlap with the states that constitute the corresponding operator @xmath159 . these are in fact states of the form @xmath443 for any @xmath444 . clearly , for the symmetric mubs that we construct , an eigenstate of the unitary @xmath47 that cycles between the different mubs has the same trace overlap with each of the states @xmath445 , for a fixed value of @xmath182 . to see this , suppose @xmath446 is an eigenvector of @xmath47 with eigenvalue @xmath447 , then for all @xmath111 and a given value of @xmath182 , @xmath448 & = & |\langle b^{(j)}|\phi\rangle|^{2 } = |\langle b^{(1)}|(u^\dagger)^{j-1}|\phi\rangle|^{2 } \nonumber \\ & = & ( |\lambda|^{2})|\langle b^{(1)}|\phi\rangle|^{2 } \nonumber \\ & = & |\langle b^{(1)}|\phi\rangle|^{2 } \ ; \end{aligned}\ ] ] this is indeed the case for @xmath63 mubs in @xmath2 , where the lower bound we derive is achieved by eigenstates of @xmath47 .

we determine effective collisional radii of positronium ( ps ) by considering ps states in hard - wall spherical cavities . -spline basis sets of electron and positron states inside the cavity are used to construct the states of ps . accurate ps energy eigenvalues are obtained by extrapolation with respect to the numbers of partial waves and radial states included in the bases . comparison of the extrapolated energies with those of a pointlike particle provides values of the effective radius of ps( ) in collisions with a hard wall . we show that for , , and states of ps , the effective radius decreases with the increasing ps center - of - mass momentum , and find a.u . , a.u . , and a.u . in the zero - momentum limit .

1702.00981

positronium ( ps ) is a light atom that consists of an electron and its antiparticle , the positron . positron- and positronium - annihilation - lifetime spectroscopy is a widely used tool for studying materials , e.g. , for determining pore sizes and free volume . for smaller pores the radius of the ps atom itself can not be neglected . this quantity was probed in a recent experiment which measured the cavity shift of the ps @xmath3-@xmath5 line @xcite , and the data calls for proper theoretical understanding @xcite . in this paper we calculate the eigenstates of ps in a hard - wall spherical cavity and determine the effective collisional radius of ps in @xmath3 , @xmath4 , and @xmath5 states as a function of its center - of - mass momentum . the most common model for pore - size estimation is the tao - eldrup model @xcite . it considers an orthopositronium atom ( @xmath9-ps ) , i.e. , a ps atom in the triplet state , confined in the pore which is assumed to be spherical , with radius @xmath10 . the ps is modeled as a point particle with mass @xmath11 in a spherical potential well , where @xmath12 is the mass of an electron / positron . collisions of @xmath9-ps with the cavity walls allow for positron two - gamma ( @xmath13 ) annihilation with the electrons in the wall , which reduces the @xmath9-ps lifetime with respect to the vacuum @xmath14-annihilation value of 142 ns . to simplify the description of the penetration of the ps wave function into the cavity wall , the radius of the potential well is taken to be @xmath15 , where the best value of @xmath16 has been empirically determined to be 0.165 nm @xcite . the model and its extensions are still widely used for pore sizes in 1100 nm range @xcite . porous materials and ps confinement in cavities also enabled a number of fundamental studies , such as measurement of ps - ps interactions @xcite , detection of the ps@xmath17 molecule @xcite and its optical spectroscopy @xcite , and measurements of the cavity - induced shift of the ps lyman-@xmath18 ( @xmath3-@xmath5 ) transition @xcite . cavities also hold prospects of creating a bose - einstein condensate of ps atoms and an annihilation - gamma - ray laser @xcite . seen in a wider context , the old subject of confined atoms @xcite has seen renewed interest in recent years @xcite . studies in this area not only serve as interesting thought experiments but also apply to real physical situations , e.g. , atoms under high pressure @xcite or atoms trapped in fullerenes @xcite . for @xmath9-ps there is a specific question about the extent to which confinement in a cavity affects its intrinsic @xmath14 annihilation rate ( see ref . @xcite and references therein ) . for smaller cavities the effect of a finite radius of the trapped particle on its center - of - mass motion can not be ignored . in fact , the radius of a composite quantum particle depends on the way this quantity is defined and probed . for example , the proton is usually characterized by its root - mean - squared charge radius . it is measured in elastic electron - proton scattering @xcite or using spectroscopy of exotic atoms , such as muonic hydrogen @xcite ( with as yet unexplained discrepancies between these experiments ) . for a particle trapped in a cavity , any practically defined radius may depend on the nature of its interaction with the walls . in the present work we consider the simple problem of a ps atom confined in a hard - wall spherical cavity . the finite size of ps gives rise to energy shifts with respect to the energy levels of a pointlike particle in the cavity . this allows us to calculate the effective _ collisional _ radius of ps that describes its interaction with the impenetrable cavity wall . ps is a hydrogenlike atom with a total mass of 2 and reduced mass of @xmath19 ( in atomic units ) . the most probable distance between the electron and positron in a free ground - state ps atom is @xmath20 , where @xmath21 is the bohr radius , while the mean electron - positron separation is @xmath22 @xcite . for excited states ps@xmath23 ) these quantities increase as @xmath24 . the ps center of mass is halfway between the two particles , so the most probable radius of ps(@xmath3 ) is @xmath25 , its mean radius being @xmath26 . one can expect that the distance of closest approach between the ps center of mass and the wall with which it collides will be similar to these values . one can also expect that this distance will depend on the center - of - mass momentum of the ps atom , as it will be `` squashed '' when colliding with the wall at higher velocities . a proper quantum - mechanical treatment of this problem is the subject of this work . a configuration - interaction ( ci ) approach with a @xmath0-spline basis is used to construct the states of ps inside the cavity . using these we determine the dependence of the effective ps radius on the center - of - mass momentum for the @xmath3 , @xmath4 , and @xmath5 states . of course , the interaction between ps and cavity walls in real materials is different from the idealized situation considered here . it can be modeled by changing the electron - wall and positron - wall potentials . on the other hand , the hard - wall cavity can be used as a theoretical tool for studying ps interactions with atoms @xcite . an atom placed at the center of the cavity will cause a shift of the ps energy levels , whose positions can be related to the ps - atom scattering phase shifts @xmath27 for the @xmath28th partial wave @xcite , @xmath29)}{y_{l+1/2}(k [ r_c-\rho(k)])},\ ] ] where @xmath30 is the ps center - of - mass momentum , @xmath31 is the bessel function , @xmath32 is the neumann function , @xmath10 is the cavity radius , and @xmath33 is the effective collisional radius of the ps atom . the paper is organized as follows . section [ sec : theory ] describes the theory and numerical implementation of the ci calculations of the energy levels and effective radii of ps in a spherical cavity . in sec . [ sec : results ] these energies and radii are presented for a number of cavity sizes and the dependence of the radii of ps(@xmath3 ) , ps(@xmath4 ) , and ps(@xmath5 ) on the ps center - of - mass momentum is analyzed . we conclude in sec . [ sec : conclusion ] with a summary . unless otherwise stated , atomic units are used throughout . the radial parts of the electron and positron states in an empty spherical cavity with impenetrable walls are solutions of the schrdinger equation @xmath34 where @xmath35 is the orbital angular momentum , that satisfy the boundary conditions @xmath36 . although eq . ( [ eq : sch_rad ] ) has analytical solutions , we obtain the solutions numerically by expanding them in a @xmath0-spline basis , @xmath37 where @xmath38 are the @xmath0 splines , defined on an equispaced knot sequence @xcite . a set of 40 splines of order 6 has been used throughout . using @xmath0 splines has the advantange that a central atomic potential can be added in eq . ( [ eq : sch_rad ] ) to investigate ps - atom interactions @xcite . we denote the electron states by @xmath39 , where @xmath40 is the spherical harmonic that depends on the spherical angles @xmath41 , and the positron states by @xmath42 , where @xmath43 ( @xmath44 ) is the position vector of the electron ( positron ) relative to the center of the cavity . for ps in an empty cavity the two sets of states are identical . the indices @xmath45 and @xmath46 stand for the possible orbital angular momentum and radial quantum numbers of each state . the nonrelativistic hamiltonian for ps inside the cavity is @xmath47 where @xmath48 is the coulomb interaction between the electron and positron . the infinite potential of the wall is taken into account through the boundary conditions at @xmath49 . the ps wave functions with a given total angular momentum @xmath50 and parity @xmath51 are constructed as @xmath52 where the @xmath53 are coefficients . the sum in eq . ( [ eq : pswfn ] ) is over all allowed values of the orbital and radial quantum numbers up to infinity . numerical calculations employ finite values of @xmath54 and @xmath55 , respectively , and we use extrapolation to achieve completeness ( see below ) . substitution of eq . ( [ eq : pswfn ] ) into the schrdinger equation @xmath56 leads to a matrix - eigenvalue problem @xmath57 where the hamiltonian matrix @xmath58 has elements @xmath59 @xmath60 ( @xmath61 ) is the energy of the single - particle state @xmath45 ( @xmath46 ) , and @xmath62 is the electron - positron coulomb matrix element . the vector @xmath63 contains the expansion coefficients @xmath53 . diagonalization of the hamiltonian matrix yields the energy eigenvalues @xmath64 and the expansion coefficients . working expressions for the wave function and matrix elements , in which the radial and angular variables are separated , are shown in appendix [ sec : app1 ] . the effective ps radius is determined from the energy shifts with respect to the states of a point particle with the same mass as ps . we employ the notation @xmath65 $ ] to label the states of ps in the cavity . here @xmath2 refers to its internal state , and @xmath66 $ ] describes the state of the ps center - of - mass motion . the means of determining the four quantum numbers @xmath67 , @xmath35 , @xmath68 , and @xmath28 for ps states is described in sec . [ sec : statident ] . also , to obtain accurate values of the effective ps radius , the energy eigenvalues @xmath64 and other expectation values are extrapolated to the limits @xmath69 and @xmath70 ; this is discussed in detail in sec . [ sec : extrapolation ] . we consider each energy eigenvalue @xmath71}$ ] as the sum @xmath72}=e_{nl}^\text{int}+e_{nl[n , l]}^\text{com},\ ] ] where @xmath73 is the internal ps bound - state energy , and @xmath71}^\text{com}$ ] is the energy of the center - of - mass motion . the latter is related to the center - of - mass momentum @xmath74}$ ] by @xmath75}^\text{com}=\frac{k_{nl[n , l]}^2}{2m},\ ] ] where @xmath76 is the mass of ps . away from the wall the ps wave function decouples into separate internal and center - of - mass wave functions , viz . , @xmath77}^\text{com}({\mathbf}{r } ) , \ ] ] where @xmath78 , @xmath79 is the position vector of the ps center of mass , and @xmath80 is the clebsch - gordan coefficient that couples the rotational state @xmath81 of the ps internal motion with that of its center - of - mass motion ( @xmath82 ) . since the center of mass is in free motion , its wave function is given by @xmath83}^\text{com}({\mathbf}{r } ) \propto \frac{1}{\sqrt r } j_{l+1/2}(k_{nl[n , l]}r ) y_{lm_l}(\omega_{\mathbf}{r}).\ ] ] for a pointlike particle , the quantization of the radial motion in the hard - wall cavity of radius @xmath10 gives @xmath74}r_c = z_{l+1/2,n}$ ] , where @xmath84 is the @xmath68th positive root of the bessel function @xmath85 . ( for @xmath86-wave ps , @xmath87 , @xmath88 . ) when the finite effective radius @xmath89}$ ] of the ps atom is taken into account , one has @xmath90}(r_c-\rho_{nl[n , l]})=z_{l+1/2,n},\ ] ] which gives the energy ( [ eq : enlnl ] ) as @xmath91}=-\frac{1}{4n^2}+\frac{z_{l+1/2,n}^2}{4(r_c-\rho_{nl[n , l]})^2}.\ ] ] this relation defines the effective collisional radius of ps , @xmath92 } = r_c - z_{l+1/2,n}{\mathopen{}\mathclose\bgroup\originalleft}(4e_{nl[n , l]}+\frac{1}{n^2}{\aftergroup\egroup\originalright})^{-1/2}.\ ] ] the ps radius thus defined may depend on the ps center - of - mass state @xmath66 $ ] , as well as the cavity radius . as we show in sec . [ sec : results ] , the radius is in fact determined only by the internal ps state @xmath2 and its center - of - mass momentum @xmath30 , i.e. , it can be written as @xmath93 . in the present work , most of the calculations were performed for @xmath94 and @xmath95 , and cavity radii @xmath96 a.u . and 12 a.u . the value of @xmath10 was kept small to assist convergence of the ci expansion ( [ eq : pswfn ] ) . after diagonalizing the hamiltonian matrix and finding the energy eigenvalues , one must determine the quantum numbers for each state before the corresponding ps radius @xmath89}$ ] can be calculated from eq . ( [ eq : rho_def ] ) . to facilitate this , the mean electron - positron separation @xmath97 and contact density @xmath98 were calculated for each state ( see appendix [ sec : app1 ] for details ) . the value of @xmath97 for free ps is twice the hydrogenic electron radius @xcite , @xmath99 thus , the expected mean separations for the @xmath3 , @xmath4 , and @xmath5 states are 3 , 12 , and 10 a.u . , respectively . in practice , the calculated separations for the @xmath4 and @xmath5 states are noticeably lower ( see sec . [ subsubsec : sep ] ) , since the free - ps values of @xmath97 are comparable to the size of the cavity . nevertheless , they are useful for identifying the individual ps states . the contact density @xmath98 is useful for distinguishing between @xmath100 and @xmath101 states , and between the @xmath100 states with different principal quantum number @xmath67 . for @xmath100 states of free ps , the contact density is given by the hydrogenic electron density at the origin @xcite scaled by the cube of the reduced - mass factor , viz . , @xmath102 hence , the expected contact densities of @xmath3 and @xmath4 states are @xmath103 and @xmath104 , respectively . for @xmath101 and higher - angular - momentum states the contact density is zero . in practice , the computed contact density for a @xmath101 state was observed to be of the order of @xmath105 or smaller . although looking at the numerical values of @xmath97 and @xmath106 calculated for finite @xmath54 and @xmath55 is often sufficient for distinguishing between the various states , their values can also be extrapolated to the limits @xmath69 and @xmath70 , as demonstrated in sec . [ sec : extrapolation ] . once the internal ps state @xmath2 has been established using the mean electron - positron separation and contact density , the angular momentum and parity selection rules , @xmath107 allow one to determine the possible values of @xmath28 . finally , for fixed @xmath67 , @xmath35 , and @xmath28 , with the energy eigenvalues arranged in increasing numerical order , the corresponding values of @xmath68 are 1 , 2 , 3 , etc . in order to obtain the most precise values possible , the calculated energy eigenvalues are extrapolated to the limits @xmath69 and @xmath70 . to this end , each calculation was performed for several consecutive values of @xmath54 with fixed @xmath55 , and this process was repeated for several values of @xmath55 . for @xmath94 we used values of @xmath10815 and @xmath10915 . due to computational restrictions , for @xmath110 it was necessary to lower the values of @xmath54 to the range 1014 , while keeping @xmath10915 ( sec . [ subsec : jneq0 ] ) . extrapolation is first performed with respect to @xmath54 for each value of @xmath55 , and afterwards with respect to @xmath55 . the convergence of ci expansions of the type ( [ eq : pswfn ] ) is controlled by the coulomb interaction between the particles . in our case , for @xmath100 states the contributions to the total energy from electron and positron states with orbital angular momentum @xmath35 behave as @xmath111 , while for @xmath101 states , as @xmath112 @xcite . denoting by @xmath113 a generic unextrapolated energy eigenvalue computed with partial waves up to @xmath114 with @xmath115 states in each , we can extrapolate in @xmath54 by using fitting curves of the form @xmath116 for @xmath100 states , and @xmath117 for @xmath101 states , where @xmath118 , @xmath0 , and @xmath119 are fitting parameters , and higher - order terms are added to improve the fit . such curves were found to give excellent fits for our data points . figure [ fig : energyextraplmax ] shows the extrapolation of the 6 lowest energy eigenvalues for @xmath94 , @xmath96 a.u . , and @xmath120 . the effect of extrapolation can be seen better in fig . [ fig : en_extr_lmax_zoom ] , which shows it for the lowest eigenvalue . for the lowest 6 energy eigenvalues for @xmath94 , @xmath96 a.u . , and @xmath120 . solid lines connect the data points to guide the eye , while the dashed lines show extrapolation to @xmath121 using eq . ( [ eq : eextr_s ] ) or ( [ eq : eextr_p ] ) . ] for the lowest energy eigenvalue for @xmath94 , @xmath96 a.u . , and @xmath120 . the solid line connects the data points to guide the eye , while the dashed line shows extrapolation to @xmath121 using eq . ( [ eq : eextr_s ] ) . ] to extrapolate with respect to the maximum number of states in each partial wave , we assume that the increments in the energy with increasing @xmath122 decrease as its negative power . the extrapolated energy eigenvalue @xmath71}$ ] is then found from the fit @xmath123 } + \alpha{n_\text{max}}^{-\beta},\ ] ] where @xmath18 and @xmath124 are fitting parameters . again , such curves produced very good fits for our data points . figure [ fig : energyextrapnmax ] shows the extrapolation of the 6 lowest energy eigenvalues for @xmath94 and @xmath96 a.u . extrapolation of the energy in @xmath55 for the lowest 6 energy eigenvalues for @xmath94 and @xmath96 a.u . solid lines connect the data points to guide the eye , while the dashed lines show extrapolation to @xmath125 using eq . ( [ eq : eextr_n ] ) . ] it is known that ci - type or many - body - theory calculations for systems containing ps ( either real or virtual ) exhibit slow convergence with respect to the number of partial waves included @xcite . accordingly , the extrapolation in @xmath54 is much more important than the extrapolation in @xmath55 , which provides only a relatively small correction . this can be seen in table [ tab : energies_radii ] , which shows the final extrapolated energy eigenvalues @xmath71}$ ] together with @xmath126 and @xmath113 obtained for the largest @xmath54 and @xmath55 values . expectation values of the electron - positron separation @xmath97 and contact density @xmath98 can also be extrapolated to the limits @xmath69 and @xmath70 . as with the energy , the extrapolation in @xmath55 makes only a small correction , which makes it superfluous here . we only use these quantities to identify ps states , so precise values are not needed . for the mean separation we used fits of the form @xmath127 } = \langle r\rangle + a { \mathopen{}\mathclose\bgroup\originalleft } ( { l_\text{max}}+ \tfrac12 { \aftergroup\egroup\originalright})^{-2 } + b { \mathopen{}\mathclose\bgroup\originalleft } ( { l_\text{max}}+ \tfrac12 { \aftergroup\egroup\originalright})^{-3}\ ] ] for @xmath100 states , and @xmath128 } = \langle r\rangle + a { \mathopen{}\mathclose\bgroup\originalleft } ( { l_\text{max}}+ \tfrac12 { \aftergroup\egroup\originalright})^{-4 } + b { \mathopen{}\mathclose\bgroup\originalleft } ( { l_\text{max}}+ \tfrac12 { \aftergroup\egroup\originalright})^{-5}\ ] ] for @xmath101 states , where @xmath129}$ ] is the value obtained in the calculation with a given @xmath114 . figure [ fig : separationextraplmax ] shows this extrapolation for @xmath94 , @xmath96 a.u . , and @xmath120 for the 6 lowest energy eigenvalues , along with tentative identifications of the quantum numbers @xmath67 and @xmath35 . for the lowest 6 energy eigenvalues for @xmath94 , @xmath96 a.u . , and @xmath120 . solid lines connect the data points to guide the eye , while the dashed lines show extrapolation to @xmath121 using eq . ( [ eq : sep_s_extrap ] ) or ( [ eq : sep_p_extrap ] ) . ] the internal ps states were determined as follows . states 1 and 2 appear to be @xmath3 states since @xmath130 for them . state 4 is also a @xmath3 state ; its @xmath97 values for smaller @xmath54 are close to 3 , but increasing @xmath54 and extrapolation lead to a higher value of @xmath131 . this distortion occurs due to level mixing between the @xmath3 state with @xmath87 and @xmath132 ( state 4 ) and @xmath4 state with @xmath87 and @xmath133 ( state 3 , see below ) . these states are close in energy , and the energy separation between them becomes smaller for @xmath134 ( see fig . [ fig : energyextraplmax ] ) . this increases the amount of level mixing and causes a noticeable decrease of the expectation value of @xmath135 with @xmath114 for state 3 . this analysis is confirmed by the values of the contact density shown in sec . [ subsubsec : contact ] . state 5 appears to be a @xmath5 state due to the much larger value of @xmath97 compared with the @xmath3 states , and also because an excellent fit of the data points is obtained by using eq . ( [ eq : sep_p_extrap ] ) , not eq . ( [ eq : sep_s_extrap ] ) . the value of the contact density confirms this ( sec . [ subsubsec : contact ] ) . states 3 and 6 can be identified as @xmath4 states because their mean separations are higher than those of the @xmath3 and @xmath5 states [ cf . ( [ eq : meanr ] ) ] , and the data points are fitted correctly by using eq . ( [ eq : sep_s_extrap ] ) , not eq . ( [ eq : sep_p_extrap ] ) . note that the calculated separations for the @xmath4 and @xmath5 states are lower than the free - ps values of @xmath136 a.u . and @xmath137 a.u . due to confinement by the cavity . expectation values of the contact density @xmath138 provide a useful check of the identification of the ps states . this quantity has the slowest rate of convergence in @xmath54 , and its extrapolation uses the fit @xcite @xmath139}=\langle \delta({\mathbf}{r})\rangle + \frac{a}{{l_\text{max}}+ \tfrac12 } + \frac{b}{{\mathopen{}\mathclose\bgroup\originalleft } ( { l_\text{max}}+ \tfrac12 { \aftergroup\egroup\originalright})^{2}}.\ ] ] figure [ fig : contactextraplmax ] shows that for @xmath96 a.u . and @xmath140 , extrapolation contributes up to 30% of the final contact - density values for the 6 lowest - energy @xmath94 eigenstates . for the 6 lowest - energy eigenstates for @xmath94 , @xmath96 a.u . , and @xmath120 . solid lines connect the data points to guide the eye , while the dashed lines show extrapolation to @xmath121 using eq . ( [ eq : contact_extrap ] ) . ] values of the contact density confirm the state identifications made in sec . [ subsubsec : sep ] . states 1 , 2 , and 4 have contact densities in the range 0.0370.041 , close to the free - ps value of @xmath141 for the @xmath3 state . state 5 has an extrapolated contact density of @xmath142 , confirming that it is a @xmath5 state ( for which the free - ps value is zero ) . states 3 and 6 , which we identify as @xmath4 sates , have contact densities 0.018 and 0.014 , respectively ; the free - ps value for the @xmath4 state is @xmath143 , i.e. , about 3 times smaller . the explanation for this difference is that the confining cavity compresses the radial extent of the ps internal wave function , thereby increasing its density at @xmath144 . this effect has been observed in calculations of radially confined ps @xcite . the effect of compression on the contact density of ps in @xmath4 states can be estimated from the ratio of the free - ps mean distance @xmath145 a.u . to the values obtained in our calculation ( fig . [ fig : separationextraplmax ] ) . the corresponding density enhancement is proportional to the cube of this ratio , giving @xmath146 and @xmath147 , for states 3 and 6 , respectively , that are close to the extrapolated densities in fig . [ fig : contactextraplmax ] . the same compression effect hardly affects @xmath3 states because they are much more compact , and the corresponding electron - positron separation values ( for states 1 and 2 ) are only a little smaller than the free - ps value of 3 a.u . as noted earlier , the @xmath148 states 3 ( @xmath3 ) and 4 ( @xmath4 ) exhibit some degree of level mixing , which reduces the contact density of the former and increases that of the latter . ccccd .. 1.8d .. 1.8d .. 1.8d .. 1.6d .. 1.3 @xmath149 & @xmath10 & state no . & @xmath65 $ ] & & & & & + @xmath148 & 10 & 1 & @xmath150 $ ] & -0.2144498&-0.215911&-0.216100 & 0.368239 & 1.469 + & & 2 & @xmath151 $ ] & -0.1174946 & -0.119914 & -0.120167 & 0.720647 & 1.281 + & & 3 & @xmath152 $ ] & 0.02472915 & 0.0241658 & 0.0239796 & 0.588148 & 4.659 + & & 4 & @xmath153 $ ] & 0.03396135 & 0.0310792 & 0.0309400 & 1.060075 & 1.109 + & & 5 & @xmath154 $ ] & 0.07686479 & 0.0768633 & 0.0768631 & 0.746627 & 3.982 + & & 6 & @xmath155 $ ] & 0.1511621 & 0.150550 & 0.150504 & 0.923047 & 3.193 + @xmath148 & 12 & 1 & @xmath150 $ ] & -0.2251509&-0.227540&-0.227819 & 0.297866 & 1.453 + & & 2 & @xmath151 $ ] & -0.1593054 & -0.162912 & -0.163350 & 0.588727 & 1.328 + & & 3 & @xmath153 $ ] & -0.05568137 & -0.0601980 & -0.0605930 & 0.870418 & 1.172 + & & 4 & @xmath152 $ ] & -0.01046464 & -0.0107571 & -0.0108531 & 0.454519 & 5.088 + & & 5 & @xmath154 $ ] & 0.02539311 & 0.0253906 & 0.0253903 & 0.592926 & 4.422 + & & 6 & @xmath155 $ ] & 0.07401803 & 0.0721989 & 0.0719415 & 0.733325 & 3.432 + & & 7 & @xmath156 $ ] & 0.08558862 & 0.0811920 & 0.0809612 & 1.150585 & 1.078 + @xmath95 & 10 & 1 & @xmath157 $ ] & -0.1787461&-0.181426&-0.181500 & 0.523450 & 1.416 + & & 2 & @xmath158 $ ] & -0.05370988 & -0.0571538 & -0.0573473 & 0.877844 & 1.200 + & & 3 & @xmath159 $ ] & 0.004981346 & 0.00497852 & 0.00497844 & 0.519532 & 3.953 + & & 4 & @xmath160 $ ] & 0.08365396 & 0.0831681 & 0.0830909 & 0.763128 & 4.112 + & & 5 & @xmath161 $ ] & 0.1112176 & 0.111200 & 0.111199 & 0.833544 & 3.086 + & & 6 & @xmath162 $ ] & 0.1228439 & 0.118529 & 0.118321 & 1.213789 & 1.016 + & & 7 & @xmath163 $ ] & 0.1344290 & 0.134426 & 0.134425 & 0.887525 & 2.921 + @xmath95 & 12 & 1 & @xmath157 $ ] & -0.2004982&-0.204725&-0.204872 & 0.424867 & 1.424 + & & 2 & @xmath158 $ ] & -0.1152794 & -0.120225 & -0.120512 & 0.719689 & 1.266 + & & 3 & @xmath159 $ ] & -0.02131484 & -0.0213199 & -0.0213201 & 0.405857 & 4.259 + & & 4 & @xmath162 $ ] & 0.006377168 & 6.0912610 ^ -4 & 1.9794210 ^ -4 & 1.000396 & 1.100 + & & 5 & @xmath160 $ ] & 0.02855653 & 0.0278117 & 0.0277551 & 0.600850 & 4.522 + & & 6 & @xmath161 $ ] & 0.05077695 & 0.0507504 & 0.0507503 & 0.673054 & 3.437 + & & 7 & @xmath163 $ ] & 0.06707912 & 0.0670723 & 0.0670721 & 0.719922 & 3.272 figures [ fig : energyextraplmax][fig : contactextraplmax ] showed how the accurate energies , electron - positron separations and contact densities of the 6 lowest - energy @xmath165 ps eigenstates in the cavity of radius @xmath10 were obtained . for this symmetry , the electron and positron orbital angular momenta @xmath166 and @xmath167 in the expansion ( [ eq : pswfn ] ) are equal , and the dimension of the hamiltonian matrix in eq . ( [ eq : hmat ] ) is @xmath168 ( 3600 in the largest calculation ) . the ground state of the system describes ps(@xmath3 ) with the orbital angular momentum @xmath87 in the lowest state of the center - of - mass motion , @xmath133 . higher - lying states correspond to excitations of the center - of - mass motion of ps(@xmath3 ) ( @xmath169 ) , as well as internal excitations of the ps atom ( @xmath4 and @xmath5 ) . for @xmath148 symmetry , the center - of - mass orbital angular momentum of ps(@xmath5 ) is @xmath170 , which is why this state ( 5 in fig . [ fig : energyextraplmax ] ) lies higher than the lowest @xmath87 , @xmath133 state of ps(@xmath4 ) ( state 4 ) . we also calculated the eigenstates for a larger cavity radius @xmath171 a.u . increasing @xmath10 lowers the energies of all states , and for @xmath165 we identify four @xmath3 states ( @xmath87 , @xmath1334 ) , two @xmath4 states ( @xmath87 , @xmath133 , 2 ) , and one @xmath5 state ( @xmath170 , @xmath133 ) ; see table [ tab : energies_radii ] . states that lie at higher energies , above the ps breakup threshold [ @xmath172 for free ps , or above @xmath173 a.u . in the cavity ] do not have the form ( [ eq : psi_nlnl ] ) but describe a relatively weakly correlated electron and positron `` bouncing '' inside the cavity . to find other states of ps(@xmath5 ) we performed calculations of @xmath174 states for both @xmath96 and 12 a.u . for this symmetry the electron and positron orbital angular momenta are related by @xmath175 , and the size of the hamiltonian matrix is @xmath176 , i.e. , about a factor of 2 larger than for @xmath165 . for computational reasons , it is convenient to define @xmath54 as the maximum angular momentum of one of the particles , e.g. , the positron . in this case the electron orbital angular momentum can be as large as @xmath177 . limiting its value by 14 , we restrict the value of @xmath54 used for extrapolation to the range 1013 . this difference aside , extrapolation of the energy eigenvalues and other quantities for the @xmath95 states is performed as described in sec . [ sec : extrapolation ] . in total , we find 7 eigenstates for @xmath110 : 3 for ps(@xmath3 ) ( @xmath87 , @xmath1333 ) , 3 for ps(@xmath5 ) ( @xmath87 , @xmath133 , 2 and @xmath178 , @xmath133 ) , and 1 for ps(@xmath4 ) ( @xmath170 , @xmath133 ) ; see table [ tab : energies_radii ] . table [ tab : energies_radii ] shows the quantum numbers and energy eigenvalues @xmath71}$ ] of the @xmath94 and @xmath95 states we found for @xmath96 and 12 a.u . alongside the corresponding ps center - of - mass momenta @xmath74}$ ] and effective ps radii @xmath89}$ ] . as expected , the values of @xmath89}$ ] for the ps(@xmath3 ) states are much smaller than those for ps in the @xmath4 and @xmath5 states . the ps radius for each internal state also displays significant variation with the ps center - of - mass quantum numbers @xmath68 and @xmath28 and with the cavity radius @xmath10 . it turns out that , to a good approximation , this variation can be analyzed in terms of a single parameter , namely , the ps center - of - mass momentum @xmath30 . figure [ fig : ps1s ] presents 13 values of the radius of ps(@xmath3 ) states from table [ tab : energies_radii ] , plotted as a function of @xmath30 . the figure shows that , to a very good approximation , the dependence of the ps radius on its center - of - mass momentum is linear , as described by the fit @xmath179 when ps collides with the wall at higher momenta , its effective radius is smaller . this could be expected from a naive picture of ps as a `` soft ball '' that gets `` squashed '' when it hits the hard wall . for higher impact velocities the distortion is stronger , and the ps center of mass gets closer to the wall . the predicted `` static '' ( i.e. , zero - incident - momentum ) radius for the @xmath3 state is @xmath180 a.u . this value is close to the mean distance between the ps center of mass and either of its constituent particles , @xmath181 a.u . ) radius @xmath182 on the ps center - of - mass momentum @xmath30 . the dashed line is the linear fit , eq . ( [ eq : rho1s ] ) . ] figure [ fig : ps2s ] shows the data for the radius of the @xmath4 state . they also suggest a near - linear relationship between the ps radius and center - of - mass momentum , with two points near @xmath183 deviating slightly from it . the larger deviation , observed for the @xmath94 , @xmath171 a.u . datum ( blue cross ) , could be , at least in part , due to level mixing between eigenstates 6 and 7 , which are separated by a small energy interval ( see table [ tab : energies_radii ] ) . a linear fit of the data gives @xmath184 the corresponding static radius @xmath185 a.u . is close to the mean radius of free ps(@xmath4 ) , @xmath186 a.u . dependence of the effective ps(@xmath4 ) radius @xmath187 on the ps center - of - mass momentum @xmath30 . the dashed line is the linear fit , eq . ( [ eq : rho2s ] ) . ] finally , fig . [ fig : ps2p ] shows the dependence of the radius on the center - of - mass momentum for the ps(@xmath5 ) states . in this case , the 6 data points for the @xmath110 states again indicate a linear dependence of @xmath188 on @xmath30 , while the points with @xmath94 appear as outliers . to understand this behavior , we performed additional sets of calculations for @xmath189 and @xmath190 , with @xmath96 and 12 a.u . for @xmath189 we used @xmath19114 and @xmath10915 , while for @xmath192 we used @xmath19313 and @xmath19413 , due to computational restrictions . for both values of @xmath10 we found one ps(@xmath5 ) state with @xmath133 and @xmath170 for each symmetry . table [ tab:1 + 2 + ] shows the data for these states . ) radius @xmath195 on the ps center - of - mass momentum @xmath30 . ] ccccd .. 1.7d .. 1.6d .. 1.3 @xmath149 & @xmath10 & state no . & @xmath66 $ ] & & & + @xmath196 & 10 & 1 & @xmath197 $ ] & 0.0477391 & 0.664045 & 3.233 + @xmath196 & 12 & 1 & @xmath197 $ ] & 0.00718021 & 0.527940 & 3.489 + @xmath190 & 10 & 3 & @xmath197 $ ] & 0.0576334 & 0.693205 & 3.518 + @xmath190 & 12 & 3 & @xmath197 $ ] & 0.0134862 & 0.551312 & 3.850 figure [ fig : ps2p1 + 2 + ] shows the dependence of the radius on the momentum for the ps(@xmath5 ) states , using the data from all of the calculations performed . as noted above , the negative - parity @xmath110 data that describe the states of ps(@xmath5 ) with the center - of - mass angular momentum @xmath87 or 2 display a clear linear trend . in contrast , the three positive - parity states with @xmath165 , @xmath196 , and @xmath190 ( for a given @xmath10 ) do not follow the trend and suggest @xmath50-dependent values of the ps radius . these states correspond to three possible ways of coupling the ps(@xmath5 ) internal angular momentum @xmath198 and its center - of - mass angular momentum @xmath170 . since neither ps(@xmath5 ) nor its center - of - mass wave function for @xmath170 [ cf . ( [ eq : psi_nlnl ] ) ] is spherically symmetric , it is not surprising that the distance of closest approach to the wall ( i.e. , the ps radius ) depends on the asymmetry of the center - of - mass motion through @xmath50 . a simple perturbative estimate of the @xmath50 splitting of these states is provided in appendix [ sec : pert ] . to define a spherically averaged collisional radius of ps(@xmath5 ) we take a weighted average of these data @xmath199j})$ ] with weights @xmath200 for each @xmath10 . the corresponding values are shown by diamonds in fig . [ fig : ps2p1 + 2 + ] . they lie close to the @xmath192 date points ( see appendix [ sec : pert ] for an analytical explanation ) , and agree very well with the momentum dependence predicted by the @xmath110 data . using these together gives the linear fit @xmath201 the static radius for the @xmath5 state @xmath202 a.u . is again close to the mean radius of free ps(@xmath5 ) , i.e. , @xmath203 a.u . ) radius @xmath195 on the ps center - of - mass momentum @xmath30 , including data for @xmath189 and @xmath190 states . the dashed line is the linear fit , eq . ( [ eq : rho2p ] ) . ] regarding @xmath110 data , the ps(@xmath5 ) radii in the states with @xmath87 ( two for each of @xmath10 ) are naturally spherically averaged . the @xmath110 states with @xmath178 are parts of the @xmath50-dependent manifold ( @xmath204 , 2 , and 3 ) . here it appears that the @xmath204 state here is close to the @xmath50-averaged value ( see appendix [ sec : pert ] ) , so that all @xmath110 data follow the same linear momentum dependence . measuments of the ps lyman-@xmath205 transition in porous silica revealed a blue shift of the transition energy @xmath206 mev @xcite . the pore size in this material is estimated to be @xmath207 @xcite . assuming spherical pores for simplicty , we find their radius @xmath208 a.u . the ps center - of - mass momentum in the lowest energy state in such pores is @xmath209 a.u . for such a small momentum one can use static values of the ps radius in @xmath3 and @xmath5 states ( see figs . [ fig : ps1s ] and [ fig : ps2p1 + 2 + ] ) . considering @xmath86-wave ps ( @xmath87 ) , we estimate the cavity shift of the lyman-@xmath205 transition energy from eq . ( [ eq : enlnlrho ] ) , @xmath210 for static radii @xmath6 a.u . and @xmath211 a.u . , and @xmath212 a.u . , we obtain @xmath213 mev . this value is close to the naive estimate using mean ps radii @xcite and is significantly larger than the experimental value . it appears from the measured @xmath214 that the radius of ps(@xmath5 ) is only slightly greater than that of ps(@xmath3 ) . this effect is likely due to the nature of ps interaction with the wall in a real material . the ps(@xmath5 ) state is degenerate with ps(@xmath4 ) , and their linear combination ( a hydrogenic eigenstate in parabolic coordinates @xcite ) can possess a permanent dipole moment . such a state can have a stronger , more attractive interaction with the cavity wall than the ground state ps(@xmath3 ) . this interaction will result in an additional phase shift of the ps center - of - mass wave function reflected by the wall . the scattering phase shift @xmath215 is related to the ps radius @xmath93 by @xmath216 [ cf . ( [ eq : def_rho ] ) ] , with the static radius @xmath217 playing the role of the scattering length . it is known that the van der waals interaction between the ground - state ps and noble - gas atoms can significantly reduce the magnitude of the scatetring length @xcite . it can be expected that a similar dispersive interaction between excited - state ps and the cavity wall can reduce the effective radius of ps(@xmath5 ) by more than that of ps(@xmath3 ) , to produce the difference @xmath218 compatible with experiment . a @xmath0-spline basis was employed to obtain single - particle electron and positron states within an otherwise empty spherical cavity . these states were used to construct the two - particle states of positronium , including only finitely many partial waves and radial states in the expansion . diagonalization of the hamiltonian matrix allowed us to determine the energy and expectation values of the electron - positron separation and contact density for each state . extrapolation of the energy with respect to the maximum orbital angular momentum @xmath54 and the number of radial states @xmath55 included for each partial wave was carried out . the electron - positron separation and contact density values were also extrapolated with respect to the number of partial waves included and used to determine the nature ( i.e. , the quantum numbers ) of each positronium state . from the extrapolated energies , the effective collisional radius of the positronium atom was determined for each state . we have found that the radius of the ps atom in the ground state has a linear dependence on the ps center - of - mass momentum , eq . ( [ eq : rho1s ] ) , the radius being smaller for higher impact momenta . the radius of ps(@xmath4 ) also displays a linear momentum dependence , eq . ( [ eq : rho2s ] ) . the static ( i.e. , zero - momentum ) collisional radii of the @xmath3 and @xmath4 states are 1.65 and 7 a.u . , respectively . determining the effective radius of ps in @xmath5 state is more complex due to its asymmetry . spherically averaged values of the collisional radius are obtained directly for ps @xmath86-wave states in the cavity , with ps @xmath219-wave states giving similar radii . however , determining the corresponding values for the ps @xmath220-wave states required averaging over the total angular momentum of the ps states in the cavity . ( see appendix [ sec : pert ] for a quantitative explanation for the @xmath50 dependence of the @xmath154 $ ] and @xmath161 $ ] energy levels . ) after this , all data points were found to follow the linear dependence on the ps momentum , giving the static radius of 5.35 a.u . in all three cases , the static values of the effective ps radius are close to the expectation value of the radius of free ps , i.e. , a half of the mean electron - positron separation . while the linear fits obtained here for the dependence of the effective ps radius on the center - of - mass momentum are clearly very good , particularly for the @xmath3 state , it must be noted that there is a certain amount of scatter around the lines . this phenomenon may be due to numerical errors in the two - particle - state calculations or in the extrapolation of the energy eigenvalues ( or both ) . the main issue here is the slow convergence of the single - center expansion for states that describe the compact ps atom away from the origin . this issue also prevented us from perfoming calculations for larger - sized cavities , which would provide effective ps radii for lower center - of - mass momenta . a possible means to reduce the scatter in the data and tackle large cavities could be to include more partial waves and radial states per partial wave in the ci expansion . however , with the hamiltonian matrix dimensions increasing as @xmath221 , this quickly becomes computationally expensive . an alternative would be to use a variational approach with explicitly correlated two - particle wave functions . although we have only considered the @xmath3 , @xmath4 , and @xmath5 states in the present work , it is possible to use the method to investigate the effective radii of higher excited states , e.g. , the @xmath222 , @xmath223 and @xmath224 states . however , this would require calculations with much larger cavities that can fit the @xmath225 ps states without significantly squeezing them . it was noted earlier that confinement can cause a ps atom to `` shrink '' from its size _ in vacuo_. this manifests in the form of a reduced electron - positron separation and increased contact density ; these effects were observed for the @xmath4 and @xmath5 states . while this is true in an idealized hard - wall cavity , in physical cavities ( e.g. , in a polymer ) there is a second , competing effect . polarization of the ps atom by the surrounding matter may actually lead to a swollen ps atom , which causes the contact density to be reduced from its value _ in vacuo _ @xcite . experimentally , it has generally been found that the net result of these two effects is that the contact density is reduced from its _ in - vacuo _ value , although increased values are not necessarily impossible @xcite . it may be possible to determine more physical effective ps radii by using realistic electron- and positron - wall potentials in place of the hard wall we have used here . such developement of the approach adopted in the present work should yield much more reliable data for the distorted ps states than crude model calculations @xcite . the technique outlined in this paper is eminently suitable for implementing a bound - state approach to low - energy ps - atom scattering . by calculating single - particle electron and positron states in the field of an atom at the center of the cavity ( rather than the empty cavity ) and constructing two - particle ps wave functions from these , a shifted set of energy levels may be found . from these , the ps - atom scattering phase shifts can be determined [ cf . ( [ eqn : phaseshifts ] ) ] using the now known collisional radius of ps . we have carried out several preliminary calculations in this area for elastic ps(@xmath3)-ar scattering at the static ( hartree - fock ) level and found a scattering length of 2.85 a.u . , in perfect agreement with an earlier fixed - core stochastic variational method calculation in the static approximation by mitroy and ivanov @xcite . this provides evidence that the linear fits presented here account for the finite size of the ps atom in scattering calculations very accurately . in our calculations we have also observed fragmented ps states at higher energies ( this manifests as a larger - than - usual electron - positron separation ) . these have been ignored in the present work , but it may be possible to use them to obtain information about inelastic scattering using this method . it is hoped that the results presented here will be of use in future studies of both confined positronium and positronium - atom scattering . we are grateful to d. g. green for helpful comments and suggestions . the work of a.r.s . has been supported by the department for the economy , northern ireland . written in terms of the angular and radial parts of the electron and positron basis states , the ps wave function is @xmath226 where @xmath227 is the clebsch - gordan coefficient , and the indices @xmath228 and @xmath229 enumerate the radial electron and positron basis states with various orbital angular momenta , @xmath230 and @xmath231 . besides the selection rules due to the clebsch - gordan coefficient , the summation is restricted by parity , @xmath232 , where @xmath233 @xmath234 for the even ( odd ) states . integration over the angular variables in the coulomb matrix elements is performed analytically @xcite , and the hamiltonian matrix elements for the ps states with the total angular momentum @xmath50 are given by @xmath235 where @xmath236 and the reduced coulomb matrix element is @xmath237[l_{\mu ' } ] [ l_\mu ] [ l_\nu ] } { \begin{pmatrix } l_{\mu ' } & l & l_\mu \\ 0 & 0 & 0 \end{pmatrix } } { \begin{pmatrix } l_{\nu ' } & l & l_\nu \\ 0 & 0 & 0 \end{pmatrix}}\nonumber \\ \label{dir } & \quad{}\times \int_{0}^{r_c}\!\!\int_{0}^{r_c } p_{\nu ' } ( r_p)p_{\mu ' } ( r_e)\frac{r_{<}^{l}}{r_{>}^{l+1 } } \nonumber\\ & \quad { } \times p_\mu ( r_e)p_\nu ( r_p ) \ , dr_e \ , dr_p,\end{aligned}\ ] ] with @xmath238\equiv 2l+1 $ ] , @xmath239 , and @xmath240 . the expectation value of the electron - positron separation @xmath241 for an eigenstate with eigenvector @xmath242 is found as @xmath243 where @xmath244[l_{\mu'}][l_\mu][l_\nu ] } { \begin{pmatrix } l_{\mu ' } & l & l_\mu \\ 0 & 0 & 0 \end{pmatrix } } { \begin{pmatrix } l_{\nu ' } & l & l_\nu \\ 0 & 0 & 0 \end{pmatrix}}\nonumber \\ & \quad{}\times \int_0^{r_c } \!\!\int_0^{r_c } p_{\nu'}(r_p ) p_{\mu'}(r_e ) \frac{r_<^l}{r_>^{l+1 } } \nonumber \\ & \quad{}\times { \mathopen{}\mathclose\bgroup\originalleft } ( \frac{r_<^2}{2l+3 } - \frac{r_>^2}{2l-1}{\aftergroup\egroup\originalright } ) p_\mu(r_e ) p_\nu(r_p ) \ , dr_e \ , dr_p .\end{aligned}\ ] ] similarly , the expectation value of the electron - positron contact density @xmath245 is @xmath246 where @xmath247[l_{\mu'}][l_\mu][l_\nu ] } { \begin{pmatrix } l_{\mu ' } & l & l_\mu \\ 0 & 0 & 0 \end{pmatrix } } { \begin{pmatrix } l_{\nu ' } & l & l_\nu \\ 0 & 0 & 0 \end{pmatrix}}\nonumber \\ & \quad{}\times \frac{[l]}{4\pi}\int_0^{r_c}p_{\nu'}(r)p_{\mu'}(r)p_\mu(r)p_\nu(r ) \frac{dr}{r^2 } .\end{aligned}\ ] ] the angular part of the ps wave function in the cavity , eq . ( [ eq : psi_nlnl ] ) , is @xmath248 the electron and positron repulsion from the wall is strongest when the vectors @xmath249 and @xmath250 are parallel or antiparallel . in the simplest approximation , we can take the corresponding perturbation as being proportional to @xmath251 , where @xmath252 is the angle between @xmath249 and @xmath250 . shifting this by a constant to make the spherical average of the perturbation zero , we write it as @xmath253 where @xmath18 is a constant that can depend on the quantum numbers @xmath67 and @xmath68 and on the cavity radius @xmath10 , and @xmath254 is the second legendre polynomial . the corresponding energy shift is @xmath255 by first - order perturbation theory . intergating over the angles , one obtains @xcite @xmath256 where @xmath257 and @xmath258 . for ps states @xmath154 $ ] and @xmath161 $ ] the possible values of @xmath50 are 0 , 1 , 2 and 1 , 2 , 3 respectively . in each case , let @xmath261 denote the calculated energy eigenvalues of the @xmath50 manifold , with the average energy @xmath262 to compare the numerical energy shifts @xmath263 with @xmath264 , we choose @xmath205 to reproduce the calculated mean - squared shift , viz . , @xmath265 ^ 2=\sum _ j(2j+1 ) { \mathopen{}\mathclose\bgroup\originalleft}[\delta e_j{\aftergroup\egroup\originalright}]^2.\ ] ] table [ tab : deltae ] shows the energy eigenvalues , @xmath50-averaged energies , numerical and perturbative energy shifts @xmath266 and @xmath264 , as well as the values of @xmath205 , for ps(@xmath5 ) states with @xmath170 and @xmath178 , for @xmath96 and 12 a.u . cccd1.8d1.7d2.7cd1.6d2.6 @xmath65 $ ] & @xmath10 & @xmath149 & & & & @xmath267 & & + @xmath154 $ ] & 10 & @xmath148 & 0.0768631 & & 0.0203912 & @xmath268 & & 0.019022 + & & @xmath196 & 0.0477391 & 0.0564719 & -0.0087328 & @xmath269 & 0.047556 & -0.009511 + & & @xmath190 & 0.0576334 & & 0.0011615 & @xmath270 & & 0.001902 + & 12 & @xmath148 & 0.0253903 & & 0.0126834 & @xmath268 & & 0.011915 + & & @xmath196 & 0.00718021 & 0.0127069 & -0.0055267 & @xmath269 & 0.029787 & -0.005957 + & & @xmath190 & 0.0134862 & & 0.0007793 & @xmath270 & & 0.001191 + @xmath161 $ ] & 10 & @xmath95 & 0.111199 & & -0.000037 & @xmath271 & & 0.008045 + & & @xmath272 & 0.103201 & 0.111236 & -0.008035 & @xmath269 & 0.040224 & -0.008045 + & & @xmath273 & 0.116992 & & 0.005756 & @xmath274 & & 0.002299 + & 12 & @xmath95 & 0.0507503 & & 0.0008605 & @xmath271 & & 0.005357 + & & @xmath272 & 0.0443476 & 0.0498898 & -0.0055422 & @xmath269 & 0.026787 & -0.005357 + & & @xmath273 & 0.0534797 & & 0.0035899 & @xmath274 & & 0.001531 to complete the multiplet for @xmath178 , calculations for @xmath275 and @xmath273 were carried out using @xmath19313 and @xmath27613 . for a given @xmath10 the values of @xmath205 for @xmath170 and 2 states are similar . on the other hand , when @xmath10 increases from 10 to 12 a.u . , the values of @xmath205 decrease as @xmath277 with @xmath278 . this is close to the expected @xmath279 dependence of the energy shifts with the cavity radius @xcite . for @xmath170 the perturbative estimates of the energy shifts are in excellent agreement with their numerical counterparts . note that the small shift of the @xmath192 level is explained by the small magnitude of the corresponding @xmath280 symbol in eq . ( [ eq : pert ] ) . for @xmath178 states , the perturbative estimate reproduces the overall @xmath50 dependence of the calculated energy shift , with the @xmath281 state being the lowest of the three . however , the relative positions of the @xmath282 and 3 states are reversed . this is probably due to higher - order corrections or level mixing not described by eq . ( [ eq : deltav ] ) . note that the numerical shift is smallest for the @xmath204 state , which justifies its use in determining the fit ( [ eq : rho2p ] ) . 51ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1103/physrevlett.106.023401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.106.209301 [ * * , ( ) ] , link:\doibase 10.1063/1.1677067 [ * * , ( ) ] link:\doibase 10.1016/0301 - 0104(81)80307 - 2 [ * * , ( ) ] and , eds . , @noop _ _ , , vol . ( , , ) link:\doibase 10.1103/physrevb.60.r5157 [ * * , ( ) ] link:\doibase 10.1016/s0301 - 0104(02)00491 - 3 [ * * , ( ) ] link:\doibase 10.1088/1742 - 6596/443/1/012003 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.213401 [ * * , ( ) ] link:\doibase 10.1038/nature06094 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.108.133402 [ * * , ( ) ] link:\doibase 10.1002/pssc.200675760 [ * * , ( ) ] link:\doibase 10.1016/s0031 - 8914(37)80196 - 2 [ * * , ( ) ] link:\doibase 10.1002/andp.19384240109 [ * * , ( ) ] link:\doibase 10.1016/0370 - 1573(95)00070 - 4 [ * * , ( ) ] link:\doibase 10.1021/jp003852u [ * * , ( ) ] in @noop _ _ ( , ) pp . in link:\doibase 10.1007/0 - 387 - 27567 - 3_1 [ _ _ ] , , ( , , ) pp . and , eds . , @noop _ _ , , vol . ( , , ) and , eds . , @noop _ _ , , vol . ( , , ) link:\doibase 10.1088/0034 - 4885/44/1/001 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/33/17/323 [ * * , ( ) ] link:\doibase 10.1038/366123a0 [ * * , ( ) ] link:\doibase 10.1088/0034 - 4885/63/6/201 [ * * , ( ) ] link:\doibase 10.1126/science.1106185 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.116.033401 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.242001 [ * * , ( ) ] link:\doibase 10.1126/science.1230016 [ * * , ( ) ] @noop _ _ , ed . ( , , ) @noop `` , '' ( ) , @noop _ _ ( , , ) @noop _ _ , ed . , , vol . ( , , ) link:\doibase 10.1063/1.462811 [ * * , ( ) ] link:\doibase 10.1088/0953 - 4075/35/2/311 [ * * , ( ) ] link:\doibase 10.1103/physreva.48.4787 [ * * , ( ) ] link:\doibase 10.1103/physreva.70.032720 [ * * , ( ) ] link:\doibase 10.1103/physreva.73.052712 [ * * , ( ) ] link:\doibase 10.1103/physreva.87.020701 [ * * , ( ) ] link:\doibase 10.1103/physreva.88.032708 [ * * , ( ) ] link:\doibase 10.1103/physreva.90.032712 [ * * , ( ) ] link:\doibase 10.1371/journal.pone.0109937 [ * * , ( ) ] link:\doibase 10.1103/physreva.81.052703 [ * * , ( ) ] link:\doibase 10.1103/physreva.65.012509 [ * * , ( ) ] link:\doibase 10.1103/physreva.67.034502 [ * * , ( ) ] link:\doibase 10.1103/physreva.90.052717 [ * * , ( ) ] in @noop _ _ , ( , , ) pp . link:\doibase 10.1139/p83 - 063 [ * * , ( ) ] link:\doibase 10.1103/physrev.120.1289 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.116.033401 [ * * , ( ) ] @noop _ _ ( , , )

we develop an instantonic calculus to derive an analytical expression for the thermally - assisted tunneling decay rate of a metastable state in a fully connected quantum spin model . the tunneling decay problem can be mapped onto the kramers escape problem of a classical random dynamical field . this dynamical field is simulated efficiently by path integral quantum monte carlo ( qmc ) . we show analytically that the exponential scaling with the number of spins of the thermally - assisted quantum tunneling rate and the escape rate of the qmc process are identical . we relate this effect to the existence of a dominant instantonic tunneling path . the instanton trajectory is described by nonlinear dynamical mean - field theory equations for a single - site magnetization vector , which we solve exactly . finally , we derive scaling relations for the spiky " barrier shape when the spin tunneling and qmc rates scale polynomially with the number of spins while a purely classical over - the - barrier activation rate scales exponentially with .

1603.01293

computationally hard combinatorial optimization problems can be mapped to classical spin glass models in statistical physics @xcite . the energy landscape of the corresponding spin hamiltonians @xmath1 possesses a large number of spurious local minima . classical optimization strategies , such as simulated annealing ( sa ) , exploit thermal over the barrier transitions for the search trajectory through the energy landscape towards low - energy spin configurations . in quantum optimization algorithms , such as quantum annealing ( qa ) @xcite ( see also @xcite for recent results ) tunneling can play a functional role by providing additional pathways to low - energy states @xcite . in an archetypical example of qa in spins models , the state evolution is determined by a time - dependent hamiltonian @xmath2 , where @xmath3 is a pauli operator for the @xmath4 spin , and @xmath5 slowly interpolates between a large value and @xmath6 . at sufficiently small values of @xmath7 , all low - energy eigenstates of @xmath8 are localized in the vicinity of the minima of @xmath1 @xcite ( pure states in the spin - glass models ) . in qa the energy landscape is time dependent and the energies of two minima can exchange orders . due to the tunneling between the minima , this results in an avoided crossing with the energy gap @xmath9 . in the absence of an environment and for sufficiently slow evolution , qa corresponds to an adiabatic evolution where the system closely follows the instantaneous ground state of @xmath8 @xcite . the state dynamics can be described as a cascade of landau zener transitions at the avoided crossings . the interaction with an environment can suppress tunneling between two given states of the system . the rate of incoherent tunneling decay is @xmath10 when the relaxation rate @xmath11 due to environment is much larger than the energy gap , i.e. , @xmath12 . however , the environment also gives rise to thermal excitations to higher - energy levels from where the system can tunnel faster @xcite . this is called thermally assisted tunneling @xcite , and has recently been discussed in applications to flux qubit qa @xcite . we assume that the system with the microscopic hamiltonian @xmath8 has a free - energy minimum associated with a thermodynamically metastable state , and that the incoherent tunneling decay rate @xmath13 of this state is much smaller than the smallest rate of relaxation @xmath11 towards the quasi - equilibrium distribution in the domain associated with this state . consider an eigenstate @xmath14 of @xmath8 localized in the metastable domain . the energy of @xmath14 acquires an imaginary part , @xmath15 , due to incoherent tunneling . thus , the partition function of the metastable state becomes a complex number @xmath16 , where @xmath17 is the inverse temperature . for @xmath18 we expand the partition function in @xmath19 and express the total tunneling decay rate @xmath13 of the metastable state in the form @xmath20 where @xmath21 is computed by neglecting the imaginary parts of the energies . the quantity @xmath13 introduced above is given by the average over the micro - canonical decay rates @xmath19 weighted with the corresponding ( quasi ) equilibrium populations . due to the entropic effects , the decay rate @xmath13 can have a very steep and non - monotonic dependence on temperatures @xcite . it can exceed the zero temperature rate or even the coherent tunneling frequency . therefore , optimal protocols of qa must also explore the finite - temperature regime . tunneling transitions represent the bottlenecks of qa @xcite since the tunneling rate @xmath13 decreases exponentially with the size of the domain @xmath22 of cotunneling spins ( typically , hamming distance between the minima ) , @xmath23 , where @xmath24 is a polynomial prefactor . in this paper , we investigate analytically thermally assisted tunneling as a _ quantum _ computational resource for an arbitrary temperatures . we compare the scaling exponent @xmath25 in the tunneling rate @xmath26 to the corresponding transition rate of path integral quantum monte carlo ( qmc ) , which is a classical algorithm that is often used to simulate qa in spin glasses @xcite , and it is the most efficient algorithm to compute exponentially small gaps at first - order phase transitions @xcite . we also provide numerical evidence to support the theoretical findings . the numerical results address the case of non - zero bias and finite ( but small ) temperatures , expanding the numerical findings presented in ref . @xcite we demonstrate in a closed analytical form for mean - field quantum spin models in transverse field that the transition rate @xmath27 of qmc and the quantum tunneling rate @xmath26 have identical exponential scaling with number of co - tunneling spins . this finding applies to the situations where the tunneling is dominated by a most probable path ( an instanton ) . we will employ the path - integral formalism , reminiscent of the one used in the case of continuous @xcite and spin @xcite systems , such as tunneling of magnetization in nano - magnets @xcite . remarkably , despite the big body of literature in this topic , spin tunneling has only been studied , using path integrals , in systems with fixed total spin . we will introduce a non - perturbative spin path - integral instanton calculus for systems where , due to thermal fluctuations , the total spin is not preserved . we augment our analytical results by detailed numerical studies . finally , we compared the quantum tunneling rate , or equivalently the classical qmc rate , with the transition rate corresponding to a classical , purely thermal activation over the barrier . we considered the case where the barrier has a form of a tall and narrow spike . we assumed that the width and height of the spike scale at most linearly with the number of spins @xmath0 . we found the relations for the parameter range of the two scaling exponents where the quantum tunneling rate ( and the classical qmc rate ) do not scale exponentially with @xmath0 while the rate of over - the - barrier thermal escape does . in sec . [ sec : wkb ] , we review the results for a thermally assisted quantum spin tunneling rate initially obtained in @xcite using the wentzel - kramers - brillouin ( wkb ) approach . in sec . [ sec : kramersqmc ] , we developed a kramers theory of the escape rate for the qmc tunneling simulations . in sec . [ sec : comp ] , we use a path integral approach to establish a detailed connection between the wkb results for the quantum tunneling rate and the associated instanton trajectory and qmc escape rate . in sec . [ sec : qmcnum ] , we provide the results of numerical studies . in sec . [ sec - spike ] , we apply the theory developed in this paper to the tunneling problem with small and narrow barriers . the thermally assisted tunneling rate @xmath26 in the mean - field models where the total spin quantum number is not conserved was analyzed in ref . @xcite using the discrete wkb approach @xcite . we shall summarize below the results in ref . @xcite for the archetypical model of a quantum ferromagnet in an @xmath0-spin system @xcite @xmath28 where @xmath7 is the strength of the transverse field ; @xmath29 is the pauli matrix of the @xmath4th spin , where @xmath30 ; @xmath31 is the @xmath25 component of the total spin operator ; and @xmath32 is an arbitrary function of @xmath33 . the mean - field interaction energy density @xmath34 in ( [ eq : h - sigma ] ) has a local and a global minimum . this class of models is known to have large free - energy barriers that lead to exponential ( in @xmath0 ) slowing down for qa @xcite . the squared total spin operator @xmath35 commutes with the hamiltonian @xmath36 . the eigenstates of @xmath36 can be expanded in the basis of the operators @xmath37 and @xmath38 , @xmath39 where @xmath40 is the fractional part of @xmath41 , and @xmath42 is the number of distinct irreducible subspaces with a given total spin quantum number @xmath43 . the coefficients @xmath44 obey the stationary schrdinger equation @xmath45 where @xmath46 . in what follows we will study the limit @xmath47 and assume that @xmath48 . we introduce the normalized total spin quantum number @xmath49 $ ] . to exponential in @xmath0 accuracy , the number of distinct irreducible subspaces @xmath50 for a given value of @xmath51 equals @xmath52 where the binary entropic factor @xmath53 takes the maximum value @xmath54 . to the leading order in @xmath43 , eq . can be written in the form @xmath55 where @xmath56 momentum @xmath57 , and the rescaled variables are @xmath58 the hamiltonian @xmath36 ( [ eq : h - sigma ] ) has quantized eigenvalues @xmath59 . their dependence on the quantum number @xmath60 for each irreducible subspace @xmath61 can be obtained using the discrete wkb approach @xcite . the eigenfunction @xmath62 in the classically forbidden region of @xmath63 is obtained from the analysis of an auxiliary classical one - dimensional system with energy @xmath64 , coordinate @xmath63 , and imaginary momentum " @xmath65 that obeys the hamilton - jacobi equation @xcite @xmath66 in the classically - forbidden region @xmath67 , leading to the condition @xmath68 > \gamma \sqrt{\ell^2-m^2 } > 0 \,.\label{eq : fr}\ ] ] from eq . ( [ eq : hcl ] ) we obtain the momentum @xmath69 at the boundaries of the forbidden region @xmath70 ( @xmath71=0,1 ) and @xmath72 . these points can be obtained from the solution of the transcendental equation @xmath73 where for the effective potential @xmath74 figure [ fig : veff ] shows plots of the effective potential @xmath75 for the curie - weiss model where interaction energy density @xmath34 takes a form @xmath76 for values of total momentum @xmath51 and energy @xmath64 where metastability exists , the eigenfunction under the barrier takes a standard wkb form @xcite @xmath77 then , to exponential accuracy in @xmath0 , the amplitude of the tunneling decay of the metastable state equals to @xmath78 where @xmath79 is the mechanical action under the barrier @xmath80 and @xmath70 are boundaries of a classically - forbidden region ( cf . [ fig : veff ] ) . in the wkb approach one can also compute the action @xmath81 by introducing the instanton trajectory in imaginary time @xmath82 for a system with coordinate @xmath63 , momentum @xmath83 , and velocity @xmath84 . since we have introduced rescaled variables ( [ eq : rescaled ] ) , we also have to rescale the time @xmath85 in order to preserve the canonical relations @xmath86 using this relation we obtain the instanton trajectory @xmath87 in imaginary time @xmath88 @xmath89 ^ 2-\gamma^2(\ell^2-m^2_{\scriptscriptstyle\rm wkb})}\;.\nonumber\end{aligned}\ ] ] then the action is @xmath81=@xmath90 where @xmath91 is the duration of motion under the barrier @xmath92 thermally assisted quantum tunneling is relevant to the study of qmc , which is always implemented at finite temperatures . the rate of the thermally - assisted tunneling @xmath26 can be written in the form ( [ eq : f ] ) @xmath93 here @xmath94 is the tunneling decay rate of the states with the set of quantum numbers @xmath51 and @xmath60 , and @xmath50 gives the number of these states . in accordance with the general prescription ( [ eq : f ] ) , each term in ( [ eq : ws ] ) is proportional to the probability of thermal activation to the states with the energies @xmath95 that subsequently undergoes a tunneling decay . in the limit of large @xmath0 one can replace the summations @xmath96 with @xmath97 and take the integrals using the method of steepest descent . to the leading order one simply needs to maximize the logarithm of the integrand @xmath98+n q_\ell - n \beta e$ ] with respect to @xmath51 and @xmath64 [ here we used eq . ( [ eq : q ] ) ] . in this way , one replaces the sum of the thermally assisted tunneling decays rates over many channels by the rate for the most probably channel ( optimal fluctuation ) . we refer to @xcite for details of this analysis and simply provide a final result @xmath99 where @xmath100 is a prefactor and @xmath101 is the difference of the effective free energies " for the thermally assisted tunneling transition . here , @xmath102 and @xmath103 is a free energy per spin of the metastable state @xmath104 here , @xmath105 is the energy of the metastable minimum . the optimal value of @xmath51 in ( [ eq : r ] ) minimizes the system free energy over the irreducible subspaces associated with the minimum @xmath106 the extremal conditions for eq . ( [ eq : amin ] ) read as @xmath107 \frac{\partial q_\ell}{\partial \ell } = \frac{\partial s_\ell}{\partial \ell}\,.\end{gathered}\ ] ] using ( [ eq : dade ] ) and ( [ eq : q ] ) we get @xmath108 @xmath109 as expected , the most probable channel for decay corresponds to the instanton @xmath110 of period @xmath111 . the partition function @xmath112 of the model at inverse temperature @xmath111 can be obtained by using the suzuki - trotter formula to map a quantum problem to a classical one with one additional ( imaginary - time ) dimension @xmath113 @xcite . in the limit of infinite number of trotter slices , @xmath112 is given by an integral over the array of spin paths @xmath114 where each path @xmath115 is periodical along the imaginary - time axis @xmath116 and parametrized by the locations of points ( kinks " ) at the axis where the sign of @xmath117 changes @xcite . the gibbs probability distribution over paths @xmath118 has the form @xmath119&={\mathcal{z}}^{-1 } \prod_{i=1}^{n}\gamma^{\kappa[\sigma_{j}(\tau)]}e^{-n \int_{0}^{\beta}g [ m(\tau ) ] d\tau},\label{eq : cf}\end{aligned}\ ] ] where @xmath120 and the function @xmath121 $ ] equals to the number of kinks in a given path @xmath117 . qmc samples from the gibbs distribution ( [ eq : cf ] ) using the metropolis - hastings algorithm by implementing a series of stochastic updates of the state vector of individual spin paths @xmath118 . at each qmc step , a transition is proposed from a current system state @xmath118 to a new candidate state @xmath122 based on a certain stochastic update rule [ often using cluster update methods for paths @xmath117 @xcite ] . the rule is specific for a given implementation of the algorithm , however , the _ ratio _ of the acceptance and rejection probabilities of the transition proposal @xmath123 is the same for any implementation and equal to @xmath124/ p_g[\underline{\sigma}(\tau)]$ ] . we assume that the qmc updates of @xmath118 occur at the sequence of random instants of time @xmath125 sampled from a poisson process with a constant rate @xmath126 . then , the stochastic time evolution of the state vector @xmath127 corresponds to the master equation for the probability distribution function @xmath128 : @xmath129 here , @xmath130 are transition probabilities that depend explicitly of the stochastic update rule . based on the above , they must obey the detailed - balance conditions @xmath131}{p_g [ \underline{\sigma}(\tau)]}\label{eq : db}\ ] ] that guarantee that gibbs distribution ( [ eq : cf ] ) is the stationary ( long - time ) solution of ( [ eq : master ] ) . because of the mean - field character of the model ( [ eq : h - sigma ] ) it is possible to obtain in a closed form a gibbs probability measure for the magnetization per spin order parameter @xmath132 : @xmath133=\frac{e^{-n \beta { \mathcal{f}}[m(\tau)]}}{z_{0}},\ , \int d m(\tau ) p_g[m(\tau)]=1\label{eq : z00}\ ] ] and the free - energy functional @xmath134 has the form @xcite@xmath135 = \frac{1}{\beta } \int_{0}^{\beta}[m(\tau ) g^\prime(m(\tau))-g ( m(\tau ) ) ] d\tau \\ - \frac 1 \beta \ln \lambda[g^\prime(m(\tau))]\label{eq : fm}\;.\end{gathered}\ ] ] here , @xmath34 is the mean - field interaction energy density ( [ eq : h - sigma ] ) , @xmath136 , and the functional @xmath137 $ ] equals @xmath138 = { \rm tr}k^{\beta,0}[{\bf b}(\tau)],\label{eq : lambda}\\ k^{\tau_2,\tau_1}[{\bf b}(\tau ) ] = { \rm t_{+}}e^{-\int_{\tau_1}^{\tau_2 } d\tau h_0(\tau)}\,,\label{eq : propagator}\\ h_0(\tau ) = -\mathbf{b}(\tau)\cdot \bm{\sigma},\quad \mathbf b(\tau)=(\gamma,0,\lambda(\tau))\label{eq : h0}\;,\end{gathered}\ ] ] where @xmath139 is the vector of pauli matrices . the propagator @xmath140 $ ] corresponds to a spin-1/2 particle evolving in imaginary time under the action of the magnetic field @xmath141 . if we consider the order parameter as static and remove the `` time '' indices , we obtain the free energy function @xmath142 we denote the minima of the free energy by @xmath143 , @xmath144 . here , the index @xmath145 ( @xmath146 ) corresponds to the local ( global ) minimum of @xmath147 . quantum tunneling can be simulated with qmc using a general approach that was first considered in a qa context @xcite . at @xmath148 the qmc state vector @xmath149 is initiated in the vicinity of the metastable state basin by uniformly sampling from the spin configurations that satisfy the conditions @xmath150 . every time when the state vector @xmath127 arrives at the vicinity of the global minimum @xmath151 , the qmc process is terminated . repeating this experiment many times , one can determine the average escape time of the qmc process from the metastable states @xmath152 , which is proportional to the inverse of the escape rate @xmath27 @xcite . the numerical studies of @xmath27 are described in sec . [ sec : qmcnum ] . it is convenient to study the stochastic trajectories @xmath127 by inspecting their projections @xmath153 onto the continuous functional space defined in ( see fig . [ fig : kramers ] ) . at low enough temperatures the distribution @xmath154 $ ] corresponding to the qmc dynamics ( [ eq : master ] ) quickly relaxes toward quasi - equilibrium ( [ eq : cf ] ) sharply localized in the vicinity of @xmath152 . this relaxation process occurs with the rate @xmath155 . the trajectory @xmath156 spends a long time ( @xmath157 ) near the metastable state @xmath152 . occasionally , a large fluctuation occurs corresponding to the escape event where the path @xmath156 moves away from @xmath152 and arrives eventually at the vicinity of @xmath158 ( see fig . [ fig : kramers ] ) . during the escape event the system initially moves uphill when the free energy @xmath159 $ ] is increasing until it reaches the saddle point of the functional @xmath134 to be denoted as @xmath160 that satisfies the equation @xmath161}{\delta m(\tau)}\right |_{m(\tau)=m_z(\tau)}=0 , \quad m_z(0)=m_z(\beta)\;.\label{eq : var}\ ] ] near the saddle point the escape path @xmath156 slows down , because the variation of @xmath162 $ ] is small . then it goes downhill almost deterministically , so that the free energy @xmath159 $ ] is decreasing until the state @xmath158 is reached . the quasi - stationary statistical distribution over @xmath132 has the gibbs form @xmath163 $ ] everywhere in the domain of the local minimum except in the small vicinity of the saddle point @xmath164-{\mathcal{f}}[m_z(\tau)]|\lesssim \beta^{-1}$ ] , where deviations from @xmath163 $ ] allow for the probability current flow away from the metastable state . this area lies inside the domain marked with dashed line in fig . [ fig : kramers ] . the probability for the path @xmath156 to reach the vicinity of saddle point that lies inside this domain is @xmath163\propto \exp\{-f[m(\tau)]-f[m_0]\}$ ] . this is precisely the boltzmann factor that determines the transition rate exponent in kramers theory of escape @xcite @xmath165-f(m_0)\;,\label{eq : wqmc}\ ] ] where @xmath166 and @xmath167 is a prefactor ( polynomial in @xmath0 ) that depends on the specific path update rule used in qmc . the connection between the saddle points of the free energy functional in a classical one - dimensional field theory and the instantons in the corresponding quantum mechanical tunneling problem was first established using the path - integral formalism in @xcite for the case of a particle in a potential . to apply the same argument to our mean - field quantum spin problem we use the partition function @xmath168 associated with the hamiltonian ( [ eq : h - sigma ] ) in a form of the path integral normalization factor for @xmath163 $ ] ( [ eq : z00 ] ) . then we express the tunneling decay rate in a standard form ( [ eq : f ] ) in terms of the imaginary part of the partition function . for a large number of spins @xmath169 this path - integral can be calculated using the saddle - point method within the instantonic calculus , where the free - energy functional @xmath162 $ ] plays the role of the mechanical action . this gives @xmath170 where @xmath171 is determined by ( [ eq : var ] ) and ( [ eq : wqmc ] ) . the dominant contribution to the path integral is given by the instanton @xmath160 . we will show below that the exponential scaling of @xmath26 and @xmath27 with @xmath0 is the same . the prefactors @xmath167 and @xmath172 are expected to be different . @xmath100 can be expressed entirely in terms of the free - energy functional @xmath173 $ ] while @xmath167 depends on the specific path update rule used in qmc . depicted with blue lines . the most probable escape path is shown with a solid blue line . it starts near @xmath152 and arrives at @xmath158 crossing the vicinity of saddle point @xmath160 shown with a red dot . the dashed line depicts the boundary of the region where @xmath174\leq { \mathcal{f}}[m_z(\tau)]$ ] . inside this region @xmath162-{\mathcal{f}}[m_z(\tau)]\gg t=\hbar/\beta$].,width=288 ] to complete our analysis , we would like to relate our findings to the wkb analysis of thermally assisted tunneling presented in sec . [ sec : wkb ] . we will now show in an explicit analytical form that the exponential scaling coefficient @xmath175 of the qmc transition rate with @xmath0 and the qmc saddle - point solution @xmath160 are identical to , respectively , the exponential scaling coefficient @xmath25 [ eq . ( [ eq : wta ] ) ] of the wkb tunneling rate with @xmath0 and the wkb instanton tunneling path @xmath176 [ eq . ( [ eq : nuh1 ] ) ] . such analysis is of interest because unlike the extrema of the free - energy functional ( [ eq : static_free_energy ] ) that obey the static " condition @xmath177 , the saddle - point ( instanton ) trajectories @xmath160 are time dependent . they also appear in the context of the qa in mean - field spin models @xcite of the type given in ( [ eq : h - sigma ] ) . using the expression ( [ eq : fm ] ) for @xmath134 , we can re - write eq . ( [ eq : var ] ) in the form of the following two equations : @xmath178,\quad \lambda(\tau)= g^\prime[m_z(\tau)]\label{eq : dm - var}\;,\ ] ] where the functional @xmath137 $ ] is defined in ( [ eq : lambda])([eq : h0 ] ) . we now introduce a vector function in imaginary time @xmath179=@xmath180 corresponding to expectation values of the operator @xmath181 for the spin-@xmath182 particle defined as @xmath183\label{eq : dm - vec}\;,\end{aligned}\ ] ] where @xmath184 is defined in eq . . we recognize that eq . corresponds to the @xmath185-component of the vectorial equation . differentiating this equation with respect to @xmath85 and using ( [ eq : h0 ] ) , we obtain @xmath186 k^{\tau,0}\right ) } { { \rm tr}\,k^{\beta,0 } } \;.\label{eq : mvec}\ ] ] one can re - write this equation in the following form @xmath187 where @xmath188 and we observe that @xmath189}/{\partial { \bf m}(\tau)}$ ] . we have to solve this equation with the periodic boundary condition @xmath190 [ cf . ( [ eq : var ] ) ] . equation ( [ eq : dmvec ] ) allows for two integrals of motion @xmath191 where @xmath192 . then , the solution of ( [ eq : dmvec ] ) can be written in the form : @xmath193 where @xmath194 and @xmath195 are given in eqs . ( [ eq : kr ] ) and ( [ eq : nuh1 ] ) , respectively . the equation for @xmath160 is identical to that for the wkb instanton trajectory @xmath196 defined in ( [ eq : nuh1 ] ) . because @xmath51 is a constant of motion , its value can be determined at @xmath197 : @xmath198 since the propagator @xmath199 depends on @xmath51 , eq . ( [ eq : l_self ] ) is a self - consistent condition . we have shown above that the saddle point of qmc satisfies the same differential equations ( [ eq : mz ] ) and ( [ eq : nuh1 ] ) as the wkb instanton . what remains to be shown is that the optimal value of @xmath51 for the wkb approach coincides with that given by the self - consistent condition of qmc ( [ eq : l_self ] ) . in the wkb approach , the optimal value of @xmath51 can be determined by the conditions ( [ eq : opt ] ) and ( [ eq : al ] ) . in this section , we express the self - consistent condition for qmc in the same form as the corresponding condition in wkb to demonstrate their equivalence . it is useful to introduce a replica qubit , which allows us to write the self - consistent condition ( [ eq : l_self ] ) as @xmath200 where @xmath201 and @xmath202 ( @xmath203 ) is the projector onto the symmetric ( anti - symmetric ) subspace of the two qubits . to analyze the double propagator @xmath204 , we consider the hamiltonian @xmath205 where @xmath206 was already introduced in eq . ( [ eq : h0 ] ) . with the bell basis @xmath207 , we have @xmath208 the anti - symmetric singlet state @xmath209 is a dark state , and the triplet states are closed under such evolution . as a consequence , the following identity always holds @xmath210 with which we have @xmath211 using the identity ( [ eq : anti - symmetric_identity ] ) , the self - consistent condition ( [ eq : l_self_d ] ) can be simplified to @xmath212 to solve @xmath51 , we need to know the trace of the double propagator in the symmetric subspace @xmath213 . consider the time evolution of a state @xmath214 in the symmetric subspace of the two qubits , @xmath215 where @xmath216 and @xmath217 take real values . using eqs . ( [ eq : phi+])([eq : psi+ ] ) , we have the following differential equations for the coefficients : @xmath218 where @xmath219 is determined by the instanton equations ( [ eq : mx])([eq : mz ] ) . equation ( [ eq : dxivec ] ) is very similar to eq . ( [ eq : dmvec ] ) , except that it is linearized . thus , a known solution to eq . ( [ eq : dxivec ] ) is the instanton solution @xmath220 . because the instanton solution is periodic , @xmath221 is an eigenvector of the propagator @xmath204 with eigenvalue @xmath222 . to evaluate the trace , we still need to know the other two linearly independent solutions to eq . ( [ eq : dxivec ] ) . we define the following symmetric bilinear form @xmath223 where @xmath224 and @xmath225 are two solutions to eq . ( [ eq : dxivec ] ) . the bilinear form takes real values , and it is not an inner product ( there is no complex conjugate on @xmath226 ) . it is readily seen that the bilinear form is a constant of motion from eq . ( [ eq : dxivec ] ) , @xmath227 let the other two solutions to eq . ( [ eq : dxivec ] ) satisfy the initial conditions @xmath228 notice that @xmath229 because all the bilinear forms in eq . ( [ eq : pseudo_product ] ) are preserved at time @xmath111 and @xmath230 , @xmath231 must proportional to @xmath232 or @xmath233 . the second possibility can be ruled out by noticing that the signs of @xmath234 can not change during the evolution , because @xmath235 for all @xmath236 $ ] . as a result , we have @xmath237 where proportional factors @xmath238 are to be determined . using eq . ( [ eq : pseudo_product ] ) , we have @xmath239 where we neglect the superscripts @xmath240 in @xmath241 to simplify the notations . introducing @xmath242 , we have @xmath243 the solution to the above quadratic equation is @xmath244 where we use the condition ( [ eq : int2 ] ) to simplify things . putting the definition of @xmath245 into eq . ( [ eq : dxivec ] ) , we have @xmath246 which can be solved exactly @xmath247 the integral in the above equation can be evaluated using the solution eq . ( [ eq : ratio ] ) , @xmath248 where we also use the relations @xmath249 and @xmath250 . the eigenvalues @xmath238 can thus be determined , @xmath251 with the three eigenvalues of the double propagator all solved , we have @xmath252 putting the above result into eq . ( [ eq : l_trace ] ) , we have @xmath253 the integral @xmath254 defined in eq . ( [ eq : si ] ) takes the same form as eq . ( [ eq : al ] ) , although there is a time scale difference of @xmath255 between the two . comparing eqs . ( [ eq : opt ] ) and ( [ eq : self_consis ] ) , we conclude that the optimal value of @xmath51 in the wkb approach equals to the self - consistent solution of @xmath51 in the qmc approach . taking into account the expression for @xmath256 defined in eq . ( [ eq : si ] ) , after some transformations the wkb action ( [ eq : a ] ) for the extremal trajectory corresponding to the instanton solution ( [ eq : mz ] ) can be re - written in the following form : @xmath257\,{d}\tau \ , . \label{eq : wkb_action}\end{aligned}\ ] ] where @xmath258 is given in ( [ eq : dm - var ] ) . using eq . ( [ eq : trace_k ] ) , we have the trace of the qmc propagator , @xmath259 or equivalently , @xmath260 with eq . ( [ eq : self_consis ] ) , the logarithm of @xmath261 can be expressed as a function of @xmath51 , @xmath262 from ( [ eq : wkb_action ] ) and ( [ eq : ln_trace_k ] ) the qmc free energy ( [ eq : fm ] ) takes the form @xmath263 & = \int_{0}^{\beta } \big(\lambda(\tau ) { { m}}_z(\tau)-g[{{m}}_z(\tau)]\big ) \ , d\tau-\ln \big(\operatorname{tr}k^\beta\big)\nonumber\\ & = { { s_\ell}}+\beta { e}+ \ell\,\tan^{-1}\ell - \ln \frac{2}{\sqrt{1-\ell^2}}\;.\label{eq : free_energy_instanton}\end{aligned}\ ] ] using the following identity from the definition of the entropic factor @xmath53 ( [ eq : q ] ) , @xmath264 we finally obtain @xmath265= { { s_\ell}}+\beta { e}-q(\ell)\,.\ ] ] after the minimization over @xmath51 and @xmath64 is performed , this has exactly the same form as the expression for the effective free - energy in the exponent of the wkb transition rate ( [ eq : amin ] ) @xmath266= \mathfrak{f}\,.\ ] ] to complete the comparison we note that the static free energy per spin @xmath267 that appears in ( [ eq : wta ] ) equals to the free energy density @xmath268 from ( [ eq : wqmc ] ) . indeed , those are merely two different expressions for the free energy per spin of the metastable state @xmath269 . connecting them requires a cumbersome but straightforward calculation given in appendix . with that we can establish that @xmath270 [ cf . ( [ eq : wta ] ) and ( [ eq : wqmc ] ) ] . thus , we have shown the equivalence of the exponential factors in wkb and qmc transition rate expressions . in this section we compare the analytical result for the tunneling escape rate based on eq . with the qmc escape rate obtained through numerical simulations performed with continuous - time path integral quantum monte carlo ( qmc ) @xcite . we study the quantum curie - weiss model with the hamiltonian ( [ eq : h - sigma ] ) , where @xmath271 is given in ( [ eq : cw ] ) . the corresponding effective potential @xmath75 for @xmath272 is depicted in fig . [ fig : veff ] . we follow the same method as in ref . @xcite to obtain the exponential scaling with @xmath0 in the numerical decay rate in qmc simulations , @xmath273 . however , we extend the study @xcite to the cases of nonzero biases and finite temperatures . we initialize all spin worldlines in the neighborhood of the higher ( left ) local minima of @xmath75 by setting @xmath274 . we measure the number of qmc sweeps ( defined as one attempted update per spin worldline ) required to decay from the metastable state . this is done by counting the number of sweeps until at least @xmath275 of the replicas reverse their magnetization to @xmath276 or @xmath277 of the replicas reverse magnetization results in only small differences in the fitting exponent @xmath25 . this difference is slightly larger for cases when the number of sweeps for a transition in qmc is small . ] . we obtain the average number of sweeps required for a given number of spins @xmath0 by repeating this measurement a large number of times . this was done for @xmath278 . the dependence of the number of sweeps on @xmath0 is well fitted by the expression @xmath279 ( see ref . we finally obtain the exponent @xmath25 from the exponential fit with @xmath280 . in figs . and the exponent @xmath25 obtained from numerical qmc simulations is compared to that obtained from the analytical approach developed in the previous sections . for the analytical value , we solve numerically the rate for thermally assisted tunneling from the expressions ( [ eq : fm ] ) and ( [ eq : wtunnqmc ] ) ( their equivalence to the wkb result ( [ eq : wta ] ) was shown in the previous section ) . we use the instanton solution eq . for a particular @xmath281 to construct the 2-dimensional matrix @xmath282 in eq . , where @xmath283 is a function of the instanton solution . the 2-dimensional matrix @xmath282 is then used to calculate the propagator eq . . knowing the propagator allows one to calculate @xmath284 using eq . . if @xmath285 , the original value @xmath286 is modified until it equals to the corresponding @xmath284 ; this value @xmath287 is the self - consistent solution @xmath288 . the free energy @xmath289 $ ] of the instanton trajectory with @xmath288 ( the saddle point ) can thus be calculated using eq . . the static free energy for the metastable state @xmath268 is calculated by minimizing @xmath147 in eq . over @xmath63 . it can be shown analytically that the extreme points of eq . equal to those of the effective potential ( after optimizing over @xmath51 ) . the escape rate @xmath290-f(m_0))$ ] given in ( [ eq : fm ] ) , ( [ eq : wtunnqmc ] ) is proportional to the difference of the instanton free energy and the static free energy corresponding to the local minimum . it is instructive to compare the quantum tunneling rate , or equivalently the classical qmc rate , with the transition rate corresponding to a classical , purely thermal activation over the barrier . the latter is the principal mechanism for the transitions between states in simulated annealing ( sa ) optimization algorithms . in the present framework , that cost function is @xmath34 , which has multiple minima separated by a barrier . as an illustrating example , we consider @xmath34 shown in fig . [ fig : spike ] , @xmath291 which is monotonically decreasing except for the small region around @xmath292 , where there is a narrow spike of the height @xmath293 and the typical width @xmath294 . problems of this type were studied previously in @xcite . the function @xmath295 describes the shape of the spike ( energy barrier ) , @xmath296 . we set the following relations @xmath297 where @xmath298 are constants . the main results will not depend on the specific form of the functions @xmath299 and @xmath295 other than they are continuous and @xmath300 for @xmath301 . we also assume that @xmath302 exponentially quickly for @xmath303 . the global minimum of the cost function @xmath34 corresponds to @xmath304 and there is a local minimum just to the left of the spike at @xmath305 as can be seen in fig . [ fig : spike ] . if we initially prepare the system in a state with @xmath306 , then thermal excitations will cause an over - the - barrier transitions with a rate that scales as @xmath307 . therefore , simulated annealing solves the corresponding optimization problem with high probability in time that scales as @xmath308 . during qa with the hamiltonian ( [ eq : h - sigma ] ) , the transverse field @xmath309 is varied in time . it starts at sufficiently large initial value of @xmath310 when all spins are polarized in the @xmath311 direction . then , @xmath5 is slowly reduced to zero at the end of the algorithm . we consider the case of low temperatures where thermal fluctuations can be neglected ( see below ) . if the evolution is adiabatic , then the system stays at the instantaneous ground state of @xmath8 at all times during qa . this state corresponds to the irreducible subspace with the maximum total spin @xmath312 . at the end of qa the system arrives at the global minimum of @xmath34 corresponding to all spins pointing in the positive @xmath185 direction . the effective potential @xmath313 [ cf . ( [ eq : ueff ] ) ] is time dependent as @xmath5 evolves during qa . the extreme points of the potential are obtained from the equation @xmath314 following the discussion in sec . [ sec : wkb ] we denote as @xmath315 the two instantaneous minima of the effective potential . the maximum of the potential @xmath316 is very close to @xmath317 for all @xmath7 : @xmath318 at the beginning of qa when the term @xmath319 in ( [ eq : uext ] ) is dominating , the system resides in the vicinity of the global minimum at @xmath320 . at the end of qa the system is expected to be at the global minimum @xmath321 . because the barrier is vary narrow , we can find the value @xmath322 at which the minima exchange orders , @xmath323 in the vicinity of this point we have @xmath324 . the value of the effective potential at the minima is @xmath325\simeq -g_0(m_b)$ ] . to calculate the tunneling rate the action under the barrier ( [ eq : a ] ) needs to be calculated at the instanton trajectory with the energy corresponding to the minima of @xmath326 at @xmath322 . the maximum value of the momentum @xmath327 ( [ eq : kr ] ) is reached at the middle " point @xmath292 under the barrier and to the leading order in @xmath293 equals @xmath328 because @xmath327 equals zero at the turning points the instantonic action can be estimated as @xmath329 with @xmath330 . the factor @xmath331 depends on the shape of the barrier function @xmath295 ( [ eq : spike ] ) . for the limiting case of a rectangular barrier @xmath332 . the tunneling transition rate has the form @xmath333 where @xmath100 is a prefactor that scales polynomially with @xmath0 . using ( [ eq : sc ] ) we get @xmath334 for the wkb analysis to be valid the tunneling action ( [ eq : a ] ) must obey @xmath335 . this implies that one of the conditions must be true : either @xmath336 or @xmath337 , @xmath338 . for a problem with small and narrow barrier . the barrier dimensions obey @xmath339 [ cf . ( [ eq : sc ] ] . [ fig : spike],width=288 ] using the eqs . ( [ eq : opt ] ) with @xmath312 , for thermal fluctuations to be neglected in qa the temperature must satisfy the condition @xmath340 where @xmath341 [ eq . ( [ eq : dade ] ) ] is the period of the instanton motion under the barrier with the energy close to the barrier top @xmath342 and @xmath322 . it can be immediately seen from the equations for the instanton velocity ( [ eq : nuh1 ] ) and ( [ eq : sc ] ) that @xmath343 is decreasing and therefore the temperature @xmath344 is increasing with @xmath0 . therefore , the rate of the tunneling transition approaches the zero - temperature limit for any fixed @xmath345 as @xmath0 increases . because the exponents in the quantum tunneling rate @xmath346 and the qmc transition rate @xmath27 are identical , the latter is also well approximated by the zero - temperature limit when @xmath347 . let us assume that the following condition is satisfied @xmath348 in this case , both the quantum tunneling and the qmc transition rates can be analyzed using the methods developed in the former sections . they scale only polynomially with @xmath0 , because the exponent in eq . ( [ eq : wpsc ] ) does not depend on @xmath0 . however , the rate of the purely thermal transition scales exponentially with @xmath0 . this implies that , while quantum tunneling is exponentially faster than classical sa algorithms , it still does not offer scaling advantages over classical algorithms for the situations where the path integral is dominated by a single path ( instanton ) . we note that the above conclusion will hold also for a broader range than that given in ( [ eq : win ] ) : @xmath349 in this case , the qmc transition rate and quantum tunneling rate can only increase compared to the case ( [ eq : win ] ) while the thermal transition rate is still exponential . an example of this situation is @xmath350 and @xmath351 corresponding to the barrier in the interaction energy @xmath352 that has for form of kronecker delta with the height that scales as @xmath0 . however , neither wkb nor kramers escape theory ( for qmc ) do work in this case and the actual transition rate expressions should be analyzed by different methods . in this paper , we considered qmc simulations of _ thermally assisted _ quantum tunneling for an @xmath0-spin mean - field model . we demonstrated , in a closed analytical form , the equivalence between the exponential scaling of the qmc transition rate and the actual physical tunneling rate with the number of co - tunneling spins . this equivalence was established numerically in a previous study @xcite in the effective zero - temperature " limit where the effect of thermal excitations on the tunneling rate can be neglected , and under the condition of zero bias . in this paper , we provided a detailed theoretical description of the results of @xcite and extended them to the case of thermally assisted tunneling at finite temperatures and arbitrary biases . we also provided numerical qmc study to complement theoretical results . the findings of the identical scaling of qmc and quantum tunneling seem counterintuitive at a first glance ; incoherent tunneling decay is a non - equilibrium process while qmc simulations describe fluctuations around equilibria . in mean - field spin models , the density of states increases exponentially with energy , therefore providing for a large number of tunneling channels at finite temperature ( [ eq : f ] ) . both the quantum system and qmc simulation process thermalize in the metastable domain much faster than a kramers transition or a tunneling decay occurs . however , it is not clear ahead of time that qmc explores the decay channels in the same way as the quantum system . the rate of quantum spin tunneling can be written as a product of a polynomial ( in @xmath0 ) prefactor and an exponential function . the exponential function dominates at the large-@xmath0 limit , and it can be determined by the change in the effective free - energy functional @xmath353 $ ] between the values calculated at the instanton and the local minimum [ see eq . ( [ eq : wta ] ) ] . tunneling can thus be described by a most probable path ( instanton ) corresponding to a single channel that minimizes the effective energy . the stochastic process in the qmc simulation samples the quasi - equilibrium distribution determined by the classical mean - field free - energy functional @xmath353 $ ] [ see eq . ( [ eq : fm ] ) ] . the kramers escape event in the stochastic process describes the transition from a local minimum to the global minimum , which is dominated by a single transition state " ( a saddle point of @xmath353 $ ] ) that the system needs to reach in order to make an escape from the metastable state . this transition state corresponds to a quantum instanton , and the change in free energy needed to reach this state is the same as that in the quantum case . this explains the equivalence in the exponential scaling of the qmc transition rate and thermally assisted quantum tunneling rate . we find the solution to the instanton and the change in free energy in a closed analytical form for a general mean - field quantum spin model with the hamiltonian @xmath354 . this is achieved by establishing a detailed connection between the @xmath353 $ ] [ eq . ( [ eq : fm ] ) ] and the analysis performed with wkb method that deals explicitly with the system eigenstates and takes into account the degeneracy of the collective spin states with different total spin . we note that despite a substantial body of work on the models of this type @xcite , the closed analytical form of the instanton based on the free - energy functional @xmath162 $ ] has not been obtained previously . we think that the spin - instanton method developed in this paper can be generalized to the case of fully connected spin - glass models in transverse field with first - order phase transition such as the @xmath355-spin models with @xmath356 . specifically , the mathematical approach developed in sec . [ sec : double ] can be used to calculate the instanton solutions in the replicated free energy within the one - step replica symmetry ansatz @xcite . we analyzed the problems with a narrow and tall barrier in the cost function . assuming that the barrier height @xmath357 and width @xmath358 we established that under the condition @xmath359 the quantum tunneling and qmc rates scale polynomially with @xmath0 while the rate of purely classical thermal activation over the barrier scales exponentially with @xmath0 . an interesting case @xmath360 corresponds to a very narrow barrier . in this case , the scaling exponent @xmath361 for the barrier height @xmath362 can be very close to 1 for qmc and quantum tunneling rates to be polynomial functions of @xmath0 . recently , brady and van dam @xcite found numerical evidence that qmc algorithms will succeed in the same regimes where quantum adiabatic optimization succeeds . more recently , crosson and harrow @xcite considered a bit - symmetric cost function with a thin , high - energy barrier . they proved that the markov chain underlying qmc finds the global minimum in polynomial time ( in @xmath0 ) if the height of the barrier scales less than order @xmath363 . they also conjectured that this is true even for higher barriers that scale as @xmath0 . we note that the result obtained in ref . @xcite refers to a specific form of the interaction energy density @xmath34 [ eq . ( [ eq : h - sigma ] ) ] with a delta - function barrier and represents a particular choice of the broad class of barriers discussed in our paper where qmc and quantum tunneling scales polynomially . in passing , we would like to briefly mention several areas open for further investigation where obstructions for the efficient simulation of quantum tunneling with qmc might exist . one of such obstructions is that qmc is not always ergodic . qmc performs dynamics of paths , which is not a representation of the dynamics of the corresponding system . a well - known problem resulting from this is that qmc might have conserved quantities not present in the physical system , such as the number of world lines ( particles , magnetization ) , braiding , or winding numbers @xcite . qmc may also be less efficient compared to qa in the optimization problems that require multidimensional tunneling to reach the solution . often in these problems the semiclassical action under the barrier @xmath364 is not purely imaginary and displays complex features due to the presence of caustics , non - integrability , and non - analyticity . in this case , _ no _ tunneling path can be defined and a huygens - type wave propagation should be carried out that involves both @xmath365 and @xmath366 @xcite . due to the highly oscillating nature of the wave function @xmath367 in the classically forbidden region , it is not clear if the associated probability @xmath368 can be faithfully recovered with qmc . it is an important open question as to how qmc will perform in comparison with qa in the problems that exhibit many - body location and delocalization ( mbld ) transitions at finite values of transverse fields @xcite . in the problems with disorder and frustration delocalized states can exist in the range of energies with exponential many local minima separated by large hamming distances from each other . a multitude of tunneling paths connects these minima together and positive interference gives rise to extended states in the space of spin configurations above the mobility edge @xcite . it is interesting to explore if the properties of the delocalized phase are important for the qa dynamics towards regions of lower energies where approximate solutions can be obtained in the vicinity of the mbld transition . in contrast , qmc tunneling , being a classical phenomena , only connects a pair of minima at a time without reproducing the positive interference among exponentially many paths . an interesting case where qa can have a scaling advantage over classical algorithms refers to the tunneling in non - stoquastic spin hamiltonians where the negative sign problem prevents a matching qmc algorithm . finally , it was found in ref . @xcite that a version of qmc with open boundary conditions can provide a quadratic speedup compared to incoherent tunneling rate for the hamiltonian ( [ eq : h - sigma ] ) ( i.e. , the scaling of qmc escape rate with @xmath0 matches that of quantum tunneling _ amplitude _ instead of the rate ) . this result can be obtained analytically by a direct extension of the present analysis , which is referred to future studies . the authors would like to acknowledge support from the nasa advanced exploration systems program and nasa ames research center . this work was supported in part by the afrl information directorate under grant f4hbkc4162g001 , the office of the director of national intelligence ( odni ) , and the intelligence advanced research projects activity ( iarpa ) , via iaa 145483 . the views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements , either expressed or implied , of odni , iarpa , afrl , or the u.s . government is authorized to reproduce and distribute reprints for governmental purpose notwithstanding any copyright annotation thereon . 54ifxundefined [ 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 10.1088/0305 - 4470/19/9/033 [ * * , ( ) ] link:\doibase 10.1103/physreve.58.5355 [ * * , ( ) ] link:\doibase 10.1126/science.284.5415.779 [ * * , ( ) ] link:\doibase 10.1126/science.1057726 [ * * , ( ) ] link:\doibase 10.1126/science.1068774 [ * * , ( ) ] http://dx.doi.org/10.1038/nphys2900 [ * * , ( ) ] @noop * * , ( ) @noop http://arxiv.org/abs/1411.4036 [ ] link:\doibase 10.1073/pnas.1002116107 [ * * , ( ) ] link:\doibase 10.1103/physrevx.6.021028 [ * * , ( ) ] http://www.jetpletters.ac.ru/ps/1494/article_22813.shtml [ * * , ( ) ] , @noop * * , ( ) link:\doibase 10.1103/physrevlett.46.388 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.100.197001 [ * * , ( ) ] link:\doibase 10.1038/ncomms2920 [ * * , ( ) ] link:\doibase 10.1038/ncomms12370 [ * * , ( ) ] link:\doibase 10.1103/physrevb.66.094203 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physreve.71.066707 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1126/science.aaa4170 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrevlett.117.180402 [ * * , ( ) ] http://journals.aps.org/prd/abstract/10.1103/physrevd.15.2929 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop _ _ , vol . ( , ) @noop * * , ( ) http://stacks.iop.org/1742-5468/2012/i=06/a=p06007 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) , @noop * * , ( ) http://www.sciencedirect.com/science/article/pii/0003491669901535 [ * * , ( ) ] @noop * * , ( ) , @noop _ _ ( , ) @noop * * , ( ) http://www.sciencedirect.com/science/article/pii/000349166790200x [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) @noop ( ) in link:\doibase 10.1145/1007352.1007428 [ _ _ ] , ( , , ) pp . in link:\doibase 10.1109/focs.2016.81 [ _ _ ] ( ) pp . @noop * * , ( ) link:\doibase 10.1103/physreva.93.032304 [ * * , ( ) ] link:\doibase 10.1080/0001873021000049195 [ * * , ( ) ] , http://dblp.uni-trier.de/db/journals/qic/qic13.html#hastings13a [ * * , ( ) ] http://journals.aps.org/pra/abstract/10.1103/physreva.41.32 [ * * , ( ) ] http://scitation.aip.org/content/aip/journal/jcp/100/1/10.1063/1.466899 [ * * , ( ) ] @noop * * , ( ) @noop * * , ( ) in this appendix , we show that the static free energy per spin @xmath267 that appears in ( [ eq : wta ] ) equals to the free energy density @xmath268 from ( [ eq : wqmc ] ) . for the static solution @xmath369 corresponding to the local minimum of the qmc free energy ( [ eq : static_free_energy ] ) , the propagator eq . ( [ eq : propagator ] ) becomes @xmath370}\;.\end{aligned}\ ] ] consequently , the self - consistent condition ( [ eq : l_self ] ) becomes @xmath371 the condition for extreme points of the static free energy eq . ( [ eq : static_free_energy ] ) is @xmath372 where we use eq . ( [ eq : l_qmc ] ) to simplify the above expression . solving @xmath373 from the above equation , we have @xmath374 thus , the pair of equations ( [ eq : l_qmc ] ) and ( [ eq : qmc_condition_b ] ) determine the local minimum @xmath269 and other extreme points of the static " free energy ( [ eq : static_free_energy ] ) . in the wkb approach , the effective potential @xmath375 ( [ eq : hcl ] ) takes the form @xmath376 the extreme points of the effective potential satisfy @xmath377 which is the same as eq . ( [ eq : qmc_condition_b ] ) . we will use the following equivalent form of eq . ( [ eq : wkb_condition_b ] ) , @xmath378 according to eq . ( [ eq : l0 ] ) the optimal values of @xmath51 satisfy @xmath379 where @xmath53 is the binary entropy . putting eq . ( [ eq : wkb_condition_c ] ) into eq . ( [ eq : wkb_condition_a ] ) , we have @xmath380 which is equivalent to eq . ( [ eq : l_qmc ] ) by using the definition of the entropic factor @xmath53 in eq . ( [ eq : q ] ) . thus , we have shown that the extreme points of the stationary solutions for qmc and wkb are the same . using the conditions ( [ eq : l_qmc ] ) and ( [ eq : qmc_condition_b ] ) , the qmc free energy ( [ eq : static_free_energy ] ) at the extremum takes the form @xmath381 from the definition of the entropy eq . ( [ eq : q ] ) , we have @xmath382 from eqs . ( [ eq : l_qmc ] ) , ( [ eq : extreme_point_static ] ) , and ( [ eq : qmc_condition_b ] ) , we also have @xmath383 putting eq . ( [ eq : another_identity ] ) into eq . ( [ eq : entropy_identity ] ) and then putting the result into eq . ( [ eq : free_erg_static_y ] ) , we have @xmath384 which is exactly the static wkb free energy @xmath385 given that we are using the optimal values of @xmath269 and @xmath51 as described above .

anomaly mediation is a popular and well motivated supersymmetry breaking scenario . different possible detailed realisations of this set - up are studied and actively searched for at colliders . apart from limits coming from flavour , low energy physics and direct collider searches , these models are usually constrained by the requirement of reproducing the observations on dark matter density in the universe . we reanalyse these bounds and in particular we focus on the dark matter bounds both considering the standard cosmological model and alternative cosmological scenarios . these scenarios do not change the observable cosmology but relic dark matter density bounds strongly depend on them . we consider few benchmark points excluded by standard cosmology dark matter bounds and suggest that loosening the dark matter constraints is necessary in order to avoid a too strong ( cosmological ) model dependence in the limits that are obtained for these models . we also discuss briefly the implications for phenomenology and in particular at the large hadron collider . cern - ph - th-2011 - 052 + lycen 2011 - 01 * anomaly mediated susy breaking scenarios in the light of cosmology and in the dark ( matter ) * a. arbey , a. deandrea and a. tarhini + _ universit de lyon , france ; universit lyon 1 , cnrs / in2p3 , umr5822 ipnl , f-69622 villeurbanne cedex , france _ + + +

1103.3244

the search for supersymmetry and its breaking , in addition to the direct searches at lep , b - factories , tevatron and the large hadron collider ( lhc ) , is actively pursued using the wmap limits on the relic density constraints . however the sensitivity of the lightest supersymmetric particle relic density calculation to the variation of the cosmological expansion rate before big - bang nucleosynthesis ( bbn ) , even if modest and with no consequences on the cosmological observations , can modify considerably the relic density , and therefore change the constraints on the supersymmetric parameter space @xcite . in the standard cosmology the dominant component before bbn is radiation , however energy density and entropy content can be modified . in the following we consider the impact of different scenarios of alternative cosmologies . the precision of the wmap data should therefore not make us forget the hypothesis which are implied by the use of standard cosmology . we discuss in the following the implications of precision b - physics , direct searches and cold dark matter relic abundance for the case of anomaly mediated models , from a minimal anomaly mediation supersymmetry breaking @xcite , to mixed moduli - anomaly mediated @xcite and to hypercharge anomaly mediation @xcite . we also discuss similar supersymmetry breaking scenarios in the case of the next - to - minimal supersymmetric standard model . in section [ sec : models ] we discuss the different anomaly mediated supersymmetry breakings in the minimal supersymmetric standard model ( mssm ) in terms of the parameter spaces for these models and discuss the limits which can be applied due to the present data from particle physics experiments and from relic dark matter density assuming the standard model of cosmology . in section [ sec : models2 ] similar scenarios are considered in the next - to - minimal supersymmetric standard model ( nmssm ) and the corresponding particle and cosmological bounds are discussed . in section [ sec : cosmologies ] we discuss how alternative cosmological models can affect dramatically the bounds on the parameter space of the models we have considered while letting unchanged the observable cosmology . four different alternatives to the standard cosmology are discussed which share this behaviour . section [ sec : constr ] and section [ sec : lhc ] discuss respectively the constraints implied by these different cosmological scenarios and the perspectives at the lhc for a list of benchmark points which are representative of the available parameter space for these amsb models . our conclusions are given in section [ sec : conclusion ] . the superconformal anomaly mediated supersymmetry breaking ( amsb ) mechanism @xcite is one of the most well - known and attractive set - ups for supersymmetry breaking . supersymmetry breaking effects in the observable sector have in this framework a gravitational origin . superconformal symmetry is classically preserved in theories without dimensional parameters and it is in general broken by the quantum effects . as anomalies only depend on the low - energy effective theory , the same will be true for the soft terms . usually the amsb scenario can not be applied to the mssm , as it leads to tachyonic sleptons . however the presence of an intermediate threshold can displace the soft terms and avoid this problem . in superconformal gravity one introduces a chiral superfield playing the role of the compensating multiplet for super - weyl transformations , called the weyl or conformal compensator . the f - term vacuum expectation value of the conformal compensator is turned on by the supersymmetry breaking in the hidden sector and the soft breaking of supersymmetry in the visible sector appears through the chiral anomaly supermultiplet . as the soft susy breaking terms arise from the anomaly , the supersymmetry breaking terms do not dominate at tree - level . several soft susy breaking scenarios can be realised starting from this setup . we discuss in the following some of these realisations . the minimal amsb ( mamsb ) scenario @xcite has very attractive properties , since the soft susy breaking terms are calculated in terms of one single parameter , namely the gravitino mass @xmath3 , and the soft terms are renormalisation group invariants which can be calculated for any scale choice . however , the amsb scenarios suffer from the problem that slepton squared masses are found to be negative , leading to tachyonic states . a solution to this problem is to consider that the scalar particles acquire a universal mass @xmath4 at the gut scale , which when added to the amsb soft susy breaking terms , makes them positive . therefore , the mamsb model relies on only four parameters : @xmath5 this scenario has been thoroughly studied in the literature , but is known to have cosmological consequences incompatible with the wmap observations of the dark matter density @xcite . we perform here an updated analysis of the mamsb parameter space constraints from flavour physics and cosmological relic density . for this study , we generate mass spectra and couplings using isajet 7.80 @xcite . the calculation of flavour observables and the computation of the relic density are performed with superiso relic v3.0 @xcite . we use the constraints given in table 9 of the latest version of the superiso manual . the first flavour observable that we consider here is the branching ratio of @xmath6 , which has been been thoroughly studied in the literature and is still under scrutiny . this observable is very interesting , as its sm contributions only appear at loop level , and its theoretical uncertainties as well as the experimental errors are now under control . it provides strong constraints on the supersymmetric parameter space , especially for large @xmath7 , where it receives large enhancements from its supersymmetric contributions . we use the following interval at 95% c.l . @xcite : @xmath8 another interesting observable is the branching fraction of @xmath9 , which is also a loop level observable , and which can receive extremely large contributions from susy at large @xmath10 , and can receive an enhancement of several orders of magnitude compared to the sm branching ratio . this decay mode has not yet been observed , and we have at 95% c.l . @xcite : @xmath11 we also consider a set of tree - level observables which are very sensitive to the charged higgs mass as well as @xmath10 , and we use the following 95% level intervals , which include the theoretical and experimental errors @xcite : @xmath12 the observable @xmath13 is related to the decay of @xmath14 and is detailed in @xcite . + for the relic density constraint , we use the wmap constraints @xcite increased by 10% of theoretical error to account for the uncertainties in the calculation of the relic density : @xmath15 in the following , we disregard the case of negative @xmath16 since it is disfavoured by the muon anomalous magnetic moment constraint , and we scan over the intervals @xmath17 $ ] gev , @xmath18 $ ] tev and @xmath19 $ ] . figure [ fig_amsb ] presents projection plots of the parameter space into the possible different planes . the green region in the plots corresponds to the parameter zone which is not excluded by flavour constraints or mass limits . the red stars corresponds to points leading to a favoured relic density but excluded by other constraints , whereas black stars are favoured by all the presented constraints , including the relic density constraint . as can be seen , no black star is visible in these plots , and the whole parameter space presented here is disfavoured either by flavour or direct constraints , or by the relic density constraint which tends to favour the low @xmath3 region . disregarding the relic density constraint , a large zone at low @xmath3 is excluded . constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive - green areas are excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green area are in agreement with all the previously mentioned constraints . the stars are points favoured by the relic density observable , in red if disfavoured by any other constraints and in black if in agreement with all the constraints simultaneously.,title="fig:",width=604 ] + constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive - green areas are excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green area are in agreement with all the previously mentioned constraints . the stars are points favoured by the relic density observable , in red if disfavoured by any other constraints and in black if in agreement with all the constraints simultaneously.,title="fig:",width=604 ] + constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive - green areas are excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green area are in agreement with all the previously mentioned constraints . the stars are points favoured by the relic density observable , in red if disfavoured by any other constraints and in black if in agreement with all the constraints simultaneously.,title="fig:",width=604 ] in fig . [ fig_amsb_omega ] , we show the relic density values as a function of the amsb parameters . the green zones correspond to regions favoured by the flavour and direct constraints , whereas the other points are either excluded by these constraints or by cosmological considerations ( charged relic or sneutrino relic , which interact therefore strongly ) . two green zones clearly appear on the @xmath4 and @xmath10 plots , for @xmath26 around @xmath27 and @xmath28 . these areas are far from the wmap dark matter allowed interval , making the mamsb scenario disfavoured by the standard cosmology . relic density in function of the amsb parameters . the green points are favoured by all the constraints , the yellow points corresponds to a charged lsp , the blue points correspond to left sneutrino lsp , and the red points are excluded by the other constraints ( flavour and direct limits).,title="fig:",width=604 ] + relic density in function of the amsb parameters . the green points are favoured by all the constraints , the yellow points corresponds to a charged lsp , the blue points correspond to left sneutrino lsp , and the red points are excluded by the other constraints ( flavour and direct limits).,title="fig:",width=604 ] + relic density in function of the amsb parameters . the green points are favoured by all the constraints , the yellow points corresponds to a charged lsp , the blue points correspond to left sneutrino lsp , and the red points are excluded by the other constraints ( flavour and direct limits).,title="fig:",width=604 ] another possibility to solve the negative slepton squared masses of the original amsb scenario has been proposed : the hypercharge anomaly mediated supersymmetry breaking ( hcamsb ) scenario @xcite , in which the mssm resides on a d - brane and the hypercharge gaugino mass is generated in a geometrically separated hidden sector @xcite . in this way , additional contribution to the gaugino mass @xmath29 is generated , and the large value of @xmath29 then increases the weak scale slepton masses beyond tachyonic values , solving the generic amsb problem @xcite . the hcamsb scenario has four parameters : @xmath30 where @xmath31 is the hcamsb contribution to @xmath29 . in order to study the parameter space , we generate mass spectra and couplings using isajet 7.80 @xcite and compute the flavour observables and relic density with superiso relic v3.0 @xcite . we use the constraints described in the previous subsection . in fig . [ fig_hcamsb ] , we scan over the whole parameter space , and project the results in two - dimensional planes . the results are somehow similar to those of the mamsb scenario : the constraints exclude low @xmath3 values . a large part of the parameter space is favoured by flavour and direct constraints , but unfortunately no part of the parameter space respects at the same time the relic density and the other constraints , so that hcamsb is also disfavoured by the standard cosmology . in fig . [ fig_hcamsb_omega ] , we show the relic density values as a function of the hcamsb parameters . again , the results are similar to those of the mamsb scenario , with two distinct zones which are not excluded by flavour and direct constraints , corresponding to a relic density @xmath26 around @xmath27 and @xmath32 . constraints on the hcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the hcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the hcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] relic density in function of the hcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the hcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the hcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] contrary to the two previous amsb scenarios , the mixed modulus anomaly mediated susy breaking ( mmamsb ) scenario @xcite provides viable dark matter candidates , in addition to solving the negative slepton mass problem naturally . this scenario is based on type - iib superstrings with stabilised moduli @xcite . in this scenario , an interesting result is that the soft susy breaking terms receive comparable contributions from both anomaly and modulus , resulting in positive slepton masses . we examine here the minimal mmamsb scenario and @xmath33 , which represent respectively the modular weights of visible sector of the matter fields and the gauge kinetic function , and which can modify the mass spectra . ] , which relies on four parameters : @xmath34 @xmath35 here parametrises the relative contributions of modulus mediation and anomaly mediation to the soft breaking terms : the largest @xmath35 is , the more mediation comes from modulus @xcite . in fig . [ fig_mmamsb ] , the parameter space is scanned over , and it is projected in two - dimensional planes . the resulting plots are different from those of mamsb and hcamsb . indeed , a large part of parameter space escapes the flavour and direct constraints , and zones around @xmath36 or at low @xmath3 also fulfil the relic density constraint . constraints on the mmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the mmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the mmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] in fig . [ fig_mmamsb_omega ] , we present the relic density values in function of the mmamsb parameters . the relic density of points favoured by flavour and direct constraints takes values between @xmath28 and @xmath37 . we can notice however that most of the points are in the interval @xmath38 $ ] , and some points fit in the wmap interval . for this reason , mmamsb is a scenario which can appear as attractive , as it fulfils simultaneously the standard cosmology and particle physics constraints . relic density in function of the mmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the mmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the mmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] an interesting extension of the mssm is the nmssm , which brings a solution to the @xmath39-problem @xcite . it has an extended higgs sector involving additional higgs bosons , modifying the relic density and flavour physics constraints . moreover , the couplings being modified , the nmssm can escape direct constraints , and new parameter zones can be allowed by the constraints used in the previous section . we consider here the simplest version of the nmssm , where the term @xmath40 of the mssm superpotential is replaced by @xmath41 in order for the superpotential to be scale invariant . the soft breaking terms @xmath42 are _ a priori _ independent . using the minimisation conditions for the potential , the scalar mass parameters @xmath43 can be replaced by the vacuum expectation values of the doublet @xmath44 and @xmath45 , with @xmath46 the singlet field mass parameter can also be replaced by the singlet expectation value @xmath47 . expanding the singlet field @xmath48 around @xmath47 gives rise to an effective parameter @xmath49 . one can also define an effective doublet mass such as @xmath50 once the mssm - like parameters ( and in particular @xmath51 ) have been fixed by the specification of the amsb scenario , we are left with four additional independent nmssm - specific parameters : @xmath52 we scan over the intervals @xmath53 $ ] gev , @xmath54 $ ] , @xmath55 $ ] gev and @xmath56 $ ] . we review in the following the differences between the different amsb scenarios when applied to the nmssm . minimal nmssm - amsb ( mnamsb ) parameter points are generated here using nmssmtools 2.3.4 @xcite , and flavour constraints and relic density are computed with superiso relic v3.0 @xcite . in fig . [ fig_namsb ] , the parameter space of mnamsb is presented . as in fig . [ fig_amsb ] , no point satisfy simultaneous flavour , direct and relic density constraints . the direct limits are less constraining than in the mssm , but the flavour constraints are stronger and exclude a larger part of the namsb parameter space in comparison with amsb . constraints on the mnamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the mnamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the mnamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] relic density in function of the mnamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the mnamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the mnamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] fig . [ fig_namsb_omega ] reveals more differences between the mamsb and mnamsb models . first , the zone not excluded by flavour and direct constraints having a relic density around @xmath28 does not exist in the namsb , a new green zone appears for @xmath57 , and its relic density around @xmath58 is much closer to the wmap constraint . however , as in the mamsb scenario , the mnamsb model is globally disfavoured by the standard cosmology . we generate the nmssm - hcamsb ( nhcamsb ) parameter points using nmssmtools 2.3.4 @xcite . the obtained constraints are shown in fig . [ fig_nhcamsb ] . again , we see that no parameter point satisfies at the same time the relic density constraint and the direct and flavour constraints . similarly to the mnamsb scenario , nhcamsb is less constrained by the direct mass limits , but is more excluded by the flavour constraints . constraints on the nhcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the nhcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the nhcamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] relic density in function of the nhcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the nhcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density in function of the nhcamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] fig . [ fig_nhcamsb_omega ] shows the relic density in function of the different parameters . the green region that exists in the hcamsb scenario for a relic density around @xmath32 disappears in the nhcamsb , and a new region opens up around @xmath58 . however , as the mnamsb scenario , the nhcamsb scenario remains also disfavoured by the standard cosmology . the nmssm - mmamsb ( nmmamsb ) scenario leads to similar results as the mmamsb scenario . as can be seen in fig . [ fig_nmmamsb ] , there exists many points satisfying all the constraints , including relic density , especially for values of @xmath35 near to 0 . constraints on the nmmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the nmmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] + constraints on the nmmamsb parameter space . the colour codes are the same as in fig . [ fig_amsb].,title="fig:",width=604 ] fig . [ fig_nmmamsb_omega ] reveals a difference , as the calculated relic density takes values between @xmath27 and @xmath59 , which is more restrictive in comparison to the mssm case . relic density as a function of the nmmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density as a function of the nmmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] + relic density as a function of the nmmamsb parameters . the colour codes are the same as in fig . [ fig_amsb_omega].,title="fig:",width=604 ] we have seen in the previous section that the relic density imposes severe constraints to the parameter spaces , excluding a major part of the amsb scenarios . however , the relic density calculation is generally based on the simplistic standard model of cosmology . in this section , we reinterpret the previous results by considering four different alternatives to the cosmological standard scenario . for this study , we focus our interest on six different benchmark points , which are described in table [ points ] and are representative of the allowed parameter space in the different models . the mass spectra associated to these points are shown in fig . [ fig_spectra ] . .benchmark points for testing alternative cosmology scenarios . all these points have @xmath60 and are in agreement with all the flavour and direct search constraints but would be excluded by wmap constraints based on the standard cosmology . for the point f extra parameters are needed to specify a point in the parameter space , which are chosen to be : @xmath61 , @xmath62 and @xmath63 gev . [ points ] [ cols="^,^,^,^,^,^,^,^ " , ] point a point b + mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302]mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302 ] + point c point d + mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302 ] mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302 ] + point e point f + mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302 ] mass spectra of the six benchmark points . note that the scales are not identical for all spectra.,title="fig:",width=302 ] + in the following , we consider that dark matter is composed of exclusively one particle produced thermally . + the density number of supersymmetric particles is determined by the boltzmann equation @xcite : @xmath64 where @xmath65 is the number density of supersymmetric particles , @xmath66 is the thermally averaged annihilation cross - section , @xmath67 is the hubble parameter , @xmath68 is the relic particle equilibrium number density . the expansion rate @xmath67 is determined by the friedmann equation : @xmath69 where @xmath70 is the total energy density of the universe . the entropy evolution reads : @xmath71 where @xmath72 is the total entropy density . solving and evolving simultaneously ( [ boltzmann ] ) , ( [ friedmann ] ) and ( [ entropy_evolution ] ) enable to compute the relic density in our present universe . in the standard cosmology , the dominant component before big - bang nucleosynthesis is considered to be radiation , which is constituted of all relativistic particles . this assumption is however relaxed in many cosmological scenarios . the last two equations can indeed be written more generally as @xcite @xmath73 @xmath74 parametrises a modified evolution of the total density of the universe , beyond radiation density @xmath75 . @xmath76 parametrises here effective entropy fluctuations due to unknown properties of the early universe . the radiation energy and entropy density evolutions are known and can be written as usual : @xmath77 in the following , we consider two scenarios in which the energy density is modified , and two scenarios with a modified entropy content . the quintessence model is one of the most well - known models for dark energy . it is based on a cosmological scalar field which has presently a negative constant pressure @xmath78 and a positive constant energy density @xmath70 such as @xmath79 @xcite . this behaviour is achieved when the kinetic term of the scalar field equilibrates the potential . however , in the early universe , the scalar field has a dominating kinetic term , leading to a positive pressure such as @xmath80 . during this period , the energy density was varying very quickly , such as @xmath81 . we study here a quintessence scenario in which the quintessence field before big - bang nucleosynthesis was dominating the expansion of the universe . in this case @xcite : @xmath82 where @xmath83 is the proportion of quintessence to radiation at the bbn temperature ( @xmath841 mev ) . we consider that @xmath83 is a free parameter , which can be constrained using the bbn abundance constraints . to compute the abundance of the elements produced during the primordial nucleosynthesis , we use the code alterbbn @xcite integrated into superiso relic . comparing the abundances to the observational constraints , we obtain limits on @xmath83 . the second scenario we consider here is a late decay of an inflaton field . the inflaton field is a scalar field which is considered to be responsible for the rapid inflation of the early universe . generally , the inflaton is considered to decay into standard particles much before the relic decoupling from the primordial soup . however , several models evoke the possibility of a late decay of the inflaton , around the time of bbn . from @xcite , there exist cosmological models in which the late decay of a scalar field reheats the universe to a low reheating temperature , which can be smaller than the freeze - out temperature , without spoiling primordial nucleosynthesis . the decay of this scalar field into radiation increases the entropy and modifies the expansion rate of the universe . we consider here such a scenario in which we neglect the eventual entropy production , and we takes @xcite @xmath85 the exponent is here increased from 6 to 8 in comparison to the quintessence field , as mentioned also in @xcite . such a modification of the expansion rate can be also achieved in a universe with extra - dimensions modifying the friedmann equations @xcite . in this third scenario , we assume that a primordial entropy production due to an unknown component occurs . in general , such an entropy production should be accompanied with energy production , but to better estimate the deviation of the relic density in such a cosmological scenario , we neglect here the energy production , and we consider that the universe has , in addition to the standard radiation entropy density , a dark entropy density evolving like @xcite @xmath86 where @xmath87 is the ratio of effective dark entropy density over radiation entropy density at the time of bbn . the corresponding entropy production is related to @xmath88 by the relation @xmath89 \ ; , \label{sigmad}\ ] ] with @xmath90 and @xmath91 in this last scenario , the reheating temperature @xmath92 is smaller than the neutralino freeze out temperature ( @xmath93 gev ) @xcite , @xmath92 should be considered as a cosmological parameter that can take any value around a few mev and then the neutralinos decoupled from the plasma before the end of the reheating process so their relic density will differ from its standard value . we consider that the inflaton decays around the time of big - bang nucleosynthesis , generating entropy . from the standard late reheating scenarios , we assume that the entropy production evolves like @xcite @xmath94 for @xmath95 mev and that this entropy production stops at the time of bbn . we again neglect the energy production or non - thermal production of particles in order to better understand the effects of a reheating entropy production on the relic density . in term of entropy density , we have @xmath96 the different scenarios do not have impact on the cosmological observations , but they can modify the abundance of the elements . we compute with alterbbn @xcite the abundance of the elements in the different scenarios for each benchmark points , varying the single parameter for a given scenario , and we apply the following conservative constraints @xcite : @xmath97 for the helium abundance @xmath98 and the primordial @xmath99 , @xmath100 , @xmath101 and @xmath102 ratios . in fig . [ fig_bbn ] , we consider the relic density calculated for each of the benchmark points ( a - f , from top to bottom ) and for different cosmological scenarios ( from left to right ) . relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] + relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] + relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] + relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] + relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] + relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162]relic density in function of the cosmological model parameters . the red lines correspond to points excluded by the bbn constraints in eq . ( [ bbnconstraints ] ) , whereas the blue lines correspond to a region with a correct abundance of the elements . the green squares correspond to points allowed by the bbn constraints and giving a relic density in agreement with the wmap dark matter interval.,title="fig:",width=162 ] these plots reveal that the quintessence and inflaton scenarios globally increase the relic density , while the entropy and reheating scenarios decrease it . the comparison with the bbn constraints is also represented , and the red part of the curves is excluded at 95% c.l . , while the blue part gives a correct abundance of the elements . as general features , the quintessence and inflaton scenarios can increase the relic density by three orders of magnitudes without interfering with the bbn constraints , and the entropy and reheating scenarios can decreased to a factor of @xmath103 . therefore , apart from point b which has an extremely low relic density value in the standard cosmological scenario , all the other benchmark points can have a relic density value compatible with the cosmological observations if a non minimal cosmological scenario is considered . we have shown with different well - known cosmological scenarios , that the relic density constraints can be very strongly relaxed . therefore , we propose to compare the relic density calculated in the standard model of cosmology to the following interval @xmath104 to take into consideration the fact that it is possible to increase any relic density calculated in the standard cosmology by three orders of magnitude , and to decrease it by six orders , with non - standard cosmological scenarios in agreement with the current cosmological data . we can re - apply the relic density constraints , and the results are shown in figs . [ fig_amsb_reloaded]-[fig_nmmamsb_reloaded ] . as in the figures of section [ sec : models ] , the green zones correspond to regions in agreement with all flavour and direct constraints , but not necessarily with the relic density constraint . we added in the figures black points , which correspond to regions also in agreement with the new dark matter interval . it is clear that the allowed regions are therefore much larger than with the initial relic density interval , but a surprising result is that even with the very large interval we use here for the relic density , the relic density constraint still excludes large part of the parameter spaces . in particular , in the mamsb and hcamsb scenarios and their nmssm counterparts , the relic density constraints clearly exclude the region @xmath105 tev . the mmamsb scenario however is not constrained anymore when using the new dark matter interval . constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive green area is excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green areas are in agreement with all the previously mentioned constraints . the black area corresponds to parameters in agreement with all constraints , including the revised relic density interval given in eq . ( [ newwmap]).,title="fig:",width=604 ] + constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive green area is excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green areas are in agreement with all the previously mentioned constraints . the black area corresponds to parameters in agreement with all constraints , including the revised relic density interval given in eq . ( [ newwmap]).,title="fig:",width=604 ] + constraints on the minimal amsb parameter space . the exclusion regions are plotted in the order given in the legend . the red zones are excluded by the inclusive branching ratio of @xmath20 , the yellow ones correspond to charged lsp , the olive green area is excluded by direct collider constraints , the light blue zones are excluded by br(@xmath21 ) , the dark blue zones by br(@xmath22 ) , the magenta zones by @xmath23 , the orange zones by br(@xmath24 ) and the grey zones by br(@xmath25 ) . the green areas are in agreement with all the previously mentioned constraints . the black area corresponds to parameters in agreement with all constraints , including the revised relic density interval given in eq . ( [ newwmap]).,title="fig:",width=604 ] constraints on the hcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the hcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the hcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] constraints on the mmamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the mmamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the mmamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] constraints on the mnamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the mnamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the mnamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] constraints on the nhcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the nhcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the nhcamsb parameter space with the revised relic density interval given in eq . ( [ newwmap ] ) . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] constraints on the nmmamsb parameter space including the revised relic density interval . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the nmmamsb parameter space including the revised relic density interval . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] + constraints on the nmmamsb parameter space including the revised relic density interval . the colour codes are the same as in fig . [ fig_amsb_reloaded].,title="fig:",width=604 ] the benchmark points selected for testing alternative cosmology scenarios are in agreement with precision flavour and direct search constraints . it turns out that the phenomenology expected at the lhc is quite peculiar . the mass spectra for the benchmark points we considered show that the lightest neutralino is the lsp and the lightest chargino is in most cases very close in mass to the neutralino ( points a , b , c , f ) . this will give rise to peculiar signatures due to the very limited number of open channels for the decay modes . we analyse in more detail in the following each of the six benchmark points previously selected . production of charginos and neutralinos takes place at the lhc via cascade decays of squark and gluinos and via the direct production channels @xmath106 where the s - channel exchange of an off - shell w or z or photon , and the contribution of susy - qcd diagrams are important . indeed these cross - sections receive important susy - qcd corrections , typical values are given in @xcite or can be obtained using a monte carlo program including the relevant k - factors or the detailed next - to - leading matrix elements . the minimal amsb scenario is disfavoured by the standard cosmology as very far from the wmap dark matter allowed interval . the region allowed in our more general analysis typically favours points in which the lightest chargino and neutralino are very close in mass and not so heavy . for point a , @xmath107 gev and @xmath108 gev so that the mass splitting is only 170 mev . due to this small mass splitting , the open decay modes for the @xmath109 are ( neglecting very small modes below the @xmath110 level ) @xmath111 , where @xmath112 so that @xmath113 and @xmath114 have the same branching through the three body decays while most of the signal with @xmath115 is through the two body production of a charged pion . the next lightest particle is the @xmath116 , decaying mainly to @xmath109 and w or @xmath117 and sm - like higgs as for the light higgs for this point ( 118.15 gev ) the decay is kinematically allowed : @xmath118 where @xmath113 , @xmath39 . since the branching of the mode @xmath119 is 100% and the mode @xmath120 is non - negligible one can study the clean trilepton signal usually suggested at hadron colliders @xcite . the benchmark point b is in the minimal amsb scenario too . the situation is similar to the previous point with in this case the lightest chargino and neutralino lighter than the sm - like higgs boson . for point b , @xmath121 gev and @xmath122 gev so that the mass splitting is only 63 mev . the decay modes for the lightest chargino are similar to the previous case @xmath123 and for the second neutralino @xmath124 the sm - higgs boson ( with a mass of 115.7 gev ) decays with a sizeable branching to pairs of lightest charginos and pairs of lightest neutralinos ( branching @xmath125 and @xmath126 respectively ) while the largest mode is @xmath127 with a branching of @xmath128 . the hcamsb scenario is disfavoured by standard cosmology and excluded by the wmap dark matter allowed interval as for the minimal amsb scenario discussed above . also in this case the lightest chargino and neutralino are very close in mass . the masses are @xmath129 gev and @xmath130 gev with a splitting of only 17 mev . in this situation the lightest chargino decays are very close to the numbers for point a. the decay modes for the lightest chargino are : @xmath131 and for the second neutralino @xmath132 in this case the trilepton mode @xmath133 is especially interesting as the branching fraction @xmath111 is 100% and the one for @xmath134 is 26% . the mixed modulus amsb supersymmetry breaking scenario allows for viable dark matter candidates . benchmark point d has the neutralino lsp with a mass @xmath135 gev . the next - to - lightest supersymmetric particle is the stau with a mass @xmath136 gev while the lightest chargino and the second lightest neutralino are heavier and almost degenerate with a mass of 1095 and 1098 gev respectively . the stau decays to tau and the lightest neutralino @xmath137 while the lightest chargino decays mainly to @xmath138 the second lightest neutralino decays to stau and tau while as the sm - like higgs is light enough to be produced ( 124.3 gev ) , the decay @xmath139 is also kinematically open @xmath140 for the same model as in the previous paragraph we have also an example of a higher mass spectrum in which all supersymmetric partners are quite heavy with the neutralino lsp above 7 tev and all other particles above 10 tev . nmssm with minimal amsb scenario is one of the models considered in the previous sections . it is disfavoured by the relic density constraints if standard cosmology is assumed . point f is in the allowed zone for flavour , low energy and direct search constraints and is allowed if standard cosmology constraints are relaxed as discussed in the previous sections . from the point of view of lhc searches , this point is similar to point a. for point f , @xmath141 gev and @xmath142 gev so that the mass splitting is 99 mev . apart from these two particles , the next supersymmetric particle ( in terms of mass ) is the second neutralino ( see fig . [ fig_spectra ] ) with a mass of 721 gev . we have considered in details the constraints on different possible realisations of superconformal anomaly mediation breaking mechanisms in supersymmetry . these constraints include the usual lep , b - factories , tevatron and lhc searches , but also precision constraints of cosmological origin , namely the wmap limits on the relic density of cold dark matter . we have discussed the standard cosmological approach and also alternative cosmological scenarios which do not change the cosmological observations but which can affect strongly the constraints on the parameter space of these supersymmetric models based on the relic abundance of dark matter . we therefore show how the dark matter constraints can be weakened in order to avoid strong model dependent assumptions in the choice of the cosmological model . based on different benchmark points for amsb models , we performed a detailed analysis of the constraints imposed by particle data and cosmology ( both standard and alternative ) and finally we gave the typical mass spectra and decay modes relevant for the lhc searches . the main lesson that can be learnt in such an exercice is that usual bounds on the parameter space of these models are too restrictive and bear a strong hidden cosmological model dependence in the assumption of the standard cosmological scenario . concerning the lhc searches , points which are excluded in the standard analysis but permitted in this more general approach , may be quite relevant for testing not only the particle theory models themselves but also alternative cosmological scenarios at the lhc , as in many cases relatively low mass supersymmetric particles are allowed with a peculiar spectrum where the lightest neutralino and chargino are very close in mass . 99 a. arbey and f. mahmoudi , `` susy constraints from relic density : high sensitivity to pre - bbn expansion rate , '' phys . b * 669 * ( 2008 ) 46 [ arxiv:0803.0741 ] . a. arbey and f. mahmoudi , `` susy constraints , relic density , and very early universe , '' jhep * 1005 * ( 2010 ) 051 [ arxiv:0906.0368 ] ; a. arbey and f. mahmoudi , `` lhc and ilc data and the early universe properties , '' nuovo cim . c * 33 * ( 2010 ) 151 [ arxiv:1002.4096 ] . l. randall and r. sundrum , `` out of this world supersymmetry breaking , '' nucl . b * 557 * ( 1999 ) 79 [ hep - th/9810155 ] ; g.f . giudice , m.a . luty , h. murayama and r. rattazzi , `` gaugino mass without singlets , '' jhep * 9812 * ( 1998 ) 027 [ hep - ph/9810442 ] ; a. pomarol and r. rattazzi , `` sparticle masses from the superconformal anomaly , '' jhep * 9905 * ( 1999 ) 013 [ hep - ph/9903448 ] ; d.w . jung and j.y . lee , `` anomaly - mediated supersymmetry breaking demystified , '' jhep * 0903 * ( 2009 ) 123 [ arxiv:0902.0464 ] . k. choi , k.s . jeong and k.i . okumura , `` phenomenology of mixed modulus - anomaly mediation in fluxed string compactifications and brane models , '' jhep * 0509 * ( 2005 ) 039 [ hep - ph/0504037 ] . r. dermisek , h. verlinde and l.t . wang , `` hypercharged anomaly mediation , '' phys . * 100 * ( 2008 ) 131804 [ arxiv:0711.3211 ] . h. baer , r. dermisek , s. rajagopalan and h. summy , `` neutralino , axion and axino cold dark matter in minimal , hypercharged and gaugino amsb , '' jcap * 1007 * ( 2010 ) 014 [ arxiv:1004.3297 ] . f. e. paige , s. d. protopopescu , h. baer and x. tata , `` isajet 7.69 : a monte carlo event generator for p p , anti - p p , and e+ e- reactions , '' hep - ph/0312045 . f. mahmoudi , `` superiso : a program for calculating the isospin asymmetry of @xmath143 in the mssm , '' comput . commun . * 178 * ( 2008 ) 745 [ arxiv:0710.2067 ] ; f. mahmoudi , `` superiso v2.3 : a program for calculating flavor physics observables in supersymmetry , '' comput . commun . * 180 * ( 2009 ) 1579 [ arxiv:0808.3144 ] ; ` http://superiso.in2p3.fr ` f. mahmoudi , `` superiso v3.0 , flavor physics observables calculations : extension to nmssm , '' comput . commun . * 180 * ( 2009 ) 1718 , ` http://superiso.in2p3.fr ` a. arbey and f. mahmoudi , `` superiso relic : a program for calculating relic density and flavor physics observables in supersymmetry , '' comput . phys . commun . * 181 * ( 2010 ) 1277 [ arxiv:0906.0369 ] , ` http://superiso.in2p3.fr/relic ` f. mahmoudi , `` new constraints on supersymmetric models from @xmath144 , '' jhep * 0712 * ( 2007 ) 026 [ arxiv:0710.3791 ] . cdf collaboration , `` search for @xmath145 and @xmath146 decays in 3.7 fb@xmath147 of @xmath148 collisions with cdf ii , '' cdf public note * 9892 * ( 2009 ) . a. g. akeroyd and f. mahmoudi , `` constraints on charged higgs bosons from @xmath149 and @xmath150 , '' jhep * 0904 * ( 2009 ) 121 [ arxiv:0902.2393 ] . m. antonelli _ et al . _ [ flavianet working group on kaon decays ] , `` precision tests of the standard model with leptonic and semileptonic kaon decays , '' arxiv:0801.1817 [ hep - ph ] . e. komatsu _ et al . _ [ wmap collaboration ] , `` seven - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological interpretation , '' astrophys . j. suppl . * 192 * ( 2011 ) 18 [ arxiv:1001.4538 ] . h. baer , r. dermisek , s. rajagopalan and h. summy , `` prospects for hypercharged anomaly mediated susy breaking at the lhc , '' jhep * 0910 * ( 2009 ) 078 [ arxiv:0908.4259 ] . h. baer , e.k . park , x. tata and t.t . wang , `` collider and dark matter searches in models with mixed modulus - anomaly mediated susy breaking , '' jhep * 0608 * ( 2006 ) 041 [ hep - ph/0604253 ] . for recent reviews , see : m. maniatis , `` the next - to - minimal supersymmetric extension of the standard model reviewed , '' int . j. mod . phys . a * 25 * ( 2010 ) 3505 [ arxiv:0906.0777 ] ; u. ellwanger , c. hugonie and a.m. teixeira , `` the next - to - minimal supersymmetric standard model , '' phys . * 496 * ( 2010 ) 1 [ arxiv:0910.1785 ] . u. ellwanger and c. hugonie , `` nmspec : a fortran code for the sparticle and higgs masses in the nmssm with gut scale boundary conditions , '' comput . commun . * 177 * ( 2007 ) 399 [ hep - ph/0612134 ] . p. gondolo and g. gelmini , `` cosmic abundances of stable particles : improved analysis , '' nucl . b * 360 * ( 1991 ) 145 ; j. edsjo and p. gondolo , `` neutralino relic density including coannihilations , '' phys . rev . d * 56 * ( 1997 ) 1879 [ hep - ph/9704361 ] . p. salati , `` quintessence and the relic density of neutralinos , '' phys . b * 571 * ( 2003 ) 121 [ astro - ph/0207396 ] . a. arbey , in preparation , ` http://superiso.in2p3.fr/relic/alterbbn ` . g.b . gelmini and p. gondolo , `` neutralino with the right cold dark matter abundance in ( almost ) any supersymmetric model , '' phys . d * 74 * ( 2006 ) 023510 [ hep - ph/0602230 ] . g. gelmini , p. gondolo , a. soldatenko and c.e . yaguna , `` the effect of a late decaying scalar on the neutralino relic density , '' phys . d * 74 * ( 2006 ) 083514 [ hep - ph/0605016 ] . gelmini and p. gondolo , `` neutralino with the right cold dark matter abundance in ( almost ) any supersymmetric model , '' phys . d * 74 * ( 2006 ) 023510 [ hep - ph/0602230 ] . g. gelmini , p. gondolo , a. soldatenko and c.e . yaguna , `` the effect of a late decaying scalar on the neutralino relic density , '' phys . d * 74 * ( 2006 ) 083514 [ hep - ph/0605016 ] . n. okada and o. seto , `` relic density of dark matter in brane world cosmology , '' phys . d * 70 * ( 2004 ) 083531 [ hep - ph/0407092 ] . k. jedamzik , `` big bang nucleosynthesis constraints on hadronically and electromagnetically decaying relic neutral particles , '' phys . rev . d * 74 * ( 2006 ) 103509 [ hep - ph/0604251 ] . w. beenakker , m. klasen , m. kramer , t. plehn , m. spira and p.m. zerwas , `` the production of charginos / neutralinos and sleptons at hadron colliders , '' phys . * 83 * ( 1999 ) 3780 [ erratum - ibid . * 100 * ( 2008 ) 029901 ] [ hep - ph/9906298 ] . h. baer and x. tata , `` probing charginos and neutralinos beyond the reach of lep at the tevatron collider , '' phys . d * 47 * ( 1993 ) 2739 . h. baer , c.h . chen , f. paige and x. tata , `` trileptons from chargino - neutralino production at the cern large hadron collider , '' phys . d * 50 * ( 1994 ) 4508 [ hep - ph/9404212 ] .

in this paper we consider the implications of the distributions of dust and metals in the disc of m31 . we derive mean radial dust distributions using a dust map created from _ herschel _ images of m31 sampling the entire far - infrared peak . modified blackbodies are fit to approximately 4000 pixels with a varying , as well as a fixed , dust emissivity index ( ) . an overall metal distribution is also derived using data collected from the literature . we use a simple analytical model of the evolution of the dust in a galaxy with dust contributed by stellar sources and interstellar grain growth , and fit this model to the radial dust - to - metals distribution across the galaxy . our analysis shows that the dust - to - gas gradient in m31 is steeper than the metallicity gradient , suggesting interstellar dust growth is ( or has been ) important in m31 . we argue that m31 helps build a case for cosmic dust in galaxies being the result of substantial interstellar grain growth , while the net dust production from stars may be limited . we note , however , that the efficiency of dust production in stars , e.g. , in supernovae ejecta and/or stellar atmospheres , and grain destruction in the interstellar medium may be degenerate in our simple model . we can conclude that interstellar grain growth by accretion is likely _ at least _ as important as stellar dust production channels in building the cosmic dust component in m31 . [ firstpage ] ism : clouds dust , extinction ism : evolution galaxies : evolution galaxies : individual : m31 galaxies : ism .

1406.7649

the life - cycle of dust is a complex process . it is expected that interstellar dust grains can grow by accretion in the interstellar medium ( ism ; see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and there is observational evidence to suggest large grains are abundant in many galactic molecular clouds @xcite . micrometer - sized dust grains may form in carbon - rich atmospheres of asymptotic giant branch ( agb ) stars @xcite , and also in oxygen - rich agb stars @xcite , but it is more likely that large interstellar grains have grown to such sizes inside molecular clouds since large grains produced in stars may not remain that large due to sputtering and shattering in the ism . in theory , shock - waves from supernovae ( sne ) should destroy dust grains as these waves propagate through the ism , but the time - scale for such dust destruction is uncertain @xcite . shock destruction of dust grains is likely efficient for carbon dust , but that may not necessarily be the case for silicates @xcite . efficient dust destruction on short time - scales also appears inconsistent with the very high dust masses detected in high-@xmath1 objects @xcite . but at least the carbon dust grains are predicted to survive in the ism for typically not more than a few hundred myr @xcite , which indicate a need for some kind of replenishment mechanism , rebuilding the dust component @xcite . it is now established that core - collapse sne are efficient dust and molecular factories @xcite with large masses of cold dust detected in sn ejecta ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) although there are significant uncertainties associated with conversion from fluxes to dust masses . it is also unclear how much of the dust actually survives and mixes with the ism . theoretical results suggest 90% of the dust produced in sne is destroyed by the reverse shock before it reaches the ism , depending on the interstellar gas density and the grain size distribution @xcite . uncertain destruction rates , the lack of suitable ( young , resolved ) remnants , in combination with possible foreground ( or background ) contamination from unrelated dust clouds along the line of sight , makes it difficult to confirm whether massive - star sne are dominant dust producers in galaxies . thus , even if the seed grains must be produced by stars , interstellar grain growth may still be needed ( e.g. * ? ? ? * ; * ? ? ? * ) , not only as a replenishment mechanism , but also for _ producing _ the bulk of the cosmic dust _ mass_. independent estimates of the efficiency of interstellar grain growth are thus still important . the dust - to - metals ratio in a galaxy may change over time as the galaxy evolves , and this can be followed using simple analytical relations based on closed - box chemical evolution models with ` instantaneous - recycling ' @xcite , or more complex modelling ( e.g. , * ? ? ? regardless of the complexity of the model , these works show that the dust abundance may not necessarily follow the metal abundances in the ism given different sources of dust and destruction . because of the stellar origin of both metals and dust grains , the dust - to - metals gradient along a galactic disc can therefore be regarded as a diagnostic for net dust growth or net destruction of dust in the ism . much of the dust mass may be the result of grain growth in the ism and passage of shocks from sne may lead to destruction by sputtering . if growth is dominating in the ism the dust - to - metals gradient is negative and if destruction is dominating it will be positive @xcite . if only stars produce all the dust ( as well as metals ) and there is no destruction of dust in the ism , the dust - to - metals gradient is essentially flat . the challenge is acquiring data with enough accuracy , resolution and sensitivity to perform this kind of test . in @xcite , this diagnostic was used on a small sample of galaxies from the _ spitzer _ infrared nearby galaxies survey ( sings ) , where dust gradients were found to be typically steeper than the corresponding metallicity gradients , suggesting very little dust destruction and significant non - stellar dust production for most of these galaxies . the dust properties of the sings sample were derived from a set of ultraviolet ( uv ) and infrared ( ir ) radial profiles obtained with _ galex _ and _ spitzer _ combined with optical data ( sdss _ ugriz ) _ in total 17 different photometric bands @xcite . dust masses were obtained by fitting standard spectral energy distribution ( sed ) models according to @xcite to the seds . however , the dust masses were derived from seds which lacked the long wavelength ( beyond 160 @xmath2 m ) observational data necessary to derive accurate dust masses , with possibility of considerable errors beyond the measurement errors , i.e. , a model - dependent uncertainty due to insufficient constraints from the data . moreover , it should be noted that the information regarding the dust distribution in the galaxy discs was limited as it was derived from surface brightness profiles . but a rising trend in dust - to - gas ratio versus metallicity has also recently been found on a global scale @xcite , which lends support to the interpretation of @xcite . in the case of m31 ( ngc 224 ; andromeda ) the situation is much improved . the launch of the european space agency s _ herschel space observatory _ , which observes in the range @xmath3 m @xcite with unprecedented sensitivity and angular resolution at these wavelengths , has produced a census of galaxies as seen through their _ dust mass_. the _ herschel _ exploitation of local galaxy andromeda ( helga ) is a survey covering a @xmath4 area centred on m31 ( further details of the helga survey can be found in * ? ? ? recently , @xcite used the helga observations to investigate the distribution of dust emission in m31 on spatial scales of @xmath5140pc , creating maps of the dust surface density and the dust emissivity index across the disk . @xcite have also used _ herschel _ data ( * ? ? ? * krause et al . , in preparation ) to constrain a detailed dust model of m31 . the unprecedented quality and spatial detail of the helga dust map makes m31 the only large spiral galaxy with a well - constrained detailed dust distribution to date . in this paper we use the helga dust map of m31 in combination with oxygen abundances obtained directly or indirectly from the literature ( section [ data ] ) . m31 is a good test case since the helga data provide extraordinary spatial resolution and detail and the metallicity of the disc can ( due to its proximity ) be constrained by both metallicities of stars and planetary nebulae , as well as h ii regions . using a simple , well tested model of galactic dust evolution ( section [ modelresults ] ) , we evaluate the importance of interstellar grain growth relative to stellar dust production in our nearest neighbour ( section [ results ] ) . here we take a closer look at the helga data , derived by @xcite and evaluate the average radial trends of dust emission across m31 . we combine these results with a derivation of the metallicity gradient based on a compilation of oxygen abundance data from h ii regions , stars and planetary nebulae to investigate the dust - to - metals gradient . _ herschel _ observations of m31 were taken in parallel - mode with the pacs @xcite and spire @xcite instruments observing at 100 , 160 , 250 , 350 and 500@xmath2 m simultaneously . full details of the observing strategy and data reduction can be found in @xcite . the final maps at each wavelength were created with pixel sizes of 2 , 3 , 6 , 8 and 12arcsec with spatial resolution of 12.5 , 13.3 , 18.2 , 24.5 , 36.0arcsec full width at half - maximum for the 100 , 160 , 250 , 350 and 500@xmath2 m maps , respectively . in addition to the _ herschel _ data , the 70@xmath2 m _ spitzer _ mips map published in @xcite was also used as an upper limit to constrain the shorter wavelength end of the sed . @xcite used this data set to create a dust surface density map of m31 by modified - blackbody fits to the fir - sed for each pixel ( using the 70@xmath2 m data as an upper limit to the hot dust component ) . only pixels with @xmath6 were used in the fits . the flux per unit area in each pixel was modelled as @xmath7 where @xmath8 is the planck distribution and the emissivity / absorptivity of the dust grains is a power law @xmath9 . they assumed a value for the coefficient @xmath10 of @xmath11 , which corresponds to a typical interstellar dust composition @xcite . keeping @xmath12 unchanged is , in principle , physically inconsistent . but we chose to do so because it yields a conservative slope of the dust profile is taken from table 5 in @xcite for a model with @xmath13 . this makes sense for our constant @xmath14 model , but for varying @xmath0 , the extrapolation does not quite hold . in m31 we have regions with @xmath15 ( inner 5 kpc ) and regions with @xmath16 ( outer regions ) . if we were to ` correct ' the dust masses here using @xmath12 scaled with @xmath17 ( the most extreme inner @xmath0 ) , @xmath12 would decrease by a factor of 2 and we would get a _ higher _ dust mass . in the outer regions , scaling @xmath12 with @xmath18 ( most extreme outer @xmath0 ) , @xmath12 would increase by a factor of 3.5 and we would get 3.5 times _ lower _ dust masses . thus , the dust - to - metals ratio for the varying @xmath0 case would become even steeper . ] a distance 0.785mpc was also assumed @xcite . in the sed fitting , @xcite initially used a fixed value of the dust emissivity index , @xmath19 ( the slope of the long - wavelength tail of the effective absorptivity / emissivity of the dust component ) across the whole galaxy , but found that with a fixed value it was impossible to adequately fit the seds . @xmath0 was therefore allowed to vary across m31 [ see @xcite , fig . 7 ] with an estimated error in @xmath0 in any pixel of @xmath20 . it is worth noting that there is a degeneracy between @xmath12 and the dust - mass density @xmath21 in the above model . it is quite likely ( if not certain ) that variations in @xmath0 correspond to variations in @xmath12 , which means that the assumed @xmath12 can put a bias on the resultant dust density @xmath21 . the decrease of @xmath0 with increasing galactocentric distance in m31 can therefore mean that we are underestimating the dust mass in its central parts . we will return to this issue later . @xcite also created a dust - to - gas map of m31 . the gas map was obtained by combining the atomic ( hi ) and molecular ( h@xmath22 ) maps ( the sum of which is adopted as the ` gas mass ' ) . the atomic hydrogen was derived from the hi moment - zero map presented in @xcite and the molecular hydrogen was derived from co(@xmath231 - 0 ) observations presented in @xcite made with the iram 30 m telescope ( full details are provided in @xcite ) . note that although the co map requires correcting to h@xmath22 [ the so - called x factor , here taken to be @xmath24 mol @xmath25(k km s@xmath26)@xmath26 ] , the correction may depend on the metallicity of the galaxy ( e.g. * ? ? ? given that the molecular gas in m31 is only 7% of the atomic hydrogen gas , we note that this correction does not affect the conclusions of this work . the sed fitting procedure described above resulted in three parameters for each pixel across m31 : the dust surface density @xmath21 , dust temperature @xmath28 and the dust emissivity index @xmath0 . combining with the gas map , we also have the gas surface density in each pixel @xmath29 . we have binned the data in terms of consecutive radial annuli ( each 2kpc wide , deprojected assuming an inclination of @xmath30 ) and computed the mean value in each bin for all of these parameters . based on the scatter in each radial bin , we have also computed the 1-@xmath31 deviation from the mean values . the resultant binned data are plotted on top of a radial projection of the dust map data in fig . [ binned ] . one can see in the upper panels that the dust is generally below 20k outside of the very centre ( @xmath32 ) , with an odd ` dip ' in the grain temperatures between @xmath33 ( see top panel of fig . [ binned ] ) . the binned @xmath0-values change significantly over the disc from @xmath34 to @xmath35 and as noted in @xcite , increases initially out to @xmath36 , then decreases radially . @xcite have also found evidence of a varying @xmath0 in m31 . it is not unexpected for the dust emissivity index ( @xmath0 ) to vary across a galaxy , indeed this may tell us something about composition of the dust at different galactocentric distances . low values ( @xmath37 ) would indicate that the dust component is dominated by amorphous carbonaceous dust @xcite , while higher values ( @xmath38 ) indicate domination by silicates or graphite @xcite . the highest values ( where @xmath39 ) can be associated with the lowest dust temperatures , which suggests these @xmath0-values may be explained by low - temperature effects in silicates . @xcite have shown that low - temperature effects occur in the laboratory at grain temperatures below @xmath40 k , which is indeed lower than the lowest grain temperatures obtained from the sed fits , but one should bear in mind that the grain temperatures obtained from the fits represent effective temperatures for the whole dust component and not a specific dust species as in the laboratory experiments . moreover , it also suggests that interstellar silicates are not necessarily iron rich , because silicate species such as pyroxenes ( [ mg , fe]sio@xmath41 ) and olivine s ( [ mg , fe]@xmath22sio@xmath42 ) are heated more efficiently due to their higher absorptivity relative to iron - free silicates . however , treating @xmath0 as a free parameter means we are at a potential risk of parameter degeneracy between the dust temperature @xmath28 and the @xmath0-value ( see discussion in section [ dtgbeta ] ) . @xcite demonstrated that while there is a @xmath43 degeneracy from the fitting algorithm this does not create any systematic offsets in the value returned and therefore can not explain the radial trends . to explore the possible effect on the dust - to - gas ratio along the disc , we have reconsidered the sed fitting to the helga dust map using a fixed @xmath14 . this value is a reasonable compromise , which is appropriate for the local ism in the galaxy @xcite and it is also in close agreement with the average @xmath0 value obtained from the varying-@xmath0 model . the resultant dust - to - gas profile is flatter for @xmath14 ( fig . [ binned_fixedb ] ) , though there is a clear , approximately exponential , profile in @xmath44 along the disc of m31 regardless of how we treat @xmath0 ( see figs . [ binned ] and [ binned_fixedb ] , bottom panels ) . the dust temperatures are also generally higher and the temperature gradient along the disc looks more as one would expect , i.e. , @xmath28 is essentially decreasing monotonously with galactocentric distance . this means the odd ` broken ' feature in the radial @xmath28-profile is gone , with @xmath28 simply decreasing with radius along the disc . in order to estimate the overall distribution of metals in m31 we have gathered metallicity data from the literature for h ii regions @xcite , young stars @xcite and planetary nebulae @xcite . we use the oxygen abundance ( o / h ) as a proxy for the overall metallicity . this works well for our purposes , but it is worth remembering that the conversion factor from oxygen abundance to total abundance of metals is different in low- and high - metallicity environments ( cf . o / fe versus fe / h in the galaxy according to , e.g. , * ? ? ? here we adopt a universal conversion factor , which is justified by the fact that m31 has relatively similar o / h ratios ( metallicity ) across its disc ( i.e. , the conversion factor can be treated as a constant ) . thus , to obtain the total metal fraction ( metallicity ) @xmath45 we first convert the number abundances of oxygen into oxygen mass fractions using the relation @xmath46 @xcite , in which we have implicitly assumed @xmath47 . furthermore , for m31 we may assume the oxygen typically makes up about a third of all metals ( which is at the low end of the possible range , see e.g. , * ? ? ? * where 45 - 60% is the suggested value ) and thus @xmath48 . an oxygen fraction as low as 40% ( @xmath49 ) is similar to the new solar value ( see , e.g. , * ? ? ? * ) , and an even lower fraction is expected at super - solar metallicity . one third of oxygen is therefore a reasonable assumption , which also ensures that we do not underestimate the metal content of m31 ( see section [ dtmgrad ] for a discussion on why this is important ) . the adopted oxygen fraction is otherwise not critical in the present study . depending on the method of derivation , the derived abundance from emission spectra can vary significantly . in particular , there is a well - known offset between strong - line calibrations from empirical data and those based on photoionization models and , in general , we have to deal with the fact that oxygen abundances derived from emission spectra have no empirical _ absolute _ scale either . to obtain a homogenous set of oxygen abundances for the h ii regions , we have re - derived o / h using the empirical strong - line calibrations by @xcite and @xcite , which are known to agree well with electron - temperature based abundances . we use the on - calibration by @xcite for all cases where the [ o ii]@xmath50 line is detected with sufficient signal - to - noise ( s / n @xmath51 ) . in the remaining cases we use the ns - calibration by @xcite , provided the [ s ii]@xmath526717 + @xmath526731 lines are measured . we add also a 0.1 dex correction for dust depletion , which seems to appear in hii regions above a certain metallicity @xcite . dust depletion can not explain the dust - to - metals gradient , however . in fig . [ m31_oh ] , we have plotted the resultant o / h ratios together with corresponding ratios derived from stars and planetary nebulae . the empirical abundances for the h ii regions agree nicely with the stellar abundances . the abundances in h ii regions agree with the abundances derived for planetary nebulae in the outer disc . a linear fit to all the metallicity data ( see black line in fig . [ m31_oh ] ) yields , @xmath53 suggesting the metallicity gradient is much flatter than the dust - to - gas gradient . the outermost data points in fig . [ m31_oh ] suggest a flat gradient beyond a certain galactocentric distance . a ` broken gradient ' fit yields almost exactly the same slope as above for @xmath54 kpc and a flat metallicity gradient beyond @xmath55 kpc ( see the blue dashed line in fig . [ m31_oh ] ) . since we are , in this work , only interested in the part of the disc inside a radius of 20 kpc ( see the region inside the white circle in fig . [ dustmap ] and the grey shaded area in fig . [ m31_oh ] ) , we will in the following adopt equation ( [ ohgrad ] ) . overall , the metallicity gradient for m31 can not be very steep regardless of the source for the metallicity data . with @xmath45 derived as above , we are faced with a problem ( which is also seen in * ? ? ? * ) : the highest metals - to - gas ratios appears to be _ lower _ than the corresponding dust - to - gas ratios , i.e , the dust - to - metals ratio ( @xmath56 ) is greater than unity . this is clearly unphysical , and may arise from underestimating the metallicity , or overestimating the dust - to - gas ratio . it is not likely that we have significantly underestimated the metallicity since the errors of the abundance data are moderate and we have assumed a relatively small oxygen fraction in order to maximize the metallicity . we are thus left with an overestimated dust abundance as the only reasonable option . as mentioned in section [ datadust ] , the emissivity law used in the modified blackbody fit to derive the dust surface density was anchored to the emissivity at @xmath57 m with @xmath58 appropriate for milky way - type grains . the dust composition of m31 may be somewhat different and variations in e.g. , the abundance of silicates relative to carbonaceous dust or the presence of grains with ice mantles , can easily account for an uncertainty of almost a factor of two in the dust surface density . we therefore choose to correct the derived dust density by an appropriate factor ( given below , in the next paragraph ) such that the dust - to - metals ratio @xmath56 never exceeds unity . we suggest that the maximum dust - to - metals ratio realistically expected to be reached in the ism is @xmath59 , i.e. , the fraction of metals in the ism locked up in dust grains can not exceed 90% . the degree of dust overabundance @xmath60 is then defined as @xmath61 the corrected dust - to - gas ratios that we will use later for our model fitting are thus @xmath62 . the required correction factor is @xmath63 for a varying @xmath0 and @xmath64 for @xmath14 . this may be interpreted as @xmath65 , which suggests the emissivity at @xmath57 m in m31 should be @xmath66 in the case of a varying @xmath0 and @xmath67 for @xmath14 . the latter value is close to the value according to the @xcite model , i.e. , @xmath68 for @xmath14 ( see eq . 5 in * ? overall , this means the dust - to - metals ratio is strongly dependent on the dust model ( i.e. dust composition ) . a correction of about a factor of 3 is also in agreement with the empirical emissivities obtained by @xcite for three nearby spiral galaxies ( ngc 891 , ngc 4013 and ngc 5907 ) . they found that the emissivity must be roughly three times the value typically adopted for the galaxy , which also means that galactic sub - mm dust emissivity may be underestimated , a conclusion that is supported by our results . but note that these are edge - on spirals , where line of sight effects are largest . the elevated emissivities can certainly be disputed ( see , e.g. , * ? ? ? * ) , but it is also interesting to note that high emissivities have been found in dense , cold molecular cores , which is thought to be the result of more efficient coagulation into complex dust aggregates @xcite . the dust composition may of course vary along the disc , but in the present study we will , for simplicity , regard @xmath60 as a constant and the slope of @xmath69 therefore remains unchanged after the correction . the resultant dust - to - gas ratios @xmath69 and the dust - to - metals ratios @xmath70 across the disc of m31 are presented in figs . [ grad ] & [ zeta ] . the dust - to - gas gradient is clearly steeper than the metallicity gradient , which indicates there is a significant _ dust - to - metals gradient _ in the disc of m31 . one could of course argue that @xmath60 may vary such that the dust - to - metals gradient flattens out and become insignificant , but there are in fact several reasons to assume the opposite , i.e. , that @xmath60 is likely larger in the less evolved outer regions of the disc than it is in the more evolved inner regions . two possible reasons stand out . first , the enrichment of carbon ( and thus carbonaceous dust ) happens on a relatively long time - scale ( see , e.g. , the models by * ? ? ? * ; * ? ? ? the ratio of silicates to carbonaceous dust is therefore probably larger in the outer disc compared to the inner disc . since carbonaceous dust grains have larger emissivity than silicate grains , @xmath60 would ( in this scenario ) _ increase _ with galactocentric distance . secondly , any @xmath43 degeneracy would also lead to an overestimate of the dust mass density in the mid / outer disc more than in the inner disc because of the differences in the sed fitting results ( see fig . 4 in * ? ? ? thus , @xmath60 would again _ increase _ with galactocentric distance rather than decrease , suggesting that variations in @xmath60 with metallicity is likely not responsible for the derived dust - to - gas slope . however , the higher @xmath0-values towards the inner disc may suggest the emissivity - law should be rescaled such that the overall emissivity is higher at small galactocentric distances . it is obviously not clear how @xmath60 may depend on galactocentric distance . assuming that the degree of dust overabundance @xmath60 is constant with @xmath45 is therefore both a reasonable and conservative approach . we have fitted the analytical models of dust - mass evolution derived by @xcite and @xcite , which are given in terms of the so - called confluent hypergeometric kummer - tricomi functions ( * ? ? ? * ; * ? ? ? * see also appendix [ theory ] for further details about the models ) to the dust - to - gas profiles derived by @xcite . we combine this with a metallicity profile derived from abundances as a function of galactocentric distance ( fig . [ m31_oh ] ) . here we briefly explain the parameters and the numerical routines for fitting the data sets in figs . [ grad ] & [ zeta ] with the models listed in table [ parameters ] . the model has four parameters : the effective dust yield @xmath71 , the corresponding total metal yield @xmath72 , the grain - growth efficiency @xmath73 and the dust - destruction efficiency @xmath74 . the metal yield @xmath72 needs to be fixed to the value obtained from the simple closed box model . the closed box model is of course not a correct model of how the metallicity in late - type galaxies evolve . but as discussed in @xcite , it is a model that works in this context since gas flows should not affect the _ dust - to - metals ratio _ very much . using the observed o / h gradient ( with a central value of o / h@xmath75 ) derived in this work ( equation [ ohgrad ] ; fig . [ m31_oh ] ) and the fact that the metallicity at a galactocentric distance @xmath76 is known to be a good proxy for the typical metallicity of a galaxy disc @xcite , we have the following relationship ( which follows from the simple closed - box model of chemical evolution , see * ? ? ? * ) for the effective metal yield , @xmath77 where @xmath2 is the global gas mass fraction of the galaxy . m31 is a mature spiral galaxy with relatively little gas left in the disc . hence , we adopt a low mean gas - mass fraction of @xmath78 , which is based on various estimates found in the literature ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? furthermore , we assume @xmath79 arcmin ( see * ? ? ? * and references therein ) such that @xmath80 . combined with the o / h gradient , this gives @xmath81 and thus @xmath82 . the effective stellar dust yield @xmath71 and @xmath73 are treated as free parameters . there are reasons to believe dust destruction plays a relatively minor role in the formation of a dust - to - metals gradient . this can be motivated as follows . we have seen in section [ data ] that there must be a significant dust - to - metals gradient in m31 , which indicates significant dust growth in the ism . m31 is also a galaxy where the metallicity is relatively similar across the disc and the dust - to - metals ratio may be close to unity in much of the inner parts ( inside the white circle in fig . [ dustmap ] ) , which suggests @xmath83 in equation . ( [ dustz2 ] ) is small . thus , we expect @xmath84 since @xmath85 must be positive and not too small in order for a dust - to - metals gradient to emerge . in case @xmath84 , it is fair to assume a model with a negligible @xmath74 , since the net effect of dust destruction would be small compared to the effect of grain growth anyway . we will therefore consider models in which @xmath86 as well as where @xmath74 is a free parameter . we also test a case where @xmath87 , which corresponds to a dust - destruction time - scale often assumed for the galaxy ( @xmath88 gyr , see * ? ? ? to compare the data and the model , we used the levenberg - markwardt scheme for @xmath89-minimization . more precisely , we used the idl - routine package mpfit @xcite in combination with a numerical implementation ( for idl ) of the kummer - tricomi functions ( see * ? ? ? * ) . the initial parameter setting is in all cases @xmath90 , @xmath91 and , where used , @xmath87 . to avoid unphysical results the parameters are forced by our fitting routine to remain non - negative numbers . in each step the fitting routine has to call a subroutine to compute the kummer - tricomi function @xmath92 for the considered argument and parameters . this may slow down the fitting considerably and even turn into a cumbersome computational effort of its own . if @xmath92 is to be computed on its integral form with high precision for any argument and parameter values , the number of iterations may in some cases be an inhibiting factor . therefore , we take a short - cut , in order to maintain a reasonable computation speed . the function @xmath92 can be defined as an infinite series , which in turn can be evaluated in terms of factorials and the so - called @xmath93-function @xcite . by truncating the series once a certain precision is obtained , we have a fast and sufficiently precise numerical implementation of @xmath92 . table [ parameters ] lists a variety of different model results fit to the derived dust - to - gas and dust - to - metal profiles in m31 ( see figs . [ grad ] & [ zeta ] ) . l|lllllllllll & & @xmath94 & @xmath95 & red . @xmath89 + & ( @xmath0 ) & ( @xmath73 ) & ( @xmath74 ) & & ( @xmath71 ) & ( @xmath96 ) & ( @xmath97 & @xmath74 ) & ( gyr ) & ( gyr ) & + ( a ) & var . & free & 0 & & @xmath98 & @xmath99 & 218 & - & 1.86 & 0.230 & 0.297 + ( b ) & var . & 0 & 0 & & @xmath100 & @xmath101 & - & - & - & - & 8.92 + ( c ) & var . & free & free & & @xmath102 & @xmath103 & 217 & @xmath104 & 1.87 & 0.231 & 0.334 + ( d ) & var . & free & 5.0 & & @xmath105 & @xmath106 & 563 & 5.0 & 0.721 & 0.0891 & 0.745 + ( e ) & 1.8 & free & 0 & & @xmath107 & @xmath108 & 123 & - & 3.30 & 0.408 & 0.278 + ( f ) & 1.8 & 0 & 0 & & @xmath109 & @xmath110 & - & - & - & - & 0.929 + ( g ) & 1.8 & free & free & & @xmath111 & @xmath112 & 148 & @xmath113 & 2.74 & 0.339 & 0.357 + ( h ) & 1.8 & free & 5.0 & & @xmath114 & @xmath115 & 579 & 5.0 & 0.701 & 0.0866 & 0.722 + reasonable model fits can be obtained for both the varying @xmath0 case and @xmath116 . there is hardly any difference between the models with @xmath86 and those which have @xmath74 as an additional free parameter ( see figs . [ grad ] and [ zeta ] and the @xmath89 values given in table [ parameters ] ) . the best fit to the dust - to - gas profiles based on a fixed , as well as a varying , @xmath0 is obtained for very small @xmath74 values , which suggests the net effect of interstellar dust destruction can not be very significant . locking @xmath74 to a certain value ( e.g. , @xmath117 ) results in a poorer , but still acceptable , fit . however , we caution the reader on the uncertainty of the simplistic model we use here - conclusions about the overall efficiency of dust destruction should not be drawn from this result . the parameter values given in table [ parameters ] should be taken with a grain of salt also because the fit is intrinsically uncertain . we have tried a boot - strap monte carlo approach to estimate the ` errors ' of the fitting parameters , but due to various uncertainties in the observational data , the resultant probability density functions ( pdfs ) can not be used as constraints ( despite 10000 iterations ) . the pdfs are patchy and multi - modal functions which give no meaningful statistical variance . but it is clear that one can easily vary each parameter by at least a factor of 2 and still stay within the error bars of the data . regardless of whether we use a fixed or varying @xmath0 there is a clear dust - to - metals gradient along the disc of m31 . this is indicative of significant interstellar dust growth @xcite and we do indeed obtain relatively large @xmath73 values from the fits ( see table [ parameters ] ) . the dust - to - metals ratio along the disc of spiral galaxies is typically not constant , but m31 ( with its flat metallicity distribution ) seems to have a steeper gradient than most of the sings spirals of comparable size @xcite . the same phenomenon is seen also in global dust - to - gas ratios and at lower metallicities in the recent results by @xcite . the favoured effective stellar dust yield @xmath71 is in all cases with @xmath118 quite small . naively , one would interpret that as stellar dust production being rather insignificant , but as mentioned above , the simplistic model we use can not provide very precise quantitative results due to its simplicity . the low @xmath71-values may be due to a possible degeneracy between stellar dust production and interstellar grain destruction due to sne in star - forming regions . since the helga dust map , in combination with the gas distribution , allows us to get a good handle on the kennicutt - schmidt ( k - s ) law in the m31 disc @xcite , we have an opportunity to estimate the grain - growth time - scale @xmath119 using a consistent data set . using the k - s law derived by @xcite we have calculated the growth time - scale @xmath119 from equation ( [ taugr ] ) in the same 2 kpc wide bins along the disc that we have used previously . the resultant @xmath119 as function of galactocentric distance is shown in fig . [ taugrplot ] for the cases of variable and constant emissivity index @xmath0 , respectively , and the maximum and minimum values are also listed in table [ parameters ] . as expected , because the growth time - scale @xmath119 is anti - correlated with the gas - mass density due to its implicit dependence on the star - formation efficiency ( see equations [ tau0 ] and [ taugr ] ) , there is a minimum where the gas distribution has its maximum ( at @xmath120 kpc ) , but we also see a significant rise in the inner disc . these properties are found both for the case with a free emissivity index @xmath0 and for @xmath14 , i.e. , the radial variation of the grain - growth time - scale ( according to the models ) are qualitatively the same . we note also that these grain - growth time - scales are in reasonable agreement with the results of more detailed models for the milky way ( e.g. , * ? ? ? * ; * ? ? ? * who find @xmath121 gyr in the solar circle ) . as discussed in @xcite , a dust - to - metals gradient can to some degree be the result of metallicity - dependent stellar dust production . more precisely , m - type giants ( on the agb ) do not produce their own raw material for dust production , which will lead to metallicity dependence . one could also argue that the inner regions should have more of evolved low - and intermediate - mass stars that have turned into carbon stars and thus alter the silicate - to - carbon - dust ratio , which in turn may affect the dust gradient we derive and cause an apparent metallicity dependence . however , both these issues are important only if a large fraction of the stellar dust is due to agb stars . since there is more and more evidence suggesting that massive stars ( which produce most of the metals ) are efficient dust producers , this is likely not the case . at very low metallicity there may be a threshold also for massive stars , though ( see the hypothesis by * ? ? ? despite it is technically possible that such metallicity dependences could lead to a dust - to - metals gradient , it can be pretty much ruled out by the result proved in appendix a in @xcite : + if the metallicity dependence of the effective stellar dust yield @xmath71 is linear ( @xmath122 ) and the dust - to - metals gradient is steeper than the metallicity gradient , then the slope of the dust - to - metals gradient can not be explained by metallicity - dependent stellar dust production . + this result is , in fact , more general than so and should hold even if @xmath71 is not linear with metallicity , and lends support to the grain - growth scenario in the present case . we note that ( compare the red dashed lines in fig . [ grad ] with the data points in fig . [ zeta ] ) the metallicity gradient of m31 is much flatter than its dust - to - metals gradient if the emissivity index @xmath0 is treated as a variable , while the difference is less if @xmath116 . thus , because of the flatness of the metallicity gradient in m31 , there is one qualitative conclusion that may be drawn from the simple model fits we present : the new helga dust map provides a good case in favour of the grain - growth scenario . despite the relatively firm qualitative results given above , there are a couple of caveats which we have to discuss in some detail . first , we have the problem of how to treat the emissivity index @xmath0 . is it a constant or a variable ? is it covariant with other parameters ? second , sed fitting can be done using various models of the sed . are there one or two grain temperatures that dominate or should we consider a continuous range of grain temperatures ? below we try to address these issues . the dust abundance data used in this paper are mainly taken from @xcite , who choose to treat the emissivity index @xmath0 as a free parameter in their one - component sed fits , which is reasonable since @xmath0 may not be the same in every environment . however , there is a parameter degeneracy between @xmath0 and the dust temperature @xmath28 , because both affects the long - wavelength slope of the model sed ( see * ? ? ? * in particular their fig . 6 ) . as one can easily see in fig . [ binned ] ( middle panel ) the mean @xmath0 values in the inner disc exceeds @xmath123 , which is the slope expected for silicate dust @xcite . it is worth emphasizing what we have already mentioned in section [ dustandgas ] : at very low temperatures @xmath124 is possible @xcite , but , as discussed by @xcite , the @xmath0@xmath28 relation has two branches associated with the inner and outer disc in m31 . this could , in principle , be a result of the aforementioned degeneracy . if such a parameter - degeneracy problem is present , it would likely put an unphysical ( and clearly unwanted ) bias on the resultant dust abundances . therefore , we have explored both fixed and varying @xmath0 in this work . the observed seds are generally better fit with a varying @xmath0 and the emissivity index @xmath0 should indeed vary depending on the ratio of carbonaceous to silicate dust : the time - scale of carbon enrichment is significantly longer than that of silicon , magnesium , oxygen and other elements relevant for silicate formation ( see , e.g. , * ? ? ? * ; * ? ? ? * ) because it is believed that carbon is mainly produced by relatively long - lived stars that become carbon stars on the agb . this suggests @xmath0 is lower in evolved parts of a galaxy than it is in younger parts . the @xmath0-trend with galactocentric distance should thus be increasing since the outer parts are usually less evolved than the inner parts . the fact that populations of cold grains may have large @xmath0 may work against the formation of such a trend , given that the characteristic grain temperature decreases with galactocentric distance , as one may naively assume . adopting a constant @xmath0 over the whole disc may therefore not be completely unjustified and , more importantly , it appears to provide a lower limit to the steepness of the dust - to - gas gradient ( cf . figs . [ binned ] and [ binned_fixedb ] ) . fitting a one - component modified blackbody model to the sed may not be an optimal way of estimating the dust mass . the validity of such a model depends on properties of the dust components . sometimes there can indeed be a single dominant dust component consisting of grains of similar temperature , in which case a one - component modified blackbody model is a very good approximation . if there are two distinct dust populations with clearly different grains temperatures , such a model would be inadequate because a single component can not capture the characteristics of a ` bimodal ' dust - temperature distribution . but in case the sed reflects a dust component with a continuous distribution of grain temperatures , a single component is more representative than a two - component fit , which may overestimate the contribution from the coldest grains . an overestimate of the cold component will also cause an overestimate of the dust mass , since cold grains emit significantly less radiation per unit grain mass . also , the _ herschel _ data provides no information about the sed beyond @xmath125 m , which makes it difficult to constrain the contribution from the coldest dust in a multi - temperature fit . as pointed out by @xcite , to use a model containing dust at more than one temperature in a reliable way , we would require additional data at longer wavelengths , e.g. , observations at @xmath126 m with scuba2 ( but see also * ? ? ? * ) . with the above in mind , one may see the radial dust - temperature profiles in a different light : the radial dust - temperature profile , and the anomalous double - branched @xmath127 relation in particular , does not necessarily reflect changes in the dust composition and heating sources in different parts of the disc only , but also a bias caused by assumptions about the dust temperatures underlying the sed model . the emissivity index @xmath0 is in practice just a ` shape parameter ' for the sed model , which is also the direct reason for the parameter degeneracy mentioned in section [ dtgbeta ] . in case one is trying to fit a single temperature model with a varying @xmath0 to a dust component which in reality has a continuous distribution of grain temperatures , that procedure may force @xmath0 to become smaller as the sed of a multi - temperature dust component is always wider than a single - temperature component . obviously , this effect will also lead to an incorrect estimate of the dust mass . hence , despite the high quality of the _ herschel _ data , the difficulty in finding a simple but general and adequate model for the seds in order to obtain the dust map requires that one uses the radial dust distributions presented here ( and in * ? ? ? * ) with some caution . we note , however , that the slope of the dust - to - gas profile we obtained with a varying @xmath0 , agrees well with the slope of the dust - to - gas profile in the inner disc of m31 derived by @xcite using different data and sed fitting technique . our result for @xmath14 , on the other hand , is in better agreement with the outer slope of the dust - to - gas profile according to @xcite . we believe the two profiles presented here ( top panels of figs . [ binned ] and [ binned_fixedb ] ) comprises the range of variation one may expect due to the uncertainty of the sed model . we have derived and modelled distributions of dust and metals in the disc of m31 with the purpose of finding indirect evidence to hopefully distinguish between one of the two competing dust production mechanisms in late - type galaxies : stellar dust production and interstellar grain growth . the data together with simple models point towards interstellar grains growth being the most important mechanism , although firm conclusions can not be drawn due to possible degeneracies between formation and destruction of dust associated with stars . we have computed mean radial dust distributions from the helga dust map based on simple sed models with a varying as well as a fixed emissivity index @xmath0 as well as an overall metal distribution derived using data collected from the literature . in a second step , we used a simple analytical model of the evolution of the dust component in a galaxy and fit this model to the radial dust - to - gas distribution . the dust - to - gas gradient in m31 is steeper than the metallicity gradient , i.e. , there is a clear negative dust - to - metals gradient along the disc , and in such cases our model suggests dust growth must be the dominant dust - formation mechanism in the ism of m31 . taken at face value , our model fit actually suggests the net stellar dust production is almost negligible and the build - up of the dust component in m31 should therefore be dominated by interstellar growth . from the above we conclude that m31 is a strong case for cosmic dust being the result of substantial interstellar grain growth , while the net stellar dust production must be limited . the efficiency of dust production in stars ( most notably sne ) and the grain destruction in the ism may be degenerate , however . consequently , we can only conclude that interstellar grain growth by accretion is _ at least _ as important as stellar dust production channels in building the cosmic dust component in nearby galaxies . however , our result is in line with the recent ( and more detailed ) dust - evolution models of late - type dwarf galaxies by @xcite , which favours low condensation efficiencies in type ii sne combined with substantial grain growth in the ism . thus , it seems worthwhile to construct a detailed model of the radial dust distribution of m31 , which is more independent of the metallicity gradient . we hope to return to this in a future publication . we thank the anonymous referee and the scientific editor ( mike barlow ) for their constructive comments which improved the final manuscript . _ herschel _ is an esa space observatory with science instruments provided by european - led principal investigator consortia and with important participation from nasa . nordita is funded by the nordic council of ministers , the swedish research council , and the two host universities , the royal institute of technology ( kth ) and stockholm university . the dark cosmology centre is funded by the danish national research foundation . hlg is supported by the science and technology facilities council . i. de looze is a postdoc researcher at the fwo - vlaanderen ( belgium ) . andersen a. c. , loidl r. , hfner s. , 1999 , a&a , 349 , 243 asano r. s. , takeuchi t. t. , hirashita h. & inoue a. k. , 2013 , ep&s , 65 , 213 asplund m. , grevesse n. , sauval a.j . & scott p. , 2009 , ara&a , 47 , 481 baes m. et al . , 2010 , a&a , 518 , l39 bianchi s. & schneider r. , 2007 , mnras 378 , 973 blair w. p. , kirshner r. p. , & chevalier r. a. , 1982 , apj , 254 , 50 bresolin f. , kennicutt r. c. , & garnett d. r. , 1999 , apj , 510 , 104 braun r. , thilker d. a. , walterbos r. a. m. , & corbelli e. , 2009 , apj , 695 , 937 carigi l. , peimbert m. , esteban c. , & garca - rojas j. 2005 , apj , 623 , 213 chapman n. l. , mundy l. g. , lai s .- p . & evans n. j. ii , 2009 , apj , 690 , 496 coupeaud a. , demyk k. , & meny c. et al . , 2011 , a&a , 535 , a124 dasyra k. m. , xilouris e. m. , misiriotis a. & kylafis n. d. , 2005 , a&a , 437 , 447 dennefeld m. , & kunth d. , 1981 , aj , 86 , 989 draine b. , 1990 , in blitzl . , ed . , asp conf . 12 , the evolution of the interstellar medium . , san francisco , p. 193 draine , b. t. 2003 , ara&a , 41 , 241 draine b. t. , 2009 , in henning th . , grn e. , steinacker j. , eds , asp conf . 414 , cosmic dust : near and far . , san francisco , p. 453 draine b. t. & lee h. m. , 1984 , apj , 285 , 89 draine b. t. & li a. 2007 , apj , 657 , 810 draine b. t. et al . , 2014 , apj , 780 , 172 dunne l. , eales s. , edmunds m. , ivison r. , alexander p. , & clements d. l. , 2000 , mnras , 315 , 115 dunne l. et al . 2009 , mnras , 394 , 1307 dunne l. et al . 2011 , mnras , 417 , 1510 dwek e. , 1998 , apj , 501 , 643 dwek e. , galliano f. & jones a.p . , 2007 , apj , 662 , 927 edmunds m.g . , 2001 , mnras , 328 , 223 edvardsson b. , andersen j. , gustafsson b. , lambert d. l. , nissen , p. e. & tomkin , j. , 1993 , a&a , 275 , 101 ford g. et al . , 2013 , apj , 769 , 55 fritz j. et al . , 2012 , a&a , 546 , a34 gall c. , andersen a. c. & hjorth j. , 2011a , a&a , 528 , a13 garnett , d.r . , 2002 , apj , 581 , 1019 gomez h. l. et al . , 2012 , apj , 760 , 96 gordon k. d. et al . 2006 , apj , 638 , l87 griffin m. j. et al . 2010 , a&a , 518 , l3 groves b. et al . 2012 , mnras , 426 , 892 hirashita h. & kuo t .- , 2011 , mnras , 416 , 1340 hfner s. , 2008 , a&a , 491 , l1 indebetouw r. et al . 2014 , apjl , 782 , l2 izotov y. i. , stasiska g. , meynet g. , guseva n. g. & thuan t. x. , 2006 , a&a , 448 , 955 jones , a. p. 2004 , in asp conf . 309 , astrophysics of dust . , san francisco , 347 jones a. p. & nuth j. a. , 2011 , a&a , 530 , a44 jones a. p. , tielens a. g. g. m. & hollenbach d. j. , 1996 , apj , 469 , 740 kamenetzky j. et al . 2013 , apj , 773 , l34 kiss , cs . , brahm p. , laureijs r. j. , mor a. & birkmann s. m. , 2006 , mnras , 373 , 1213 kummer e. e. , 1837 , j. reine angew . , 17 , 228 kwitter k. b. , lehman e. m. m. , balick b. & henry r. b. c. , 2012 , apj , 753 , 12 larson r. b. 1998 , mnras , 301 , 569 lee c .- h . et al . , 2013 apj , 777 , 35 mcconnachie a. w. , irwin m. j. , ferguson a. m. n. , ibata r. a. , lewis g. f. , tanvir n. , 2005 , mnras , 356 , 979 mckee c. f. , 1989 , in allamandola l. j. , tielens a. g. g. m. , eds , proc . iau symp . 135 , interstellar dust . kluwer , dordrecht , p. 431 markwardt c.b . , 2009 , in bohlender d. , durand d. , & dowler p. , eds . , asp conf 411 , data analysis software and systems xviii , astron . , san francisco , p.251 matsuura m. et al . , 2011 , science , 333 , 1258 mattsson l. , 2010 , a&a , 515 , a68 mattsson l. , 2011 , mnras , 414 , 781 mattsson l. & hfner s. , 2011 , a&a , 533 , a42 mattsson l. , andersen a. c. & munkhammar j. d. , 2012 , mnras , 423 , 26 mattsson l. & andersen a. c. , 2012 , mnras , 423 , 38 mattsson l. , de cia a. , andersen a. c. & zafar t. , 2014 , mnras , 440 , 1562 morgan h.l . & edmunds m.g . , 2003 , mnras , 343 , 427 morgan h.l . , dunne l. , eales s.a . , ivison r.j . , edmunds m.g . , 2003 , apj , 597 , l33 muoz - mateos j. c. et al . 2009a , apj , 701 , 1965 muoz - mateos j. c. et al . , 2009b , apj , 703 , 1569 nieten c. , neininger n. , gulin m. , ungerechts h. , lucas r. , berkhuijsen e. m. , beck r. , wielebinski r , 2006 , a&a , 453 , 459 norris b. r. m. , et al . , 2012 , nature , 484 , 220 ormel c. w. , paszun d. , dominik c. & tielens a. g. g. m. , 2009 , a&a , 502 , 845 ossenkopf v. , 1993 , a&a , 280 , 617 pagani l. , steinacker j. , bacmann a. , stutz a. & henning t. , 2010 , science , 329 , 1622 pagel b.e.j . , 1997 , `` nucleosynthesis and chemical evolution of galaxies '' , cambridge univ . press , cambridge paradis d. , bernard j .- ph . & mny c. , 2009 , a&a , 506 , 745 pilbratt g.l . et al . , 2010 , a&a , 518 , l1 pilyugin l. s. , vlchez j. m. & contini t. , 2004 , a&a , 425 , 849 pilyugin l. s. , vlchez j. m. & thuan t. x. , 2010 , apj , 720 , 1738 pilyugin l.s . & mattsson l. , 2011 , mnras , 412 , 1145 planck collaboration , planck early results xix , 2011 , a&a 536 , a19 poglitsch a. et al . 2010 , a&a , 518 , l2 rmy - ruyer a. et al . 2014 , a&a , 563 , a31 ridderstad m. , juvela m. , lehtinen k. , lemke d. , & liljestrm t. , 2006 , a&a , 451 , 961 rho j. , jarrett t. h. , reach w. t. , gomez h. & andersen m. , 2009 , apj , 693 , 39 sandstrom k. m. et al . , 2013 , apj , 777 , 5 serra daz - cano l. & jones a. p. , 2008 , a&a , 492 , 127 slavin j. d. , jones a. p. , & tielens a. g. g. m. , 2004 , apj , 614 , 796 smartt s. j. , crowther p. a. , dufton p. l. , lennon d. j. , kudritzki r. p. , herrero a. , mccarthy j. k. & bresolin f. , 2001 , mnras , 325 , 257 smith m. w. l. et al . , 2012 , apj , 756 , 40 steinacker j. , pagani l. , bacman a. & guieu s. , 2010 , a&a , 511 , a9 stepnik b. et al . , 2003 , a&a , 398 , 551 tamm a. , tempel e. , tenjes p. , tihhonova o. , tuvikene t. , 2012 , a&a , 546 , a4 tricomi f. g. , 1947 , ann . mat . pura appl . quarta , 26 , 141 trundle c. , dufton p. l. , lennon d. j. , smartt s. j. , & urbaneja m. a. , 2002 , a&a , 395 , 519 venn k. a. , mccarthy j. k. , lennon d. j. , przybilla n. , kudritzki r. p. & lemke m. , 2000 , apj , 541 , 610 viaene s. et al . 2014 , a&a , 567 , a71 worthey g. , espaa a. , macarthur l. a. & courteau s. , 2005 , apj , 631 , 820 zhukovska s. , gail h .- & trieloff m. , 2008 , a&a , 479 , 453 zhukovska s. & henning t. , 2013 , a&a , 555 , a99 zhukovska s. , 2014 , a&a , 562 , a76 zurita a. & bresolin f. , 2012 , mnras , 427 , 1463 in the present paper we use the analytical closed - box model described in @xcite , which is based on the equation @xmath128 where @xmath45 and @xmath69 are the metals - to - gas ( metallicity ) and dust - to - gas ratios , respectively , @xmath129 is the dust destruction time scale , @xmath119 is the interstellar dust - growth time scale and @xmath71 , @xmath72 denote the yields ( dust and metals , respectively , as defined in * ? ? ? according to @xcite the dust destruction time - scale can be parameterized as @xmath130 where @xmath131 is the effective gas mass cleared of dust by each sn event , and @xmath132 is the sn rate . the time scale @xmath129 may be approximated as @xcite @xmath133 where @xmath74 will be referred to as the dust destruction parameter , which is a measure of the efficiency of dust destruction . @xmath134 and @xmath135 denotes surface density by mass of stars and gas , respectively . note that there no explicit dependence on the gas mass density @xmath135 or the stellar mass density @xmath134 . for a @xcite imf and @xmath136 @xcite , then @xmath137 ( see * ? ? ? * ) which can be regarded as an upper limit @xcite . the dust - destruction efficiency @xmath74 can also be calibrated to the expected dust - destruction timescale for the galaxy , which one can assume is approximately 0.7gyr @xcite . given that the effective galactic gas - consumption rate is @xmath138 , and the gas density is @xmath139 , which implies @xmath140 @xcite . as discussed in @xcite , one can modify this timescale so that the indirect effects of grain shattering are included . small grains tend to be more easily destroyed @xcite and it is therefore reasonable to assume the dust destruction time scale should depend on the amount of grain shattering as well . the shattering rate is to first order proportional to the square of the dust - grain density in the ism . thus , we may approximate the destruction time scale with the expression @xmath141 where @xmath142 is the present - day galactic dust - to - gas ratio . the timescale of grain growth can thus be expressed as ( see @xcite ] ) : @xmath143 where , to first order , @xmath144 is essentially just a simple function of the metallicity and the growth rate of the stellar component , i.e. , @xmath145 where @xmath73 is a free parameter of the model . adopting the above scenario , with the dust - destruction timescale defined as in eq . ( [ taud2 ] ) , we arrive at the equation @xmath146 \right\},\ ] ] where @xmath72 is the metal yield . with @xmath147 as a basic requirement , solutions for the dust - to - gas ratio @xmath69 in terms of the metallicity @xmath45 can be expressed in terms of the confluent hypergeometric kummer - tricomi functions of the first and second kind ( denoted @xmath148 and @xmath92 ) , respectively @xcite . we refer to @xcite for further details on how such solutions are obtained . the solution to eq . ( [ dustz2 ] ) can be written @xmath149 \over m\left[{1\over 2}{y_{\rm d}\over y_z}\left(1+{1\over z_{\rm d,\,g}}{\delta \over \epsilon}\right ) , { 1\over 2 } ; { 1\over 2}{\epsilon z^2 \over y_z}\right ] } \,z,\ ] ] where @xmath150 is the kummer - tricomi function of the first kind , which is identical to the confluent hypergeometric function @xmath151 . for comparison , we will also consider the case there is neither grain growth , nor destruction of dust in the ism i.e. @xmath152 . we then have the trivial solution , @xmath153 corresponding to dust produced only by stars ( e.g. from sne and/or agb ) , or a scenario where the interstellar grain growth and dust destruction are exactly balanced .

an efficient time - stepping algorithm is proposed based on operator - splitting and the space - time discontinuous galerkin finite element method for problems in the non - classical theory of thermoelasticity . the non - classical theory incorporates three models ; the classical theory based on fourier s law of heat conduction resulting in a hyperbolic - parabolic coupled system , a non - classical theory of a fully hyperbolic extension , and a combination of the two . the general problem is split into two contractive sub - problems , namely the mechanical phase and the thermal phase . each sub - problem is discretised using space - time discontinuous galerkin finite element method resulting each to be stable which then leads to unconditional stability of the global product algorithm . a number of numerical examples are presented to demonstrate the performance and capability of the method . operator - splitting , space - time discontinuous galerkin finite element , non - classical theory of thermoelasticity , fourier s law , second sound , contractive .

1505.00708

in some solids thermal energy can be transmitted by the mechanism of _ wave - like _ propagation of heat , unlike the usual mechanism of conduction by diffusion . this phenomenon of heat conduction as waves , known as _ second sound _ , has been observed experimentally ( see , for example , @xcite for an extensive survey of experimental works involving propagation of heat as a thermal wave ) . the classical theory of heat conduction based on _ fourier s law _ , fails to model the second sound phenomenon . moreover , the classical theory permits _ infinite speed _ of propagation of parts of a localized initial heat pulse , which is paradoxical from a physical point of view . as a result , efforts have been made in an attempt to find a consistent model of heat propagation that is capable of capturing the second sound phenomenon with finite speed ( see , for example , @xcite for a review of models of heat conduction as waves ) . one of the alternative theory for formulating the propagation of heat in a general way that aims at capturing the second sound phenomenon was proposed by green and naghdi @xcite . the theory of green and naghdi is based on three types of constitutive relations for the heat flux : type i is equivalent to the classical theory based on fourier s law . type ii permits the propagation of a localized heat signal as thermal wave without dissipation ( see @xcite for a remark on the appropriateness of this classification ) . type iii is the most general theory , which includes both type i and ii as special cases , in which second sound phenomenon is supported while dissipation is incorporated in the process . the thermomechanical coupling of non - classical heat conduction with classical elasticity is the subject of non - classical thermoelasticity . extensive overviews of the non - classical thermoelasticity of green and naghdi can be found in @xcite . theoretical results concerning the non - classical theory have been addressed in several research works . in @xcite exact solutions are obtained for thermal wave propagation in one dimension . results on the existence and uniqueness of solutions of non - classical problem of thermoelasticity can be found , for example , in @xcite and the references therein . designing a robust and efficient numerical solution strategy for strongly coupled problems of hyperbolic - type is challenging . this is particularly the case for the non - classical theory of thermoelasticity where hyperbolic ( or nearly hyperbolic ) heat conduction equation is coupled with the classical hyperbolic elasticity problem . a standard approach for solving such time - dependent problems is the method of lines ( mol ) in which the governing partial differential equation is first discretised in space using the finite element method ( fem ) leading to a system of ordinary differential equations , which can then be solved using the finite difference method . despite its popularity , mol struggles to accurately solve problems involving propagation of sharp gradients or discontinuities @xcite . recently , a great deal of attention has been invested in designing a spatial discontinuous galerkin ( dg ) approach for convection - dominated problems ; see for example @xcite . however , these methods , like mol , are based on decoupling space and time in the sense that space and time are treated differently . hulbert and hughes @xcite introduced a powerful scheme based on a space - time dg finite element methodology for linear elastodynamics problems . in their approach , space and time are treated simultaneously and the unknown fields are allowed to be discontinuous in time while continuous in space . recently , the space - time dg method has been used in @xcite for classical thermoelasticity , using a monolithic approach where all the unknown fields are solved for simultaneously . recently , in @xcite a numerical solution approach based on mol was proposed for non - classical thermoelasticity in which time integration was done in two ways : continuous galerkin fem for type ii and iii , while mixed - discountinuous galerkin fem for the classical problem based on the fourier s law of heat conduction . in their approach a streamline - upwind numerical stabilization was added to localize numerical oscillations due to the propagation of sharp thermal wave . in the current work , we extend the existence and uniqueness results otained in @xcite for type ii theory to the more general problem of type iii thermoelasticity . we also present a novel numerical algorithm for the non - classical thermoelasticity based on an operator - splitting technique motivated by armero and simo @xcite for classical thermoelasticity , coupled with a space - time dg methodology that extends the work of hulbert and hughes @xcite which was formulated for linear elastodynamics . the major contributions of this work are twofold : ( i ) the adaptation of the operator - splitting strategy for classical thermoelasticity first proposed by @xcite to the non - standard theory , in which the operator defining non - standard thermoelasticity is split in a way that the resulting sub - operators retain the same contractive behaviour as the global operator ; and ( ii ) the time - dg formulation in which continuity of the unknown fields is enforced weakly by using an @xmath0-inner product in contrast to the energy - norm used in @xcite . the rest of this paper is organized as follows . in section [ sec : gnt ] , the governing equations of the non - classical theory are summarized in a general framework of type iii thermoelasticity . well - posedness and physically meaningful boundary and initial conditions are also discussed in this section . an operator - splitting strategy for the problem of type iii thermoelasticity is proposed and the resulting sub - operators are analysed in section [ sec : alg - os ] . in section [ sec : t - dg ] , time - dg formulation is proposed for the sub - problems and stability of the individual algorithms and the global one is analysed in detail . a number of numerical examples both in 1d and 2d are presented in section [ sec : numresults ] to demonstrate the excellent performance and capability of the proposed numerical scheme . finally , concluding remarks and some open problems are discussed in section [ sec : conclusion ] . this section summarizes the equations governing the non - classical theory of thermoelasticty of type iii as proposed by green and naghdi . well - posedness of the problem is analysed . results obtained here will serve as the basis of design and analysis of the numerical algorithm that will be presented in later sections . let @xmath1 , with @xmath2 be the reference placement of a continuum body @xmath3 with smooth boundary @xmath4 . following green and naghdi s theory of thermoelasticity of type iii , the system of partial differential equations governing the thermomechanical interaction in the solid @xmath3 are @xmath5 + \rho\mathbf{b}\\ \dot{\alpha } & = \theta\\ \rho c \dot{\vartheta } & = \mathrm{div}[\mathbf{k}_{2}\nabla \alpha + \mathbf{k}_{3}\nabla \theta ] - \theta_{0}\mathbf{m}:\pmb{\bm{\varepsilon}}(\dot{\bm{u } } ) + \rho r \end{aligned } \right\rbrace \text { in } \omega \times \mathbb{i},\ ] ] where @xmath6 $ ] is the time interval of interest of length @xmath7 . superimposed dots denote time derivatives . the displacement and the velocity vector fields are denoted by @xmath8 and @xmath9 respectively . the scalar field @xmath10 denotes the relative temperature with respect to a uniform reference value @xmath11 such that the absolute temperature @xmath12 is given by @xmath13 . the quantities @xmath14 and @xmath15 are the prescribed body force and heat source . green and naghdi s theory of non - classical thermoelasticity is based on the inclusion of a state variable , known as the _ thermal displacement _ @xmath16 , that is defined in terms of an _ empirical temperature _ @xmath17 ( which is assumed to coincide with the absolute temperature @xmath12 ) through equation @xmath18 , see , for example , @xcite and the references therein . the symbol @xmath19 denotes the small strain tensor associated with a displacement @xmath8 . it is assumed that the elasticity tensor @xmath20 has the following properties : @xmath21 equations and are minor and major symmetries of @xmath20 , while equation is the positive definiteness of @xmath20 . the coupling second - order tensor @xmath22 is of the form @xmath23 where @xmath24 , @xmath25 , and @xmath26 denote , respectively , the thermal expansion coefficient , the bulk modulus and the identity second - order tensor , and @xmath27 and @xmath28 are the lam constants . it is assumed that the tensor @xmath29 is symmetric and positive - definite , and that @xmath30 are symmetric and positive - semidefinite . the scalars @xmath31 and @xmath32 denote the material density and heat capacity . * remarks : * * the non - classical theory of thermoelasticity of type iii is the most general one in that it contains both type i and ii as a special cases . if @xmath33 is omitted form , then one obtains type i ( or classical thermoelastic model ) where a parabolic heat conduction equation is coupled with the hyperbolic mechanical equation . on the other hand , if @xmath30 is set to zero , one obtains the type ii thermoelastcity where , now , part of the system , that is responsible for heat conduction , is hyperbolic ( non - classical heat conduction ) . * under the assumption of mechanical and thermal isotropy the elasticity tensor @xmath20 , and the tensors @xmath29 and @xmath30 become @xmath34 where @xmath35 denotes the forth - order identity tensor , and @xmath36 , and @xmath37 are constants . * the free energy @xmath38 , and hence the stress tensor @xmath39 and the entropy density @xmath40 are given by @xmath41 where @xmath42 is the absolute entropy density . * the heat flux vector @xmath43 within the non - classical theory of thermoelasticity of type iii is defined as @xmath44.\ ] ] using the entropy constitutive relation @xmath18 the coupled system can be written in terms of @xmath40 as @xmath45 + \rho\mathbf{b}\\ \dot{\alpha } & = \theta\\ \rho \theta_0\dot{\eta } & = \mathrm{div}[\mathbf{k}_{2}\nabla \alpha + \mathbf{k}_{3}\nabla \theta ] + \rho r \end{aligned } \right\rbrace \text { in } \omega \times [ 0,~t].\ ] ] it is this form of the dynamical system which is crucial in designing the computational scheme based on operator - splitting in latter sections . let @xmath46 and @xmath47 be two partitions of @xmath4 , each contains mutually disjoint subsets ; that is , @xmath48 let @xmath49 @xmath50 @xmath51 and @xmath52 be prescribed displacement , traction , thermal displacement and flux fields . thus the boundary conditions are given by @xmath53 where @xmath54 denotes the outward unit normal field to @xmath4 . it is easy to observe the analogy between the two set of equations : the mechanical part @xmath55 and the thermal part @xmath56 . in such analogy , we clearly see that the displacement @xmath8 goes with the thermal displacement @xmath16 ( in fact , it is this analogy that motivated the name _ thermal displacement _ @xcite ) , and the velocity @xmath9 goes with the absolute temperature @xmath12 , and hence with @xmath10 . as a consequence , however , one would expect a thermal dirichlet boundary condition is given in terms of @xmath16 as it is customarily the case in the mechanical part that @xmath8 is prescribed as dirchlet boundary condition . the thermal dirichlet boundary condition , in this case , is given via the relative temperature @xmath10 ( and hence absolute temperature @xmath12 ) . the reason for this is that , usually , boundary conditions are prescribed in terms of physical quantities , which can be measured , which , in the thermal case , is the relative temperature , @xmath10 ( or @xmath12 ) . furthermore , the initial conditions read @xmath57 where @xmath58 , @xmath59 , @xmath60 , and @xmath61 are prescribed initial displacement , velocity , thermal displacement and absolute temperature respectively . in prescribing an initial thermal state , a thermal configuration is assumed so that the initial thermal displacement @xmath16 is homogeneous , that is @xmath62 , while , the physically observable quantity , the relative temperature can be initiated at a non - zero value . let @xmath63 , @xmath64 , @xmath65 , and @xmath66 be characteristic scalar quantities with the dimensions of length , time , mass , and temperature , respectively . define the dimensionless variables as 4 |&= , & | & = , & |&= , & |t & = t , + after introducing the dimensionless variables , the non - dimensional form of become @xmath67 + \bar{\rho}\bar{\mathbf{b}},\\ \dot{\bar{\alpha } } & = \bar{\theta},\\ \bar{\rho } \bar{c } \dot{\bar{\theta } } & = \mathrm{div}[\bar{\mathbf{k}}_{2}\nabla \bar{\alpha } + \bar{\mathbf{k}}_{3}\nabla \bar{\theta } ] - \bar{\theta}_{0}\bar{\mathbf{m}}:\pmb{\bm{\varepsilon}}(\dot{\bar{\bm{u } } } ) + \bar{\rho } \bar{r } , \end{aligned } \right.\ ] ] where the spatial and time derivatives are with respect to the dimensionless space and time variables , and 4 | & = , & |&= , & |&= , & |c&=c , + if we drop the bars in the notations of equation , similar expressions as in equation is obtained . for the reminder of this section , whenever the system is mentioned , unless stated otherwise , it refers to its non - dimensional form , and the initial and boundary conditions should also be understood accordingly . the positive - definiteness property of @xmath20 and @xmath29 , and the positive - semidefiniteness of @xmath30 imply that the system together with the initial and boundary conditions and define an evolution equation of a general form @xmath68 where @xmath69 is a closed linear operator with dense domain @xmath70 defined in some suitable banach space @xmath71 . for the case of non - classical linear thermoelasticity , for the sake of simplicity , we consider homogeneous dirichlet boundary condition with respect to both @xmath8 and @xmath16 and the space @xmath71 @xmath72^{d}\times [ l^2(\omega)]^{d}\times \mathbf{h}^{1}(\omega)\times l^2(\omega):\bm{u}=\bm{0},\alpha = 0 \text { on } \gamma\right\rbrace,\ ] ] is a hilbert space . the abstract solution vector @xmath73 , while the linear operator @xmath69 and the source term @xmath74 in are defined by @xmath75\\ \theta\\ \dfrac{1}{\rho c}\mathrm{div}[\mathbf{k}_{2}\nabla \alpha+\mathbf{k}_{3}\nabla \theta ] -\dfrac{\theta_{0}}{\rho c}\mathbf{m}:\bm{\varepsilon}(\bm{v } ) \end{bmatrix},\quad \mathbf{f}{:=}\begin{bmatrix } \bm{0}\\ \bm{b}\\ 0\\ \dfrac{1}{c}r \end{bmatrix}.\ ] ] we consider an inner product , @xmath76 on @xmath71 defined by @xmath77 where @xmath78 denotes the standard @xmath0-inner product pairing of tensor , vector , or scalar fields , that should be understood in context , and @xmath79 and @xmath80 . the norm on @xmath71 induced by the inner product @xmath76 is denoted by @xmath81 . note that the linear differential operator @xmath82 is closed and the space @xmath83^d\times[\mathbf{h}^{1}_{0}(\omega)]^d\times(h^{1}_{0}(\omega)\cap h^{2}_{0}(\omega))\times h^{1}_{0}(\omega)\subset\mathcal{d}(\mathbf{a}),\ ] ] is dense in @xmath71 . hence @xmath84 is dense in @xmath71 . a very important inequality concerning the evolution equation is _ dissipativity property _ of the defining operator @xmath69 . an operator @xmath69 on closed subspace @xmath84 of a hilbert space @xmath71 endowed with an inner product @xmath76 is said to be dissipative if it satisfies the inequality @xmath85 for each @xmath86 @xcite . if the operator @xmath69 is dissipative , the norm of the solution of the corresponding evolution equation is monotonically decreasing in time , which is referred to as _ contractivity _ of the solution . that is , for a solution @xmath87 of the evolution problem , assuming dissipativity of @xmath69 and @xmath88 , we have @xmath89 now , we shall show that the operator @xmath69 that defines the problem is dissipative . let @xmath90 be in the domain of @xmath69 , @xmath84 , satisfying the homogeneous boundary condition . then @xmath91 - \frac{\theta_{0}}{c}\mathbf{m}:\pmb{\bm{\varepsilon}}(\bm{v})\rangle\\ \nonumber = & \;\langle \pmb{\bm{\varepsilon}}(\bm{u}),\mathbb{c}\pmb{\bm{\varepsilon}}(\bm{v})\rangle - \langle \pmb{\bm{\varepsilon}}(\bm{v}),\mathbb{c}\pmb{\bm{\varepsilon}}(\bm{u})\rangle+ \langle\pmb{\bm{\varepsilon}}(\bm{v}),\mathbf{m}\vartheta\rangle+\langle \dfrac{\rho c } { \theta_{0}}\mathbf{k}_2\nabla \alpha , \nabla \theta \rangle\\ \nonumber & -\langle \dfrac{\rho c}{\theta_{0}}\mathbf{k}_2\nabla \theta , \nabla \alpha \rangle -\langle \dfrac{\rho c } { \theta_{0}}\mathbf{k}_3\nabla \theta , \nabla \theta \rangle - \langle \vartheta,\mathbf{m}:\pmb{\bm{\varepsilon}}(\bm{v } ) \rangle \\ = & \;-\langle \dfrac{\rho c}{\theta_{0}}\mathbf{k}_3\nabla \theta , \nabla \theta \rangle\leq 0 . \end{aligned}\ ] ] in the general context , since @xmath30 is positive - semidefinite equation leads to dissipation . in the limiting case where @xmath92 ( type ii ) the above argument implies conservation of energy - norm @xmath93\mathrm{d } \omega.\ ] ] this is the reason why type ii is also referred to as the theory of thermoelasticity _ without energy dissipation _ , see , for example , @xcite . another important relation concerning the operator @xmath69 is that it should satisfy the following : for all @xmath94 , there exists @xmath87 in @xmath84 such that @xmath95 in other words , the operator @xmath96 is onto . to show that @xmath69 satisfies the relation , we proceed as follows : let @xmath97 and @xmath98 then from the definition of @xmath69 equation , implies that @xmath99 & = \bm{v}^*,\\ \alpha -\vartheta & = \alpha^*,\\ \vartheta-\dfrac{1}{\rho c}\mathrm{div}[\mathbf{k}_{2}\nabla \alpha + \mathbf{k}_{3}\nabla\theta]+\dfrac{\theta_{0}}{c}\mathbf{m}:\pmb{\bm{\varepsilon}}(\bm{v})&= \vartheta^{*}. \end{aligned } \right.\ ] ] substitution of equations @xmath100 and @xmath18 into the remaining equations of leads to the ( equilibrium ) problem : find @xmath101 such that @xmath102 , @xmath103 and satisfying @xmath104 & = \accentset{\circ}{\bm{u } } , \\ \rho c\alpha-\mathrm{div}[\mathbf{k}\nabla \alpha]+\rho\theta_{0}\mathbf{m}:\pmb{\bm{\varepsilon}}(\bm{u})&= \accentset{\circ}{\alpha } , \end{aligned } \right.\ ] ] where @xmath105 $ ] , @xmath106+\rho \theta_{0}\mathbf{m}:\pmb{\bm{\varepsilon}}(\bm{u}^*)$ ] , and @xmath107 . the weak form of equation reads : find @xmath108 such that @xmath102 , @xmath103 and satisfying @xmath109 for all @xmath110 . the bilinear form @xmath111 and the right hand side functional @xmath112 are given by @xmath113 note that @xmath114^d$ ] and @xmath115 and here the symbol @xmath116 in equation represents duality pairing in their respective spaces . from the definition of @xmath117 , we can easily see that it is a bounded bilinear form . since @xmath118 then @xmath117 is @xmath119^d\times\mathbf{h}^{1}_{0}(\omega))$]-elliptic . by applying lax - milgram theorem we conclude that there exists @xmath120 which solves the weak problem , and hence solves equation . therefore , this proves the ontoness of the resolvent operator @xmath121 . in conclusion , we have seen that the operator @xmath69 defining the non - classical linear thermoelasticity ( type iii ) * is closed , * has dense domain @xmath84 in @xmath71 , * is dissipative , and * is such that @xmath122 is onto . therefore , by the lumer - phillips theorem , @xmath69 generates a strongly continuous semigroup of contractions in @xmath71 , see , for example @xcite and the references therein . in other words , the problem is well - defined and contractive . this also means that the dynamical system represented by the equation of non - classical thermoelasticity of type iii is , in general , stable in the sense of lyapunov . consider an abstract evolutionary problem of the form . assume that @xmath69 can be expressed additively as @xmath123 such that the operators @xmath124 , @xmath125 define sub - problems @xmath126 let @xmath127 be consistent and stable time - stepping algorithms corresponding to the sub - problems with @xmath128 representing the time step length that the algorithms step up the state vectors , @xmath129 , from a given time @xmath130 to @xmath131 . a time - stepping algorithm , @xmath132 , for the global problem is obtained by taking products of the algorithms as @xmath133 the algorithm @xmath132 is referred to as _ lie - trotter - kato product formula _ it is sometimes called a _ sequential split algorithm _ or _ single pass algorithm_. in addition to the discretisation error in the individual algorithms @xmath134 , the application of the operator - splitting strategy introduces another source of error known as the _ splitting error_. the splitting error associated to the lie - trotter - katto product formula is of order of magnitude @xmath135 ( see , for example , @xcite ) . it means that @xmath132 is only first order accurate . higher - order algorithms based on operator - splitting strategies include : @xmath136 the marchuk strang split is second - order accurate , while the double - pass split is only first - order . note that lie - trotter - kato product formula depends on the order of operations , for example in , @xmath137 is applied first followed by @xmath138 , while the order of operations does not matter in the other two operator - splitting algorithms given in . the non - classical theory of thermoelasticity is a coupling of two dynamical systems : the classical linear elasticity and the non - fourier thermal conduction . a naive splitting of the system into a mechanical problem under constant thermal states ( isothermal ) and a thermal problem with a fixed configuration will result in at most a conditionally stable algorithm even if the sub - algorithms for the two processes are unconditionally stable @xcite . rather , care must be taken in splitting the two systems ; this is usually dictated by an understanding of the underlying physics . in this respect , it makes sense if we split the operator @xmath69 in so that in the mechanical phase the entropy is held fixed ( isentropic ) while the temperature is allowed to vary , and in the thermal phase heat is allowed to be conducted while the configuration is fixed . in fact , in this split , it can be shown that each sub - process defines an evolution problem which is contractive , as is the global problem . furthermore , consistent and stable algorithms for the sub - processes render a consistent and stable algorithm for the global problem by the way of operator - splitting strategy . to this end , inspired by the work of armero and simo @xcite , taking the 4-tuple @xmath139 as the state variables we consider the splitting of the system of equation into @xmath140 + \rho \mathbf{b},\\ \dot{\alpha } & = 0,\\ \rho \theta_{0 } \dot{\eta } & = 0 , \end{aligned } \right . \quad \text { and } \quad \left\ { \begin{aligned } \dot{\bm{u } } & = 0,\\ \rho \dot{\bm{v } } & = 0,\\ \dot{\alpha } & = \theta,\\ \rho \theta_{0 } \dot{\eta } & = \mathrm{div}[\mathbf{k}_2\nabla\alpha + \mathbf{k}_3\nabla\theta ] + \rho r. \end{aligned } \right.\ ] ] this corresponds to the additive splitting of the operator @xmath141 in @xmath142\\ 0\\ 0 \end{bmatrix } , \hspace{2em } \mathbf{a}_2\bm{\sigma}=\begin{bmatrix } \bm{0}\\ \bm{0}\\ \theta\\ \dfrac{1}{\rho c}\mathrm{div}[\mathbf{k}_{2}\nabla \alpha+\mathbf{k}_{3}\nabla \theta ] \end{bmatrix}.\ ] ] using the same calculation as for the dissipativity of @xmath69 in , we obtain the estimates such that for each @xmath143 , @xmath144 in general case , since @xmath145 is positive - semidefinite , both operators @xmath146 and @xmath147 are dissipative . in particular , if @xmath148 , the systems that accounts for thermal conduction , @xmath149 , is energy conserving , i.e. it represents heat conduction without energy loss in a rigid body . moreover , it can be shown that both of the sub - operators satisfy additional conditions in order to generate strongly continuous semigroups of contraction just like how it was done in section [ sec : well - posed ] . now , having two sub - operators @xmath146 and @xmath147 generating contractive semigroups let us assume that there corresponds two algorithms @xmath150 , @xmath125 which are _ b - stable _ ( _ non - linearly stable _ ) ; that is @xmath151 in the linear case this means that each of the algorithms satisfy the estimate ( see @xcite and the references therein ) : @xmath152 where @xmath153 is the dual of @xmath71 . let @xmath154 and @xmath155 be two sequences in @xmath71 generated by the lie - trotter - kato product formula @xmath132 corresponding to two initial conditions @xmath156 and @xmath157 respectively . then the product formula @xmath132 satisfies the stability estimate @xmath158-\mathbb{a}^{\delta t}_2[\mathbb{a}^{\delta t}_1\tilde{\bm{\chi}}^{n}\vert_{\mathcal{v}}\\ \nonumber&\leq \vert\mathbb{a}^{\delta t}_1\bm{\chi}^{n}-\mathbb{a}^{\delta t}_1\tilde{\bm{\chi}}^{n}\vert_{\mathcal{v}}\qquad\qquad\text{($\vert\mathbb{a}^{\delta t}_2\vert_{\mathcal{v}}\leq 1$)}\\ & = \vert\bm{\chi}^{n}-\tilde{\bm{\chi}}^{n}\vert_{\mathcal{v}}\;\,\quad\qquad\qquad\qquad\text{($\vert\mathbb{a}^{\delta t}_1\vert_{\mathcal{v}}\leq 1$)},\end{aligned}\ ] ] which proves the _ non - linear stability _ of the global algorithm corresponding to the product formula @xmath132 . the numerical scheme that is going to be formulated in the subsequent sections is based on the operator - splitting approach and time - discontinuous galerkin finite element method . a lie - trotter - kato product formula is applied to merge algorithms for the two phases . hence , the product formula can also be viewed in the sense of _ predictor corrector _ regime , where the sub - algorithm for the mechanical phase is used as predictor and that of the thermal phase as a corrector . let @xmath159 be a triangulation of @xmath160 , where @xmath160 denotes the closure ( the union of the interior and boundary ) of @xmath161 , such that @xmath162 denote the space of scalar piecewise polynomials on the mesh @xmath163 by @xmath164 : @xmath165 where @xmath166 denotes the set of polynomials of degree at most @xmath167 defined on @xmath168 . consider a partition of the time domain @xmath169 $ ] into the collection @xmath170\right\}_{n=0}^{n-1}$ ] , @xmath171 of non - overlapping subintervals . the time step length is @xmath172 for @xmath173 with @xmath174 for each time sub - domain @xmath175 we consider the space - time domain of the form @xmath176 referred to as the @xmath177 _ space - time slab_. admissible scalar functions , @xmath178 , that we consider in t - dg fem will be polynomials in time @xmath179 with coefficients from the spatial function space @xmath180 ; i.e. @xmath181 where @xmath182 is a monomial in @xmath179 of order @xmath183 . denote by @xmath184 , the space of admissible functions on the space - time domain @xmath185 of degree @xmath186 ( that is , @xmath167 in space and @xmath187 in time ) : @xmath188 in fact , it is easy to observe that the space @xmath184 is generated by the tensor products of the basis elements of the spaces @xmath164 and @xmath189the set of polynomials in time of degree at most @xmath167 . * the space - time mesh for @xmath185 is composed of cells with one element thickness in the time direction ; and in each cell in the slab , time and space are orthogonal to each other . i.e. each cell is of the form @xmath190 . nevertheless , the formulation being developed here can be easily modified in terms of non - orthogonal space - time elements and with slabs composed of more than one element thickness in time direction as well . * the approach would readily accommodate the use of an adaptive mesh refinement procedures . in such cases , there may be cells with time direction thickness less than @xmath191 embedded in slab @xmath185 . in this case , at each hanging node the solution must be constrained so that the hanging nodes will be condensed out later . let @xmath192 and @xmath38 be functions defined on space time domain slab @xmath185 . some frequently used notations are * spatial @xmath0 inner product @xmath193 * space - time @xmath0 inner product @xmath194 * space - time boundary integrals on @xmath195 and @xmath196 @xmath197 * right / left limit of a discontinuous function in time at @xmath198 @xmath199 * temporal jump of a discontinuous function at @xmath198 @xmath200\!]}}_{n } = \varphi(t_{n}^{+})-\varphi(t_{n}^{-}).\ ] ] in the mechanical phase the entropy is held fixed , so that the last equation of the left hand system in ; that is the equation @xmath201 = 0\hspace{1em}\text { on } \quad\omega\times [ t_n , t_{n+1})\ ] ] is solved in closed form to obtain an _ intermediate temperature _ @xmath202 . this leads to an explicit formula for @xmath202 in terms of the state variables at time step @xmath198 from the left and at time value @xmath203 , that is , @xmath204 we substitute this result into the mechanical problem @xmath100 to obtain @xmath205 + \bm{f } , \end{aligned } \right.\ ] ] where @xmath206 , in the terminology used in @xcite , is referred to as _ adiabatic elasticity tensor _ and @xmath207 $ ] and @xmath208 and @xmath209 denote the temperature at the end of the previous space - time slab , @xmath210 . note that the adiabatic elasticity tensor @xmath211 remains positive - definite and symmetric as @xmath20 . on the current space - time slab , @xmath185 , we use same boundary conditions as given in for the mechanical fields but the initial conditions , in this case , are the solution for @xmath8 and @xmath9 at the end of the previous slab ; that is @xmath212 and @xmath213 . to define the t - dg fem formulation of the mechanical problem , we first define the trial and weight function spaces for displacement @xmath8 and velocity @xmath9 vector fields as @xmath214^{d}:\bm{u}^{h}=\bar{\bm{u } } \text { on } \gamma_{\bm{u}}\times i_n\right\},\\ \mathcal{t}^{^v}_{_h } & = \left\{\bm{v}^{h}\in[\mathcal{s}_{_{h}}(q_n;j , l)]^{d}:\bm{v}^{h}=\dot{\bar{\bm{u } } } \text { on } \gamma_{\bm{u}}\times i_n\right\},\\ \mathcal{w}^{^u}_{_h } & = \left\{\bm{w}^{h}\in[\mathcal{s}_{_{h}}(q_n;j , l)]^{d}:\bm{w}^{h}=\bm{0 } \text { on } \gamma_{\bm{u}}\times i_n\right\},\\ \mathcal{w}^{^v}_{_h } & = \left\{\bm{\varphi}^{h}\in[\mathcal{s}_{_{h}}(q_n;j , l)]^{d}:\bm{\varphi}^{h}=\bm{0 } \text { on } \gamma_{\bm{u}}\times i_n\right\ } , \end{aligned}\ ] ] where @xmath215 and @xmath216 , @xmath217 and @xmath218 are trial and weight function spaces for displacement and velocity vector fields respectively . the t - dg fem is formulated as : find @xmath219 such that for all @xmath220 @xmath221 where @xmath222 * the main difference between the dg formulation presented here in and that of @xcite is the inner product used to enforce the equation of motion weakly . in our formulation we use the @xmath0-inner product to weakly enforce the mechanical problem while in @xcite an energy - inner product is used . * the formulation is consistent in the sense of a time - stepping algorithm . this can be seen from the euler - lagrange form of given by @xmath223-\bm{f},\bm{\varphi}^h ) \hspace{2em}\text { ( equation of motion)}\\ \nonumber & \quad+\langle{\ensuremath{[\![\bm{u}^{h}]\!]}}_{n},\bm{w}^{h}(t_n^+)\rangle\hspace{16em}\text{(displacement continuity)}\\ & \quad+ \langle{\ensuremath{[\![\rho\bm{v}^{h}]\!]}}_{n},\bm{\varphi}^{h}(t_n^+)\rangle , \hspace{14em}\,\;\;\;\;\text{(velocity continuity)}\end{aligned}\ ] ] that upon substitution of a sufficiently smooth solution pair @xmath224 of the strong form into , the weak forms of the jumps and the equation of motion vanish . * the jump terms are used to improve the stability of the scheme without degrading the accuracy . as a result , the formalism can be readily extended to the non - linear case without eliminating the jump term from the displacement - velocity relation . * one of the consequences of using the @xmath0-inner product is that a dirchlet - type boundary condition may not be necessary to define the velocity trial and weight function spaces . instead we can use @xmath225^{d}.\ ] ] for the sake of simplicity , we assume a homogeneous source term @xmath226 and boundary conditions , i.e. @xmath227 and @xmath228 . we claim that the formulation renders an unconditionally stable time - stepping algorithm . that is to say : @xmath229 where @xmath230 is the total mechanical energy of @xmath231 at time @xmath130 , given by @xmath232~\mathrm{d}\omega.\ ] ] for the analysis we use elliptic and @xmath0 interpolation operators @xmath233^d\rightarrow \mathcal{w}^{^u}_{_h}(t)$ ] and @xmath234^d\rightarrow\mathcal{w}^{^u}_{_h}(t)$ ] respectively defined as : for @xmath235^d$ ] and @xmath236^d$ ] @xmath237,\bm{\varepsilon}(\bm{\varphi}^h)\rangle & = \langle\mathbf{c}_{_{ad}}\bm{\varepsilon}(\bm{u } ) , \bm{\varepsilon}(\bm{\varphi}^h)\rangle,\qquad\forall \bm{\varphi}^h\in\mathcal{w}^{^u}_{_h}(t),\\ \langle\hat{\bm{\pi}}\bm{w},\bm{\psi}^h\rangle&= \langle\bm{w},\bm{\psi}^h\rangle , \qquad \hspace{4em}\forall\bm{\psi}^h\in\mathcal{w}^{^u}_{_h}(t ) , \end{aligned}\ ] ] where @xmath238 referees to the space of functions in @xmath217 at a fixed but arbitrary @xmath179 . we also use the fact that @xcite @xmath239 = \hat{\bm{\pi}}\mathrm{div}[\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u})].\ ] ] now , given the solution @xmath240 of at the end of the previous space - time slab , @xmath210 , and let @xmath241 be the solution of in the current space - time slab , @xmath242 . replace @xmath243,\bm{0})^t$ ] in to obtain @xmath244)_{q_n}-(\bm{v}^{h},\hat{\bm{\pi}}\mathrm{div}[\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u}^h)])_{q_n}\\ & \quad + \langle{\ensuremath{[\![\bm{u}^{h}]\!]}}_n,\hat{\bm{\pi}}\mathrm{div}[\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u}^h(t_n^+))]\rangle , \end{aligned}\ ] ] applying in and the definition of the projection operator @xmath245 and using integration by parts ( note the homogeneous boundary conditions ) leads to @xmath246\!]}}_n,\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u}^h(t_n^+))\rangle . \end{aligned}\ ] ] again substituting @xmath247 into yields @xmath248\!]}}_n,\bm{v}^h(t_{n}^+)\rangle.\ ] ] adding the equations and we obtain @xmath249\!]}}_n,\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u}^h(t_n^+))\rangle+\langle{\ensuremath{[\![\rho\bm{v}^h]\!]}}_n,\bm{v}^h(t_{n}^+)\rangle=0.\ ] ] taking the time derivative out of the space integral , becomes @xmath250\!]}}_n,\mathbb{c}_{_{ad}}\bm{\varepsilon}(\bm{u}^h(t_n^+))\rangle+\langle{\ensuremath{[\![\rho\bm{v}^h]\!]}}_n,\bm{v}^h(t_{n}^+)\rangle=0 . \end{aligned}\ ] ] after some algebraic manipulation we obtain @xmath251\!]}}_n)=\mathscr{e}_{_m}(\bm{u}^{h}(t_{n}^{-})).\ ] ] since @xmath252\!]}}_n)$ ] is non - negative the energy equation leads to the estimate that renders the scheme for the mechanical phase unconditionally stable . in fact , the total numerical dissipation added is precisely equal to @xmath253\!]}}_n).\ ] ] the use of projection operators in and reveals an important point , that is , the current t - dg formulation can be converted into the one given in @xcite . the solution @xmath254 of the mechanical phase is known at the end of the current space - time slab . the objective , in the present phase , is to solve for the thermal states @xmath255 , @xmath179 . hence , the global solution at the end of the current slab will be @xmath256 . the operator - splitting is performed based on the state vector @xmath139 . as a result , we enforce the problem in the thermal phase using the conservation form @xmath257 + \rho r. \end{aligned}\ ] ] here , recall that the entropy density @xmath40 is obtained from the relation @xmath258 . in the thermal phase , the displacement @xmath8 and the velocity @xmath9 are fixed at the corresponding values at the end of the current slab in the mechanical phase ; that is , @xmath259 where the subscripts @xmath260 and @xmath261 represents the values of the fields in the thermal and mechanical phases , respectively . thus , the time derivative of the terms @xmath262 vanishes , and consequently the left hand side of equation @xmath263 becomes @xmath264 in addition , the jump in the entropy , in this case , reads as @xmath265\!]}}_{n } & = \rho\theta_0\eta(t_n^+)-\rho\theta_0\eta(t_n^-)\\ & = \big[\rho c \vartheta(t_{n}^+)+\rho\theta_0\mathbf{m}:\bm{\varepsilon}(\bm{u}_{m}(t_{n+1}^{-}))+s_0\big]-\big[\rho c \vartheta(t_{n}^{-})+\rho\theta_0\mathbf{m}:\bm{\varepsilon}(\bm{u}(t_{n}^{-}))+s_0\big]\\ & = { \ensuremath{[\![\rho c \vartheta]\!]}}_{n}+\rho\theta_0\mathbf{m}:[\bm{\varepsilon}(\bm{u}_{m}(t_{n+1}^-))-\bm{\varepsilon}(\bm{u}(t_n^- ) ) ] . \end{aligned}\ ] ] it should not cause any confusion if we drop the superscript @xmath261 in the equation so that the jump term can be written as @xmath266\!]}}_{n } = { \ensuremath{[\![\rho c \vartheta]\!]}}_{n}+\rho\theta_0\mathbf{m}:[\bm{\varepsilon}(\bm{u}(t_{n+1}^-))-\bm{\varepsilon}(\bm{u}(t_n^-))].\ ] ] to define the t - dg formulation for the thermal phase we first define the thermal displacement and temperature trial and weight function spaces @xmath267 , @xmath268 and @xmath269 , @xmath270 respectively based on @xmath184 and the boundary condition requirements . formally , the t - dg fem formulation of the thermal phase on the domain @xmath185 is defined as : find @xmath272 such that for each @xmath273 @xmath274 where @xmath275 , \nabla \sigma^h)_{q_n},\\ & \quad + \langle\alpha^h(t_n^+),\beta^h(t_n^+)\rangle + \langle\rho c \vartheta^h(t_n^+ ) , \sigma^{h}(t_n^+)\rangle\\ b_{_n}^{^t}(\bm{\lambda}^{h})&=\langle\alpha^h(t_n^-),\beta^h(t_n^+)\rangle + \langle\rho c \vartheta^h(t_n^- ) , \sigma^{h}(t_n^+)\rangle+\langle\rho\theta_0[\bm{\varepsilon}(\bm{u}(t_{n}^-))-\bm{\varepsilon}(\bm{u}(t_{n+1}^- ) ) ] , \sigma^h(t_{n}^{- } ) \rangle\\ & + ( \bar{h},\sigma^h)_{f_n}+(\rho r , \sigma^h)_{q_n}. \end{aligned}\ ] ] the relation between the dg formulation and the point - wise form is apparent from the euler - lagrange form @xmath276+\rho r,\sigma^h)_{q_n}\quad\,\hspace*{5.7em } \text{(equation of motion)}\\ & \quad + \langle{\ensuremath{[\![\alpha^h]\!]}}_{n},\beta^h(t_n^+)\rangle\;\,\hspace*{18em}\text{($\alpha$-continuity)}\\ & \quad + \langle{\ensuremath{[\![\rho c \vartheta^h]\!]}}_n,\sigma^h(t_n^+)\rangle,\hspace*{17em}\;\text{($\vartheta$-continuity ) } \end{aligned}\ ] ] which reveals that a sufficiently smooth solution of the strong problem also satisfies , and vice versa , while the jump terms are vanished at the smooth solution . this also proves the consistency of the t - dg scheme of the thermal problem . the unconditionally stability of the scheme can also be shown along the same line of argument used for the mechanical case . * again , the use of the @xmath0-inner product to enforce the thermal problem allows one to omit the boundary restriction when we define the thermal displacement trial and weight function space . i.e. @xmath277 this is a very important observation in terms of practical implementation . as we have seen from section [ sec : alg - os ] that consistent and stable sub - algorithms render a consistent and stable global algorithm in the sense of time - stepping algorithms based on operator - splitting . both the mechanical and the thermal phase algorithms are shown to be consistent and unconditionally stable . therefore , the algorithm for the global problem based on lie - trotter - kato product formula is consistent and unconditionally stable . moreover , the convergence of the global scheme follows from the well known result stated below . for * consistent * numerical approximations , * stability * is equivalent to * convergence*. in this section , we present a range of results for type ii and iii problems of non - classical thermoelasticity . we start by comparing convergence of the proposed splitting scheme against a monolithic approach in which all the governing equations are discretised simultaneously using the time - dg finite element method . for this , a 1-d problem of non - dimensional form is considered . the result shows excellent agreement between the monolithic and the splitting scheme . then we go on to present various results in 1-d and 2-d . the examples in this case are designed to illustrate two key features of the time - dg scheme : ( i ) its performance in solving problems that involves the propagation of sharp gradients without creating spurious oscillations ; and ( ii ) its capability in capturing the unique aspects of non - classical theory , for example , propagation of thermal wave and its complex response due to the coupling of elasticity problem . the family of problems considered in this section are organized as follows . to analyse the rate of convergence and capability of the proposed scheme , non - dimensional form of a 1-d non - classical thermoelastic problem is presented in section [ sec : nondim ] . the performance of the splitting algorithm is examined in section [ sec : initprop ] for an initial temperature pulse propagation in a two dimensional square plate under plane strain condition . finally , in section [ sec : expcylin ] , a quasi - static expansion of a thick walled , infinitely long cylinder in plane strain condition is presented , which is modelled as type i and type iii theory of thermoelasticity and the remarkable difference of thermal responses between the two models are also analysed . the non - dimensional form of 1-d gnt problem given in is @xmath278+\bar{b},\\ \partial_{_\tau}\bar{\alpha } & = \bar{\vartheta},\\ \partial_{_\tau}\bar{\vartheta } & = \partial_{_\xi}[\partial_{_\xi}\alpha+k\partial_{_\xi}\bar{\vartheta}]-\varepsilon_{_{2}}\partial_{_\xi}\bar{v } + \bar{s},\\ \end{aligned}\ ] ] with the dimensionless parameters @xmath279 where @xmath280 denotes the square of ratio of uncoupled velocities of the mechanical wave ( or first sound ) and thermal wave ( or second sound ) , @xmath281 denotes the strength of the thermomechanical coupling , @xmath282 represents the non - dimensional classical heat conductivity . the speed of first sound @xmath283 is actually the speed of sound in the medium , that is @xmath284 where @xmath285 denotes the young s modulus of the medium , while that of the second sound @xmath286 is a characteristic feature of the theory of non - classical heat conduction by green and naghdi that represents the speed in which a thermal disturbance travels through the medium : @xmath287 the non - dimensionless variables are given by @xmath288 where @xmath289 , @xmath290 , @xmath291 , @xmath292 are characteristic quantities having the same dimension as @xmath293 , @xmath130 , @xmath294 , @xmath16 respectively that can be chosen according to the relations @xmath295 from the above equations we can observe that there are infinitely many ways of choosing the characteristics constants without changing the form of the system . the nondimensional energy counterpart of , also referred to as the @xmath296-norm , is given by @xmath297 ^ 2+\bar{v}^2+\frac{1}{\varepsilon_{_2}}[\partial_{_\xi}\bar{\alpha}]^2+\frac{1}{\varepsilon_{_2}}\bar{\vartheta}^2\bigg]\mathrm{d}\xi,\ ] ] and the @xmath0-norm @xmath298\mathrm{d}\xi,\ ] ] where @xmath299 is the state vector at a given time . for the purpose of the convergence analysis an exact solution to problem is obtained in such a way that source terms @xmath300 and @xmath301 are suitably prescribed such that a given state vector @xmath302 be an exact solution @xcite . to this end let the source terms be @xmath303,\\ \bar{s } & = \frac{\pi^2}{4}\big[k\pi\sin(\pi \xi)\cos(\pi t)+\varepsilon_{_2}\cos(\pi \xi)\cos(\pi \tau)\big ] , \end{aligned}\ ] ] so that the exact solutions are @xmath304 defined on the space - time domain @xmath305\times[0,\bar{t}]$ ] . for convergence analysis the values of the non - dimensional parameters at taken at @xmath306 , @xmath307 , @xmath308 . such set of values represents a strongly coupled problem of two purely hyperbolic systems ( type ii thermoelasticity ) . the space - time domain corresponds to @xmath309 and @xmath310 . bilinear finite element functions , @xmath311 , are used in each space - time slab with each element having an aspect ratio of one ( i.e. @xmath312 ) . [ fig : convrates ] ( a ) reports the _ spatial convergence _ results of the monolithic and operator - splitting approaches with error norms of the approximate solutions at @xmath313 are computed using the @xmath296 and @xmath0norms as given in equations and , respectively . by construction the monolithic algorithm is only first - order accurate . remarkably , it is shown that the error norms for both approaches are seen to overlap showing the error associated with the splitting is almost negligible . this demonstrably shows an increment in efficiency of the splitting algorithm , while maintaining the accuracy of the monolithic scheme . while fig . [ fig : convrates ] ( b ) presents the result of _ temporal convergence _ results of the two approaches with errors of the approximate solutions computed at the mid - point , @xmath314 , and @xmath315 using the @xmath316 vector norm . again errors corresponding to both of the approaches are almost identical . similarly , this shows the temporal error associated with the operator - split strategy is minimal . [ cols="^ " , ] this problem deals with the quasi - static thermo - mechanical interaction in a thick walled cylinder as it expands as a result of an inner wall dirichlet - type boundary condition , in the linear and plane strain case . the material considered is isotropic both thermally and mechanically . the thermal variation is purely the result of mechanical changes ( the expansion of the cylinder ) unlike in the previous examples ( sections [ sec : laserprop ] and [ sec : initprop ] ) in which thermal variations cause mechanical effects . the cylinder has cross section occupying the region @xmath317 with inner and outer radii @xmath318 mm and @xmath319 mm , respectively . a zero heat flux boundary condition is maintained on the inner wall , while the outer wall is kept at the reference temperature @xmath320 . the inner wall is dynamically prescribed a radial displacement of @xmath321 mm per second , while the outer wall is mechanically free . the problem is analysed for @xmath322 seconds until the inner wall reaches a radius of @xmath323 . the time - dg finite element mesh consists of trilinear shape functions of @xmath324 elements around the circumference of the cylinder by @xmath325 elements radially with one element thickness in the temporal direction with step length @xmath326 s for each space - time slab . in this quasi - static case , since only the thermal equations contains temporal derivatives then the thermal fields are allowed to be discontinuous while the displacement field is continuous across the interfaces of each space - time slab . this implies that the numerical dissipation comes from the weak enforcement of the continuity of the thermal fields only . we consider two cases : the first is classical or type i thermoelasticity with @xmath327 and the other is type iii thermoelasticity with @xmath328 and @xmath329 . [ fig : temprofile ] shows temperature variations over time for each case sampled at the equally spaced points along the radial direction labeled a - e as shown in the fig . [ fig : cylindermesh ] . as expected , in both cases , the temperature of the entire cylinder is converging to the reference temperature as time increases . the sinusoidal thermal response of type iii is due to a temperature wave moving back and forth indicating second sound phenomenon . c + ( a ) * type i * + + ( b ) * type iii * an operator - splitting strategy coupled with a space - time discontinuous galerkin finite element method for the solution of transient and fully coupled initial - boundary problem of generalized thermoelasticity was presented . well - posdeness of the problem in the general setting ( type iii ) is proven using the theory of semigroups . the defining operator is split additively so that the first sub - operator represents an isentropic ( adiabatic ) elasticity in which the entropy density is held fixed , and the other is a non - standard heat conduction at fixed configuration . both of the sub - problems are also shown to inherit the same contractivity property as the full problem . each sub - problem is then discretised separately using a time - discontinuous galerkin finite element method where the unknown fields are allowed to be discontinuous along the interfaces of each space - time slab . weak continuity of the unknown fields is enforced using an @xmath0-inner product which differs from the original time - discontinuous formulation using an energy - inner product @xcite which was formulated for linear elastodynamics problem . the unconditionally stability behaviour of each of the algorithms is proven without the need to add extra ` artificial viscosity ' . the algorithm for the global problem is finally obtained by way of lie - trotter - kato product formula , leading to an unconditional stability . the results presented in this paper are demonstrated by a number of numerical examples in both one and two dimensional cases . the efficiency of the current numerical scheme were examined by comparing the rate of convergence of with the corresponding monolithic approach . the result shows that the splitting scheme not only it retains the accuracy of the monolithic scheme but also it improves the efficiency as two smaller problems are solved sequentially at each time - step . the capability of the splitting algorithm is tested using problems involving propagation of heat waves driven by a pulsing laser heat source and an initial temperature disturbance in one and two dimensions respectively . furthermore , the capability of the non - standard thermoelasticity and the proposed numerical method to model the phenomenon of second sound in some solids is demonstrated by considering the quasi - static expansion of an infinitely long thick walled cylinder in plane stress . the dg formulation proposed in this work may be extended to the non - linear regime without the need to eliminate the displacement - velocity relation in the formulation . hence , a full recovery of the numerical dissipation in the non - linear case is possible . this will be the subject of a forthcoming work in which issues such as non - linear stability and the existence of lyapunov function are discussed . * acknowledgments . * the work reported in this paper has been supported by the national research foundation of south africa through the south african research chair in computational mechanics . this support is acknowledged with thanks . the authors also thank professor s. bargmann for discussions which led to various improvements in the work . 10 url # 1`#1`urlprefixhref # 1#2#2 # 1#1 d. k. khalmonova , f. costanzo , a space - time discontinuous galerkin finite element method for fully coupled linear thermo - elasto - dynamic problems with strain and heat flux discontinuities , computer methods applied mechanics and engineering 197 ( 2008 ) 13231342 . f. armero , j. c. simo , a new unconditionally stable fractional step method for non - linear coupled thermomechanical problems , international journal for numerical methods in engineering 35 ( 1992 ) 737766 . j. c. simo , nonlinear stability of the time - discrete variational problem of evolution in nonlinear heat conduction , plasticity and viscoplasticity , computer methods in applied mechanics and engineering 88 ( 1991 ) 111131 . c. johnson , discontinuous galerkin finite element methods for second order hyperbolic problems , computer methods in applied mechanics and engineering 107 ( 1993 ) 117129 .

the detection and quantification of narrow emission lines in x - ray spectra is a challenging statistical task . the poisson nature of the photon counts leads to local random fluctuations in the observed spectrum that often results in excess emission in a narrow band of energy resembling a weak narrow line . from a formal statistical perspective , this leads to a ( sometimes highly ) multimodal likelihood . many standard statistical procedures are based on ( asymptotic ) gaussian approximations to the likelihood and simply can not be used in such settings . bayesian methods offer a more direct paradigm for accounting for such complicated likelihood functions but even here multimodal likelihoods pose significant computational challenges . the new markov chain monte carlo ( mcmc ) methods developed in and , however , are able to fully explore the complex posterior distribution of the location of a narrow line , and thus provide valid statistical inference . even with these computational tools , standard statistical quantities such as means and standard deviations can not adequately summarize inference and standard testing procedures can not be used to test for emission lines . in this paper , we use new efficient mcmc algorithms to fit the location of narrow emission lines , we develop new statistical strategies for summarizing highly multimodal distributions and quantifying valid statistical inference , and we extend the method of posterior predictive p - values proposed by to test for the presence of narrow emission lines in x - ray spectra . we illustrate and validate our methods using simulation studies and apply them to the _ chandra _ observations of the high redshift quasar pg1634 + 706 .

0808.3164

modern x - ray observations show complex structures in both the spatial and spectral domains of various astrophysical sources . nonetheless , active galalactic nuclei ( agn ) including quasars nuclei remain spatially unresolved even with the highest - resolution x - ray telescopes . most of their energy is released within the unresolved core , and only spectral and timing information is available to study the nature of the x - ray emission . generally speaking , emission and absorption lines constitute an important part of the x - ray spectrum in that they can provide information as to the state of plasma . one of the goals of x - ray data analysis is to understand the components present in the spectrum , and to obtain information about the emission and absorption features , as well as their locations and relation to the primary quasar emission . the detection of weak lines in noisy spectra is the main statistical problem in such analyses : is a bump observed in the spectrum related to a real emission line or is it simply an artifact of the poissonian noise ? although quasars x - ray spectra are usually featureless as expected based on the comptonization process ( see for example * ? ? * ; * ? ? ? * ; * ? ? ? * ) , an important x - ray emission feature identified in agn and quasars spectra is the iron k emission line ( see recent review by * ? ? ? determining the origin and the nature of this line is one of main issues in agn and quasar research . this line is thought to come directly from illuminated accretion flow as a fluorescent process @xcite . the location of the line in the spectrum indicates the ionization state of iron in the emitting plasma , while the width of the line tells us the velocity of the plasma @xcite . the iron line provides a direct probe of the innermost regions of accretion flow and matter in close vicinity of a black hole . absorption features associated with the outflowing matter ( warm wind , partial covering absorber ) have also been observed in recent x - ray observations @xcite . although the location and width of absorption lines provide information as to the velocity of the absorber and its distance from the quasar , this article focuses on statistical issues in fitting the spectral location of narrow emission lines , i.e. , identifying the ionization state . there are two parts to the fe - k - alpha emission line observed in agn @xcite : one is a broad component thought to be a signature of a relativistic motion in the innermost regions of an accretion flow ; the other is a narrow component that is a result of a reflection off the material at larger distances from the central black hole . a detection of the broad component is challenging as it requires a spectral coverage over a large energy range , so the continuum emission is well determined and the broad line can be separated @xcite . the relativistic line profile is broad and skewed , and two strong peaks of the emission line that originates in a relativistic disk can be prominent and narrow . while the full profile of the broad line may not be easily separable from the continuum , these two peaks may provide a signature for this line in the x - ray spectrum . the broad fe - line gives an important diagnostic of the gas motion and can be used to determine the spin of a black hole @xcite ; see also an alternative model for the red wing " component by @xcite . the narrow component of the line gives diagnostics of the matter outside the accretion disk and conditions at larger distances from the black hole ; see fe - line baldwin effect discussion in @xcite . both line components are variable and the line may `` disappear '' from the spectrum @xcite . the spectral resolution of x - ray ccd detectors ( for example 100 - 200 ev in acis on _ chandra _ or epic on xmm-_newton _ ) is relatively low with respect to the predicted width of narrow ( @xmath0 km s@xmath1 ) emission or absorption lines in agn and quasars . observations with grating instruments ( rgs or heg ) can provide high resolution x - ray spectra , but the effective area of the present x - ray telescopes is too low for efficient agn detections , and only a handful of bright low redshift sources have been observed with gratings to date @xcite . therefore mainly the x - ray ccd spectra of lower resolution are used to study large samples of agn and quasars ( see for example * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . using these relatively low resolution x - ray detectors , the fe - k - alpha emission line can be narrow enough to be contained entirely in a single detector bin . in some cases ( for example in _ chandra _ ) the line may occupy a few bins . in this article we focus on the statistical problem of fitting the spectral location of an emission line or a set of emission lines that are narrow . this is a common objective in high - energy analyses , but as we shall discuss fitting these relatively narrow features poses significant statistical challenges . in particular we find evidence that using line profiles that are narrower than we actually expect the emission line to be can improve the statistical properties of the fitted emission line location . x - ray spectra , such as those available with the _ chandra x - ray observatory _ carry much information as to the quasar s physics . taking advantage of the spectral capacity of such instruments , however , requires careful statistical analysis . for example , the resolution of such instruments corresponds to a fine discretization of the energy spectrum . as a result , we expect a low number of counts in each bin of the x - ray spectrum . such low - count data make the gaussian assumptions that are inherent in traditional minimum @xmath2 fitting inappropriate . a better strategy , which we employ , explicitly models photon arrivals as an inhomogeneous poisson process @xcite . in addition , data are subject to a number of processes that significantly degrade the source counts , e.g. , the absorption , non - constant effective area , blurring of photons energy , background contamination , and photon pile - up . thus , we employ statistical models that directly account for these aspects of data collection . in particular , we design a highly structured multilevel spectral model with components for both the data collection processes and the complex spectral structures of the sources themselves . in this highly structured spectral model , a bayesian perspective renders straightforward methods that can handle the complexity of _ chandra _ data @xcite . as we shall illustrate , these methods allow us to use low - count data , to search for the location of a narrow spectral line , to investigate its location s uncertainty , and to construct statistical tests that measure the evidence in the data for including the spectral line in the source model . the energy spectrum can be separated into two basic parts : a set of continuum terms and a set of several emission lines . we begin with a standard spectral model that accounts for a single continuum term along with several spectral lines . throughout this paper , we use @xmath3 as a general representation of model parameters in the spectral model . the components of @xmath4 represent the collection of parameters for the continuum , ( emission ) lines , absorption , and background contamination , respectively . ( notice that the roman letters in the superscripts serve as a mnemonic for these four processes . ) because the x - ray emission is measured by counting the arriving photons , we model the expected poisson counts in energy bin @xmath5 , where @xmath6 is the set of energy bins , as @xmath7 where @xmath8 and @xmath9 are the width and mean energy of bin @xmath10 , @xmath11 is the expected counts per unit energy due to the continuum term at energy @xmath9 , @xmath12 is the set of free parameters in the continuum model , @xmath13 is the number of emission lines , @xmath14 is the expected counts due to the emission line @xmath15 , and @xmath16 is the proportion of an emission line centered at energy @xmath17 and with width @xmath18 that falls into bin @xmath10 . there are a number of smooth parametric forms to describe the continuum in some bounded energy range ; in this article we parameterize the continuum term @xmath19 as a power law , i.e. , @xmath20 where @xmath21 and @xmath22 represent the normalization and photon index , respectively . the emission lines can be modeled via the proportions @xmath16 using narrow gaussian distributions , lorentzian distributions , or delta functions ; the counts due to the emission line are distributed among the bins according to these proportions . while the gaussian or lorentzian function parameterizes an emission line in terms of center and width , the center is the only free parameter with a delta function ; the width of the delta function is effectively the width of the energy bin in which it resides . while the model in equation [ park : eq : ideal ] is of primary scientific interest , a more complex statistical model is needed to address the data collection processes mentioned in [ park : sec : hea ] . we use the term _ statistical model _ to refer to the model that combines the _ source _ or _ astrophysical model _ with a model for the stochastic processes involved in data collection and recording . thus , in addition to the source model , the statistical model describes such processes as instrument response and background contamination . specifically , to account for the data collection processes , equation [ park : eq : ideal ] is modified via @xmath23 where @xmath24 is the expected observed poisson counts in detector channel @xmath25 , @xmath26 is the set of detector channels , @xmath27 is the probability that a photon that arrives with energy corresponding to bin @xmath10 is recorded in detector channel @xmath28 ( i.e. , @xmath29 is the so - called redistribution matrix or rmf commonly used in x - ray analysis ) , @xmath30 is the effective area ( i.e. , arf , a calibration file associated with the x - ray observation ) of bin @xmath10 , @xmath31 is the probability that a photon with energy @xmath9 is _ not _ absorbed , @xmath32 is the collection of parameters for absorption , and @xmath33 is a poisson intensity of the background counts in channel @xmath28 . while the scatter probability @xmath27 and the effective area @xmath30 are presumed known from calibration , the absorption probability is parameterized using a smooth function ; see @xcite for details . to quantify background contamination , a second data set is collected that is assumed to consist only of background counts ; the background photon arrivals are also modeled as an inhomogeneous poisson process . unfortunately , the statistical methods and algorithms developed in @xcite can not be directly applied to fitting _ narrow _ emission lines . there are three obstacles that must be overcome in order to extend bayesian highly structured models to spectra containing narrow lines . in particular , we must develop ( 1 ) new computational algorithms , ( 2 ) statistical summaries and methods for inference under highly multimodal posterior distributions , and ( 3 ) statistical tests that allow us to quantify the statistical support in the data for including an emission line or lines in the model . our main objective in this paper is to extend the methods of @xcite in these three directions , and to evaluate and illustrate our proposals . here we discuss each of these challenges in detail . [ [ challenge-1-statistical - computation . ] ] challenge 1 : statistical computation . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + fitting the location of narrow lines requires new and more sophisticated computational techniques than those developed by @xcite . indeed , the algorithms that we develop require a new theoretical framework for statistical computation : they are not examples of any existing algorithm with known properties . although the details of this generalization are well beyond the scope of this article , we can offer a heuristic description ; a more detailed description is given in appendix [ ap : alg ] . readers who are interested in the necessary theoretical development of the statistical computation techniques are directed to @xcite and @xcite . the algorithms used by @xcite to fit the structured bayesian model described in [ park : sec : model ] are based on the probabilistic properties of the statistical models . for example , the parameters of a gaussian line profile can be fit by iteratively attributing a subset of the observed photons to the line profile and using the mean and variance of these photon energies to update the center and width of the line profile . the updated parameters of the line profile are used to again attribute a subset of the photons to the line , i.e. , to stochastically select a subset of the photons that are likely to have arisen out of the physical processes at the source corresponding to the emission line . these algorithms are typically very stable . for example , they only return statistically meaningful parameters because the algorithms themselves mimic the probabilistic characteristics of the statistical model . the family of expectation / maximization ( em ) algorithms @xcite and markov chain monte carlo ( mcmc ) methods such as the gibbs sampler @xcite are the examples of statistical algorithms of this sort . a drawback of these algorithms is that in some situations they can be slow to converge . when fitting the location of a gaussian emission line , for example , the location is updated more slowly if the line profile is narrower this is because only photons with energies very close to the current value of the line location can be attributed to the line . updating the line location with the mean of the energies of these photons can not result in a large change in the emission line location . the situation becomes chronic when a delta function is used to model the line profile : the line location parameter sticks at its starting value throughout the iteration . it is to circumvent this difficulty that we develop both new em - type algorithms @xcite and new mcmc samplers specially tailored for fitting narrow lines . our new samplers are motivated by the gibbs sampler , but constitute a non - trivial generalization of gibbs sampling known as partially collapsed gibbs sampling @xcite ; see appendix [ ap : alg ] . our updated versions of both classes of algorithms are able to fit narrow lines by avoiding the attribution of photons to the emission line during the iteration . such algorithms tend to require fewer iterations to converge regardless of the width of the emission line . because they involve additional evaluation of quantities evolving the large dimensional redistribution matrix , @xmath34 , however , each iteration of these algorithms can be significantly more costly in terms of computing time . a full investigation of the relative merit of the algorithms and a description of how the computational trade - offs can be played to derive optimal algorithms are beyond the scope of this paper . except in appendix [ ap : alg ] , we do not discuss the details of the algorithms further in this article ; interested readers are directed to @xcite and @xcite . [ [ challenge-2-multimodal - likelihoods . ] ] challenge 2 : multimodal likelihoods . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the likelihood function for the emission line location(s ) is highly multimodal . each mode corresponds to a different relatively likely location for an emission line or a set of emission lines . standard statistical techniques such as computing the estimates of the line locations with their associated error bars or confidence intervals implicitly assume that the likelihood function is unimodal and bell shaped . because this assumption is clearly and dramatically violated , these standard summary statistics are unreliable and inadequate . unfortunately , there are no readily available and generally applicable simple statistical summaries to handle highly multimodal likelihoods . instead we must develop summaries that are tailored to the specific scientific goals in a given analysis . because general strategies for dealing with multimodal likelihood functions are little known to astronomers and specific strategies for dealing with multimodal likelihood functions for the location of narrow spectral lines do not exist , one of the primary goals of this article is to develop and illustrate these methods . a fully bayesian analysis of our spectral model with narrow emission lines is computationally demanding , even with our new algorithms . thus , we develop techniques that are much quicker and give similar results for the location of emission lines . these methods based on the so - called _ profile posterior distribution _ do not stand on as firm of a theoretical footing as a fully bayesian analysis , but are much quicker and thus better suited for _ exploratory data analysis_. the profile posterior distribution along with our exploratory methods are fully described and compared with the more sophisticated bayesian analysis . [ [ challenge-3-testing - for - the - presence - of - narrow - lines . ] ] challenge 3 : testing for the presence of narrow lines . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + in addition to fitting the location of one or more emission lines , we often would like to perform a formal test for the inclusion of the emission lines in the statistical model . that is , we would like to quantify the evidence in a potentially sparse data set for a particular emission line in the source . testing for a spectral line is an example of a notoriously difficult statistical problem in which the standard theory does not apply . there are two basic technical problems . first , the simpler model that does not include a particular emission line is on the boundary of the larger model that does include the line . that is , the intensity parameter of an emission line is zero under the simpler model and can not be negative under the larger model . an even more fundamental problem occurs if either the line location or width is fit , because these parameters have _ no value _ under the simpler model . the behavior ( i.e. , sampling distribution ) of the likelihood ratio test statistic under the simpler model is not well understood and can not be assumed to follow the standard @xmath2 distribution , even asymptotically . @xcite propose a monte - carlo - based solution to this problem based on the method of posterior predictive p - values @xcite . in this article we extend the application of protassov _ et al_. s solution to the case when we fit the location of a narrow emission line , a situation that was avoided in @xcite . the remainder of the article is organized into four sections . [ park : sec : model - based ] reviews bayesian inference and monte carlo methods with an emphasis on multimodal distributions , outlines our computation methods , proposes new summaries of multimodal distributions , and describes exploratory statistical methods in this setting . we introduce illustrative examples in [ park : sec : model - based ] , but detailed spectral analysis is postponed in order to allow us to focus on our proposed methods . in [ park : sec : simul ] , a simulation study is performed to investigate the statistical properties of our proposed methods , with some emphasis placed on the potential benefits of model misspecification . [ park : sec : quasar ] presents the analysis of the high redshift quasar pg1634 + 706 , and how to test for the inclusion of the line in the spectral model . concluding remarks appear in [ park : sec : conclusion ] . an appendix outlines the computational methods we developed specifically for fitting the location of narrow emission lines . using a poisson model for the photon counts , the likelihood function of the parameter in the spectral model described in [ park : sec : model ] is given by @xmath35 $ ] where @xmath36 denotes the observed photon counts . with likelihood - based methods , the parameter value that maximizes the probability of the observed data is generally chosen as an estimate of @xmath3 ; this estimate is called the maximum likelihood estimate ( mle ) . in bayesian methods , prior knowledge for @xmath3 can be combined with the information in the observed data . prior distribution _ can be used to quantify information from other sources or to impose structure on a set of parameters . the prior distribution is combined with the likelihood to form a _ posterior distribution_. the prior distribution is denoted by @xmath37 and the posterior distribution by @xmath38 . bayesian inferences for @xmath3 are based on the posterior distribution . using bayes theorem , the prior distribution and the likelihood function are combined to form the posterior distribution via @xmath39 where the last proportionality holds because @xmath40 does not depend on @xmath3 and , given the observed data , is considered a constant . bayesian statistical inferences are made in terms of probability statements , which are quantified using various numerical summaries of the posterior distribution . to illustrate this , we consider a stylized right - skewed distribution ; see figure [ park : fig : skewed ] . this distribution is similar to the posterior distribution of the expected counts due to the emission line @xmath15 , @xmath14 , because this parameter is necessarily non - negative . although from a bayesian perspective the posterior distribution of a model parameter is the complete summary of statistical inference for that parameter , it is often useful to summarize the posterior distribution using point estimates or intervals . commonly used bayesian point estimates of a parameter are the mean , median , and mode(s ) of the posterior distribution . error bars for the point estimates can be computed based on the variation of the posterior distribution . the equal - tail interval and the highest posterior density ( hpd ) interval are both commonly used summaries of uncertainty . for example , a 68% equal - tail posterior interval is the central interval of the posterior distribution and corresponds to the range of values of the parameter above and below which lies exactly 16% of the posterior probability . a 68% hpd interval , on the other hand , is the interval of values that contains 68% of posterior probability and within which the density is never lower than that outside the interval . the 68% hpd interval is the shortest possible interval that accounts for 68% of the posterior probability . this is illustrated in figure [ park : fig : skewed](a ) and ( b ) . the equal - tail interval achieves the same probability as the hpd interval by excluding a more likely region and by including a less likely region . when the posterior distribution is unimodal and symmetric , the equal - tail interval and the hpd interval are identical . in addition to computing parameter estimates and their error bars , it is often important to check if model assumptions are supported by the data . one way to do this is to generate simulated data under the model and compare the simulated data with the observed data ; refer to @xcite where this strategy is used to determine whether emission line profiles should be included in the spectral model for a gamma - ray burst . if the simulated data vary systematically for the observed data , it is an indication that the model used to simulate the data may not be adequate . in a bayesian analysis , we might generate such simulated data from the _ posterior predictive distribution _ , denoted by @xmath41 , i.e. , @xmath42 where @xmath43 represents an unknown future observation and the last equation follows because @xmath43 and @xmath44 are conditionally independent given @xmath3 . in words , the posterior predictive distribution averages the likelihood function over the posterior distribution of @xmath3 . data are simulated from the posterior predictive distribution and then used to make predictive inferences ; see [ park : sec : model - check ] for details . posterior simulation plays a central part in applied bayesian analysis because of the usefulness of a simulation that can often be relatively easily generated from a posterior distribution . in performing simulations , given a large enough sample , a histogram of a monte carlo simulation can provide practically complete information about an actual posterior distribution . figure [ park : fig : skewed](c ) shows that the histogram of the monte carlo simulation carries the same information as the posterior distribution itself . thus , once a monte carlo simulation is obtained , it can be used to compute the mean , variance , percentiles , and other summaries of the posterior distribution . in particular , with a random simulation of size @xmath45 , the posterior mean can be approximated as @xmath46 where @xmath47 is a simulation from the posterior distribution , @xmath38 . a 68% equal - tail posterior interval is computed by generating a monte carlo simulation of size @xmath45 from the posterior distribution , sorting the simulated values into increasing order , and choosing the @xmath48$]th and the @xmath49$]th values in the list . with the monte carlo simulation , a 68% hpd region is computed by segmenting the range of possible parameter values into bins , approximating the posterior probability of each bin as the proportion of the simulated values in that bin , and computing a region by beginning with the bin with the largest posterior probability and adding additional bins in the order of their posterior probabilities until the resulting region contains at least 68% of the posterior probability . our search for a narrow emission line begins by finding modes of its posterior distribution that correspond to plausible locations of an emission line . to find the modes ( and to compte the profile posterior distribution ) , we use an algorithm optimized for this problem ( i.e. , the rotation(9 ) em - type algorithm , see appendix [ ap : alg ] and @xcite for details ) . because the posterior distribution of the line location is highly multimodal ( see [ park : sec : mm ] ) , the algorithm is run using multiple starting values selected across the entire energy range of possible line locations , e.g. , 50 starting values for the line location equally spaced between 1.0 kev and 6.0 kev . using multiple starting values enables us to identify the important local modes of the posterior distribution . it is important to use enough starting values to ensure all of the important modes are identified . this is a standard strategy , long advocated in texts on bayesian data analysis @xcite and is closely related to the computation of the profile posterior distribution described in [ park : sec : advise ] . the profile posterior distribution is computed by fixing the line location at each value of a fine grid , finding the posterior modes of the other model parameters , and plotting the resulting maximum posterior probability as a function of the line location . this procedure corresponds to the projected delta - chi - square method in the chi - square setting , see @xcite and @xcite . mode finding also begins with a fine grid of starting values , but we run the mode finder allowing all parameters including the line location to be fit . after the modes are found , monte carlo simulation techniques optimized to this problem can be run to further investigate the uncertainty of the possible line locations . we employ state - of - the - art mcmc samplers , i.e. , partially collapsed gibbs ( pcg ) i for the delta function emission line and pcg ii for the gaussian emission line to obtain the posterior distribution of line location ; see appendix [ ap : alg ] , @xcite , and @xcite for details . to ensure the convergence of a markov chain constructed by the mcmc samplers , we run multiple chains with overdispersed starting values ( e.g. , 1 kev , 3 kev , and 5 kev for the line location parameter ) and monitor the convergence qualitatively and quantitatively . for example , we compute the estimate of the potential scale reduction @xcite , denoted by @xmath50 , for all parameters of interest . if @xmath50 is near 1 ( e.g. , below 1.2 ) for each of the parameters , we collect the second halves of the chains together and use these monte carlo draws for inference ; see @xcite for theoretical justification and discussion . there are of course many strategies that one might employ to construct efficient monte carlo samplers . methods based on annealing or tempering , or that use explicitly parallel methods are often useful for exploring multimodal posterior distributions . given the low autocorrelation of the simulated values produced by our method , we have not pursued these strategies . fitting a narrow emission line often tends to yield a highly multimodal posterior distribution of the line location , as shown in figure [ park : fig : hpd ] . thus , we are interested in the two types of intervals for the posterior distributions : an equal - tail posterior interval and an hpd region . figure [ park : fig : hpd ] illustrates a 68% equal - tail posterior interval and a 68% hpd region for the line location when its posterior distribution is highly multimodal . this is only an example of a highly multimodal posterior distribution . we come to the details of our analysis of a line location in [ park : sec : simul ] . the 68% equal - tail posterior interval is a central interval , so that it includes segments with nearly zero posterior probability here and there within the interval . because the posterior distribution of the line location in figure [ park : fig : hpd ] is multimodal , the 68% hpd region consists of the four shaded disjoint intervals ; notice that the height of each of the histogram bars outside the hpd region is less than that of those within the region . in such a multimodal posterior distribution , the hpd region not only is shorter in length but also conveys more information as to likely locations of the line than does the equal - tail interval . in figure [ park : fig : hpd ] , we use a coarse binning to illustrate the distinction between the equal - tail posterior interval and hpd region . this results in a posterior region that is relatively imprecise ; in our spectral analysis , we use a finer binning to construct a more precise region from a monte carlo simulation . when the posterior distribution of the line location is plotted with the same fine binning as the _ chandra _ energy spectrum , however , it may not be smooth due to monte carlo errors , as shown in the top panel of figure [ park : fig : hpd - illust ] . an hpd region computed from the unsmoothed posterior distribution may result in a combination of too many posterior intervals . to avoid such fragmentation of hpd regions , we use gaussian kernel smoothing to smooth the posterior distribution before we compute hpd regions . the middle panel of figure [ park : fig : hpd - illust ] presents the smoothed posterior distribution resulting from applying gaussian kernel smoothing with standard deviation equal to the bin size of the _ chandra _ energy spectrum , i.e. , 0.01 kev . the smoothed posterior distribution smooths out lower posterior probabilities but does not flatten higher posterior probabilities too much . we propose a new graphical summary to better describe the hpd regions of a ( smoothed ) multimodal posterior distribution . an _ hpd graph _ is constructed by plotting a series of hpd regions against their corresponding hpd levels . for example , for the data set used to compute the posterior distribution in figure [ park : fig : hpd - illust ] , we compute 100 hpd regions , one for each of 100 levels , 1% , 2% , @xmath51 , and 100% . each of these regions is a union of possible values of the line location . we can plot the line location on the horizontal axis and the level on the vertical axis . each of the 100 hpd regions can then be plotted as a union of horizonal line intervals at the appropriate level of the hpd region on the vertical axis . the resulting hpd graph lets us visualize many hpd regions with varying levels , so that all the important modes of a multimodal distribution can be effectively summarized with their relative posterior probabilities . as an illustration , we computed the three hpd regions with levels 50% , 68% , and 95% , and plotted them in the middle panel of figure [ park : fig : hpd - illust ] along with the smoothed posterior distribution . the solid lines indicate the disjoint intervals that compose each hpd region , and the dotted lines the intervals outside the hpd region . the bottom panel in figure [ park : fig : hpd - illust ] shows the hpd graph in grey with the three hpd intervals from the middle plot superimposed . , where each marginal standard deviation corresponds to the size of an energy bin in the _ chandra _ spectrum . based on the smoothed posterior distribution , we compute ten 2-d hpd regions with levels ranging from 10% to 100% . these regions are plotted with different shades of grey in the 2-d hpd graph shown in the right panel.,width=576 ] when there are two parameters of interest , the 1-d hpd graph can be extended to a 2-d hpd graph . that is , a joint posterior distribution is computed from a monte carlo simulation by using bivariate gaussian kernel smoothing and used to construct 2-d hpd regions with various levels . these hpd regions can then be plotted with different shades of grey . for example , figure [ park : fig : twod - hpd - illust ] shows the 2-d hpd graph computed for the joint posterior distribution of two possible line locations ; see [ park : sec : simul2 ] for further discussion on the scatter plot . the left panel of figure [ park : fig : twod - hpd - illust ] is an unsmoothed joint posterior distribution . after applying bivariate gaussian kernel smoothing , 10 hpd regions ( one for each of 10 levels , 10% , 20% , ... , and 100% ) are computed and plotted with different shades of grey from black ( 10% ) to white ( 100% ) , as shown in the right panel of figure [ park : fig : twod - hpd - illust ] . hpd regions with lower levels contain pixels with higher posterior probabilities , so that darker pixels indicate more probable regions of the two line locations . multimodal likelihoods and posterior distributions pose another challenge for statistical analysis . the calibration of many standard statistical procedures is based on the asymptotic gaussian nature of the likelihood and posterior distribution . thus , standard methods for computing error bars and confidence intervals rely on the likelihood being at least approximately gaussian . when the likelihood function exhibits the multimodal features that we see in figures [ park : fig : hpd - illust ] and [ park : fig : twod - hpd - illust ] , these standard results simply do not apply . one consequence of this is that the nominal level of an interval may not match its frequency coverage . that is , if we were able to repeat our observation many times , we would find that percentage of intervals that contain the true line location might differ from the nominal level of the interval . in [ park : sec : simul ] , we find that the posterior distribution is highly multimodal when the true emission line is weak or there are several emission lines in a spectrum . in this case , we also find that by _ misspecifying _ the model we are able to improve the frequency properties of our proposed method . in particular , we find that using a delta function line profile improves the properties of our procedure even when the true line has appreciable width . as discussed in [ park : sec : mm ] , the posterior distribution of a narrow emission line location tends to be highly multimodal . in this case , the profile posterior distribution can be used as a handy and quick - to - compute summary of the posterior distribution to explore the possible locations of a spectral line . the profile posterior distribution is the posterior distribution of a parameter evaluated at the values of the other parameters that maximize the posterior distribution . the profile posterior distribution is a bayesian analogue to the profile likelihood , a standard likelihood method for dealing with nuisance parameters ; see @xcite or @xcite for applications of the profile likelihood . in the context of parameter estimation , the distribution of the minimum @xmath2 statistic described by @xcite is closely related to the profile posterior distribution . generally we do not advocate using the profile posterior distribution as a substitute for the marginal posterior distribution because interval or region estimates computed with the profile posterior distribution have rather unpredictable statistical properties . the marginal posterior distribution is obtained by integrating out ( i.e. , averaging over ) the other parameters and computing the marginal posterior distribution requires sophisticated numerical integration methods such as mcmc especially when the dimension of the nuisance parameters is large . however , the profile posterior distribution can be computed without mcmc and , as an analogue to the marginal posterior distribution , can be used to roughly examine the posterior distribution of a model parameter . thus , we believe the profile posterior distribution is well suited for the initial exploration of the data because it gives a clear and reliable set of potential locations for spectral lines . the profile posterior distribution can be computed on a fine grid of the possible values of the line location by running an optimizer to maximize over the other model parameters for each value of the line location on the grid ; we recommend using stable optimizers such as em - type algorithms @xcite . the profile posterior distribution of the line location computed in this way is computationally less demanding and cheaper in terms of cpu time than the marginal posterior distribution produced by monte carlo methods such as mcmc . our simulation study is conducted to assess the validity of the highly structured multilevel spectral model discussed in [ park : sec : model ] , to illustrate the connection between the multimodal posterior distribution of a single narrow emission line and evidence for multiple lines , to illustrate the possible advantage of the misspecification of an emission line width , and to illustrate the relationship among gaussian line parameters . we consider the following six cases that we believe are representative of the cases that are of general interest : case 1 : : : there is no emission line in the spectrum . case 2 : : : there is a narrow gaussian emission line at 2.85 kev with @xmath52 kev and equivalent width$ ] . ] , @xmath53 kev in the spectrum . case 3 : : : there is a moderate gaussian emission line at 2.85 kev with @xmath54 kev and @xmath55 kev in the spectrum . case 4 : : : there is a narrow gaussian emission line at 2.85 kev with @xmath56 kev and @xmath57 kev in the spectrum . case 5 : : : there are two narrow gaussian emission lines with @xmath56 kev , one at 1.20 kev with @xmath58 kev and the other at 2.85 kev with @xmath59 kev in the spectrum . case 6 : : : there are two narrow gaussian emission lines with @xmath56 kev , one at 1.20 kev with @xmath58 kev and the other at 2.85 kev with @xmath57 kev in the spectrum . that is , each spectrum has either 0 , 1 , or 2 lines . data are recorded in an energy grid with bin width 0.01 kev . thus , the narrow emission line ( i.e. , cases 2 , 4 , 5 , and 6 ) corresponds to about 17 energy bins and the moderate one ( i.e. , case 3 ) about 85 energy bins , using @xmath60 standard deviations . as compared to the _ chandra _ resolution , the delta function emission line profile corresponds to 1 energy bin and thus it does not correctly specify the width of the gaussian emission lines in this simulation . through the simulation study , however , we illustrate possible advantage of this model misspecification in producing valid and efficient estimates and associated uncertainties for the line location . we show that using delta functions emission lines in the model is a useful strategy even when the true line occupies multiple bins . for each of the six spectra , we generate twenty test data sets ( 120 data sets in total ) each with about 1500 counts similar to the observed number of counts in the _ chandra _ x - ray spectrum of pg1634 + 706 analyzed in [ park : sec : quasar ] , mimicking the real data situation . each spectrum has a power law continuum with @xmath61@xmath625 and @xmath63 . our simulation is done with sherpa software @xcite in ciao , assuming the _ chandra _ responses ( effective area and instrument response function ) and no background contamination . . the vertical dashed lines represent the locations of the true emission lines.,width=576 ] for each of the test data sets , we run state - of - the - art mcmc samplers to fit a spectral model with a single delta function emission line . based on the monte carlo draws collected from the multiple chains of the mcmc samplers , the top two rows of figure [ park : fig : simul ] present the marginal posterior distribution of the delta function line location for one simulation under each of the six cases ; the vertical dashed lines represent the true line locations . the marginal posterior density is smoothed using gaussian kernel smoothing with standard deviation 0.01 kev , as described in [ park : sec : mm ] . when there is no emission line in the spectrum ( i.e. , case 1 ) , the posterior distribution of the delta function line location is highly multimodal . in the case of a weak narrow gaussian emission ( i.e. , case 2 ) , the marginal posterior distribution often remains highly multimodal , but one mode typically identifies the true line location . in practice , the local mode(s ) of such a highly multimodal posterior distribution may suggest plausible line locations and show evidence for multiple lines ; see [ park : sec : simul2 ] . even with a moderate line ( i.e. , case 3 ) , the true line location appears well estimated with the marginal posterior distribution of the delta function line location . as we shall see in table [ park : tbl : simulhpd ] , however , in this case the resulting posterior region under - covers the true values because the true line is 85 times wider than the specified model . with the strong narrow gaussian line ( i.e. , case 4 ) , the posterior distribution of the line location tends to be unimodal , and the posterior mode correctly identifies the true line location . the posterior distribution for case 5 in figure [ park : fig : simul ] is bimodal , with the modes corresponding to the two true line locations . when multiple lines are present in a spectrum , the posterior distribution of the single line location can be multimodal , as shown in the case 5 of figure [ park : fig : simul ] . thus the multiple modes may be indicative of multiple lines ; see [ park : sec : simul2 ] for details . when one of two narrow gaussian emission lines is much stronger ( i.e. , case 6 ) , the single delta function line model tends to identify only one of the two true line locations . to visualize the uncertainty of the fitted delta function line location(s ) , the bottom two rows of figure [ park : fig : simul ] show the hpd graphs constructed with 100 hpd regions as described in [ park : sec : mm ] . the multimodality in the marginal posterior distribution of a single line location may indicate the existence of multiple lines in a spectrum , provided the lines are well separated . when a model is fitted with one emission line , modes in the likelihood function of the line location correspond to ranges of energy with excess emission relative to the continuum . multiple modes in the likelihood indicate that there are multiple ranges of energy with such excess emission . the height of the mode is indicative of the degree of excess . thus , if there are several emission lines , we might expect to see several corresponding modes in the likelihood . if there is one energy range that dominates in terms of excess emission , however , it corresponds to the dominate mode of the likelihood . thus , if there are lines of very different intensities , only the strongest ones may show up as a mode of the likelihood . this can be seen by comparing case 5 and case 6 in figure [ park : fig : simul ] . if there is evidence for multiple lines in a spectrum or if we suspect multiple lines a priori , we can fit a model with two or more lines . we illustrate this using simulated data under case 2 ( one narrow line ) and case 5 ( two narrow lines ) . beginning with case 2 , the actual spectrum has only one line , but we investigate what happens when we fit two lines to this data . a scatterplot of the two fitted line locations identified when fitting two emission lines to one of the data sets generated under case 2 is presented in the top left panel of figure [ park : fig : simul - twolines ] . there is a label switching problem between the two fitted line locations because of the symmetry of the emission lines in the model . we can remove the symmetry by imposing a constraint on line locations . to do this , we first fit the model with a single delta function line profile and compute the posterior mode of its line location . returning to the model with two fitted delta functions , we separate the two fitted line locations by setting the first " line location to be the one closest to the posterior mode . in case 2 , the posterior mode for the single line location is 2.815 kev , so that the first line location is the line location closest to 2.815 kev and the second line is the other location . the two panels in the top right corner of figure [ park : fig : simul - twolines ] show the resulting marginal posterior distributions of the two fitted line locations . as shown in the figure , the marginal posterior distribution of the first line location correctly identifies the true line location . the spectrum used to generate the data under case 2 has no second line , so that the marginal posterior distribution of the second line is highly multimodal . in practice , we may take the local mode(s ) in the second marginal posterior distribution as candidates for another line location . however , the resulting hpd regions for the second line are wide , indicating that either there is no second line or if there is , it can not be well identified . when two emission lines are present in a spectrum , we follow the same procedure as illustrated using the data generated under case 5 . the bottom left panel of figure [ park : fig : simul - twolines ] shows the scatterplot of two line locations identified in the spectrum of case 5 . label - switching is handled as above using the posterior mode for the single line location , which is computed as 2.855 kev . the fitted marginal posterior distributions of the first and second line locations are given in the bottom right corner of figure [ park : fig : simul - twolines ] . when there are two emission lines in a spectrum , the two true line locations are precisely specified by the two marginal posterior distributions . this is a verification of what is suggested by the multiple modes in the marginal posterior distribution of the single line location shown in figure [ park : fig : simul ] . the possibility of model misspecification when using a delta function to model an emission line depends on both the width of the true line and the resolution of the detector . misspecification only occurs when the line is not contained in one energy bin , and we shall illustrate that such misspecification only has statistical consequences for the fitted line location if it is very severe . indeed there can be a possible statistical advantage of using a delta function rather than a gaussian line if we know the spectral line is not too wide . as a toy example , consider a simple gaussian model with known standard deviation : when @xmath64 , a 95% confidence interval for @xmath65 is given by @xmath66 . if we misspecify the modal as @xmath67 with @xmath68 , the resulting interval for @xmath65 is shorter and has lower coverage . we similarly underrepresent the error bars of an emission line location when we use a delta function for a line that is not contained in one energy bin . we expect this to reduce both the length and coverage of the confidence regions . the advantage or disadvantage of this strategy is not immediately clear , however , since the nominal coverage of the intervals is based on an asymptotic gaussian approximation to the posterior distribution which clearly does not apply in this setting . nonetheless , our simulation study illustrates that the use of a delta function line profile can result in a shorter and more informative hpd region while maintaining good coverage . we now turn to the computation of hpd regions and the possible statistical advantage of using delta functions in place of narrow gaussian emission lines . we fit a spectral model that includes a single delta function line or a single narrow gaussian line to the twenty simulated data sets generated under each of the 6 cases . after smoothing the marginal posterior distribution using gaussian kernel smoothing , we construct 95% hpd regions for the line location , as shown in figure [ park : fig : simul-10datasets ] . for visual clarity , we present results only for the first ten simulated data sets in figure [ park : fig : simul-10datasets ] ; results from all twenty simulated data sets are discussed in table [ park : tbl : simulhpd ] . because there is no emission line in the spectrum used to simulate data under case 1 , the 95% hpd regions for the line location are very wide and show large uncertainties for the fitted line location . when there is at least one strong emission line ( i.e. , cases 3 , 4 , and 6 ) , both line models produce comparable hpd regions , although those computed under the gaussian line model appear somewhat wider . the tradeoff between the two line models becomes evident when there is no strong emission line in the spectrum ( i.e. , cases 2 and 5 ) . in case 2 , the 95% hpd regions for a single gaussian line location are somewhat wider . with the same nominal level , the delta function line model yields more compact and informative hpd regions . an added advantage of the delta function line model occurs in case 5 when the 95% hpd regions for a single delta function line location consist of two disjoint hpd intervals which simultaneously contain the two true line locations ; this behavior is more often observed with the delta function line model ( 2 times out of 20 ) than the gaussian line model ( 1 time out of 20 ) . .summary of 95% hpd regions for the line location in the simulation study . [ cols="^,^,^,^,^,^,^ " , ] for example , the 95% hpd regions of the delta function line location are presented in table [ park : tbl : hpds ] along with local modes of the posterior distribution associated with each interval . each of the 95% hpd regions is composed of a number of disjoint intervals . only the intervals that have posterior probabilities greater than 5% are presented in table [ park : tbl : hpds ] , so that the probabilities may sum to less than 95% . for example , the two intervals of obs - id 47 presented in table [ park : tbl : hpds ] have a combined posterior probability of 80.96% and the other eleven intervals not shown in the table have a posterior probability of about 14.04% , for a total of 95% . the posterior modes of the delta function line location that are located near 2.74 kev are indicated in bold face in table [ park : tbl : hpds ] . and the right panel over a range near 2.85 kev.,width=576 ] the six observations of pg1634 + 706 were independently observed with _ chandra_. thus , under the flat prior distribution on @xmath65 , @xmath69 , the posterior distribution of the line location given all six observations is given by @xmath70 where @xmath71 denotes the six observations , @xmath65 denotes the delta function line location parameter , @xmath72 denotes the set of model parameters other than @xmath65 for the six observations , and @xmath73 represents a likelihood function of @xmath74 given @xmath44 . ( here we allow @xmath75 to vary among the six observations ; i.e. , we do not exclude the possibility that the six observations have somewhat different power law normalizations and photon indexes . ) the values of the posterior distribution given one of the individual data sets is sometimes indistinguishable from zero because of numerical inaccuracies . thus we add 1/15000 to the posterior probability of each energy bin and renormalize each of the posterior distributions . this allows the product given in equation [ park : eq : combined ] to be computed for each energy bin and is somewhat conservative as it increases the posterior uncertainty corresponding to each of the individual data sets . figure [ park : fig : combined ] presents the marginal posterior distribution of the delta function line location given all six observations computed in this way ; the left panel examines the whole range of the line location while the right panel focuses on the range near 2.74 kev . as shown in figure [ park : fig : combined ] , the posterior distribution given all six observations is fairly unimodal and symmetric , except the little local mode near 2.655 kev . thus , using a nominal 95% hpd region , the most probable delta function line location is summarized as @xmath76 kev with posterior probability of 95.3% . posterior predictive methods @xcite can be employed to check the specification of the spectral model . this methods aim to check the self - consistency of a model , i.e. , the ability of the fitted model to predict the data to which the model is fit . to evaluate and quantify evidence for the inclusion of an emission line in the spectrum , we extend the method of posterior predictive p - values proposed by @xcite and @xcite . with the _ chandra _ observations of pg1634 + 706 , we consider the same spectral model discussed in [ park : sec : model ] except that we compare three models for the emission line : model 0 : : : there is no emission line in the spectrum . model 1 : : : there is a delta function emission line with location fixed at 2.74 kev but unknown intensity in the spectrum . model 2 : : : there is a delta function emission line with unknown location and intensity in the spectrum . we could equally well consider a gaussian line profile in models 1 and 2 ; either line profile model results in a valid test . we consider a delta function line profile simply because we are looking for evidence of a narrow emission line . we use ppp - values to compare the three models and quantify the evidence in the data for the delta function emission line ; see @xcite for details of this method and its advantages over the standard f - test , the standard cash statistic @xcite ( or likelihood ratio test statistic ) , and bayes factors . in the posterior predictive checks , model 1 fixes the delta function line location at 2.74 kev using prior information as to the location of the fe - k - alpha emission line . in order to combine the evidence for the line from all six observations with different exposure areas and exposure times , we base our comparisons on the test statistic that is the sum of the six loglikelihood ratio statistics for comparing model @xmath77 and model 0 , i.e. , @xmath78 where @xmath79 , @xmath80 , and @xmath81 represent the parameter spaces under models 0 , 1 , and 2 , respectively , and @xmath82 denotes the collection of six data sets simulated under model 0 . specifically , we generate 1000 replications of each data set from the posterior predictive distribution under model 0 and compute @xmath83 for @xmath84 . histograms of @xmath85 and @xmath86 appear in figure [ park : fig : pp - check ] . comparing the histogram of the simulated test statistics with the observed value of the test statistic yields the ppp - values @xcite shown in figure [ park : fig : pp - check ] . the ppp - value is the proportion of the simulated test statistics that are as extreme as or more extreme than the observed test statistic . smaller ppp - values give stronger evidence for the alternative model , i.e. , model 1 or model 2 , thereby supporting the inclusion of the line in the spectrum in our case . as shown in figure [ park : fig : pp - check ] , there is evidence for the presence of the spectral line given all six observations . the comparison between models 0 and 1 shows stronger evidence for the line location because we are using extra a priori information about the plausible line location . this article presents methods to detect , identify , and locate narrow emission lines in x - ray spectra via a highly structured multilevel spectral model that includes a delta function line profile . modeling narrow emission lines with a delta function causes the em algorithms and mcmc samplers developed in @xcite to break down and thus requires more sophisticated statistical methods and algorithms . the marginal posterior distribution of the delta function emission line location tends to be highly multimodal when the emission line is weak or multiple emission lines are present in the spectrum . because basic summary statistics are not appropriate to summarize such a multimodal distribution , we instead develop and use hpd graphs along with a list of posterior modes . testing for an emission line in the spectrum is a notoriously challenging problem because the value of the line intensity parameter is on the boundary of the parameter space , i.e. , zero , under a model that does not include an emission line . thus , we extend the posterior predictive methods proposed by @xcite to test for the evidence of a delta function emission line with unknown location in the spectrum . using the simulation study in [ park : sec : simul ] , we demonstrate the potential advantage of model misspecification using a delta function line profile in place of a gaussian line profile . we show that the delta function line profile may provide more precise and meaningful summaries for line locations if the true emission line is narrow . when multiple lines are present in the spectrum , the marginal posterior distribution of a single delta function line location may indicate multiple lines in the spectrum . our methods are applied to the six different _ chandra _ observations of pg1634 + 706 in order to identify a narrow emission line in the x - ray spectrum . given all the six observations , the most probable delta function line is identified at @xmath76 kev in the observed frame . the corresponding rest frame energy for the line is @xmath87 kev , which may suggest the high ionization of iron in the emitting plasma . there is some recent evidence that high ionization iron line can be variable on short timescales ( see for example mkn766 in * ? ? ? such variability would explain no detection of the emission line in one of the six _ chandra _ observations . the authors gratefully acknowledge funding for this project partially provided by nsf grant dms-04 - 06085 and by nasa contract nas8 - 39073 and nas8 - 03060 ( cxc ) . this work is a product of joint work with the california - harvard astrostatistics collaboration ( chasc ) whose members include j. chiang , a. connors , d. van dyk , v. l. kashyap , x .- meng , j. scargle , a. siemiginowska , e. sourlas , t. park , a. young , y. yu , and a. zezas . the authors also thank the referee for many helpful and constructive comments . in this section we give an overview of the mode finding and posterior simulation methods used to fit the spectral model with a narrow emission line . our summary is brief , and thus readers who are interested in a more detailed description should refer to @xcite and @xcite . in order to illustrate our computational strategy , consider a simplified example of an _ ideal instrument _ that is not subject to the data contamination processes . in particular , the redistribution matrix is an identity matrix , the effective area is constant , there are no absorption features , and there is no background contamination . in addition , we assume that the continuum is specified with no unknown parameters and that there is a single gaussian emission line that has a known width @xmath88 and a known intensity @xmath89 . thus , the line location is the only unknown parameter , the source model given in equation [ park : eq : ideal ] simplifies to @xmath90 and the counts are modeled as @xmath91 . this model can be fit using the method of data augmentation @xcite by setting @xmath92 , where @xmath93 and @xmath94 are the counts due to the continuum and the emission line in bin @xmath10 , respectively . in particular , the em iteratively split the counts into continuum counts and emission line counts . given the current iterate of the line location , @xmath95 , the e - step updates the line counts via e - step : : : compute @xmath96 $ ] for each bin @xmath10 . that is , @xmath97 = y_j \frac{{\lambda}\pi_j(\mu^{(t)},\nu_0 ) } { \delta_j f(e_j)+{\lambda}\pi_j(\mu^{(t)},\nu_0)}. \label{park : eq : toy - estep } \end{aligned}\ ] ] which is the weighted average of the bin energies and uses the emission line counts as weights . although this em algorithm is simple , it breaks down when fitting the location of a _ narrow _ emission line , i.e. , when @xmath88 is small relative to the size of the bins . in the extreme , the gaussian line profile becomes a delta function , so that @xmath100 is zero for all bins except the bin containing @xmath95 . this results in an e - step that computes zero line counts in all bins except the bin containing @xmath95 and finally an m - step that computes @xmath101 . this means that em will return the same line location at each iteration and that the algorithm will not converge to a mode . in this simplified example , this difficulty can be avoided by directly maximizing the posterior distribution @xmath102 because of the binning of the data , we can treat possible line locations within each bin as indistinguishable and compute the mode by evaluating equation [ park : eq : toy - obs - like ] on the fine grid that corresponds to the binning of the data . the situation is more complicated in the full spectral model described in [ park : sec : model ] . the method of data augmentation can be used to construct efficient algorithms that both fit the parameters in the continuum and the lines and account for instrument response and background contamination . in the case of the narrow emission line , however , we must implement a strategy that uses less data augmentation when updating the line location / width than when updating the other model parameters . the expectation / conditional maximization either ( ecme ) algorithm @xcite allows us to use no data augmentation when updating the line location / width , but the resulting m - step is time consuming owing to the multiple evaluations of the conditional posterior distribution of the line location / width which involve the large dimensional redistribution matrix , @xmath103 . an intermediate strategy uses the standard data augmentation scheme to adjust for instrument response and background contamination but does not separate continuum and line photons when updating the line location / width . this strategy is an instance of the alternating expectation / conditional maximization ( aecm ) algorithm @xcite and each iteration is much quicker than with ecme but more iterations are required for convergence . the algorithms we use for mode finding aim to combine the advantages of ecme and aecm by running one ecme iteration followed by @xmath77 aecm iterations and repeating until convergence . @xcite call this a _ rotation(@xmath77 ) _ algorithm and illustrate the computational advantage of the strategy . returning to the simplified example of appendix [ ap : toyem ] , we can formulate a gibbs sampler using the same data augmentation scheme . given the current iterate , @xmath95 , step 1 simulates the line counts in bin @xmath10 via when this algorithm is applied with a narrow emission line , it breaks down just as the em algorithm does . when a delta function is used for the line profile , the simulation in equation [ park : eq : step1 ] results in no line counts in any bin except the one containing the line , and again @xmath65 does not move from its starting value . when a narrow gaussian line profile is used , the situation is less extreme , but the sampler exhibits very high autocorrelations and typically can not jump among the posterior modes . just as with em , the difficulty can be avoided by computing the posterior distribution of @xmath65 on a find grid and directly simulating @xmath112 when the method of data augmentation is used to account for data contamination processes described in [ park : sec : model ] , however , this approach should be modified in a manner analogous to the ecme and aecm algorithms . this leads to the strategy of using conditional distributions from different data augmentation schemes . in this case , however , the resulting set of conditional distributions used to construct the gibbs sampler may be _ incompatible _ and there may be no joint distribution that corresponds to this set of conditional distributions . although such a sampler may result in efficient computation , care must be taken to be sure the sampler delivers simulations from the target posterior distribution . this is formalized through the partially collapsed gibbs ( pcg ) sampler of @xcite which outlines the steps that should be taken to ensure proper convergence ; refer to @xcite for the applications and illustrations of pcg samplers . the pcg samplers can be viewed as the stochastic version of ecme and aecm , thereby allowing us to sample the line location with no data augmentation ( as in ecme ) or partial data augmentation ( as in aecm ) . thus , the pcg sampler differs from the gibbs sampler developed in @xcite in sampling the line location ( and line width ) , and in the order of sampling steps . @xcite design two pcg samplers to fit the spectral model with a delta function emission line or a narrow gaussian emission line , which are called pcg i and pcg ii , respectively . as compared to pcg i , pcg ii requires one additional sampling step for the line width , so that fitting the narrow gaussian emission line is computationally more demanding than fitting the delta function emission line . yaqoob , t. , george , i. m. , nandra , k. , turner , t. j. , serlemitsos , p. j. , & mushotzky , r. f. 2001 , , 546 , 759 yaqoob , t. , padmanabhan , u. , dotani , t. , george , i. m. , nandra , k. , tanaka , y. , turner , t. j. , & weaver , k. a. 2001 , x - ray emission from accretion onto black holes , yaqoob , t. 2007 , astronomical society of the pacific conference series , 373 , 109

spectral properties of coupled cavity arrays in one dimension are investigated by means of the variational cluster approach . coupled cavity arrays consist of two distinct `` particles , '' namely , photons and atomiclike excitations . spectral functions are evaluated and discussed for both particle types . in addition , densities of states , momentum distributions and spatial correlation functions are presented . based on this information , polariton `` quasiparticles '' are introduced as appropriate , wave vector and filling dependent linear combinations of photon and atomiclike particles . spectral functions and densities of states are evaluated for the polariton quasiparticles , and the weights of their components are analyzed .

0912.4459

the experimental progress in controlling quantum optical and atomic systems , which has been achieved over the last few years , prompted ideas for new realizations of strongly correlated many body systems , such as ultracold gases of atoms trapped in optical lattices @xcite or light - matter systems.@xcite the latter consist of photons , which interact with atoms or atomiclike structures . normally , the interaction between photons and atoms is very weak , since the interaction time is small . however , a strong interaction can be achieved when photons are confined within optical cavities . in this case , the coupling between photons and atoms leads to an effective repulsion between photons , which means that it costs energy to add additional photons to the cavity . the arrangement of such cavities on a lattice , see fig.[fig : intcav ] , allows the photons to `` hop '' between neighboring sites , provided the cavities are coupled . quantum mechanically the coupling of adjacent cavities means that their photonic wave functions overlap . due to the strong interaction between photons and atoms , and the introduction of a lattice of coupled cavities , a strongly correlated phase emerges where photons are present . the light - matter models share some basic properties with the bose - hubbard ( bh ) model , @xcite such as the quantum phase transition from a mott phase , where particles are localized on the lattice sites , to a superfluid phase , where particles are delocalized on the whole lattice . @xcite yet the physics of the light - matter models is far richer because two distinct particles , namely , photons and atomiclike excitations , are present . a major advantage of these man - made realizations of strongly correlated many - body systems is that they can be tailored to correspond to a many - body model , whose parameters can be directly controlled in the experiment . furthermore local quantities , such as the particle density at a specific lattice site , can be addressed individually due to the mesoscopic scale of the cavities and both lattice size and geometry can be controlled in the fabrication process . an experimental realization of these light - matter systems is still missing but there are several promising approaches , such as photonic crystal cavities or toroidal and disk - shaped cavities . @xcite if light - matter systems can be realized , they will undoubtedly provide fascinating insight in the physics of strongly correlated many - body systems . the realizations might be used as quantum simulators for other quantum mechanical problems or even more intriguing for quantum information processing applications.@xcite recently , there has been a lot of research activity in the field of light - matter systems . most of the work has been devoted to investigate the quantum phase transition from the mott to the superfluid phase . some basic characteristics of the quantum phase transition have been evaluated from small systems of a few cavities by means of exact diagonalization.@xcite results are available at mean - field level @xcite as well or more accurately from analytical strong coupling perturbation theory calculations,@xcite and from simulations based on the density matrix renormalization group@xcite ( dmrg ) , the variational cluster approach@xcite and quantum monte carlo.@xcite spectral properties of light - matter systems have been investigated in refs . , , and . in the present paper , we study in detail the spectral properties of a one - dimensional light - matter system . in particular , we evaluate both photonic as well as atomic - excitation spectral functions . the investigation of both spectral functions allows us to characterize the polariton excitations in light - matter models . in addition to the spectral functions , we present densities of states , momentum distributions and spatial correlation functions . for completeness we also show the first two lobes delimiting the mott transition . this paper is organized as follows : in sec . [ sec : model ] , we introduce the light - matter model . section [ sec : method ] contains both the description of the numerical method as well as the exploration of the polaritonic properties . section [ sec : results ] is devoted to spectral properties of the light - matter system . here , we present our results for spectral functions , densities of states , momentum distributions and spatial correlation functions . finally , we summarize and conclude our findings in sec . [ sec : conclusion ] . from the great variety of possible theoretical descriptions of light - matter systems @xcite we concentrate on the simplest one , which consists of an array of cavities each of which contains a two - level system.@xcite the physics of the @xmath0-th cavity can be described by the jaynes - cummings ( jc ) hamiltonian , @xcite which for @xmath1 is given by @xmath2 where @xmath3 is the resonance frequency of the cavity , i.e. , the frequency of the confined photons , @xmath4 is the energy spacing of the two - level system , and @xmath5 is the atom - field coupling constant . the operator @xmath6 creates a photon with frequency @xmath3 , whereas @xmath7 annihilates one . the two - level system can be mathematically described by pauli spin algebra . thus , we identify the ground state of the two - level system with @xmath8 and the excited state with @xmath9 . with that the atomic raising operator is defined as @xmath10 and the atomic lowering operator as @xmath11 , respectively . in order to obtain the jc hamiltonian the rotating wave approximation , which is justified for @xmath12,@xcite has been assumed . the deviation between the resonance frequency and the energy spacing of the two - level system , @xmath13 , is termed detuning . for the jc hamiltonian the particle number @xmath14 is a conserved quantity , as @xmath15 = 0 $ ] . this is a consequence of the rotating wave approximation.@xcite the full model consists of an array of @xmath16 cavities , which form a lattice and hence we refer to this model as the jaynes - cummings lattice ( jcl ) model . due to the coupling of the cavities , photons are allowed to hop between neighboring lattice sites . this leads to the jcl hamiltonian @xmath17 where @xmath18 is the hopping strength and @xmath19 the chemical potential , which controls the total particle number @xmath20 of the system . the first sum with the angle brackets around the summation indices is restricted to nearest - neighbor sites . in the case of the jcl model , the particle number of a specific cavity @xmath21 is not conserved anymore . however , the total particle number @xmath22 is a conserved quantity . in summary , the jcl hamiltonian can be rewritten as @xmath23 from eq.([eq : spe : jclrewritten ] ) and from the fact that we consider the coupling strength @xmath5 as unit of energy , it follows that the physics only depends on three independent parameters , namely the hopping strength @xmath18 , the detuning @xmath24 and the modified chemical potential @xmath25 . in order to fulfill the condition for the rotating wave approximation the resonance frequency @xmath3 has to be large in comparison with the detuning @xmath24 , which can be always satisfied theoretically as solely the difference between the chemical potential and the resonance frequency appears in the grand - canonical hamiltonian @xmath26 . in order to investigate the properties of the jcl model , we use the variational cluster approach @xcite ( vca ) , which has been formulated for bosonic systems in ref . . previous work on the jcl model within vca was carried out in ref . . the basic concept of vca is that the grand potential @xmath27 is expressed as a functional of the self - energy @xmath28 and that dyson s equation for the exact green s function @xmath29 is recovered at the stationary point of the self - energy functional @xmath30$].@xcite in order to evaluate @xmath30 $ ] , the self - energy @xmath28 of the investigated system is approximated by the self - energy of an exactly solvable , so - called reference , system . in practice , this means that the self - energy @xmath28 becomes a function of the set @xmath31 of single - particle parameters of the reference system , i.e. , @xmath32 . for bosonic systems the approximated grand potential reads@xcite @xmath33 where primed quantities correspond to the reference system and @xmath34 is the noninteracting green s function . the stationary condition on @xmath35 is given by @xmath36 this condition can be evaluated numerically by varying some or all of the single - particle parameters . in order to guarantee that a given physical quantity ( such as the number of particles ) is thermodynamically consistent , it is necessary that @xmath27 is stationary with respect to the associated coupling constant ( here the chemical potential).@xcite therefore , varying @xmath3 ensures that the total number of photons is thermodynamically consistent . on the other hand , it would be advisable for a conserved quantity , i.e. , @xmath20 to be consistent as well . otherwise uncommon situations could occur . for example , as we show below , the total particle density @xmath37 , evaluated as a trace of the green s functions is not integer in the mott phase . this effect becomes stronger close to the tip of the mott lobe , see fig.[fig : comparison ] . the noninteger particle density , occurring when @xmath19 is not taken as a variational parameter , clearly introduces an uncertainty in the determination of the phase boundary . in principle , however , there is a formal difficulty in taking @xmath19 as a variational parameter . the problem is related to the coupling of @xmath19 with atomic excitations , which , in contrast to photons , can not be seen as noninteracting particles . this is , in general , not allowed within vca , whereby the reference system can differ from the physical one by a single - particle hamiltonian only . the solution is readily overcome by observing that the two - level atomic system can be mapped onto a hard - core boson model . in this way , @xmath19 couples to the total number of `` atomic '' bosons plus photons , i.e. , a noninteracting hamiltonian . the hard - core constraint simply becomes a local ( in principle infinite ) interaction , which is common to the reference and to the physical system . the mapping of the two - level excitations onto hard - core bosons is mathematically achieved by the following replacements @xmath38 this is valid provided one excludes states with double occupation of @xmath39 particles _ even as intermediate states_.@xcite with this mapping the jcl hamiltonian reads @xmath40 where we have formally implemented the hard - core constraint by introducing an infinite interaction for @xmath39 particles . in the restricted hilbert space of zero or one hard - core boson per lattice site , the matrix elements of the two representations are identical . in principle , states with higher occupation number @xmath41 have to be considered in the bosonic version as well . however , the occupation of such states would cost infinite energy and , therefore , they do not influence the energies obtained from the hilbert space sector with occupation numbers @xmath42.@xcite we have checked this aspect numerically for very large @xmath43 . it can also be verified easily when the sector of the hilbert space with @xmath44 is included perturbatively . these considerations can be straightforwardly extended to light - matter models with more than one atom or atomiclike structure ( with two relevant levels ) per cavity . in this case , one introduces a boson species for each atom and the hard - core constraint is enforced for each boson species . in our calculation we take both parameters @xmath3 and @xmath4 of the reference system as variational parameters ( @xmath19 is just a linear combination ) , which ensures thermodynamic consistency for the particle number of both species , and , consequently , of the total particle number . we show below , that varying both parameters instead of just @xmath3 provides an improvement in the accuracy of the phase boundaries for a given cluster size , see tab.[tab : spe : quality ] . the present formulation of the vca can not address the superfluid phase , and is , thus , restricted to the mott phase.@xcite outside of the mott lobes , peaks with negative ( positive ) spectral weight appear in the positive ( negative ) @xmath45 region , signaling the instability towards the superfluid phase . a treatment of the superfluid phase requires a nambu green s function treatment analogous to the fermionic case . the boundary of the mott phase could be , in principle , determined by this criterion . however , it is simpler ( and , of course , equivalent ) to identify it , for a given @xmath18 , as the region between the ground - state energies of the @xmath46 and @xmath47-particle states , which can be directly inferred from the single - particle spectral function . in vca , the reference system is chosen to be a decomposition of the total system into identical clusters , which means that the total lattice of @xmath16 sites is divided into clusters of size @xmath48 . mathematically this can be described by introducing a superlattice , such that the original lattice is recovered when a cluster is attached to each lattice site of the superlattice . the reference system defined on a cluster is solved by means of the band lanczos method . @xcite the initial vector of the iterative band lanczos method for the single - particle excitation term of the cluster green s function contains @xmath49 elements and is given by @xmath50 where @xmath51 is the @xmath52 particle ground state . for the single - hole excitation term the initial vector of the band lanczos method is obtained by replacing the creation operators in eq.([eq : init ] ) by annihilation operators . to evaluate the grand potential and the single - particle green s function of the original system we use the bosonic @xmath53-matrix formalism.@xcite this formalism yields the green s function @xmath54 in a mixed representation , partly in real space and partly in reciprocal space , see app . [ app : c ] . the matrix @xmath54 is of size @xmath55 and @xmath56 belongs to the first brillouin zone of the superlattice . due to the specific order of the creation operators in the initial vector of the band lanczos method we are able to extract the green s function for photons @xmath57 and the green s function for two - level excitations @xmath58 from @xmath59 in the following way @xmath60 where @xmath61 $ ] . the application of the periodization prescription proposed in ref . ( green s function periodization ) yields the fully @xmath62 dependent green s functions @xmath63 and @xmath64 . from that we are able to evaluate the single - particle spectral function @xmath65 the density of states @xmath66 and the momentum distribution @xmath67 where @xmath68 can be either @xmath69 for photons or @xmath70 for two - level excitations . we use the @xmath53-matrix formalism to evaluate the momentum distribution , since this approach yields particularly accurate results.@xcite furthermore we calculate the spatial correlation functions @xmath71 which just depend on the distance between two cavities @xmath0 and @xmath72 , i.e. , @xmath73 . notice that the poles of the hard - core boson green s function coincide with the poles of the two - level excitation green s function as the energies of both representations are identical . however , the hard - core boson green s function exhibits additional poles located at energies of the order @xmath74 . the additional poles which have finite weight result from the fact that excitations such as @xmath75 are in principal allowed but cost infinite energy , whereas the corresponding excitation @xmath76 is strictly forbidden . therefore , the single - particle correlation functions @xmath77 and @xmath78 differ only by contributions from frequencies of the order @xmath74 . yet it should be mentioned that the single - hole correlations function of hard - core bosons is not affected by these considerations as @xmath79 is always equivalent to @xmath80 . this also implies that the spectral weight of the poles with negative energy are identical for both representations and that the particle density of the two - level system is equal to the particle density of the hard - core bosons . in the following , we will always speak loosely about two - level excitation green s functions but we have to keep in mind that there are differences in the single - particle spectral weight of the hard - core boson and two - level excitation green s functions at infinite energies . in the next step , we want to investigate the polaritonic properties of the jcl model , which arise due to the coupling between the photons and the two - level excitations . adding a particle or hole to the many - body ground state may result in quasiparticle or collective excitations which are built up by the @xmath81-particle eigenstates of the many - body system entering the green s function . these many - body eigenstates for the infinite system can be extracted within the vca framework from the vca green s function . as shown in app . [ app : c ] , they are linear combinations of the particle and hole excitations of the cluster green s function weighted by the eigenvector matrix @xmath82 , defined in app . [ app : c ] . our goal is to describe the eigenvectors of the ( @xmath83)-particle hilbert space , which form the quasiparticle excitations of the green s function by polaritonic quasiparticles added to the exact @xmath52 particle groundstate @xmath51 . to this end , we introduce the polariton creation operators @xmath84 for particle excitations and @xmath85 for hole excitations as appropriate linear combinations of photons and two - level excitations @xmath86 [ eq : polcreationops ] it should be stressed that the hole creation operator is not the adjoint of the particle creation operator or its annihilation counterpart , which it would be in the case of noninteracting particles . as we will see , the coefficients or weights of the linear combinations @xmath87 and @xmath88 depend on the wave vector @xmath62 , the quasiparticle band @xmath89 , and additionally on the filling @xmath90 , which is not explicitly written in eq.([eq : polcreationops ] ) , since the filling dependence is not important for the present discussions . the normalized polariton quasiparticle states are defined by applying the polaritonic operators on the exact @xmath52 particle ground state @xmath51 yielding [ eq : xxx1 ] @xmath91 [ eq : polwavefu ] respectively . the normalization terms can be rewritten as [ eq : xxx2 ] @xmath92 [ eq : norm ] in eq.([eq : norm ] ) the vectors @xmath93 are defined as @xmath94 and @xmath95 are the overlap matrices of single - particle excitations and single - hole excitations , respectively . the overlap matrix for the hole excitations is given by @xmath96 where the static correlation functions are evaluated in the @xmath52 particle ground state @xmath51 . all quantities entering @xmath97 are correctly evaluated in the hard - core boson model as no excitations of the `` two - level bosons '' into the @xmath98 sector occur . for the particle case the situation is different , as we need to evaluate @xmath99 the term @xmath100 of the two - level system can not be directly evaluated in the hard - core boson model . using the commutator property @xmath101=0 $ ] for @xmath102 and the local anticommutation relation @xmath103 , we end up with an expression that only contains static correlation functions which can be computed correctly within the hard - core boson model @xmath104 in order to derive a formalism to construct the optimal polariton weights , we start out with the analysis of an exact eigenvector @xmath105 of the hamiltonian in the @xmath46-particle sector . for the sake of clarity we will suppress in the following considerations the index @xmath62 for all quantities , and the indices @xmath89 and @xmath106 for quasiparticle weights and wave functions . the optimality criterion in this case is clearly the overlap of the exact eigenvector with the approximate ( normalized ) vector given in eqs.([eq : xxx1 ] ) and ( [ eq : xxx2 ] ) @xmath107 where @xmath108 denote the components of the two - dimensional vectors , and @xmath109 and @xmath110 , see app . [ app : c ] . the maximization of @xmath111 leads to the generalized eigenvalue problem @xmath112 where the elements of the @xmath113 matrix @xmath114 are @xmath115 in eq.([eq : gen_ev ] ) we replaced @xmath116 by @xmath117 as the eigenvalues are just determined except for a constant @xmath118 , which will be specified later . as the eigenvalue corresponds to the value of the overlap squared @xmath119 , the deviation of the eigenvalue from one is a measure of the quality of the polariton approximation . it also points out that the eigenvector corresponding to the largest eigenvalue determines the optimal polariton coefficients . interestingly , @xmath120 is the contribution of the excitation @xmath121 to the corresponding spectral function , i.e. , its _ quasiparticle weight_. in general , the quasiparticle peak is a superposition of several exact many - body eigenstates . hence , the obvious generalization of the optimality criterion is to sum over all eigenstates @xmath121 , which contribute to the quasiparticle excitation @xmath89 . to this end we define an energy window @xmath122 in which the quasiparticle peak @xmath89 is located and we integrate the spectral density in this energy window resulting in @xmath123 the polariton coefficients are again obtained by the generalized eigenvalue problem @xmath124 and the eigenvalue is given by @xmath125 the eigenvalues are still restricted to the unit interval @xmath126 $ ] . the lower limit is due to the positivity of @xmath127 and @xmath128 . the upper limit follows from the property that a summation of the integrated spectral density over all nonoverlapping energy intervals @xmath122 is given by @xmath129 of course , @xmath130 and , hence , the polariton operators will depend on the wave vector @xmath62 , the quasiparticle band index @xmath89 and the filling @xmath90 , i.e. , the mott lobe . the discussion so far was for the particle case only , however , it is straightforward to iterate the procedure for the hole case . eventually , we merely need the integrated spectral density @xmath131 determined within the vca framework , which is given by @xmath132 details are presented in app . [ app : c ] as well as the proof that all contributions of the sum have the same sign , which is necessary for the optimality criterion to make sense at all . the optimality criterion as well as the eigenvalue problem only fix the coefficient vector @xmath133 up to a normalization factor @xmath118 , i.e. , @xmath134 . the latter is determined by the condition that the total spectral weight should be conserved @xmath135 as the excitations can now be described by wave vector , band and filling dependent polaritonic quasiparticles , it remains to evaluate the polariton spectral function @xmath136 , which is due to the invariance of the trace in eq.([eq : trconv ] ) equal to the sum of the photon spectral function @xmath137 and the two - level excitation spectral function @xmath138 . in this section , we present the results of our calculations . specifically , in sec . [ qpt ] , we discuss the quantum phase transition from mott phase to superfluid phase occurring in the jcl model and investigate the impact of the variational parameter space on the accuracy of the results . in sec . [ spec ] , we study the spectral properties of both photons as well as two - level excitations . the first two subsections refer to results obtained for zero detuning @xmath139 , whereas nonzero detuning is considered in the third subsection , sec . [ nonz ] . finally , in sec . [ pola ] , we study the polaritonic properties of the jcl model . in particular , we introduce polariton quasiparticles as wave vector and filling dependent linear combinations of photons and two - level excitations and analyze the weights of their constituents . the jcl model exhibits , comparable to the bh model,@xcite a quantum phase transition from a localized mott phase to a delocalized superfluid phase . for integer particle density and small hopping strength @xmath18 , the ground state of the system is a mott state . the first two mott lobes of the one - dimensional ( 1d ) jcl model for zero detuning @xmath139 obtained by means of vca with the variational parameters @xmath140 are shown in fig.[fig : pd ] . . ( a ) vca results for the variational parameters @xmath140 and various cluster sizes of the reference system . the gray shaded area indicates dmrg data.@xcite ( b ) phase boundaries obtained for the largest cluster ( @xmath141 for the first mott lobe and @xmath142 for the second mott lobe ) . the marks refer to parameters where spectral functions are evaluated . , scaledwidth=48.0% ] as discussed in the previous section , including @xmath4 in the set of variational parameters is nontrivial and is solely possible since the two - level excitations can be mapped onto hard - core bosons . the gray shaded area in fig.[fig : pd ] indicates dmrg results for the phase boundary obtained by d. rossini _ et al . _ in ref . . we find excellent agreement between the phase boundary evaluated by means of vca with the variational parameter set @xmath143 and the dmrg results , even at the lobe tip , where quantum fluctuation effects are most important , and even for moderate cluster sizes @xmath144 . figure [ fig : comparison ] compares the phase boundaries at the tip of the first mott lobe for different variational parameters . and @xmath145 , respectively , for small clusters of size @xmath146 and @xmath147 . ( a ) phase boundaries at the tip of the first mott lobe . the gray shaded area indicates dmrg results.@xcite ( b ) total particle density @xmath90 , which is the sum of the photon density and the two - level excitation density , across the first mott lobe . , scaledwidth=48.0% ] the results obtained with @xmath143 are connected by lines , whereas the open symbols correspond to @xmath145 . we observe that using both on - site energies as variational parameters improves the results for the phase boundary and also yields a better approximation for the slope of the lobe tip . a quantitative measure for the quality @xmath148 of the calculated phase boundary is given by the absolute deviation from the dmrg data per phase boundary point @xmath149 where @xmath150 and @xmath151 are corresponding phase boundary points calculated by means of vca and dmrg , respectively , and @xmath152 is the number of phase boundary points , which contribute to the sum . in tab.[tab : spe : quality ] we compare the quality @xmath153 of the phase boundary between the two sets of variational parameters for various cluster sizes . . quality @xmath153 of the phase boundary for @xmath154 and @xmath155 , respectively . the quality @xmath148 is evaluated using eq.([eq : spe : chiquality ] ) . [ cols=">,^ , < " , ] when using the augmented set of variational parameters @xmath155 in contrast to @xmath154 we observe an improvement in the quality of the phase boundary which ranges from @xmath156 to @xmath157 depending on the cluster size of the reference system . using both the resonance frequency @xmath3 of the cavities and the energy spacing @xmath4 of the two - level system as variational parameters thus provides a significant improvement with respect to the case of a single variational parameter.@xcite as discussed in refs . and , a correct particle density in the original system can only be obtained when the corresponding on - site energies are included in the set of variational parameters , i.e. , in the case of the jcl model @xmath140 . this is demonstrated in fig.[fig : comparison ] , where the total particle density @xmath90 , which consists of a photon and a two - level excitation contribution , is evaluated along the first mott lobe . for @xmath158 the deviation of the particle density from one is growing with increasing hopping strength @xmath18 but shrinking with increasing cluster size @xmath48 . however , when @xmath4 is included as variational parameter the total particle density @xmath90 is as desired equal to one across the whole first mott lobe . a deviation of about @xmath159 can be observed for @xmath160 . yet , the hopping strength @xmath160 is probably even slightly above the critical hopping strength @xmath161 , which indicates the tip of the mott lobe.@xcite the phase diagram of the 1d jcl model is in many aspects similar to the phase diagram of the 1d bh model.@xcite particularly , the mott lobes are point shaped and a reentrance behavior can be observed , which means that for certain values of @xmath19 upon increasing @xmath18 the system leaves the mott phase and later on enters it again . yet a very important difference is that the width of the lobes of the jcl model at zero hopping is shrinking with increasing particle density . this comes from the fact that the effective on - site repulsion of the jcl lattice model , which is hidden in the interaction between photons and two - level excitations , is not constant , as in the bose - hubbard model . the exact location of the phase boundaries at zero hopping is derived as a by - product in app . [ app : a ] , whose major intention is , however , to introduce the notation used for the dressed states @xmath162 and for the corresponding energies @xmath163 , where @xmath164 describing the ground state and the excited state in the corresponding constant particle number sector of the single - cavity hilbert space . the spectral function for photons @xmath137 , the spectral function for two - level excitations @xmath138 and the corresponding densities of states @xmath165 and @xmath166 evaluated by means of vca for parameters belonging to the first mott lobe are shown in fig.[fig : sflobe1 ] . , first row , and density of states @xmath165 , second row . two - level excitation spectral function @xmath138 , third row , and density of states @xmath166 , fourth row . the spectral functions are evaluated for the parameters ( a ) @xmath167 , @xmath168 , @xmath139 and ( b ) @xmath169 , @xmath170 , @xmath139 , which belong to the first mott lobe . the dashed lines in the spectral functions in ( a ) correspond to first - order degenerate perturbation theory results , see app . [ app : b ] . the roman numerals in the captions of the subfigures refer to the marks in fig.[fig : pd].,scaledwidth=48.0% ] we use an artificial broadening @xmath171 and the variational parameter set @xmath155 for the numerical evaluation of the spectral functions . both spectral functions @xmath137 and @xmath138 have the same gap as the photons and the two - level excitations are coupled . the spectral functions of the jcl model generally consist of four bands . this can best be understood in terms of the analytic solution of the jcl model for zero hopping strength @xmath172 . the ground state @xmath51 of the jcl model in the mott phase with particle density @xmath90 for zero hopping is given by the tensor product state @xmath173 where @xmath174 is the dressed @xmath90 particle ground state of lattice site @xmath121 . the states with a single - particle excitation are those , where @xmath175 sites remain in the dressed state @xmath176 and one site is excited to the state @xmath177 . similarly , for the single - hole excitation @xmath175 sites remain in the state @xmath176 and one site is excited to the state @xmath178 . in both cases , the excited states are @xmath16 fold degenerate as the particle / hole excitation can be located on any of the @xmath16 lattice sites . the degenerate states have thus the structure @xmath179 [ eq : psidegenerate ] respectively . two of the four bands , we refer to them as lower modes @xmath180 , emerge from the excitation of site @xmath0 from the dressed state @xmath181 to the states @xmath182 , which are ground states of the corresponding hilbert - space sector with constant particle number . analogously , we refer to the bands which emerge from the excitation of site @xmath0 from @xmath181 to the excited states in the corresponding particle sector @xmath183 as upper modes @xmath184 . the presence of the upper modes has been first noted by s. schmidt _ et al . _ in ref . and has been numerically observed in latest qmc calculations@xcite as well . the two upper modes @xmath184 indicate a clear deviation from the bh physics , which emerges due to the composition of two distinct particles . as discussed in the previous section , the two particle bands @xmath185 , @xmath186 , determine the polariton particle creation operators @xmath84 whereas the two hole bands @xmath187 specify the hole creation operators @xmath85 . in the spectral functions of fig.[fig : sflobe1 ] , the lower modes @xmath180 correspond to the cosinelike shaped bands centered around @xmath188 . the intensities of the lower modes @xmath180 are contrary for the photon spectral function @xmath189 and the two - level excitation spectral function @xmath190 . for @xmath189 the particle band @xmath191 is more intense than the hole band @xmath192 whereas the hole band is more intense than the particle band for @xmath190 . for the first mott lobe the upper hole mode @xmath193 does not exist as this would require to excite a single - site @xmath0 from the dressed state @xmath194 to the non - existing state @xmath195 . thus , only the upper particle mode @xmath196 can be observed in the spectral functions shown in fig.[fig : sflobe1 ] , which corresponds to the essentially flat band located at @xmath197 . [ app : b ] , we evaluate the single - particle and single - hole excitation bands by means of first - order degenerate perturbation theory , which yields @xmath198 [ eq : om ] respectively , where @xmath199 is the renormalized hopping strength . figure [ fig : sflobe1 ] shows , additionally to the spectral functions obtained by means of vca , the perturbation results for the bands . for small hopping strength we observe , as expected , good agreement between the two approaches . from the analytic solution of the bands we are able to extract their width , which is given by @xmath200 . the renormalization factor in @xmath199 essentially consists of a square of the form @xmath201 , see eqs.([eq : teffp ] ) and ( [ eq : teffh ] ) . evaluating these expressions shows that @xmath202 for the lower modes @xmath180 but @xmath203 and @xmath204 for the upper modes @xmath184 . therefore , @xmath205 and @xmath39 almost cancel each other in the latter case , which yields a small renormalized hopping strength of the upper modes @xmath206 in comparison to the one of the lower modes @xmath207 and thus , essentially flat upper particle / hole bands @xmath184.@xcite plugging in the value of the modified chemical potential @xmath208 , which has been used to evaluate the spectral function shown in fig.[fig : sflobe1 ] , into eq.([eq : omparticle ] ) yields @xmath209 , where we neglected the dependence on the wave vector . this matches perfectly with the vca results . in addition to previous work@xcite we evaluate the upper modes not only for photons but also for two level - excitations . interestingly , the spectral weight differs significantly for the two types of particles . in particular , the upper particle mode @xmath210 has a very large intensity in the two - level excitation spectral function @xmath211 , but is almost not visible in the photon spectral function @xmath137 . of the photon spectral functions @xmath137 shown in fig.[fig : sflobe1 ] , where the parameters ( a ) @xmath167 , @xmath168 , @xmath139 and ( b ) @xmath169 , @xmath170 , and @xmath139 have been used . vca results ( density plot ) are compared with bands evaluated by means of first - order degenerate perturbation theory ( dashed lines ) . , scaledwidth=48.0% ] for the spectral function shown in fig.[fig : sflobe1 ] a different chemical potential @xmath170 has been used . thus , the upper particle mode is shifted slightly upwards in comparison to fig.[fig : sflobe1 ] and is located at @xmath212 . figure [ fig : sfpert ] shows the lower particle band @xmath213 of the photon spectral function @xmath137 for the same parameters as in fig.[fig : sflobe1 ] . in this figure we compare the vca results for different hopping strengths with the results obtained by means of first - order degenerate perturbation theory . for small hopping strength , @xmath167 , see fig.[fig : sfpert ] , the perturbative results agree very well with the vca results in both the width as well as the shape of the band . however , for large hopping strength @xmath169 , which is already close to the tip of the mott lobe , the lower particle band does not exhibit a simple cosine shape anymore , see fig.[fig : sfpert ] . in addition the width of the band is slightly overestimated by first - order degenerate perturbation theory . in the spectral functions shown in fig.[fig : sflobe1 ] there is additional spectral weight located at @xmath214 . we can exclude that this additional weight stems from the periodization prescription used in vca or from any other vca internal processes as it also appears in the cluster green s function , which is solved by exact diagonalization . this can be verified best by comparing the density of states obtained from the vca green s function with the density of states obtained from the cluster green s function , see fig.[fig : doscluster ] . and ( b ) density of states of two - level excitations @xmath166 . the parameters used for these plots are the same as in fig.[fig : sflobe1 ] . , scaledwidth=48.0% ] , first row , and density of states @xmath165 , second row . two - level excitation spectral function @xmath138 , third row , and density of states @xmath166 , fourth row . the spectral functions are evaluated for the parameters ( a ) @xmath215 , @xmath216 , @xmath139 and ( b ) @xmath217 , @xmath218 , and @xmath139 , which belong to the second mott lobe . the roman numerals in the captions of the subfigures refer to the marks in fig.[fig : pd ] . , scaledwidth=48.0% ] both densities of states , the one obtained from the cluster green s function and the one obtained from the vca green s function , exhibit a peak located at @xmath219 . the additional peak can be revealed in the framework of perturbation theory . first - order local particle fluctuations in the ground state will have contributions of the form @xmath220 where @xmath221 correspond to nearest - neighbor sites . due to the energy denominator @xmath222 the predominant terms are those with @xmath223 . the correction term @xmath224 is proportional to the hopping strength @xmath18 , which explains , why the additional peak is not present in fig.[fig : sflobe1 ] . the particle excitation couples to final states with an additional particle either on site @xmath225 , @xmath226 or on one of the remaining sites . a detailed analysis shows that the excitation , responsible for the additional peak at about @xmath214 , is @xmath227 the corresponding excitation energy is given by @xmath228 for zero detuning and @xmath170 the energy is @xmath229 . as discussed before the upper hole mode @xmath193 does not exist in the first mott lobe . yet , the mode @xmath230 is present in spectral functions of the second mott lobe , see fig.[fig : sflobe2 ] . according to eq.([eq : om ] ) the upper modes are located at @xmath231 and @xmath232 for the parameters used in fig.[fig : sflobe2 ] . this matches very well the results obtained by means of vca . the chemical potential of the spectral function shown in fig.[fig : sflobe2 ] differs from the one of ( a ) merely about 0.01 . thus , the bands @xmath184 are located at rather the same position in both spectral functions . the momentum distribution for photons @xmath233 and two - level excitations @xmath234 in the first and second mott lobe are shown in fig.[fig : n ] . . solid lines correspond to the momentum distributions of photons @xmath233 and dashed lines to the momentum distributions of two - level excitations @xmath234 . , scaledwidth=48.0% ] for increasing hopping strength @xmath18 the momentum distribution becomes more peaked for both the photons and the two - level excitations . in the first mott lobe the momentum distributions @xmath233 and @xmath234 are centered around @xmath235 , which means that the cavities are on average equally occupied by photons and two - level excitations . in the second mott lobe @xmath233 is centered around @xmath236 . however , @xmath234 is still centered around @xmath235 , as the maximum local occupation number of the two - level systems is restricted to one . in order to display the slowing down of correlations upon approaching the boundary of the mott phase , we evaluate the spatial correlation function @xmath237 in the first mott lobe ( fig.[fig : clobe1 ] ) . . , scaledwidth=48.0% ] the spatial correlation function can be obtained from the fourier transform of the momentum distribution . for small distances @xmath238 between sites @xmath0 and @xmath72 the correlation function is a superposition of multiple exponential functions with distinct strengths of decay . for large distances , however , the exponential function with the smallest decay dominates and thus the correlation function is of the form @xmath239 as expected in the insulating phase . using vca we are able to extract the correlation length @xmath240 , as data are available for large distances between two sites @xmath0 and @xmath72 . from a linear fit for sufficiently large distances we obtain @xmath241 for the parameters i , see marks in fig.[fig : pd ] , and @xmath242 for the parameters ii . therefore , the slope of the correlation function is the same for the two particle species , which is due to the coupling between the photons and the two - level excitations . as in the bh model @xcite the absolute slope @xmath243 of the correlation function shrinks with increasing hopping strength , which is a precursor of the superfluid phase , where the correlation between sites persists up to long distances . the detuning @xmath24 , which is the difference between the resonance frequency @xmath3 of the cavities and the energy spacing @xmath4 of the two - level systems , is a very important parameter of the jcl model . by varying the detuning it is possible to change the width of the mott lobes . phase boundaries obtained by means of vca with the set of variational parameters @xmath244 for @xmath245 and @xmath246 are shown in fig.[fig : pddet ] . and ( b ) @xmath246 . the marks refer to the parameters where spectral functions are evaluated . , scaledwidth=48.0% ] for the parameters marked with @xmath31 we evaluate the spectral function of photons @xmath137 and two - level excitations @xmath138 , see fig.[fig : sfdet ] . , first row , and density of states @xmath165 , second row . two - level excitation spectral function @xmath138 , third row , and density of states @xmath166 , fourth row . the spectral functions are evaluated for the parameters ( a ) @xmath247 , @xmath248 , @xmath245 and ( b ) @xmath160 , @xmath249 , @xmath246 . the roman numerals in the captions of the subfigures refer to the marks in fig.[fig : pddet ] . , scaledwidth=48.0% ] an interesting effect can be observed in the spectral functions @xmath138 . namely , the intensity of the upper band @xmath210 depends significantly on the detuning @xmath24 . for negative detuning @xmath245 , the upper mode in @xmath138 is very intense , see fig.[fig : sfdet ] , whereas it is almost not visible for positive detuning @xmath246 . this behavior remains valid when the spectral functions for positive and negative detuning are evaluated for identical hopping strength . the zero - hopping result for the energy of the upper mode is @xmath250 for the spectral function shown in fig.[fig : sfdet ] and @xmath251 for the spectral function shown in fig.[fig : sfdet ] . the momentum distributions of photons @xmath233 and two - level excitations @xmath234 for the parameters marked in fig.[fig : pddet ] are shown in fig.[fig : ndet ] . , where ( a ) corresponds to the parameters v , i.e. , negative detuning @xmath252 and ( b ) to the parameters vi , i.e. , positive detuning @xmath246.,scaledwidth=48.0% ] for negative detuning it is energetically more expensive to excite the two - level system than to add a photon to the cavity . thus , the momentum distribution of photons @xmath233 dominates over the momentum distribution of two - level excitations @xmath234 . for positive detuning the situation is reversed and @xmath234 is larger than @xmath233 for all values of the momentum . up to now we investigated the photon properties and the two - level excitation properties of the jcl model separately , by extracting the green s function of photons @xmath253 and the green s function of two - level excitations @xmath254 from the compound green s function @xmath255 , which is a @xmath256 matrix of the form @xmath257 next we will discuss the polaritonic properties of the jcl model . we start out with the first mott lobe for zero detuning and focus again on the parameter set marked as ii in fig.[fig : pd ] , i.e. , @xmath169 , @xmath170 and @xmath139 . the polaritonic spectral function @xmath136 and the corresponding density of states @xmath258 , which is by construction identical to the total density of states of photons plus two - level excitations , is shown in fig.[fig : sfdospollobe1 ] . , @xmath170 , and @xmath259 . in ( b ) the polariton density of states @xmath258 is compared with the sum of the photon density of states @xmath260 and the two - level excitation density of states @xmath261 , which coincide by definition . , scaledwidth=48.0% ] for the first mott lobe the hole case is special since both , @xmath262 and @xmath263 yield the exact zero - particle state . consequently , the polariton can be chosen ad libitum , it will always be exact . therefore in fig.[fig : qwweightsslobe1 ] only the particle part of the polaritonic weights is depicted . and two - level excitation contribution @xmath264 to the polariton quasiparticles . ( a ) shows the results for @xmath265 corresponding to the upper particle band @xmath210 and ( b ) for @xmath266 corresponding the lower particle band @xmath213 . additionally to the weights @xmath267 and @xmath264 the overlap @xmath268 is shown . , scaledwidth=48.0% ] the right panel represents the result for the lower particle excitation . the polariton has very pronounced photonic character and the weights of photons and two - level system have opposite sign . interestingly , the lower particle excitation can very well be mimicked by a single polariton on top of the @xmath52-particle ground state , as can be inferred from the fact that @xmath269 . moreover , a slight @xmath62-dependence of the weights is observed . contrarily in the upper particle band , the polariton has pronounced two - level - system character , the weights have the same sign , there is almost no @xmath62-dependence , and the polariton description is poor ( @xmath270 ) . now we turn to the second mott lobe , which allows us to study the hole polariton as well . the polariton spectral function and the corresponding density of states evaluated for the parameters iv , i.e. , @xmath217 , @xmath218 , and @xmath139 , are shown in fig.[fig : sfpol ] . , @xmath218 , and @xmath139 . , scaledwidth=48.0% ] the weights are shown along with the overlap @xmath268 in fig.[fig : qwbands ] . the lower bands @xmath180 are well described by the quasiparticles as the overlap @xmath268 is almost one for both bands . the upper bands @xmath184 , however , are not described that well . in particular @xmath271 for the upper particle band and @xmath272 for the upper hole band . the weights @xmath267 and @xmath264 are significantly more wave vector dependent , especially for the upper bands @xmath184 , i.e. , @xmath273 . apart from the more pronounced @xmath62-dependence , the weights for the particle case are rather similar to those of the first mott lobe . however , there are striking differences in the weights for the particle and hole part within the second mott lobe . first , the @xmath62 dependence is more pronounced . second , the sign of the relative weights is positive for both bands @xmath274 , and finally , the composition of the polariton in the two bands is reverse . the lower band has predominantly photonic character , while opposite holds for the upper band . eventually , we want to compare the vca results with those of the single - site problem , which are derived in app . [ app : d ] . in the single - site problem the sign of the relative polaritonic weights is the same as that observed in the lattice . in the first mott lobe the relative weights for the particle case are for the upper band @xmath275 and for the lower band the reciprocal relation holds @xmath276 . there is agreement in the relative signs and the composition of the polariton between the single - site limit and the lattice system , but the reciprocal property is strongly violated in the lattice case . this might be understood as follows . the itinerant particles are the photons . in order to gain kinetic energy it is convenient for the system to increase the photonic character in the dispersive lower band , depicted in fig.[fig : sfpol ] . the upper band , on the other hand , has little dispersion and behaves more like the single - site limit . and two - level excitation contribution @xmath264 to the polariton quasiparticles . ( a ) shows the results for @xmath265 corresponding to the upper particle band @xmath210 , ( b ) for @xmath277 corresponding the lower particle band @xmath213 , ( c ) for @xmath278 corresponding to the lower hole band @xmath279 and ( d ) for @xmath280 corresponding to the upper hole band @xmath230 . additionally to the weights @xmath267 and @xmath264 the overlap @xmath268 is shown . , scaledwidth=48.0% ] in the second mott lobe , the relative weights obtained in the single - site limit for particle excitations are @xmath281 and @xmath282 . like in the lattice case , the weights of the particle part are comparable in the first and second mott lobe . quantitatively , the relative weight @xmath283 is roughly @xmath284 larger in the second mott lobe , which is also the case in the lattice system . as far as the hole part is concerned , the single - site limit nicely corroborates all observations of the lattice model . in the single - site problem , the exact many - body eigenstates @xmath285 can be generated exactly by suitable polariton operators acting on the state @xmath286 . this is no longer the case in the lattice due to local particle number fluctuations induced by particle motion . already in the single - site limit , the polariton operators are , however , not universal , they depend on the filling @xmath90 and in the lattice case even on the wave vector @xmath62 . on top of that , the polariton operator for holes is not the adjoint of the corresponding polariton creation operator of the particle type , or in other word its annihilation operator . in this paper we presented and discussed the spectral properties of the jaynes - cummings lattice model in one dimension obtained within the variational cluster approach . using the resonance frequency @xmath3 of the cavities and the energy spacing @xmath4 of the two - level systems as variational parameters in the variational cluster approach procedure provides a significant improvement with respect to the case of a single variational parameter . on the one hand , varying both @xmath3 as well as @xmath4 ( or , at least @xmath19 ) is necessary to achieve a correct particle density in the original system and on the other hand improved results for the phase boundaries , and thus , for the spectral functions as well , are obtained due to the augmented set of variational parameters . in order to apply the variational cluster approach and include @xmath4 as variational parameter the two - level systems have been mapped onto hard - core bosons , which yields correct poles of the green s function in the relevant energy range . we evaluated and discussed spectral functions for photons and two - level excitations . the spectral functions generally consist of four bands , cosinelike shaped lower particle / hole bands , which are centered around zero energy , and essentially flat upper particle / hole bands . an exception are the spectral functions in the first mott lobe , which contain the two lower bands but from the upper bands only the particle part . using first - order degenerate perturbation theory , we evaluated analytical expressions for the bands , which allowed us to explain why the upper modes are essentially flat whereas the lower modes exhibit a pronounced cosinelike shape . additionally , we compared the analytical solution for the bands with the variational cluster approach results . for small hopping strength @xmath18 we observe , as expected , good agreement between the two approaches . however for parameters located close to the tip of the mott lobe , first - order degenerate perturbation theory yields results that differ from the exact ones in both , shape and width of the bands . furthermore , we evaluated densities of states , momentum distributions and spatial correlation functions for photons and two - level excitations . we also investigated detuning effects on the spectral properties and found indications that the intensity of the upper particle band of the two - level excitation spectral function depends strongly on the detuning . based on the information obtained from the photons and two - level excitations we investigated the polaritonic properties of the jaynes - cummings lattice model . therefore we introduced wave vector and filling dependent polariton particle creation and hole creation operators , which are linear combinations of photon and two - level excitation creation operators . we evaluated spectral functions and densities of states based on the polariton quasiparticles and analyzed the weights of their constituents . we have seen that the polariton operators are nontrivial combinations of photon and two - level system operators , which depend on the wave vector , the quasiparticle band , and the filling , or rather the mott lobe . on top of that , the polariton operators of particle and hole type are not adjoint operators . it is therefore not possible to describe the jcl model by a simple single - band polariton model . here , we will derive some properties of the vca green s function for bosons . to begin with , we define new operators @xmath287 with @xmath288 and @xmath289 , where @xmath290 stands for the site number within the clusters and @xmath56 is the wave vector of the first brillouin zone of the superlattice . the vca green s function will still be diagonal in the latter index due to the periodicity of the clusters . the spectral representation of the cluster green s function @xmath291 can be written in the compact form using the so - called @xmath53-matrices @xcite @xmath292 here , @xmath53 is a @xmath293 matrix , where @xmath294 is twice the number of cluster sites ( the factor 2 stems from the two species of operators ) and @xmath295 , where @xmath296 and @xmath297 is the dimension of the hilbert space for @xmath298 and @xmath299 particles , respectively . the @xmath53-matrix is defined as follows @xmath300 the diagonal matrix @xmath301 contains the individual poles @xmath302 of the cluster and @xmath303 is a diagonal sign matrix with @xmath304 for particle excitations ( @xmath305 ) and @xmath306 for hole excitations ( @xmath307 ) . the vca green s function in @xmath53-matrix representation for bosons @xcite reads @xmath308 where @xmath309 is the diagonal matrix of the individual poles at the vca energies . these energies and the corresponding eigenvector matrix @xmath82 are determined via the generalized eigenvalue problem @xmath310 where @xmath311 is the difference of the matrices of the single - particle part of the hamiltonian for the original and the reference system ( i.e. , the cluster ) . a general feature of such eigenvalue equations for hermitian matrices @xmath312 is that both @xmath313 and @xmath314 are diagonalized , but @xmath82 is not unitary . we can exploit this fact as follows @xmath315 the matrix @xmath316 is diagonal as we just discussed and can be combined with @xmath317 resulting in @xmath318 moreover , the pole strengths @xmath319 are real since @xmath320 when the vca parameters are determined consistently , the stability of the @xmath52-particle system requires that the sign of @xmath319 coincides with the sign of the excitation energies @xmath321 , like in the exact spectral representation . so far the green s function still depends on the intra cluster indices @xmath322 . the purely @xmath62-dependent green s function is commonly obtained by green s function - periodization.@xcite invoking the periodization prescription yields the green s function matrix merely in the indices @xmath323 for the two particle species @xmath324 see also eq.([eq : gf2by2 ] ) . equation ( [ eq : vcaxxx ] ) corresponds to the spectral representation of the exact green s function and it allows to extract the vca approximation of the many - body eigenstates of the infinite system , which are obviously a linear combination of the cluster eigenstates for both , particle and hole excitations . as described in the text we need the integrated spectral density , i.e. , @xmath325 we readily recognize , that the integrated spectral density is either positive or negative definite , depending on whether the quasiparticle under consideration is of particle or hole type . equivalently , in the original representation @xmath326 for the polariton discussion it is convenient to suppress the minus sign arising in the hole case and we define the strictly positive integrated spectral densities as @xmath327 for zero - hopping strength @xmath172 the jcl model can be solved exactly , as it reduces to a single - site problem , i.e. , to the jc model . including the chemical potential yields the single - site hamiltonian @xmath328 where we dropped the site index @xmath0 . it can be evaluated with respect to the bare states @xmath329 , where @xmath330 is the number of photons and @xmath331 . next , we sketch the most important steps for solving the single - site jcl model . a detailed discussion can be found for example in refs . or . as the jc hamiltonian conserves the particle number the hamiltonian @xmath332 is block diagonal . each block corresponds to a certain particle number @xmath90 and thus we use the bare states @xmath333 and @xmath334 to evaluate the block , which yields @xmath335 when using as denoted in sec . [ sec : model ] the coupling @xmath5 as unit of energy . the eigenvalues of the block @xmath336 are @xmath337 where @xmath164 and @xmath338 . for a certain particle number @xmath90 the energy @xmath339 is always smaller than @xmath340 and thus @xmath339 is the ground state energy of the sector with @xmath90 particles , i.e. , of the block @xmath336 . the eigenvectors @xmath162 of the matrix @xmath336 are termed dressed states and are given by @xmath341 where @xmath342 , @xmath343 and @xmath344 with the following relations @xmath345 and @xmath346 . an exception is the bare state @xmath347 , which forms a @xmath348 block of zero particles and has the eigenvalue @xmath349 , independently of the detuning @xmath24 . according to the notation used in eq.([eq : dressedstates ] ) , we denote this state as @xmath350 . in order to obtain the phase boundary for zero hopping between two adjacent mott lobes , the energies @xmath339 and @xmath351 have to be set equal . the energies of the states @xmath352 are used , as the phase diagram is evaluated for the ground state . the comparison of the energies yields @xmath353 for the location of the phase boundary at zero hopping . in this appendix we evaluate the results of first - order degenerate perturbation theory for the single - particle and single - hole excitation bands of the jcl model . to apply first - order degenerate perturbation theory the matrix elements of the perturbation @xmath354 , where @xmath355 is the hopping matrix , have to be evaluated with respect to the degenerate states @xmath356 and @xmath357 , see eq.([eq : psidegenerate ] ) . as the hopping term @xmath358 does not change the total particle number and does not effect the excitation @xmath89 , the following two matrices have to be evaluated ; one for single - particle excitations @xmath359 and one for single - hole excitations @xmath360 plugging eq.([eq : psi1 ] ) in eq.([eq : mp ] ) yields @xmath361 due to the orthogonality of the eigenvectors of sectors with different particle number , the conditions @xmath362 and @xmath363 hold , which reduce the matrix elements to @xmath364 in the second step , we dropped the site index as the expectation value does not depend on the specific lattice site . the corrected matrix elements are thus the old ones with renormalized hopping strength @xmath365 analogously , one obtains @xmath366 for the matrix elements defined in eq.([eq : mh ] ) . from that the renormalized hopping strength for single - hole excitations is evaluated as @xmath367 the eigenvalues of the matrices @xmath368 are the first - order corrections and thus the corrected energies @xmath369 of the one - dimensional jcl model are given by @xmath370 respectively , where @xmath371 is a wave vector of the first brillouin zone . within first - order degenerate perturbation theory we obtain @xmath372 for the single - particle excitation band and @xmath373 for the singe - hole excitation band . in this appendix , we want to analyze the polaritonic feature in the single - site limit for zero detuning . in the single - site limit it is exactly possible to construct a polariton operator which , applied to the many - body eigenstate @xmath286 , generates the eigenstates @xmath285 . the polaritonic weights follow from @xmath374 here , we explicitly include the filling @xmath90 as index . so the relative weights are @xmath375 for the upper band ( @xmath273 ) and @xmath376 for the lower band ( @xmath377 ) . this means that in the lower band the weights have opposite sign and the polaritons are of predominant photonic character , while the opposite applies to the upper band . the modulus of relative weight is just the inverse , i.e. , @xmath378 . next , we study the hole case for @xmath98 @xmath379 which is positive for both bands @xmath274 . again we have the reciprocal property @xmath380 and the lower band has predominantly photonic character , while the opposite is the case in the upper band . now we want to scrutinize the generalized eigenvalue problem of the green s function . the single - site green s function reads @xmath381 where we introduced the operators @xmath382 and @xmath383 . for the single - particle term we obtain @xmath384 with the definition @xmath385 the integrated spectral density for the particle part can be expressed as @xmath386 the overlap matrix @xmath387 is readily obtained by the spectral theorem @xmath388 and the generalized eigenvalue problem for the polariton weights according to eq.([eq : gen_ev ] ) reads @xmath389 the eigenvalues are zero and one . for the polariton weights we are interested in the latter . the corresponding eigenvector is simply given by the vector orthogonal to @xmath390 @xmath391 with that one obtains for the ratio of the weights @xmath392 which is in agreement with the exact result in eq.([eq : appyy ] ) . now we address the hole case , again for @xmath98 , @xmath393 we proceed as in the particle case with the definition of @xmath394 . the remaining steps are the same as before and we end up with @xmath395 which is also in agreement with the exact result , see eq.([eq : appyy2 ] ) . so we see that the determination of the polaritonic weight via the generalized eigenvalue problem is reasonable . in the single - site limit , the exact many - body eigenstates @xmath285 can be generated correctly by suitable polariton operators acting on the state @xmath286 . the operators are , however , not universal , they depend on @xmath90 and in the lattice case even on @xmath62 . on top of that , the polariton creation operator for holes is not the adjoint of the corresponding polariton creation operator of the particle type , or in other words its annihilation operator . we are grateful to d. rossini for providing us the dmrg data of the phase boundaries used in figs.[fig : pd ] and [ fig : comparison ] . we made use of parts of the alps library ( ref . ) for the implementation of lattice geometries and for parameter parsing . we acknowledge financial support from the austrian science fund ( fwf ) under the doctoral program `` numerical simulations in technical sciences '' w1208-n18 ( m.k . ) and under project no . p18551-n16 ( e.a . ) .

we present a scheme for the deterministic generation of-photon fock states from three - level atoms in a high - finesse optical cavity . the method applies an external laser pulse that generates an-photon output state while adiabatically keeping the atom - cavity system within a subspace of optically dark states . we present analytical estimates of the error due to amplitude leakage from these dark states for general , and compare it with explicit results of numerical simulations for . the method is shown to provide a robust source of-photon states under a variety of experimental conditions and is suitable for experimental implementation using a cloud of cold atoms magnetically trapped in a cavity . the resulting-photon states have potential applications in fundamental studies of non - classical states and in quantum information processing .

quant-ph0211134

the generation of non - classical states of light has been central to the confirmation and elucidation of the quantum theory of radiation . today , such work takes on an added importance as part of the advancing field of deliberate quantum state engineering , motivated in part by applications in quantum communication and computation . for instance , deterministically generated single - mode single - photon states could greatly advance the efficiency and security of quantum cryptography.@xcite they are also a crucial resource for implementing a quantum computing scheme using linear optics , single - photon states , and photodetection @xcite . these immediate potential applications have spurred the development of devices which can produce single photons on demand , such as solid state devices which use the coulomb interaction between strongly confined electrons to produce single exciton states which then decay optically @xcite , or devices in which the fluorescence from single , isolated and optically - pumped molecules is collected @xcite . the stream of pulses produced from each of these devices has been shown to contain either zero or one , but rarely more than one , photon per pulse , thus differing radically from a classical coherent - state pulse which would contain a poisson distribution of photon number . furthermore , while these recent devices all produce fluorescence from a single optical emitter on - demand , none outputs this fluorescence into a practical single mode of the optical field . it has been shown recently that the effects of cavity quantum electrodynamics ( cqed ) can in principle be used to overcome this limitation and produce single - mode single photons on demand @xcite . in such a scheme , a single three - level atom is induced to fluoresce with high probability into a single resonant mode of a high - finesse optical cavity . experimental evidence for such cavity - induced raman transitions has been obtained @xcite , and a variant of this scheme has very recently been used to generate a sequence of single photons @xcite . by creating a highly controlled , single - quantum - level interaction of atoms and light , cqed can be used quite generally to produce non - classical states of the electromagnetic field in a single cavity mode . in this paper , we analyze a particular desired function of a cqed device that goes beyond the production of single- photon states , namely , the deterministic production of a fock state of the electromagnetic field containing an exact number of photons ( @xmath0 ) . such non - classical states are of interest for fundamental tests and applications of the theories of quantum optics ( such as quantum state tomography , as performed recently on the single - photon state @xcite ) , as a resource for heisenberg - limited quantum measurements made possible by the production of two orthogonal @xmath0-photon states @xcite , and quite generally as a starting point for the controlled engineering of more complex quantum states . starting with a precisely counted number of @xmath0 atoms trapped in the confines of a high - finesse optical cavity , we consider a scheme in which a classical pump field is ramped up to induce deterministic raman emission into a single cavity mode by each of the trapped atoms , resulting in an optical field of exactly @xmath0 photons that are emitted from the cavity in a single pulse . the atom - cavity system is constrained to remain within a subspace of optically dark @xmath0-atom states , resulting in a high fidelity of production . the present scheme provides a generalization of that proposed for the production of single photons from single atoms @xcite and indicates a systematic route to generation of other non - classical states . several other theoretical and experimental works have discussed the use of high - finesse cavities for the quantum engineering of mesoscopic non - classical optical states . the possibility of producing both fock states and arbitrary coherent superpositions of these inside a cavity by exploiting adiabatic transfer of atomic ground - state zeeman coherences in single atoms was already explored in ref . @xcite . a scheme for producing an arbitrary quantum optical state using a single two - level atom in a high - finesse cavity has also been presented by law and eberly , based on the arbitrary real - time control of a classical pump field and the coupling to a cavity field @xcite . experimental evidence for fock states of a microwave cavity field has recently been obtained as a dynamical equilibrium for a stream of rydberg atoms passing through a micromaser @xcite . another approach toward the construction of a fock state was proposed in ref . @xcite in which a rydberg atom with a stark - tunable level splitting is used to transfer photons one - by - one from a classically - populated cavity field to another initially empty cavity field . initial experimental steps towards this goal have been recently demonstrated @xcite . in contrast to the schemes of refs.@xcite which require delicate temporal control of the atom - cavity coupling , our present scheme yields the desired @xmath0-photon state for quite arbitrary temporal profiles of the classical pump field . this is achieved through a rapid adiabatic passage which transfers the initial `` non - classical '' state of @xmath0 atoms to the non - classical state of @xmath0 photons within a short period of time . thus , the quantum nature of the photon field is already guaranteed by the initial state of the atoms trapped in an initially empty cavity . the initial conditions can be achieved by lowering a cloud of cold atoms into the cavity , and pulse control is then entirely contained in the time dependence of the ramping field . this avoids the need to make use of atomic motion in controlling the coupling to the cavity field , as was required in ref . @xcite , and is one of the key elements allowing _ deterministic _ production of @xmath0-photon states to be achieved . the number of atoms in the cavity can be determined by detection of excited state atoms at the single - atom level , as a result of the cavity - mediated shifts of atomic levels . such single - atom detection has already been demonstrated @xcite , and can be readily extended to larger numbers of atoms . consequently , the present scheme opens the way to deterministic generation of more complicated quantum states of light by first producing non - classical states of trapped atoms ( such as spin - squeezed states produced through interatomic interactions @xcite or by measurement @xcite ) , and then transferring that state onto the optical field using cqed . the remainder of the paper is constructed as follows . a brief review of the deterministic single - photon generation schemes of refs . @xcite is given in section [ sec : single ] , which establishes some common concepts with the present work . a discussion of the @xmath0-atom / cavity system in a single - mode external field follows in section [ sec : nphoton ] . we demonstrate here the existence of a family of optically - dark coupled @xmath0-atom / cavity states and show generally how adiabatic ramping of an external field can be used to generate @xmath0-photon emission from the cavity . detailed analysis of the energy spectrum of the closed @xmath0-atom / cavity system as a function of the ramp time provides estimates of the populations in the bright states and also leads to analytic estimates of the energy gap required for limiting adiabatic state transfers . in section [ sec : open ] we then present analysis of the cavity decay responsible for the @xmath0-photon emission , treating in detail the effects of spontaneous emission and non - adiabaticity on the output states . we obtain analytic estimates of error rates deriving from these contributions that scale linearly in the total number of atoms , resulting in a constant relative error in the output @xmath0-photon states and guaranteeing production of an @xmath0-photon state with high fidelity . in section [ sec : quantumtraj ] we present numerical simulations of the open system using the quantum jump approach . these numerical calculations are used to explore the sensitivity of the scheme to critical experimental parameters , as well as to explore the limits of our analytical estimates of the error bounds . finally , in section [ sec : conclusions ] we summarize and indicate directions for further work and for experimental implementation . our @xmath0-photon generation scheme uses the same internal atomic structure used in the single - photon proposals of refs . @xcite , namely atoms having three internal levels , labeled @xmath2 , @xmath3 and @xmath4 , arranged in a @xmath5 configuration such that states @xmath2 and @xmath4 are non - radiating atomic states while @xmath3 is an excited state connected to states @xmath2 and @xmath4 by allowed transitions ( figure [ fig : three_level_atom ] ) . levels @xmath2 and @xmath3 are typically hyperfine levels of the electronic ground state . we assume that the allowed transitions to state @xmath3 can be addressed selectively . this can be achieved , _ e.g. _ , as a result of polarization selection rules , or merely due to a large energy difference between states @xmath2 and @xmath4 . the three - level atoms are located in a high - finesse optical cavity which supports a resonant mode having vacuum rabi frequency @xmath6 ( _ i.e. _ , the rabi frequency due to the presence of a single photon in the cavity mode ) that couples states @xmath3 and @xmath4 . the cavity mode has frequency @xmath7 which can in general be detuned from the atom resonance @xmath8 by @xmath9 ( see figure [ fig : three_level_atom ] ) . in the schemes of refs . @xcite and @xcite , single - photon generation is accomplished by exposing a single atom in internal state @xmath2 to a classical laser field of frequency @xmath10 and rabi frequency @xmath11 , which is controlled dynamically . the laser frequency , @xmath10 , is chosen to be resonant with the cavity - mediated raman transition between states @xmath2 and @xmath4 , _ i.e. _ , @xmath12 or @xmath13 . this laser connects the states @xmath2 and @xmath4 through a cavity - mediated raman transition and induces the fluorescence of a cavity photon by the atom . the photon exits the cavity into a single cavity - output mode , and thereby generates the desired single - photon state . @xcite and @xcite showed that under suitable conditions on the external pulse field , the single photon can be emitted deterministically . to understand the operation of such a deterministic device , it is helpful to first consider the atom / cavity system as a closed quantum system , _ i.e. _ , we ignore the decay of cavity photons to cavity - output modes that actually produces the desired fock state outside the cavity , as well as the possible spontaneous decay from the excited state @xmath3 to modes other than the cavity mode . spontaneous emission lowers the fidelity of deterministic photon generation , and clearly needs to be avoided or at least minimized . we write the basis states of the single atom / cavity system as @xmath14 , @xmath15 and @xmath16 , where the first index refers to the atomic state and the second index gives the number of photons in the cavity , _ i.e. _ , the cavity field is implicitly assumed to be quantized . the interaction hamiltonian of this closed system is given by @xmath17 \label{eq : hint}\ ] ] where @xmath18 is the time - dependent coupling to the external ( classical ) laser pump field , @xmath19 . unless essential for the analysis , we shall omit the explicit time dependence of @xmath20 to streamline the notation . as discussed by kuhn _ et al . _ @xcite , the dynamics of this system are governed by the presence of a null - valued eigenstate @xmath21 , which is a `` dark state '' containing no population in the excited atomic state @xmath3 , and which is therefore immune to spontaneous decay . this dark state exists for all values of @xmath11 and is given by @xmath22 it is the presence of this dark state that allows the high - fidelity generation of a single photon in response to a suitable choice of @xmath11 . the single - atom / cavity system starts initially in the state @xmath23 , which is the dark state for the initial condition @xmath24 . during a sufficiently slow ramp of @xmath11 ( _ i.e. _ , a broad pulse of the classical pump laser ) , the atom - cavity system can adiabatically follow the dark state , eq . ( [ eq : dark_single ] ) . for sufficiently large values of @xmath11 ( @xmath25 ) , @xmath26 , _ i.e. _ , a single cavity photon is produced with a high degree of certainty . this cavity photon then rapidly decays from the cavity , resulting in a deterministic single - photon source that acts within a time interval specified by the period of the external pump . law and kimble @xcite and kuhn _ _ et al.__@xcite have presented numerical calculations to assess deviations from this ideal behaviour caused by spontaneous emission , and have explored the extent to which the photon emission probability can be controlled by modifying the trigger pulse . recent experiments by kuhn , hennrich , and rempe have produced single photons by this method @xcite . for the generation of arbitrary fock states of the electromagnetic field , _ i.e. _ , with arbitrary large photon number @xmath0 , we now consider @xmath0 such three - level atoms confined within the optical cavity . we assume that the atoms are indistinguishable in all respects . this provides an important experimental simplification relative to other cqed schemes in which the atoms are required to be individually addressable @xcite . each atom interacts individually with the laser field and the cavity mode , just as for the single atom case . we make the simplifying assumption that the atoms do not interact directly with one another . however , they do experience a second order interaction via the cavity mode . the cavity coupling @xmath6 and the classical pump rabi frequency @xmath20 are taken to be identical for each of the @xmath0 atoms , consistent with their indistinguishability @xcite . thus the @xmath0-atom / cavity interaction hamiltonian is given , for the closed system , by @xmath27 where @xmath28 describes the atom / field coupling , eq . ( [ eq : hint ] ) , for the @xmath29 atom . from now on , we shall use this in the operator form @xmath30 where the operators @xmath31 and @xmath32 are the annihilation and creation operators for the quantized cavity mode . since this hamiltonian is symmetric under the exchange of any two atoms , a symmetric initial state remains symmetric as it evolves . we may thus reduce our effective hilbert state to consider only the states that are completely symmetric with respect to atomic interchange . we may thus use a number representation of the atomic state , namely @xmath33 , where @xmath34 gives the number of atoms in state @xmath35 ( @xmath36 ) , and @xmath37 the number of photons in the cavity . we allow arbitrary values for @xmath38 , @xmath39 , @xmath40 and @xmath37 , and employ the bose creation and annihilation operators @xmath41 and @xmath42 for the atomic states @xmath43 . we can then rewrite the interaction hamiltonian for the closed system as @xmath44 this many - body @xmath0-atom / cavity hamiltonian conserves the total number of atoms , represented by the operator @xmath45 , as well as the difference between the number of atoms in state @xmath4 and the number of cavity photons , represented by the operator @xmath46 . when cavity decay is added to this description , the operator @xmath47 gains the significance of referring to the number of photons that have escaped the cavity . since the operators @xmath48 and @xmath47 commute , we define subspaces @xmath49 composed of the eigenstates with simultaneous eigenvalues @xmath50 and @xmath51 of the operators @xmath48 and @xmath47 , respectively . this is summarized schematically in figure [ fig : n_atoms ] . we find that each manifold @xmath49 contains a null - valued eigenstate @xmath52 , given explicitly by @xmath53 where @xmath54 is a normalization constant . this state is the analog of the null - valued dark eigenstate for the single - atom / cavity system @xcite . it contains no population in the @xmath3 internal state and is thus an @xmath0-atom dark state immune to spontaneous decay from any atom . ( [ eq : dark ] ) represents a continuous family of dark states that are transformed into each other by the time dependence of @xmath11 . for nearly all values @xmath55 , we can show that the null - valued eigenspace of @xmath56 is non - degenerate , _ i.e. _ , each of these dark eigenstates possesses a unique energy . we will discuss exceptions later ( section [ sec : deltadelta ] ) . one can also show that acting on eq . ( [ eq : dark ] ) with the cavity annihilation operator @xmath31 , produces the corresponding dark state having one less photon , _ i.e. _ , any such dark eigenstate @xmath52 decays to @xmath57 by cavity emission . thus cavity decay does not take the system out of the family of dark states . conversely , the only way to directly couple dark states in different manifolds , is either to spontaneously lose a photon or to add a particle . this representation suggests that adiabatic evolution might be used for @xmath0-photon generation in an analogous manner to that proposed for single - photon generation in refs . we illustrate this here with the manifold corresponding to @xmath58 . initially the pump laser is off , @xmath59 , and the corresponding initial state has @xmath0 atoms in the @xmath2 state and no photons in the cavity , @xmath60 . one can then imagine slowly ramping the value of the pump laser until the pump laser coupling is much larger than the atom coupling to the cavity , @xmath25 . at this final time , @xmath61 , one finds that approximately all @xmath0 atoms are in the state @xmath4 and there are @xmath0 photons in the cavity , @xmath62 . this procedure will generate @xmath0 photons in a closed cavity . we will show that this procedure can be coupled to cavity decay to produce an @xmath0-photon state . in practice , realization of this ideal sequence requires that two key issues be adequately addressed . first , the need for adiabatic evolution through the family of dark states via couplings to the excited atomic level @xmath3 places constraints on how we vary the strength of the pump pulse , @xmath20 , based on the energy spectrum of the @xmath0-atom / cavity system . second , spontaneous emission of photons from the cavity will provide perturbations to the adiabatic evolution that may be non - negligible . we examine these issues in detail in section iv . before this , we first analyze the energy spectrum of the closed @xmath0-atom / cavity system in the remainder of this section . this will allow us to establish the critical parameters limiting the adiabatic evolution , that are required in order to estimate the errors due to non - adiabaticity and spontaneous emission within the open system approach employed in section [ sec : open ] . while the dark state of each @xmath49 manifold has a succinct description , eq . ( [ eq : dark ] ) , the general eigenstates for the @xmath0-atom / cavity system are quite complicated . however , in the limit of both large and small @xmath11 , the eigenstates are found to have familiar forms that render them susceptible to analytic investigation . we examine first the strong pump ( large @xmath20 ) limit . here the eigenstates can be interpreted in terms of the familiar angular momentum states . in the limit where @xmath20 is large relative to the other parameters ( @xmath6 , @xmath63 ) , we can neglect the terms in eq . ( [ eq : h0 ] ) that are proportional to the cavity coupling parameter @xmath6 . we then make a transformation from the atomic modes @xmath64 and @xmath65 to generalized angular momentum operators , @xmath66 , using the schwinger representation.@xcite this gives @xmath67 and @xmath68 , where @xmath69 denotes the total angular momentum , @xmath70 . in this limit we find that the system hamiltonian becomes @xmath71 where @xmath72 , @xmath73 , @xmath74 , and @xmath75 . the corresponding energy levels are now identical to those of the generalized angular momentum @xmath66 . therefore the eigenstates are simply given by @xmath76 where @xmath77 is the eigenvalue of total angular momentum , @xmath78 is the angular momentum projection along the axis @xmath79 , @xmath40 is the number of atoms in atomic state @xmath4 , and @xmath37 is the number of cavity photons as before . the cavity coupling term , @xmath80 , can now be considered to act perturbatively on eq . ( [ eq : rlarge ] ) , to mix states differing by @xmath81 , and to change the value of the cavity photon number by unity . relating these states in the large @xmath20 limit to our invariants @xmath48 and @xmath47 , we find that the @xmath49 manifold is composed of states @xmath82 , where @xmath83,@xmath84 , and @xmath85 ( see figure [ fig : ang_mom ] ) . our dark state at large @xmath20 is @xmath86 . in the number representation this is simply @xmath87 ( eq.([eq : dark ] ) ) . we note that the loss of a photon only reduces the photon number @xmath37 , and conserves @xmath88 and @xmath40 . the eigenstates in the large @xmath20 limit are shown schematically in figure ( [ fig : ang_mom ] ) . we now consider the weak pump ( small @xmath20 ) limit . here we find that the eigenstates may also be interpreted in terms of a known set of states , but these are now the less well known eigenstates of the tavis - cummings hamiltonian @xcite . we proceed in this limit by starting from the the system hamiltonian , eq . ( [ eq : h0 ] ) , at @xmath89 . this is simply @xmath90 we again make a transformation to a schwinger representation , but this time we choose the transformation to be made between modes @xmath91 and @xmath65 . the generalized angular momentum operators that are created from these two modes will be denoted here by @xmath92 , _ i.e. _ @xmath93 , @xmath94 and @xmath95 , where @xmath96 denotes the corresponding total angular momentum , @xmath97 . ( [ eq : h_r0 ] ) then becomes @xmath98 which is recognized to be the off - resonant tavis - cummings hamiltonian @xcite . note that this hamiltonian conserves the generalized angular momentum @xmath96 . it also conserves sum of the number of photons , @xmath37 , and the angular momentum in the z direction , @xmath99 . in the small @xmath20 limit we can then investigate the effect of finite @xmath20 using a perturbative analysis . this perturbative analysis has two consequences for the energetics . first , the perturbation due to @xmath20 , which is of the form @xmath100 , will only couple states whose total @xmath92 value differs by 1/2 . second , as a result of this , the resulting change in energy of the eigenstates of eq . ( [ eq : h0 ] ) is only second order in @xmath20 . the eigenstates of the tavis - cummings hamiltonian are not trivial , but we note that in the limit of large @xmath63 they are approximately eigenstates of @xmath101 with a number of photons in the cavity given by @xmath102 where @xmath103 is the total angular momentum and @xmath104 is an eigenvalue of @xmath47 . the states of the tavis - cummings model are described in detail in @xcite and qualitatively in appendix a. figure [ fig : tc_levels ] provides a schematic of these states in the small @xmath20 limit . one useful advantage of these schwinger angular momentum representations of the atomic states for our analysis is that in both of these limits of large and small @xmath20 , _ i.e. _ , whether for a fixed @xmath103 or for fixed @xmath77 , the corresponding eigenvalues @xmath99 and @xmath105 provide a measure of the population in the excited state . this population is given by @xmath106 . this property will be used in section [ sec : deltadelta ] to make estimates of the population in the scattering state _ i.e. _ , in the atomic state @xmath3 , that is susceptible to spontaneous emission , and hence of the errors due to spontaneous decay . the above description of the @xmath0-atom / cavity system as a closed quantum system is clearly incomplete , since a proper assessment of the operation of an @xmath0-photon generator requires the consideration of this cqed device as an open quantum system . we must take into account the two channels by which the @xmath0-atom - cavity system interacts with its environment . these are : i ) the possibility of spontaneous decay from atoms in the excited state @xmath3 to optical modes outside the cavity , determined by the spontaneous decay rate @xmath107 , and ii ) the coupling of cavity photons to electromagnetic modes outside the optical cavity , characterized by the cavity decay rate @xmath108 @xcite . the latter provides the required coupling to transfer an @xmath0-photon state from the cavity mode to an external mode . cavity decay thus plays two different roles here . first , in allowing emission of the @xmath0-photon state , and , second , affecting the dynamics inside the cavity as discussed below . we describe the @xmath0-atom - cavity system as an open system within a quantum wavefunction formulation @xcite . to characterize its action as an n- photon generator , we evaluate the cavity flux @xmath109 . in this section , we use the quantum wave function formulation in a perturbative regime to derive analytic estimates for the error rates of its action as an @xmath0-photon generator . in section [ sec : quantumtraj ] we then make numerical simulations of the full open quantum system that allow us to ascertain the extent of validity of these perturbative error estimates . in this open quantum system analysis , we will show that the structure of the closed quantum system , namely the identification of manifolds of states @xmath49 , each of which contains a dark state that is immune to spontaneous emission and that connects adiabatically to the initial state , still plays a critical role . when @xmath11 is varied in the open quantum system , the @xmath0- atom / cavity system evolves primarily within the family of accessible dark states @xmath52 ( @xmath110 ) from which no spontaneous emission occurs , just as in the closed quantum system . the system can fail to produce the desired @xmath0-photon output state _ only if photons are actually lost to spontaneous emission_. thus a crucial part of assessing the failure rate of the @xmath0-photon generator is to quantify the extent of `` non - darkness , '' _ i.e. _ , the probability that the system will evolve towards a bright state from which spontaneous emission may indeed occur . non - darkness can arise from two factors : non - adiabatic evolution when the rate of change of @xmath11 is too fast , and the conditional dynamics resulting from cavity decay which can couple the dark state @xmath52 to a non - dark state . in the quantum trajectory approach , @xcite the dynamics of the @xmath0-atom - cavity system are characterized by deterministic non - unitary evolution , interspersed with random jumps determined by photon losses from the cavity and by spontaneous emission ( scattering ) from the excited state . the non - unitary evolution is given by the conditional hamiltonian @xmath111 the imaginary terms @xmath112 and @xmath113 describe the back - action on the quantum system from that accumulates between instances of cavity decay and spontaneous emission jumps , respectively . the corresponding quantum jump operators are given by @xmath114 and @xmath115 . for a full discussion of the quantum jump approach see ref . @xcite . for the numerical calculations presented in section [ qts ] we use the full conditional hamiltonian , as required in the quantum trajectory formulation.@xcite in order to obtain analytic error estimates here , we proceed first by assuming that the adiabatic errors are small . therefore , at all times the state of the system is now regarded as deviating only slightly from the dark state . in this situation we may neglect the spontaneous emission term , @xmath116 , treating it implicitly as a higher order perturbation than the cavity decay term . the cavity decay term @xmath108 is treated in first order perturbation theory . before describing the details of this analysis , we note again that the cavity decay , characterized by the jump operator @xmath114 , connects dark states in the manifold @xmath49 to dark states in a lower manifold @xmath117 , i.e.,@xmath118 ( see section [ sec : nphoton ] above . ) thus neither of the quantum jump operators @xmath114 or @xmath119 will lead to errors ( i.e. to non - darkness ) in the operation of our cqed device , as long as the system is maintained within the dark states @xmath120 . consequently , in order to quantify the failure rate of the @xmath0-photon generator under these conditions , one need only consider the evolution of the system under the conditional hamiltonian @xmath121 . we will now make a detailed analysis of the errors , starting with an estimate of the extent of non- darkness introduced by cavity decay , then estimating the adiabatic error in following the dark state as @xmath11 is varied , and finally estimating the spontaneous emission flux rate due to population in the @xmath3 internal state that is introduced by a combination of cavity decay and non - adiabaticity . we first analyze the errors due to cavity dynamics . for this purpose , we treat the @xmath122 term in eq . ( [ hcond ] ) as a perturbation to the closed - system hamiltonian , eq . ( [ eq : h0 ] ) . in particular , under the realistic scenario when cavity decay is weaker than the cavity coupling ( @xmath123 ) , we find that the cavity decay term causes the dark states @xmath52 of the closed - system hamiltonian to be modified to the states @xmath124 of the conditional hamiltonian , according to @xmath125 here we have modified our notation to define the states @xmath126 as the instantaneous eigenvalues of the closed - system hamiltonian @xmath121 with corresponding energies @xmath127 . using this expression , we find the degree of `` non - darkness '' due to cavity decay , @xmath128 , to be equal to @xmath129 from this expression we can generate an upper bound on the degree of non - darkness , @xmath130 . we first note that @xmath131 where @xmath132 is the variance in the cavity photon number in the dark state @xmath133 . our upper bound is then obtained by replacing the bohr frequencies @xmath134 in eq . ( [ eq : ecavity ] ) with the minimal bohr frequency , and replacing the time - varying variance @xmath132 with its maximum . this results in @xmath135 the maximum variance of @xmath136 is bounded by @xmath137 ( see appendix b ) while the minimal bohr frequency depends on @xmath63 , and will be discussed detail in section [ sec : deltadelta ] . our expression for the bound on the extent of non - darkness due to cavity decay thus becomes @xmath138 we now examine errors that arise due to non - adiabatic evolution resulting from a non - zero derivative , @xmath139 . using the standard treatment @xcite we estimate the population leakage from the adiabatic state at a time t , @xmath140 here @xmath141 . we then apply the standard upper bound for adiabatic error @xcite , given by the square of the maximum angular velocity of the state , divided by the square of the minimal bohr frequency , @xmath142 one can bound the maximum angular velocity , @xmath143 , to be smaller than @xmath144 ( see appendix c ) . this results in the following upper bound : @xmath145 eqs . ( [ admax ] ) and ( [ cmax ] ) provide an upper bound for the `` non - darkness '' in the limit that our perturbative approach is appropriate . when the minimal energy separation , @xmath146 is independent of @xmath0 , both equations suggest that the `` non - darkness '' scales at worst linearly with @xmath0 , in the worst case . furthermore , we note that the maximum `` non - darkness '' decreases with increasing @xmath104 . we now proceed to estimate the effect of spontaneous emission , or equivalently , the rate of spontaneous emitted flux . the spontaneous emission flux rate is equal to the product of the population in the spontaneously emitting state and twice the decay rate of that state , _ i.e. _ , @xmath147 . unfortunately , we are unable to analytically calculate the average population in the excited atomic state @xmath148 . the simplest way to estimate an upper bound on the spontaneous flux rate is then to use the maximum possible value of @xmath149 , and the maximum probability of the system being in a non - dark state , @xmath150 . we can thus obtain an upper bound on the spontaneous emission flux rate as @xmath151 . naturally , by detuning from resonance , we expect to minimize the population that will leak to population in the excited state as is the case in both off - resonant rayleigh and raman scattering . hence , we expect the spontaneous emission flux rate to change as a function of the detuning @xmath63 . the value of @xmath63 affects the spontaneous error rate in two ways . first , it controls the value of the minimal bohr frequency , and can thereby either reduce or enhance the first order population leakage from the dark state . second , we know by analogy with the three level system ( section [ sec : single ] and refs . @xcite ) that the `` non - darkness '' should contain less excited state character as the absolute magnitude of @xmath63 increases . our scheme works independent of the sign of @xmath63 . however , for concreteness , @xmath63 is assumed to be negative ( red detuning ) throughout the rest of the paper . referring to the discussion in section iii , for small values of @xmath63 , _ i.e._,@xmath152 , there exist states with energy only @xmath153 away from the dark state . in both the small and large @xmath20 limits , these states are the @xmath154 states ( section [ sec : nphoton ] ) . hence , for @xmath155 , the minimal bohr frequency is equal to @xmath63 . furthermore , the eigenstates are effectively eigenstates of @xmath156 or @xmath157 and therefore have average value @xmath158 . therefore , whatever states the dark state falls into due to either adiabatic error or cavity decay , these states will have large population in the excited atomic state @xmath159 . so in the small @xmath160 limit our upper bound described above is reasonable . consequently , we use the `` non - darkness '' factors calculated in eqs . ( [ cmax ] ) and ( [ admax ] ) times @xmath161 . hence , we find that the error rate for small values of detuning @xmath160 is always less than @xmath162 in contrast , for large values of @xmath160 , the states are effectively eigenstates of @xmath163 or @xmath101 ( section [ sec : nphoton ] ) . for these states , the smallest valued eigenstate is the @xmath164 or @xmath165 state , and the corresponding perturbative energy scales in second order as @xmath166 . however , this smallest energy eigenstate has a population in the excited state that scales at worse as @xmath167 . we believe that it is a reasonable assumption to multiply this smallest eigenstate estimate of the excited state population by our calculated `` non - darkness '' factor @xmath168 , in order to arrive at a better estimate of the excited atomic state population . this procedure yields the followng expression for large values of @xmath63 , namely an error rate that is independent of @xmath63 : @xmath169 careful analysis shows that in the large @xmath160 limit , the state of energy @xmath170 is not always the closest state to the dark state . indeed states can be identified for particular settings of @xmath63 and @xmath20 which are degenerate with the null - valued dark state . however , these states are shown to be characterized by higher @xmath103 or @xmath77 , and therefore in the perturbative limit they do not couple directly with the dark state . numerical simulations suggest that the situation is similar for states which cross the null state in the non - perturbative limit . . [ sec : small ] having obtained error estimates due to cavity decay , non - adiabaticity , and spontaneous emission , we now consider the dependence on the cavity photon number @xmath37 . in the limit that one very slowly ramps up @xmath20 , the average number of photons in the cavity is always much smaller than 1 . we assume that our dark state is simply @xmath171 . for this state @xmath172 , where @xmath173 ( eq . ( [ eq : dark ] ) ) . in the limit of small @xmath174 , the variance of photons in the cavity equals the number of photons in the cavity . since the ramp is very slow , we expect that the error rate will be due to the cavity dynamics and , therefore , proportional to the variance , eq . ( [ cavar ] ) . the rate of photon output is proportional to the number of photons in the cavity . hence , in the case of a small number of photons @xmath37 in the cavity , we expect at worse for the error to scale proportional to the number of photons output . this implies a total error that scales at worse proportional to the number of atoms @xmath0 , when @xmath175 . one might ask how can we increase @xmath20 to be of equal magnitude with @xmath6 , and still maintain a small number of photons in the cavity . the reason is that as @xmath20 increases , there is an increasing chance of a photon being emitted . if one varies @xmath20 slowly enough , the chance of that occurring before @xmath20 becomes too large is quite high . at this point , we then change manifolds @xmath49 . examining the equation for the dark state ( eq . ( [ eq : dark ] ) ) and looking at only the first two terms , one sees that changing manifolds is equivalent to reducing the effective value of @xmath20 . compared to the @xmath58 manifold , the effective @xmath176 of the @xmath104 manifold is equal to @xmath177 . as @xmath0 increases , for small values of @xmath104 one then needs smaller and smaller values of @xmath20 for the approximation to hold . for example , when @xmath58 , @xmath178 . however , as the value of @xmath104 increases , @xmath20 can eventually reach a value comparable to the product @xmath179 and still imply a small number of photons in the cavity . to summarize , in the limit of a small number of photons @xmath37 in the cavity and @xmath180 , we expect the ratio of spontaneous emission flux to cavity flux to be either a constant , or a decreasing function with @xmath0 . explicitly , we expect spontaneous flux to be smaller than @xmath181 ( see eq . ( [ eq : gg ] ) and eq . ( [ cavar ] ) ) . the ratio of the spontaneous flux , @xmath182 , to the cavity flux , @xmath183 , is then @xmath184 which is larger than the fractional spontaneous loss , the spontaneous flux divided by the expected number of photons out , @xmath0 . therefore , for current state of the art optical cavity technology for which @xmath185 and @xmath186 @xcite , we expect that only 1 of every 200 photons will be lost to spontaneous emission . [ sec : quantumtraj ] the quantum trajectory or quantum jump approach allows one to calculate the properties of open quantum systems by averaging over individual quantum trajectories . @xcite the basic elements of this approach were described briefly in section [ sec : open ] . technical details of the algorithm are given in ref . the calculations described below average over 5000 trajectories per simulation . the simulations were run until the cavity flux and spontaneous emission flux were less than @xmath187 . this results in 2000 - 5000 steps of length @xmath188 , depending on the specific parameters of the system . the basic quantity we calculate is the cavity flux , 2@xmath189 , which describes the number of photons emitted from the cavity . for these simulations , we have chosen as our figure of merit the fractional spontaneous loss . the fractional spontaneous loss is defined as the number of photons lost to spontaneous emission , @xmath190 , divided by the expected number of photons out , @xmath0 . we calculate @xmath190 as the product of @xmath191 and the integral of @xmath148 . due to the statistical error resulting from the finite number of trajectories , we have estimated a numerical uncertainty of @xmath192 in @xmath193 . using a simple driving field which increases linearly with time , and realistic cavity parameters based on the cavity of kimble @xcite , ( @xmath194 ) , we find that the deterministic production of @xmath0-photons within a single pulse with small loss is indeed possible . figure [ fig : nflux ] shows the output pulse as a function of time for @xmath195 . we see the expected linear increase in the area of the output pulse , _ i.e _ , the cavity flux 2@xmath196 , with @xmath0 . we then calculated the loss of photons into the spontaneous emitted modes , shown in figure [ fig : nerr ] . when the simulation was run in the limit of a small number of photons in the cavity and with @xmath197 , we expected that the total photon loss will grow at worst linearly , as explained above in section [ sec : small ] . therefore , the fractional spontaneous loss should either be constant or decrease with increasing @xmath0 . this expectation is confirmed by simulations , see figure [ fig : nerr ] and figure [ fig : nexp ] . furthermore , the error is smaller than our expected bound of @xmath198 fractional spontaneous loss , see section [ sec : small ] . although , we have made calculations with only a small number of @xmath0 values here , it appears that our analytic estimates coupled with this numeric evidence for @xmath1 suggest that the deterministic production of large @xmath0 fock states when @xmath199 is indeed possible . the numerical results summarized in figure 6 show that for @xmath200 we have minimal fractional spontaneous losses of approximately 0.3actually decrease with increasing @xmath0 . since our analytical results suggest an upper bound of 0.5expect to be able to produce a fock state containing on the order of 100 photons before the total integrated loss due to spontaneous flux equals a single photon . in the previous subsection , we used current experimental cavity parameters @xcite and , for simplicity , a linear driving pulse . we now show how one can minimize the spontaneous loss by tuning various cavity parameters and modifying the driving field . for linear ramping with a small rate of change , one finds that the fractional error actually decreases with increasing @xmath0 ( figure [ fig : nerr ] ) . this is in line with our analytical bounds which suggested that at worse the fractional error should be constant for small occupation of photons in the cavity ( section [ sec : small ] ) . figure [ fig : rerr ] shows that as the ramp speed increases beyond the adiabatic regime , the errors increase rapidly . one can further reduce the error by using a more sophisticated time - dependent driving field . for example , the application of a gaussian pulse , @xmath201 , reduces the fractional spontaneous loss by a factor of 2 , relative to the fractional spontaneous loss resulting from a linear ramp , as illustrated in figure [ fig : nexp ] . one expects that appropriate pulse shaping could further lead to an order of magnitude increase in the fidelity . however , the minimal fractional spontaneous loss obtained by pulse shaping is limited by errors due to cavity dynamics . in figure [ fig : nerr ] , one sees that this limit is approached by ramp speeds of @xmath202 . the ability to detune from resonance is one of the basic tools of atomic physics . therefore , it is important to determine how the fidelity of our operation scales with the detuning , @xmath63 . in section [ sec : deltadelta ] , our analytical predictions suggest that as @xmath160 is increased from @xmath203 , the initially large fractional spontaneous loss should decrease , eventually reaching a constant non - zero value for @xmath204 . for larger @xmath160 , our numerical calculations confirm this prediction for @xmath205 atoms , figure [ fig : nerr ] . in figure [ fig : nerr ] , one sees that for detunings , @xmath160 , smaller than the cavity coupling constant , @xmath6 , the error decreases as the detuning increases . one then sees a relatively flat region , followed by an increase in error as @xmath160 increases . this rise is outside of the predictive ability of our analytical model and represents a breakdown in the first order perturbative methods when the energy separation between the dark state and nearest neighbor state becomes smaller than the ramping speed . the main conclusion from our numerical simulations is that a wide region of detunings are nevertheless possible . in terms of efficiency , ( number of photons out per time / fractional spontaneous loss ) , our data suggest that a detuning of @xmath206 would be optimal . for large values of @xmath160 , one can reduce the error by increasing the value of @xmath6 . as predicted analytically in section [ sec : deltadelta ] , the total error will reduce as @xmath6 increases . furthermore , if one varies @xmath6 and @xmath207 proportionally , one can numerically observe the @xmath208 dependence of fractional loss predicted by eq . ( [ eq : gg ] ) . this is shown in figure [ fig : gerr ] . we have presented a scheme for the deterministic production of @xmath0-photon fock states from coupling of @xmath0 three - level atoms in a high - finesse optical cavity to an external field . the method relies on adiabatic passage through a continuous family of dark states that is controlled by the ramping up of the external field . we have shown that this procedure can reliably output pulses of @xmath0-photons from the cavity . we have made a detailed analysis of the errors involved in the @xmath0-photon state production . these result when there is a breakdown of adiabaticity and from spontaneous emission . this error analysis yields analytical bounds on the errors which are well reproduced by the results of numerical simulations for up to @xmath0=5 photons . our estimates for physically realistic cavity and atom / field parameters indicate that this scheme may be used reliably to generate states with up to @xmath0=100 photons . one way in which such a deterministic fock state generator may be constructed in the near future is using a combination of an ultra - cold atomic source and a high - finesse optical resonator ( with parameters such as those used for the calculations in section [ sec : quantumtraj ] ) such as those used in the works of kimble and colleagues @xcite . modern techniques of laser and evaporative cooling in magnetic traps can readily produce cold and tightly confined atomic gases which can be transported in vacuum from a production region to the confines of an optical resonator . using ultra - cold alkali atoms , the states @xmath2 and @xmath4 can be chosen to be two ground state hyperfine levels which are both magnetically trapped , such as the @xmath209 and @xmath210 hyperfine states of @xmath211rb . these may be connected by a @xmath212 raman transition using excited levels on the @xmath213 or @xmath214 transitions while the strength of this transition is suppressed at large detunings @xmath160 from the atomic resonance , our work shows that only moderate detunings ( several @xmath6 ) are necessary for high - fidelity operation . the initial @xmath0-atom state may be generated by following a weak microwave excitation with atom - number - sensitive selection , or , alternately , by real - time observation of the atomic number in a given hyperfine level . this would set proper initial conditions for @xmath0-photon generation . several additional comments are in order . throughout this work , we have assumed that the atoms are identically coupled to the cavity and pump fields . similarly , we assume that these atoms are indistinguishable in their spontaneous emission to optical modes outside the cavity . it would be desirable for the n - photon generator to operate similarly if these restrictions are eased . toward this end , it is still possible to define a family of dark states for atoms which are not identically coupled to the cavity and pump fields . as such , our analytical approach to estimating the photon losses to spontaneous emission could be similarly extended to such cases . we believe this would provide similar performance to the idealized case we have provided herein , i.e. the deterministic production of fock states containing as many as @xmath215 photons should be possible . however , one would expect that while different spatial arrangements of @xmath0 atoms in a cavity would all produce @xmath0-photon fock states , the specific optical mode occupied by those @xmath0-photons would differ . applications requiring many orthogonal , simultaneous pulses of @xmath0-photons ( such as heisenberg - limited interferometry ) would be constrained by these differences . we envisage a number of applications for deterministically - produced @xmath0-photon states . as alluded to in the above discussion , two orthogonal @xmath0-photon states can be combined using linear beam - splitters to create an interferometer which is sensitive to differential phase shifts between the two arms of the interferometer which are on the order of @xmath216 ( the heisenberg limit ) , rather than the typical @xmath217 sensitivity ( the standard quantum limit ) obtained with classical light pulses @xcite . two orthogonal @xmath0-photon states can also be used to create a highly entangled state using only measurement and linear optics @xcite . such highly entangled photon states can be used to perform precision measurements @xcite . the cqed device described in this work could thus be used for demonstrations of this interferometric method . while the low photon numbers deterministically available from this cqed device ( given current state - of - the - art cavity parameters ) would not yield the precision available from the use of much more intense classical light sources , there may be applications requiring high precision at low light levels which are enabled by this fock - state generator , e.g. coherent control @xcite . other applications in the field of quantum information , communciation , cryptography , and computation are also possible , and we intend to investigate such potential applications in future work . finally , a deterministic n - photon source would find both basic and applied use for the absolute calibration of optical detectors , particularly those designed to be sensitive to multiple photons ( as opposed to conventional avalanche photodiode devices ) . similarly , the @xmath0-photon generator can be used as the light source for novel multiphoton spectroscopy , e.g. rempi @xcite , although the optical frequency range which can be produced by a given atom - based system would be quite limited . the authors effort was sponsored by the defense advanced research projects agency ( darpa ) and air force laboratory , air force materiel command , usaf , under agreement number f30602 - 01 - 2 - 0524 , the office of naval research under grant no . fdn 00014 - 01 - 1 - 0826 , and the nsf . dmsk acknowledges support by the sloan foundation . kbw thanks the miller institute for basic research in science for a miller research professorship for 2002 - 2003 . the work of krb was also supported by the fannie and john hertz foundation . we would like to thank neil shenvi for useful discussions . after this work was completed we learned that similar @xmath0-atom dark states have been proposed in ref . . in this appendix , we present a qualitative picture of the eigenstates of our system in the small @xmath20 limit , i.e. , a description of the eigenstates of the non - resonant tavis - cummings hamiltonian , eq . ( [ eq : h_r0 ] ) . first , note that the creation / annihilation operators associated with the cavity were not present in eq . ( [ eq : h_r0 ] ) , the tavis- cummings hamiltonian would become again simply a hamiltonian describing angular momentum ( @xmath92 ) about some axis ( @xmath218 ) . in fact , for large enough @xmath63 , one sees that is indeed the case , and we do merely have eigenstates of angular momentum about the @xmath219 axis . in order to get a qualitative picture of the tavis - cummings eigenstates for general values of the cavity coupling @xmath6 and finite values of @xmath37 , it is useful to treat the cavity creation / annihilation operators as operators that act to enhance the `` effective magnetic field '' in the `` @xmath220 '' direction . for example , when one has a large number of photons , @xmath37 , in the cavity a standard approximation is to replace the cavity photon creation / annihilation operators with @xmath221 . applying this transformation to eq . ( [ eq : h_tavis ] ) , one finds that the hamiltonian becomes simply @xmath222 , _ i.e. _ , the cavity coupling has been transformed into an effective magnetic field in the @xmath220 direction . the eigenstates of eq . ( [ eq : h_tavis ] ) are constructed as follows . first , one starts with atomic states that are the eigenstates of @xmath101 , namely , @xmath223 . we know that the hamiltonian conserves @xmath47 , the difference in the number of atoms in state @xmath224 and the number of photons , @xmath37 , in the cavity . in this representation , the conserved quantity is @xmath225 . we can then append a photon state that reflects this conservation symmetry to the atomic states of modes @xmath3 and @xmath4 identified via @xmath101 . this results in the atom - photon states @xmath226 . the only missing component is now an atomic state of mode @xmath2 . we then append the appropriate number state of the atomic mode @xmath227 such that @xmath0 is conserved , according to @xmath228 . this then specifies the remaining index @xmath38 . we have thereby arrived at a set of states @xmath229 , that may now be latex npb.used as a basis for expansion of the exact eigenstates of eq . ( [ eq : h_tavis ] ) . since the hamiltonian eq . ( [ eq : h_tavis ] ) conserves @xmath230 , and @xmath231 , the exact eigenstates can be formally written as @xmath232 . naturally , determining the coefficients @xmath233 is not trivial @xcite . for our purposes here , it is not necessary to find explicit solutions for the coefficients . we require only the main features of the energy spectrum , which are available from the identification of the above basis and are described below . it is important to note that in order for the photon number @xmath37 to be positive , we must have @xmath234 . for fixed @xmath103 and @xmath104 values , this can lead to a truncation of the possible accessible states . one finds that for positive @xmath104 , the angular momentum projection , @xmath235 , can have only @xmath236 values , whereas for negative @xmath104 , @xmath235 can take all @xmath237 values . this implies that for @xmath238 , the eigenstate degeneracy for given values of @xmath103 , @xmath0 , and @xmath104 , equals the degeneracy one would expect for a total angular momentum value of @xmath103 ( _ i.e. _ , @xmath239 ) . in contrast , for values @xmath240 , the eigenstate degeneracy equals the degeneracy expected for a total angular momentum value of @xmath241 , _ i.e. _ , @xmath236 . note that in our scheme we always assume that there are initially no photons in the cavity , i.e. , @xmath242 . in summary , we see that one think of the states @xmath243 as being effectively `` angular momentum '' states possessing a total angular momentum @xmath96 equal to @xmath241 and angular momentum projection of @xmath235 about the `` magnetic axis '' . although this analogy is not perfect because of the spread of actual eigenstates over this basis , it does contain the following important feature . this is , that for @xmath63 larger than @xmath244 , where @xmath245 is the maximum number of photons in the cavity , the eigenstates are eigenstates of @xmath101 , with eigenvalue @xmath246 . so the large @xmath63 limit is simplified . in contrast , for small and medium size values of @xmath63 , the eigenstates are superpositions over a wide range of @xmath99 states . in this appendix , we derive an upper bound on the variance of cavity photon number for the dark state . in order to reach this bound , we need to first outline other properties of the dark state . we begin with a few definitions . the dark state in the @xmath247 manifold is given by , @xmath248 where @xmath249 and @xmath250 is the normalization constant . let @xmath251 be the state given by @xmath252 . thus , @xmath253 where @xmath254 is the normalization constant . we further define the state @xmath255 given by the normalized state @xmath256 . therefore , @xmath257 where @xmath258 is the normalization constant . we now present some properties relating these normalization constants to each other and their derivatives . so let @xmath259 be the average number of photons in the cavity for the darkstate in the @xmath104 manifold . in other words , @xmath260 @xmath261 _ proof : _ by definition , @xmath250 is the normalization constant of the dark state @xmath262 . hence , @xmath263 dividing both sides by @xmath264 gives us , @xmath265\end{aligned}\ ] ] @xmath266 _ proof : _ @xmath267 is the normalizaton constant for the state @xmath268 . hence by definition , @xmath269 dividing both sides by @xmath264 , we get the result , @xmath270 from eq . ( [ avgl ] ) , we prove the required result . if the system is in the dark state @xmath262 , then @xmath271 _ proof : _ it is trivial to see that property 3 is true when @xmath272 . let @xmath273 . consider the state @xmath251 given by eq . [ phik ] . since @xmath273 we can extend the summation to @xmath274 . thus , @xmath275 } \\ & = & \frac{1}{y_k z_k } \sum_{l=0}^{n - k } \left ( l-{\langle l \rangle}_k \right ) \frac{(-x)^{l-1}}{\sqrt{(n - k - l)!(l+k)!l ! } } { |(n - k - l)\;\;0\;\;(l+k)\;\;l\rangle } \\ & { } & \qquad \qquad \qquad \mbox{[using property 1 ] } \\\end{aligned}\ ] ] @xmath254 is the normalization constant of @xmath251 . thus , @xmath276 it easy to see that this is merely @xmath277 . * property 4 * @xmath278 _ proof : _ @xmath279 using property 2 we get the required result . @xmath280 _ proof : _ we need to show that @xmath281 which is equivalent to showing that @xmath282 since @xmath283 is positive for all @xmath77 . @xmath284 we note that @xmath285 thus , @xmath286 in order to evaluate the sum in eq . ( [ eq:5.1 ] ) , we choose two integers between 0 and @xmath287 , @xmath31 and @xmath288 . we then evaluate the sum of the two terms corresponding to @xmath289 , @xmath290 , and @xmath291 , @xmath292 . we see that @xmath293 since @xmath294 , we simply need to determine whether @xmath295 . if @xmath296 , @xmath297 is positive and @xmath298 is positive , so the product is therfore also positive . if @xmath299 , @xmath297 is negative and @xmath300 is also negative , so therefore the product is still positive . hence , for all @xmath31 and @xmath288 @xmath301 which implies that @xmath302 by eq . ( [ eq:5.1 ] ) and eq . ( [ eq:5.2 ] ) , we have @xmath282 and thus @xmath303 + if the system is in the dark state @xmath262 , @xmath304 _ proof : _ @xmath305}\\ & \leq & { \langle l \rangle}_k\\ & { } & \qquad \mbox{[using property 5]}\\ & \leq & n - k.\end{aligned}\ ] ] we determine here an upper bound on the angular velocity of the dark state , @xmath306 . we first note that using the chain rule the @xmath307 is equivalent to determining the maximum value of @xmath308 where @xmath254 is the normalization constant defined in equation [ eq : yk ] ) . for convenience , we will define @xmath309 and @xmath310 ( throughout this appendix , we will use @xmath311 to represent the derivative of @xmath103 with respect to @xmath312 , and @xmath313 to represent the derivative f with respect to @xmath314 ) . the second step is to determine the maximum value of @xmath317 . using the notation from appendix b , we first note that @xmath318 , since @xmath319 is normalized . from eq . ( [ eq : yk ] ) , this implies that @xmath320 hence , we can write the following expression for @xmath317 , @xmath321 since @xmath317 must be positive and both @xmath322 and @xmath323 are positive , we can write the following inequality @xmath324 note that @xmath317 is an even function of @xmath312 , so the value of @xmath317 at @xmath272 must be a local minimum or maximum . when one takes the derivative of @xmath254 with respect to @xmath312 , one finds only a single zero at the origin . the identity , @xmath325 , implies that @xmath317 must go to zero as @xmath312 increases in order to maintain a finite variance . therefore , the maximum value of @xmath254 occurs at the origin . explicitly calculating the limit at the origin yields @xmath326 consequently , we have an expression for the maximum angular velocity @xmath327 , , , , phys . lett . * 75 * , 3788 ( 1995 ) . we note that when identical values of g are not possible for all @xmath0 atoms one can still reliably produce n - photon states using a pump laser which does not distinguish between the atoms . the analysis would be similar and rely on the dark states discussed in ref . @xcite and coupling strength , @xmath20 . the atom is also coupled to a cavity mode of frequency @xmath7 and coupling strength @xmath6 . both the cavity and the driving fields are detuned from the atomic transition resonance by a common frequency @xmath63 . the atom / cavity states are denoted here as @xmath328 where @xmath329 are the three atomic levels and @xmath37 is the number of cavity photons . ] . the @xmath49 manifold is composed of the eigenstates with simultaneous eigenvalues @xmath50 and @xmath51 where @xmath330 is the number of atoms in atomic state @xmath331 and @xmath37 is the number of photons in the cavity . the transition strength between levels is proportional to either the driving field , @xmath20 , or the cavity coupling , @xmath6 . the energy of the states in the absence of all couplings is given by -@xmath332 . for red detuning , @xmath333 , the states with a higher occupation of the exited atomic state , @xmath39 , will have a greater energy . the dark state is the superposition of states with @xmath334 described in eq . ( [ eq : dark ] ) . a transition from the manifold _ e_(5,3 ) to _ e_(5,2 ) occurs when a photon is emitted from the cavity . this transition preserves the dark state ( see text ) . ] and @xmath335 and assuming red detuning , @xmath336 . this representation is appropriate when the external driving field , @xmath20 , is much larger than the cavity coupling , @xmath6 . the eigenstates to first order are eigenstates of angular momentum about an axis defined by the effective magnetic fields @xmath337 and @xmath338 . the cavity coupling , @xmath6 , acts as a perturbation , mixing states differing by a total schwinger angular momentum of 1 . ] and @xmath224 and assuming red detuning , @xmath339 . this representation is appropriate when the cavity coupling , @xmath6 , is much larger than the external driving field , @xmath20 . the eigenstates of the tavis - cummings hamiltonian are complicated ( see text and ref . @xcite ) but preserve total schwinger angular momentum @xmath103 obtained from modes @xmath3 and @xmath224 . the coupling to the external field , @xmath20 , acts as a perturbation , mixing states differing by a total schwinger angular momentum of 1 . ] , for @xmath205 atoms plotted versus dimensionless time , @xmath340 . these simulations were done using a linear ramp @xmath341 where @xmath342 . the other cavity parameters were @xmath343 , and @xmath344 . the integrated cavity flux , @xmath190 , increases linearly with @xmath0 as expected . ] as a function of @xmath63 . a linear ramp was used with @xmath345 and @xmath346 . as expected one sees that in the large @xmath63 limit that the fractional spontaneous loss scales less than n ( see text ) . our analytical bounds in the large @xmath160 limit suggest that the fractional spontaneous loss should be less than @xmath347 ( dotted line ) . the discrepancy at large @xmath160 represents parameters for which our perturbative approach is invalid . for the small @xmath160 limit the increasing fractional spontaneous loss is consistent with leakage to states that have a higher occupation of the excited state . ] , and detuning @xmath63 ( @xmath348 and @xmath346 ) . as expected , larger ramp rates lead to an increased loss when the system can no longer adiabatically follow the dark state . note how the dependence on @xmath63 varies with @xmath349 . in the adiabatic limit , small @xmath349 , large @xmath63 yields the lowest fractional spontaneous loss . however , when one ramps the system quickly , _ i.e. _ , at large @xmath349 , the dependence on @xmath63 dependence is reversed and small @xmath63 yields the lowest loss . ] and @xmath346 ) . for the linear ramp ( circles ) , @xmath350 . for the gaussian ramp ( diamonds ) , @xmath351 , @xmath352 . the gaussian width , @xmath353 , was chosen such that the process of emitting a single photon would occur in the same time as the linear ramp . one expects that a more sophisticated pulse could result in an order of magnitude reduction of the fractional spontaneous loss.,title="fig : " ] n , and detuning , @xmath63 ( @xmath354 ) . the values of @xmath63 in the legend are in units of @xmath355 . one sees that at all values of @xmath63 there is a significant decrease in fractional spontaneous loss for increasing @xmath6 . in the large @xmath160 limit , one finds that the decrease in spontaneous loss with @xmath6 has a @xmath208 dependence , as predicted by our pertubative analysis ( see section [ sec : small ] ) . ]

we present 630010100 spectra for a sample of 13 t dwarfs observed using lris mounted on the keck i 10 m telescope . a variety of features are identified and analyzed , including pressure - broadened k i and na i doublets ; narrow cs i and rb i lines ; weak cah , crh , and feh bands ; strong ho absorption ; and a possible weak ch band . h emission is detected in three of the t dwarfs , strong in the previously reported active t dwarf 2mass 1237 + 6526 and weak in sdss 1254 and 2mass 1047 + 2124 . none of the t dwarfs exhibit li i absorption . guided by the evolution of optical spectral features with near - infrared spectral type , we derive a parallel optical classification scheme , focusing on spectral types t5 to t8 , anchored to select spectral standards . we find general agreement between optical and near - infrared types for nearly all of the t dwarfs so far observed , including two earlier - type t dwarfs , within our classification uncertainties ( subtype ) . these results suggest that competing gravity and temperature effects compensate for each other over the 0.62.5 spectral region . we identify one possible means of disentangling these effects by comparing the strength of the k i red wing to the 9250 ho band . one of our objects , 2mass 0937 + 2931 , exhibits a peculiar spectrum , with a substantial red slope and relatively strong feh absorption , both consequences of a metal - deficient atmosphere . based on its near - infrared properties and substantial space motion , this object may be a thick disk or halo brown dwarf .

astro-ph0305139

current classification schemes for the two latest spectral classes , l dwarfs @xcite and t dwarfs @xcite , are defined in different spectral regions . l dwarf classification is tied to the red optical region ( 630010000 ) , enabling straightforward comparison with m dwarf classification schemes @xcite , and hence a well - defined demarcation of the m and l spectral classes . t dwarf classification is tied to the near - infrared region ( 12.5 @xmath5 ) , taking advantage of both greater relative brightness ( @xmath4 75% of the emergent flux ) and the presence of strong h@xmath0o and defining ch@xmath1 bands . classifications in different spectral regions can not be expected to be identical , however , due to the competing influences of temperature , gravity , and composition at different wavelengths . this is particularly the case for late - type m , l , and t dwarfs , whose molecular - rich atmospheres also contain condensate clouds , predominantly affecting near - infrared spectral flux @xcite . the influence of these chemical and physical variables muddle the characterization of the l / t transition , a spectromorphological boundary that encompasses a dramatic shift in near - infrared colors ( j k @xmath6 2 @xmath7 0 ) , condensate opacity , and overall spectral energy distribution . clearly , the evolution of both near - infrared and optical spectra must be examined to fully characterize this critical transition . the red optical spectra of t dwarfs are of interest in their own right , as they encompass unique features and diagnostics of physical parameters . early investigation of the prototype of this class , gliese 229b @xcite , revealed strong h@xmath0o and weak ch@xmath1 molecular absorption ( counterparts to the strong bands that dominate the near - infrared spectrum ) , cs i alkali lines , and a steep red spectral slope shortward of 1 @xmath5 @xcite . the latter feature , attributable to the pressure - broadened wings of the 7665/7699 k i resonant doublet lines @xcite , is responsible for the extremely red optical / near - infrared colors of this object ( r j @xmath4 9.1 ; matthews et al . 1996 , golimowski et al . 1998 ) , a property that has been exploited in the search for t dwarfs in wide - field surveys such as the two micron all sky survey ( * ? ? ? * hereafter 2mass ) and the sloan digital sky survey ( * ? ? ? * hereafter sdss ) . @xcite have shown that these heavily - broadened lines and their 5890/5896 na i counterparts are sensitive to both effective temperature ( t@xmath8 ) and specific gravity . the much weaker cs i lines are key tracers of atmospheric rainout @xcite . additional molecular bands of feh and crh reported in the spectra of mid - type t dwarfs @xcite are possible tracers of condensate cloud opacity across the l / t transition @xcite . finally , persistent h@xmath2 emission detected in the t dwarf 2mass 1237 + 6526 @xcite , whose atmosphere is predicted to be too cool to support a substantial chromosphere @xcite , is not yet understood and has been the subject of some debate @xcite . in this article , we continue the spectroscopic investigation of t dwarfs initiated in ( * ? ? ? * paper i ) by examining their red optical spectra . this article complements that of @xcite , which discusses the optical l / t transition in detail . in @xmath92 we describe the acquisition and reduction of the data , obtained with the low resolution imaging spectrograph ( * ? ? ? * hereafter lris ) mounted on the keck i 10 m telescope . in @xmath93 we present the reduced spectra , analyzing in detail the spectral features observed . in @xmath94 we present a classification scheme for mid- and late - type t dwarfs using red optical spectra that is parallel to , but independent of , the near - infrared scheme of @xcite . in @xmath95 we discuss our results , including the physical properties of the peculiar t dwarf 2mass 0937 + 2931 , a possible means of disentangling gravity and temperature in red optical spectra , and a comparison between t dwarf optical and near - infrared classifications . we summarize our results in @xmath96 . we acquired lris red optical spectra for a sample of 13 t dwarfs identified in 2mass @xcite and sdss @xcite on three nights , 2000 march 5 , 2001 february 20 , and 2003 january 2 ( ut ) . an observing log is given in table 1 . conditions during our march 2000 and january 2003 observations were clear with seeing @xmath10 1@xmath11 ; our february 2001 observations were made through light clouds with 1@xmath11 seeing . lris is nominally a dual spectrograph with a 5500 dichroic mirror separating incoming light into two ( red and blue ) channels . for all observations , we used only the red channel with a mirror bypassing the dichroic . the og570 order - blocking filter was employed to suppress higher - order light , and a 1@xmath11 ( 4.7 pixels ) slit was used with the 400 lines mm@xmath12 grating blazed at 8500 , yielding 630010100 spectra with @xmath47 resolution ( r @xmath4 1200 ) . dispersion on the chip was 1.9 pixel@xmath12 . targets were nominally acquired using the facility guide camera ; objects that were too optically faint to be seen in the guider image were placed into the slit by blind offset from a nearby visible source , using offset values determined from 2mass coordinates . individual exposures ranged from 1200 to 1800 sec , with up to four observations obtained on a particular night . observations of the b1 v flux standard hiltner 600 @xcite were obtained for flux calibration on all runs using the same instrumental configuration . to correct for telluric absorption , we observed either da / dc / dz white dwarfs selected from @xcite or late - f / early - g stars before or after the target observations and at similar airmass . these stars typically have few or no features in the red optical telluric absorption bands . quartz lamp flat - field exposures ( reflected off of the interior dome ) were observed at the start of each run to calibrate detector response , and near arc lamp exposures were obtained immediately after the target observations for wavelength calibration . science data were reduced using standard iraf routines . first , we removed the image bias . for the march 2000 data , this was done by median - combining a series of one - second bias frames , and subtracting this image from the science and calibration data . for the february 2001 and january 2003 data , the bias level for each row was derived from the mean counts in the overscan region and subtracted . images were trimmed and divided by the median - combined and normalized flat - field exposures to correct for detector response . spectra were then extracted using the iraf apextract package . curvature of the dispersion line was determined from either the telluric or flux standard star exposures and used as a template for the target extractions . all spectral data were optimally extracted with a cleaning algorithm to eliminate cosmic rays and spurious pixels . wavelength calibration was done using the arc lamp exposures and line identifications from the nist atomic line database . we then computed a telluric correction from the telluric calibrator spectra by interpolating over atmospheric o@xmath0 ( @xmath468506900 b - band , @xmath476007700 a - band ) and h@xmath0o ( @xmath471507300 , @xmath481508350 , and @xmath489509650 ; stevenson 1994 ) bands , and then dividing this modified spectrum by the original calibrator spectrum . this correction spectrum was applied to both target and flux standard data to eliminate telluric absorption from the spectrum . we note that we acquired the wrong star as the telluric calibrator for 2mass 0937 + 2931 . wd 0924 + 199 ( a.k.a pg 0924 + 199 , ton 1061 ) is classified dc5 @xcite , with a possible indication of h@xmath2 absorption and an 11,000 k blackbody continuum . our observations show a roughly 5000 k blackbody continuum , weak and narrow h@xmath2 absorption , a strong ca ii resonance triplet ( 8498 , 8542 , & 8662 ) and a ba ii line at 6497 . this object is likely a background f / g - type dwarf , and fortuitously works fine as a telluric calibrator . finally , flux calibration was done by correcting our hiltner 600 observations to the spectrophotometric data given by @xcite , being careful to interpolate over telluric absorption regions and the balmer h@xmath2 line . this correction was applied to the t dwarf data to produce the final calibrated spectra . if multiple spectra were obtained on a particular night , these were coadded with a sigma - clipping algorithm to improve signal - to - noise . reduced spectra for the t dwarfs are shown in figure 1 in order of their near - infrared spectral classifications . features present in the spectra are noted in figure 1 and listed in table 2 ; we discuss these features in detail below . the strongly pressure - broadened na i and k i resonance doublets suppress most of the optical flux of t dwarfs @xcite . the broad wings of both alkali doublets can be seen clearly in the highest signal - to - noise spectrum of our sample , that of 2mass 0559@xmath31404 , shown in figure 2 with absolute flux density ( calibrated using i@xmath13 photometry and parallax measurements from dahn et al . 2002 ) plotted on a logarithmic scale . the peak - up in flux between these features is clearly evident , and their shapes are consistent with heavily pressure - broadened atomic lines @xcite . other alkali line features present in these spectra include the 8521/8943 cs i and 7800/7948 rb i resonant doublets . examination of figure 1 shows that the cs i lines are clearly present in all of the t dwarfs observed , although they become rather weak in the latest - type objects . rb i lines lie very close to the core of the strong k i doublet , and are generally only seen in our higher signal - to - noise spectra ( e.g. , figure 2 ) . we have measured pseudo - equivalent widths ( pews ) for each of the cs i and rb i lines as follows : first , we subtracted a linear fit to the local pseudo - continuum , @xmath14 , to yield the line profile , @xmath15 . next , we fit each line profile to a lorentzian function of the form @xmath16 where @xmath17 , @xmath18 ( the transition wavelength in ) , and @xmath19 ( the effective line width in ) were left as free parameters . fits were made iteratively and confirmed by eye to eliminate spurious matches to noise spikes . finally , pseudo - equivalent widths were derived from @xmath20 uncertainties , or upper limits in the cases where no line could be visually identified , were derived from the uncertainty in the continuum fit and residuals in a 40 region of the spectrum around each alkali line after the line profile fit . in cases where the continuum was essentially undetected , no pew upper limits could be derived . values are listed in table 3 . early- and mid - type t dwarfs have cs i pews @xmath6 79 , with the exception of 2mass 0755 + 2212 , whose short wavelength spectrum is partly filled in by flux from a nearly - aligned background galaxy @xcite . both cs i lines weaken in the later - type objects , particularly the 8521 line , which is weakest and only marginally detected in the t8 2mass 0415@xmath30935 ( [email protected] ) . the residual strength of the 8943 line may be the result of contamination by ch@xmath1 absorption ( @xmath93.2 ; figure 4 ) . these trends are consistent with chemical equilibrium calculations by @xcite , which predict a gradual chemical depletion of cs to cscl at t @xmath10 1300 k , accelerated at t @xmath10 1000 k as nacl ( and atomic na ) converts to na@xmath0s and liberates elemental cl . note that k is depleted to kcl around the same temperature @xcite , weakening the k i wing and reducing the suppression of the local pseudo - continuum around the cs i lines . that the cs i lines nevertheless wane in the later t dwarfs provides supporting evidence that elemental cs is in fact depleted in their photospheres . the 7948 rb i line could be measured in about two - thirds of the objects in our sample . line strengths for the early- and mid - type t dwarfs vary substantially , likely due to the decreased signal - to - noise in this spectral region . surprisingly , many of the latest - type t dwarfs shown detectable , albeit marginal , 7948 rb i absorption . figure 3 shows a close - up around this line for five t dwarfs with positive detections ; for 2mass 0727 + 1710 , gliese 570d , and 2mass 0415@xmath30935 , these detections are clearly marginal . nonetheless , the resurgence of the rb i line could be a real effect . as with k , atomic rb is depleted to rbcl for t @xmath10 1000 k @xcite . the lower depletion temperature for rb implies a much higher relative abundance than cs in the later t dwarfs , particularly as rb is roughly 20 times more abundant in a solar metallicity mixture @xcite . furthermore , the weakening of the k i red wing is likely to enhance rb i pews . the 7800 rb i line is only detected in our two brightest t dwarfs , 2mass 0559@xmath31404 and 2mass 1503 + 2525 , with pews of 6@xmath215 and 12@xmath215 , respectively . one glaring absence among the alkali features is the 6708 li i resonance doublet . 2mass 0559@xmath31404 shows a possible feature close to this line ( figure 2 ) , but it is offset at 6690 and is likely a residual noise spike . all of the objects in our sample are presumably substellar based on their low t@xmath8s @xcite , and a good fraction likely have masses below the 0.06 m@xmath22 limit for li core fusion depletion @xcite . however , none exhibit li i absorption . atomic li is chemically depleted to licl , lih , and lioh for t @xmath10 1500 @xcite , at hotter temperatures than na , k , rb , and cs . a general weakening of the li i line is observed in substellar late - type l dwarfs @xcite , and the absence of this line in the t dwarfs is consistent with a temperature effect . however , we note that these observations are not consistent with current spectral models that use the @xcite chemical abundances , which continue to exhibit detectable li i absorption down to t@xmath8 @xmath6 600800 k @xcite . confirmation of chemical depletion for all of the alkali species will require examination of their respective chloride bands at far - infrared wavelengths . the most prominent molecular features in the red optical spectra of t dwarfs are the h@xmath0o bands at 9250 and 9450 . these bands strengthen considerably toward the latest - type objects in our sample , but are generally weaker than their near - infrared counterparts . as is the case in the near - infrared , the steam bands overlap telluric h@xmath0o absorption at 89509700 ( strongest from 93009650 ; stevenson 1994 ) ; however , the hotter brown dwarf bands clearly extend blueward to 9250 . the 9896 wing - ford feh band is seen quite clearly in many of the spectra of figure 1 ; figure 2 of @xcite shows a close - up of this feature and the 9969 crh band . both bands are strongest in mid - type t dwarfs like 2mass 1534@xmath32952ab , but are weak or absent in the latest - type t dwarfs . the identification of the crh band may be in error , as @xcite find that absorption in this region for m and l dwarfs is entirely attributable to feh absorption . while opacity data from @xcite clearly show a crh bandhead at this wavelength , its visibility depends on the relative abundance of crh to feh , as their opacities are roughly equal in strength at these wavelengths @xcite . @xcite predicts an equilibrium abundance crh / feh @xmath6 10@xmath23 for 1800 @xmath10 t @xmath10 2500 k , but a higher relative abundance of @xmath40.3 for t @xmath10 1500 k. hence , it is entirely possible that the 9969 crh band is by feh absorption in m and l dwarf spectra , but detectable in the cooler t dwarf spectra . higher - resolution observations are required to test this possibility . the shorter - wavelength counterparts of the feh and crh bands at 8692 and 8611 , respectively , are present in the spectrum of sdss 1254@xmath30122 , but absent in the remaining spectra . a weak signature of cah at 67507150 , seen in m and l dwarf spectra , also appears to be present but weak in the spectra of 2mass 0559@xmath31404 ( figure 2 ) , sdss 1254@xmath30122 , and 2mass 1503 + 2525 , and is not seen in any of the other spectra . in chemical equilibrium , cah and elemental ca are expected to be depleted to the solid minerals perovskite ( catio@xmath24 ) , grossite ( caal@xmath0o@xmath1 ) , hibonite ( caal@xmath25o@xmath26 ) , and gehlenite ( ca@xmath0al@xmath0sio@xmath27 ) at t @xmath10 1900 k @xcite . the detection of this weak band in mid - type t dwarfs is therefore unexpected and may be indicative of non - equilibrium processes . however , improved opacities for cah are needed before this scenario can be properly explored . while ch@xmath1 is ubiquitous in the near - infrared spectra of t dwarfs , only the relatively weak 8950 ch@xmath1 band , noted by @xcite in gliese 229b , is likely to be present in the red optical spectra of the latest - type t dwarfs . figure 4 shows a close - up of this spectral region for six t dwarfs including gliese 229b . absorption at the base of the 8943 cs i line clearly becomes very broad in the later - type objects , and is strongest in 2mass 0415@xmath30935 . as shown in this figure , the breadth of this absorption is much greater than that expected for a pressure - broadened line , which exhibits a lorentzian line profile at detunings of a few 10s of angstroms @xcite . furthermore , cs i pews are weakening in the latest - type t dwarfs , while the broad absorption in this region is clearly strengthening . below the spectra we plot a ch@xmath1 opacity spectrum generated from the hitran database ( * ? ? ? * see references in burrows et al . 1997 ) . while the broad trough around the cs i line , and a weak feature at 8835 in the spectrum of 2mass 0415@xmath30935 , are generally consistent with the ch@xmath1 opacity data , none of the t dwarf spectra , including that of gliese 229b , exhibit the strongest feature in the opacity data at 8875 . however , it is important to consider that the ch@xmath1 opacity data is based on laboratory and planetary measurements made at temperatures significantly below those of a typical t dwarf photosphere ; hence , the overall band shape may be somewhat different in the latter environment . pending improved opacity data , we consider the detection of ch@xmath1 at 8950 to be tentative . three other ch@xmath1 features at 7300 , 8000 , and 10000 seen in planetary and laboratory spectra @xcite are intrinsically weaker than the 8950 band and are not seen in the t dwarf data . h@xmath2 emission at 6563 is detected in three objects in our sample : 2mass 1047 + 2124 , 2mass 1237 + 6526 ( previously reported in burgasser et al . 2000a ) , and sdss 1254@xmath30122 . figure 5 shows a close - up of the h@xmath2 spectral region for these three objects . note that the emission lines in 2mass 1047 + 2124 and sdss 1254@xmath30122 are very weak , detected at the 2.2 and 3@xmath28 levels , respectively , while the emission in 2mass 1237 + 6526 is substantial . none of the other t dwarfs exhibit discernible emission at 6563 . emission fluxes and 3@xmath28 upper limits are listed in table 4 , along with @xmath29 estimates , calculated from 2mass j - band magnitudes and using a linear interpolation of the j - band bolometric correction as a function of near - infrared spectral type , @xmath30 ( spt(t5 ) = 5 , etc . ) . this relation is based on luminosity estimates for 2mass 0559@xmath31404 ( * ? ? ? * @xmath31 ) , gliese 229b ( * ? ? ? * @xmath32 ) , and gliese 570d ( * ? ? ? * @xmath33 ) . for sdss 1254@xmath30122 , we used an average @xmath34 between 2mass 0559@xmath31404 and the latest - type l dwarfs ( * ? ? ? * @xmath35 ) . note that pew measurements of the h@xmath2 line is prohibited in most of these objects by the absence of a detected continuum . the detection of h@xmath2 emission in three t dwarfs appears to stand contrary to the reduced ionization fractions predicted in their very cool atmospheres @xcite . the lack of ionized material prevents the coupling of magnetic field lines to the upper atmosphere , thereby discouraging the formation of a substantial chromosphere via collisional heating @xcite . however , the relative emission luminosity of 2mass 1047 + 2124 and sdss 1254@xmath30122 is roughly 2 orders of magnitude lower than that of active mid - type m dwarfs ( * ? ? ? * @xmath36 ) , and is substantially lower than that of active l dwarfs @xcite , as shown in figure 3 of @xcite . hence , the emission in these objects is consistent with reduced , but existent , chromospheric activity . the strong , persistent emission in 2mass 1237 + 6526 is discussed in further detail in @xcite . one other emission feature at 7740 is detected in the spectrum of 2mass 0755 + 2212 , noted in figure 1 . as described in @xcite , this object is aligned with a background galaxy , and the observed emission line is consistent with h@xmath2 redshifted to @xmath37 . the background galaxy also contributes continuum flux in the blue portion of the spectrum of 2mass 0755 + 2212 , filling in some of the alkali absorption features . the spectral features described above show clear trends with near - infrared spectral type : an increasing spectral slope and strengthening h@xmath0o absorption ; a peak in the cs i line strengths in the early-/mid - type t dwarfs ; loss of feh and crh absorptions at 8692 and 8611 in the early - type t dwarfs , and at 9896 and 9969 in the late - type t dwarfs ; and the possible emergence of ch@xmath1 at 8950 in the latest - type t dwarfs . hence , the makings of a spectral sequence in the red optical that parallels the near - infrared sequence are apparent . the low signal - to - noise in most of these spectra implies that a detailed scheme is generally not possible . we have therefore aimed at deriving a rough classification that parallels the near - infrared scheme , enabling a general comparison between near - infrared and optical spectral morphologies . we stress that our optical scheme is based entirely on red optical features , and that we have no _ a priori _ expectation that near - infrared and optical spectral morphologies must necessarily coincide . we have augmented our observed spectral data with t dwarf lris data from @xcite and @xcite . all of the spectra in our full sample have therefore been acquired and reduced in an identical manner ( with the exception of telluric absorption correction ) and can therefore be reliably compared . our procedures for classification followed the basic tenets of the mk method @xcite . we chose four t dwarfs sdss 1254@xmath30122 ( t2 ) , 2mass 0559@xmath31404 ( t5 ) , sdss 1624 + 0029 ( t6 ) , and 2mass 0415@xmath30935 ( t8 ) and the l8 2mass 1632 + 1904 @xcite as our spectral standards . three of the t dwarfs are near - infrared standards in the @xcite classification scheme , while sdss 1624 + 0029 replaces the t6 near - infrared standard 2mass 1225@xmath32739ab , which has been identified as an unequal - magnitude binary @xcite . note that the spectra of 2mass 1632 + 1904 and sdss 1624 + 0029 have not been corrected for telluric absorption . to improve the signal - to - noise of the spectra of sdss 1254@xmath30122 and 2mass 0415@xmath30935 , we combined our data with those of @xcite . we did not use the t1 and t3 near - infrared standards sdss 0837@xmath30000 and sdss 1021@xmath30304 because of their low signal - to - noise data . this choice leaves sdss 1254@xmath30122 as the only early - type t dwarf standard , so classifications between l8 and t5 are not intended to be robust ; see @xcite for a more detailed discussion on the optical spectra of early - type t dwarfs . we have nevertheless left placeholders for the unassigned intermediate subtypes ( t0 , t1 , t3 , t4 , and t7 ) to allow a consistent comparison with near - infrared classifications . to avoid confusion , we hereafter designate near - infrared and optical classifications as t@xmath38 and t@xmath39 , respectively . all of the spectral standards are plotted together in figure 6 in both linear and logarithmic scales . the trends discussed above are readily apparent over this short sequence . more importantly , the various standards are clearly distinguishable , and hence represent distinct morphological classes . in other words , the selected standards would have been chosen to represent separate optical classes regardless of their near - infrared types . classification of the remaining spectra in our full sample was first done by visual comparison to the spectral standards . we have found that the optimal means of doing this is by comparison on a logarithmic scale ( e.g. , right panel in figure 6 ) . this method permits simultaneous examination of spectral slopes ranging over two orders of magnitude in flux density with weaker molecular and atomic features . we have made visual classifications for all objects with lris spectra previously typed t@xmath380 or later ; values are listed in table 7 . in general , visual types for the t dwarfs correspond well with the near - infrared classifications , agreeing to within @xmath211 subtype for nearly every source . one exception is sdss 0423@xmath30414 , which has an earlier - type spectrum than our l8 standard ; @xcite , who classify this object as l7.5 in the optical , discusses this mismatch and its possible physical basis in detail . 2mass 0937 + 2931 is assigned a peculiar visual classification because of its extreme red spectral slope , far in excess of any of the standards . we discuss this object in further detail in @xmath95.2 . a more quantitative approach to classification is through the use of spectral indices , ratios of flux or flux density that measure the strengths of particular absorption or pseudo - continuum features . we examined the behavior of 20 indices sampling the cs i , feh , crh , h@xmath0o , and spectral slope features on our t dwarf spectra . some of these were taken from the literature and possibly modified @xcite , and some are of our own construction . the most useful indices are those that show the clearest trends and greatest contrast with spectral type in the standards . all of the indices were compared , and those that best satisfied these criteria were chosen . eight promising indices were identified , defined in table 5 . the cs i indices are essentially identical to those defined by @xcite , measuring the strengths of the 8521 and 8943 cs i lines , respectively . these indices appear to peak around t@xmath392 , then weaken toward the later subclasses , consistent with the behavior of the corresponding pews . the h@xmath0o index samples the 9250 band and may be weakly affected by telluric absorption , although it has been defined to avoid the strongest absorption longward of 9300 @xcite . this index increases dramatically beyond roughly t@xmath392 . the feh(a ) and crh(a ) indices measure the shorter wavelength bands of these molecules , which disappear around type t@xmath392 , after which the indices essentially measure a limited range of spectral slope . the feh(b ) and crh(b ) indices measure the longer wavelength bands , and show rather complex behavior , weakening in the late l and early t dwarfs , strengthening again in the mid - type t dwarfs , and finally weakening in the late t dwarfs . @xcite have proposed that this trend is attributable to the disruption of clouds across the l / t transition . their complex behavior limits the use of the feh(b ) and crh(b ) indices to spectral types t@xmath395 and later . finally , the color - e index measures the red spectral slope due to k i absorption , but is defined at longer wavelengths than the color - d @xcite or pc3 @xcite indices because of the absence of detectable flux shortward of 8000 in many of the spectra . this index is useful earlier than roughly t@xmath392 ( i.e. , around the l / t transition ) , but appears to saturate ( with considerable scatter ) in the mid- and late - type t dwarfs , possibly due to a turnover in the strength of the k i red wing ( however , see @xmath95.2 ) . taking into account similar behavior between pairs of indices and differing sensitivity for early and late t types , we chose to use the csi(a ) , feh(b ) , color - e , and hybrid crh(a)/h@xmath0o ratios for our classification . these ratio values are plotted versus optical type in figure 7 . ratio values for the five spectral standards are listed in table 6 , and values for the remaining t dwarf spectra in our sample , as well as the associated standard types for each ratio , are listed in table 7 . final classifications were derived by averaging those ratio types ( feh(b ) was not used for types earlier than t@xmath395 , and color - e was not used for types later than t@xmath392 ) and rounding off to the nearest whole subclass . individual ratio types for each object agree within roughly one standard class . note that sdss 0837@xmath30000 and sdss 1021@xmath30304 are assigned uncertain classifications because of the substantial gaps between the nearest standard subclasses . with the exception of these two objects , all of our spectral ratio - derived classifications agree with our visually - derived classifications within one subclass , implying that our suite of indices accurately represent the overall spectral morphology . furthermore , there is excellent agreement between the optical ratio classifications and the near - infrared classifications , again with the singular exception of sdss 0423@xmath30414 . based on these results , we conclude that classification of mid- and late - type t dwarfs in the optical is feasible , either by visual comparison to the selected standards or through the use of spectral indices defined in table 5 . however , our scheme is only accurate to within one subclass , and is therefore not as precise as current near - infrared schemes . the behavior of the color - e and crh(a)/h@xmath0o indices also suggest that early - type t dwarfs may be delineated in the optical , suggesting that a clear segregation of l and t dwarfs at these wavelengths is possible . optical spectral observations of additional objects in this spectral type range will have to be made , however . we note that disagreement between the optical and near - infrared spectral morphologies of the key l / t transition object , sdss 0423@xmath30414 , does point to possible difficulties in this regime @xcite . the only object in our sample whose red optical spectrum is truly unique in comparison to the spectral standards is the t@xmath386p / t@xmath397p 2mass 0937 + 2931 . this object stands out in the near - infrared because of its enhanced cia h@xmath0 absorption ( resulting in a highly suppressed k - band peak ) and absence of the 1.25 @xmath5 k i lines , both of which point to a metal - poor and/or high gravity atmosphere @xcite . as noted in @xmath9 4.2 , the optical spectrum is peculiar because of its strong spectral slope , also a consequence of low atmospheric metallicity and/or high surface gravity @xcite . to examine the relative contribution of these secondary effects , we have compared the spectrum of 2mass 0937 + 2931 to theoretical models from @xcite . these models do not incorporate the most recent line broadening theory of @xcite , nor do they include feh absorption , a relatively strong feature in the spectrum of 2mass 0937 + 2931 . nevertheless , as the current state of the art , they serve to illustrate general trends . as discussed in @xcite , increased surface gravity and decreased metallicity both result in a higher pressure photosphere , and hence enhancement of the pressure - sensitive k i and na i line wings . this effect is strong enough to overcome the reduced column depth and decreased chemical abundance resulting from a higher gravity or lower metallicity atmosphere , respectively , which tend to weaken other chemical features , such as the 9250 h@xmath0o band . decreased temperature also strengthens the na i and k i lines , but results in stronger h@xmath0o bands as well ( see @xmath9 5.2 ) . when we compare these trends to the optical spectrum of 2mass 0937 + 2931 ( figure 8) , it becomes clear that the relatively weak h@xmath0o band seen in these data requires a moderately high t@xmath8 , while the strong na i and k i require both a high gravity and low metallicity atmosphere . the best fitting models from @xcite , t@xmath8 = 800 and 1000 k , g = @xmath40 cm s@xmath41 , and @xmath42 are shown in figure 8 along with the observed spectrum . both models show somewhat stronger inner k i wings than the data , consistent with the analysis of @xcite that the line - broadening theory used in @xcite overestimates the line opacity by as much as a factor of 10 shortward of 9000 . nevertheless , the reasonable fit of the models to the data indicate a low metallicity , high gravity atmosphere for this object . note that the constraint on t@xmath8 is fairly weak ; a robust estimate could be obtained by measurement of this object s parallax . the substantial proper motion of 2mass 0937 + 2931 ( @xmath43 ; burgasser et al . 2003a ) suggests that it is either quite close to the sun or has a high space velocity . @xcite have estimated a j - band spectrophotometric distance of roughly 9 pc , implying @xmath44 @xmath6 60 km s@xmath12 , which suggests ( but does not guarantee ) membership in the galactic thick disk or halo populations . such an object would have both a high surface gravity ( as it must be both old and therefore more massive for its temperature ) and a metal - deficient atmosphere @xcite . @xcite have recently identified an l - type metal - deficient dwarf which , like 2mass 0937 + 2931 , exhibits strong k i and metal hydride absorption , relatively blue near - infrared colors , and substantial space motion . determining whether 2mass 0937 + 2931 is a cool analog to this substellar subdwarf requires additional astrometric data and refined spectral models . nonetheless , the observed peculiarities in its optical and near - infrared spectra clearly indicate that 2mass 0937 + 2931 is a unique t dwarf requiring further investigation . if we ignore metallicity variations ( i.e. , excluding 2mass 0937 + 2931 ) , then differences amongst the spectra in our sample are determined largely by temperature and gravity effects . the na i and k i resonant doublet lines , which generally dominate this spectral region , are highly sensitive to both parameters . so too is the 9250 h@xmath0o band . however , while the alkali lines strengthen with increasing gravity , h@xmath0o absorption decreases because of the reduced column depth of the atmosphere @xcite . the contrasting behavior of these features suggests that they may be used to distinguish temperature and gravity effects in t dwarf spectra . as our sample likely includes objects with ages of roughly 15 gyr ( assuming they are drawn from the galactic disk population ; reid & hawley 2000 ) and 700 @xmath10 t@xmath8 @xmath10 1200 k @xcite , and hence surface gravities ( @xmath45 ) spanning the range @xmath46 cm s@xmath41 @xcite , demonstrative gravity effects may be present . as noted in @xmath9 4.3 , the color - e index ( which measure the k i red wing ) shows substantial scatter with optical spectral type for classes later than t@xmath392 ( figure 7d ) , while the combined crh(a)/h@xmath0o index is reasonably monotonic over this region . figure 9 compares the color - e and crh(a)/h@xmath0o indices for all of our telluric - calibrated spectra . the general trends with t@xmath8 and gravity are noted at the bottom , derived by measuring the indices on the spectral models of @xcite . for crh(a)/h@xmath0o @xmath10 0.7 ( spt @xmath47 t@xmath392 ) , there is a clear spread in the color - e values , with 2mass 0937 + 2931 being the most significant outlier . the simplest explanation for this divergence is intrinsic scatter in the indices , due to perhaps poor signal - to - noise in some of the spectra . however , we note two suggestive trends . first , the four objects with h@xmath2 emission ( including sdss 0423@xmath30414 ; kirkpatrick et al . 2003 ) generally have steeper spectral slopes ( larger color - e ) than other objects with similar crh(a)/h@xmath0o values . @xcite have noted that for late - type m and l field dwarfs , h@xmath2 emission is generally found amongst the older , and hence more massive , objects . if this relation follows through into the t dwarf regime , then we expect the active objects to have higher surface gravities , and therefore systematically larger color - e ratios . second , one object in our sample has an independent age , and hence gravity , determination , the companion t@xmath388/t@xmath397 gliese 570d @xcite . at 25 gyr , this brown dwarf has a surface gravity of ( 12)@xmath48 cm s@xmath41 @xcite . below this object is the [email protected]/t@xmath397 2mass 1217@xmath30311 @xcite , which @xcite have suggested may be a low - gravity t dwarf based on its somewhat brighter k - band peak ( implying weaker cia h@xmath0 absorption ) . again , the relative positions of these sources in figure 9 are consistent with the expected trends . these examples suggest that temperature- and gravity - based features could be disentangled , enabling a means of resolving the brown dwarf age / mass / temperature degeneracy . independent age and/or gravity information ( e.g. , kinematics , companionship to well - studied stars , or cluster membership ) or improved spectral models are needed to verify and calibrate these potentially useful diagnostics . one of our motivations for deriving a classification scheme for t dwarfs in the optical is to compare the evolution of spectral morphology between optical and near - infrared wavelengths . for this discussion , we focus on the mid- and late - type t dwarfs , t@xmath385/t@xmath395 and later . as discussed above , classifications for these objects in the two wavelength regimes considered here are remarkably consistent , implying a coherent evolution of features over 0.62.5 @xmath5 with spectral type . this may be somewhat surprising , given the competing temperature and gravity dependencies of defining h@xmath0o and ch@xmath1 bands @xcite and k i and na i resonance line wings @xcite , the relatively weak gravity dependence of the optical cs i lines @xcite , and the strong gravity dependence of cia h@xmath0 in the near - infrared @xcite . it may be the case that , despite the wide range of possible surface gravities in our sample , these particular t dwarfs have fairly similar masses and ages and hence little variation in @xmath45 . such a case would imply that temperature is the @xmath49 discriminant in current t dwarf classification schemes . however , the suggestive trends noted in the previous section do point to separable gravity / temperature diagnostics being present within our sample , and hence a non - negligible range of surface gravities . this possibility leads us to consider that competing gravity and temperature effects may compensate for each other amongst all of the optical and near - infrared diagnostic spectral features , and that t subtypes in either regime may not in fact map directly onto a temperature scale . the latter point is an important problem , as classifications are commonly used as proxies for t@xmath8 and luminosity in population studies . our results indicate that comparing optical and near - infrared morphologies can not , at the precision of our near - infrared and optical classifications , resolve dependencies between gravity and temperature in the spectra of mid- and late - type t dwarfs . however , developing the spectral diagnostics described in previous section may lead to a two - dimensional system that could be tied directly to temperature and gravity scales . the agreement between optical and near - infrared classifications for mid- and late - type t dwarfs also suggests that two possible causes for disagreements amongst late - type l and early - type t dwarf classifications condensate cloud opacity and/or duplicity @xcite may not be important for later - type t dwarfs . in general , t dwarf spectra and photometry are well - matched to models without condensate cloud opacity ( * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * however , see marley et al . 2002 ; burrows et al . 2002a ) , although low - level near - infrared photometric @xcite and spectroscopic @xcite variability , attributed to condensate clouds , has been observed in some t dwarfs . these time - dependent variations are typically quite small , however , and do not significantly change the overall spectral morphology @xcite . as for duplicity , later - type t dwarf doubles are not affected by the substantial redistribution of spectral flux that occurs across the l / t transition @xcite , which allows a late - type l dwarf to dominate optical emission while its early - type t dwarf companion dominates near - infrared emission . this scenario may explain the presence of ch@xmath1 at 2.2 @xmath5 in the l6.5 unequal - magnitude binary 2mass 0920 + 3517ab @xcite , as well as the near - infrared / optical classification discrepancy for sdss 0423@xmath30414 @xcite . in contrast , both of the known binaries in our sample , 2mass 1225@xmath32739ab and 2mass 1534@xmath32952ab @xcite , have essentially identical optical and near - infrared types . it is important to consider , however , that a larger , and perhaps time - resolved , optical sample is required to rigorously examine the influence of variability and duplicity on differences between optical and near - infrared t dwarf spectra . we have examined lris red optical spectra for a sample of t dwarfs . these spectra are largely dominated by pressure - broadened na i and k i features , which are highly sensitive to temperature , gravity , and metallicity . weaker lines of cs i and rb i are also present , and appear to follow the trends of current chemical equilibrium calculations . molecular bands of h@xmath0o , cah , crh , feh , and possibly ch@xmath1 are also present . h@xmath2 emission is seen in three objects , weakly in sdss 1254@xmath30122 and 2mass 1047 + 2124 , but strong in the active t dwarf 2mass 1237 + 6526 . none of the spectra exhibit the 6708 li i line , consistent with chemical depletion of this atomic species for t @xmath10 1500 k. trends in the observed strengths of these features have allowed us to define a classification scheme in the red optical , accurate to within one subclass . tying our scheme to selected near - infrared spectral standards , we are able to make a consistent comparison between optical and near - infrared types , and hence a critical examination of spectral morphologies in these two spectral regions . we find excellent agreement between these classifications amongst the mid- and late - type t dwarfs , suggesting that either current classification schemes are largely temperature - based , or that temperature and gravity effects compensate for each other throughout the 0.62.5 @xmath5 region . disentangling these physical parameters may be possible by comparing the strength of the k i red wing and the 9250 h@xmath0o band , potentially leading to a two - dimensional classification system that can be mapped onto temperature and gravity scales . both gravity and metallicity effects help explain the peculiar spectrum of 2mass 0937 + 2931 , which may be an old , massive , and slightly metal - poor thick disk or halo brown dwarf . further astrometric data and improved spectral models will enable a more rigorous characterization of this unique brown dwarf . clearly , while very little of the total luminosity of t dwarfs is emitted shortward of 1 @xmath5 , there are an abundance of diagnostics present in this spectral region that can be used to examine the physics of these cool brown dwarfs . we are grateful to our keck observing assistants joel aycock , julie rivera , and terry stickel , and instrument specialists / support astronomers paola amico , tom bida , and bob goodrich for their assistance during the data acquisition ; and to the nasa tac for its allocation of time for this project . we also thank our anonymous referee for her / his insightful comments and criticisms . a. j. b. acknowledges support provided by nasa through hubble fellowship grant hst - hf-01137.01 awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , incorporated , under nasa contract nas5 - 26555 . j. d. k. acknowledges the support of the jet propulsion laboratory , california institute of technology , which is operated under contract with the national aeronautics and space administration . a. b. acknowledges funding through nasa grants nag5 - 10760 and nag5 - 10629 . this publication makes use of data from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center , funded by the national aeronautics and space administration and the national science foundation . 2mass data were obtain through the nasa / ipac infrared science archive , which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . portions of the data presented herein were obtained at the w. m. keck observatory which is operated as a scientific partnership among the california institute of technology , the university of california , and the national aeronautics and space administration . the observatory was made possible by the generous financial support of the w. m. keck foundation . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . leggett , s. k. , golimowski , d. a. , fan , x. , geballe , t. r. , & knapp , g. r. 2003 , in proceedings of the 12th cambridge workshop on cool stars , stellar systems , and the sun , ed . a. brown , t. r. ayres , & g. m. harper ( boulder : univ . colorado press ) , p. 120 lllccllc 2mass j04151954@xmath30935066 & t@xmath388/t@xmath398 & 2001 feb 20 & 6000 & 1.201.38 & wd 0413 - 074 & da4 & 1 + 2mass j05591914@xmath31404488 & t@xmath385/t@xmath395 & 2000 mar 5 & 3600 & 1.201.22 & wd 0552 - 041 & dc9 & 2 + 2mass j07271824 + 1710012 & t@xmath387/t@xmath398 & 2000 mar 5 & 1800 & 1.00 & wd 0747 + 073.1 & dc9 & 1 + 2mass j07554795 + 2212169 & t@xmath385:/t@xmath396 & 2001 feb 20 & 2400 & 1.131.19 & sao 79820 & g0 v & 1 + 2mass j09373487 + 2931409 & t@xmath386p / t@xmath397 & 2000 mar 5 & 3600 & 1.01 & ( wd 0924 + 199 ) & ( dc5 ) & 1 + 2mass j10475385 + 2124234 & [email protected]/t@xmath397 & 2001 feb 20 & 4800 & 1.061.22 & hd 93583 & g0 v & 3 + 2mass j12171110@xmath30311131 & [email protected]/t@xmath397 & 2000 mar 5 & 1800 & 1.111.16 & wd 1225 - 079 & dza5 & 3 + 2mass j12255432@xmath32739466ab & t@xmath386/t@xmath396 & 2000 mar 5 & 3600 & 1.481.50 & wd 1225 - 079 & dza5 & 3 + 2mass j12373919 + 6526148 & [email protected]/t@xmath397 & 2000 mar 5 & 1800 & 1.46 & & & 3 + & & 2001 feb 20 & 2400 & 1.491.52 & sao 15828 & f8 v & 3 + sdssp [email protected] & t@xmath382/t@xmath392 & 2001 feb 20 & 3600 & 1.081.13 & hd 111942 & g0 v & 4 + gliese 570d & t@xmath388/t@xmath397 & 2000 mar 5 & 3600 & 1.331.36 & wd 1444 - 174 & dc9 & 5 + & & 2001 feb 20 & 4800 & 1.081.13 & hd 131878 & g0 v & 5 + 2mass j15031961 + 2525196 & [email protected]/t@xmath396 & 2003 jan 2 & 1200 & 1.351.40 & bd+26 2652 & g0 & 6 + 2mass j15344984@xmath32952274ab & [email protected]/t@xmath396 & 2001 feb 20 & 2400 & 1.581.63 & hd 138874 & f7 v & 1 + lccl na i & 5890 & 3s @xmath50s@xmath51 @xmath3 3p @xmath50p@xmath52 & 1,2 + na i & 5896 & 3s @xmath50s@xmath51 @xmath3 3p @xmath50p@xmath51 & 1,2 + h@xmath2 & 6563 & 3d @xmath50d@xmath53 @xmath3 2p @xmath50p@xmath52 & 1 + cah & 67507050 & 0 - 0 band of a@xmath50@xmath54-x@xmath50@xmath55 & 3 + k i & 7665 & 4s @xmath50s@xmath51 @xmath3 4p @xmath50p@xmath52 & 1,2 + k i & 7699 & 4s @xmath50s@xmath51 @xmath3 4p @xmath50p@xmath51 & 1,2 + rb i & 7800 & 5s @xmath50s@xmath51 @xmath3 5p @xmath50p@xmath56 & 1 + rb i & 7948 & 5s @xmath50s@xmath51 @xmath3 5p @xmath50p@xmath56 & 1 + ch@xmath1 & 88009200 & 4(@xmath57@xmath58,@xmath57@xmath24 ) & 4 + cs i & 8521 & 6s @xmath50s@xmath51 @xmath3 6p @xmath50p@xmath52 & 1 + crh & 8611 bandhead & 0 - 0 band of a@xmath59@xmath55@xmath60-x@xmath59@xmath55@xmath60 & 5 + feh & 8692 bandhead & 1 - 0 band of a@xmath61@xmath62-x@xmath61@xmath62 & 6 + cs i & 8943 & 6s @xmath50s@xmath51 @xmath3 6p @xmath50p@xmath51 & 1 + h@xmath0o & 92509400 & 3(@xmath57@xmath58,@xmath57@xmath24 ) & 7 + h@xmath0o & 94509800 & 2(@xmath57@xmath58,@xmath57@xmath24 ) + 2@xmath57@xmath0 & 7 + feh & 9896 bandhead & 0 - 0 band of a@xmath61@xmath62-x@xmath61@xmath62 & 6 + crh & 9969 bandhead & 0 - 1 band of a@xmath59@xmath55@xmath60-x@xmath59@xmath55@xmath60 & 5 + llccccccccccc sdss 1254@xmath30122 & t@xmath382/t@xmath392 & 8520@xmath212 & [email protected] & & 8942@xmath212 & [email protected] & & ... & @xmath63 7 & & 7947@xmath218 & [email protected] + 2mass 0559@xmath31404 & t@xmath385/t@xmath395 & 8522@xmath212 & [email protected] & & 8945@xmath212 & [email protected] & & 7800@xmath2130 & 6@xmath215 & & 7950@xmath2114 & [email protected] + 2mass 0755 + 2212 & t@xmath385:/t@xmath396 & 8521@xmath213 & [email protected] & & 8944@xmath212 & [email protected] & & ... & @xmath63 2 & & 7949@xmath2111 & [email protected] + 2mass 1503 + 2525 & [email protected]/t@xmath396 & 8519@xmath212 & [email protected] & & 8941@xmath212 & [email protected] & & 7800@xmath2130 & 12@xmath215 & & 7944@xmath2115 & [email protected] + 2mass 1534@xmath32952ab & [email protected]/t@xmath396 & 8517@xmath212 & [email protected] & & 8940@xmath212 & [email protected] & & ... & @xmath63 130 & & 7943@xmath218 & 15@xmath2112 + 2mass 1225@xmath32739ab & t@xmath386/t@xmath396 & 8522@xmath212 & [email protected] & & 8944@xmath212 & [email protected] & & ... & @xmath63 24 & & ... & @xmath63 10 + 2mass 0937 + 2931 & t@xmath386p / t@xmath397 p & 8520@xmath212 & [email protected] & & 8942@xmath212 & [email protected] & & ... & ... & & 7949@xmath2117 & 10@xmath216 + 2mass 1047 + 2124 & [email protected]/t@xmath397 & 8522@xmath213 & [email protected] & & 8946@xmath214 & [email protected] & & ... & @xmath63 25 & & ... & @xmath63 7 + 2mass 1237 + 6526 & [email protected]/t@xmath397 & ... & @xmath63 9 & & 8941@xmath213 & 8@xmath214 & & ... & ... & & ... & @xmath63 60 + & & 8522@xmath213 & [email protected] & & 8943@xmath212 & [email protected] & & ... & @xmath63 40 & & ... & @xmath63 14 + 2mass 0727 + 1710 & t@xmath387/t@xmath398 & 8519@xmath213 & [email protected] & & 8943@xmath212 & [email protected] & & ... & @xmath63 50 & & 7949@xmath2110 & 16@xmath217 + 2mass 1217@xmath30311 & [email protected]/t@xmath397 & 8521@xmath212 & [email protected] & & 8944@xmath213 & [email protected] & & ... & @xmath63 18 & & ... & @xmath63 8 + gliese 570d & t@xmath388/t@xmath397 & 8519@xmath212 & [email protected] & & 8942@xmath212 & [email protected] & & ... & @xmath63 18 & & ... & ... + & & 8523@xmath213 & [email protected] & & 8944@xmath212 & [email protected] & & ... & @xmath63 18 & & 7951@xmath217 & 11@xmath217 + 2mass 0415@xmath30935 & t@xmath388/t@xmath398 & 8520@xmath219 & [email protected] & & 8943@xmath212 & [email protected] & & ... & @xmath63 12 & & 7948@xmath217 & 11@xmath216 + llcc sdss 1254@xmath30122 & t@xmath382/t@xmath392 & [email protected] & @xmath64 + 2mass 0559@xmath31404 & t@xmath385/t@xmath395 & @xmath65 & @xmath66 + 2mass 0755 + 2212 & t@xmath385:/t@xmath396 & @xmath67 & @xmath68 + 2mass 1503 + 2525 & [email protected]/t@xmath396 & @xmath69 & @xmath70 + 2mass 1534@xmath32952ab & [email protected]/t@xmath396 & @xmath71 & @xmath72 + 2mass 1225@xmath32739ab & t@xmath386/t@xmath396 & @xmath73 & @xmath74 + 2mass 0937 + 2931 & t@xmath386p / t@xmath397 p & @xmath75 & @xmath76 + 2mass 1047 + 2124 & [email protected]/t@xmath397 & [email protected] & @xmath77 + 2mass 1237 + 6526 & [email protected]/t@xmath397 & [email protected] & @xmath78 + & & [email protected] & @xmath79 + 2mass 0727 + 1710 & t@xmath387/t@xmath398 & @xmath80 & @xmath81 + 2mass 1217@xmath30311 & [email protected]/t@xmath397 & @xmath82 & @xmath83 + gliese 570d & t@xmath388/t@xmath397 & @xmath84 & @xmath83 + & & @xmath85 & @xmath68 + 2mass 0415@xmath30935 & t@xmath388/t@xmath398 & @xmath63 7.9 & @xmath74 + lcccl csi(a ) & @xmath86+@xmath87 & @xmath88 & 8521 cs i & 1,2 + csi(b ) & @xmath89+@xmath90 & @xmath91 & 8943 cs i & 1,2 + h@xmath0o & @xmath92 & @xmath93 & 9250 h@xmath0o & 2 + crh(a ) & @xmath94 & @xmath95 & 8611 crh & 3 + crh(b ) & @xmath96 & @xmath97 & 9969 crh & 2,3 + feh(a ) & @xmath94 & @xmath98 & 8692 feh & 3 + feh(b ) & @xmath96 & @xmath99 & 9896 feh & 2,3 + color - e & @xmath100 & @xmath101 & spectral slope & 2 + llccccl 2mass 1632 + 1904 & l8 & 1.70 & 1.02 & 1.11 & 1.88 & 1 + sdss 1254@xmath30122 & t@xmath392 & 2.01 & 0.78 & 1.13 & 4.02 & 2,3 + 2mass 0559@xmath31404 & t@xmath395 & 1.77 & 0.63 & 1.37 & 4.24 & 2 + sdss 1624 + 0029 & t@xmath396 & 1.68 & 0.47 & 1.15 & 3.83 & 4 + 2mass 0415@xmath30935 & t@xmath398 & 1.19 & 0.25 & 0.94 & 4.20 & 2,3 + lccccllll sd 0423@xmath30414 & 1.95 ( t@xmath392 ) & 1.20 ( @xmath63 l8 ) & 1.44 ( @xmath63 t@xmath395 ) & 2.21 ( l8 ) & l8 : & @xmath63 l8 & t@xmath380 & 1 + sd 0837@xmath30000 & 1.80 ( l8/t@xmath392 ) & 0.89 ( l8/t@xmath392 ) & 0.90 ( @xmath63 t@xmath395 ) & 3.35 ( l8/t@xmath392 ) & t@xmath390@xmath102 & t@xmath392 & t@xmath381 & 1 + sd 1021@xmath30304 & 1.98 ( t@xmath392 ) & 0.67 ( t@xmath392/5 ) & 1.37 ( t@xmath395 ) & 3.85 ( @xmath103 t@xmath392 ) & t@xmath394@xmath102 & t@xmath392 & t@xmath383 & 1 + 2 m 1503 + 2525 & 1.61 ( t@xmath396 ) & 0.50 ( t@xmath396 ) & 1.30 ( t@xmath395/6 ) & 4.75 ( @xmath103 t@xmath392 ) & t@xmath396 & t@xmath395/6 & [email protected] & 2 + 2 m 1534@xmath32952ab & 1.70 ( t@xmath396 ) & 0.54 ( t@xmath395/6 ) & 1.38 ( t@xmath395 ) & 4.47 ( @xmath103 t@xmath392 ) & t@xmath396 & t@xmath395/6 & [email protected] & 2 + 2 m 0755 + 2212 & 1.50 ( t@xmath396/8 ) & 0.52 ( t@xmath395/6 ) & 1.33 ( t@xmath395 ) & 3.78 ( @xmath103 t@xmath392 ) & t@xmath396 & t@xmath395/6 & [email protected] : & 2 + 2 m 1225@xmath32739ab & 1.82 ( t@xmath395 ) & 0.41 ( t@xmath396 ) & 1.17 ( t@xmath396 ) & 4.28 ( @xmath103 t@xmath392 ) & t@xmath396 & t@xmath396 & t@xmath386 & 2 + sd 1346@xmath30031 & 1.43 ( t@xmath396/8 ) & 0.35 ( t@xmath396/8 ) & 1.01 ( t@xmath396/8 ) & 5.01 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath396/8 & t@xmath386 & 3 + 2 m 0937 + 2931 & 1.55 ( t@xmath396/8 ) & 0.38 ( t@xmath396/8 ) & 1.09 ( t@xmath396/8 ) & 5.99 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath396/8 p & t@xmath386 p & 2 + 2 m 1047 + 2124 & 1.46 ( t@xmath396/8 ) & 0.37 ( t@xmath396/8 ) & 1.00 ( t@xmath396/8 ) & 4.41 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath396/8 & [email protected] & 2 + 2 m 1237 + 6526 & 1.58 ( t@xmath396 ) & 0.37 ( t@xmath396/8 ) & 0.92 ( t@xmath398 ) & 4.93 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath396/8 & [email protected] & 2,3 + 2 m 0727 + 1710 & 1.28 ( t@xmath398 ) & 0.36 ( t@xmath396/8 ) & 0.92 ( t@xmath398 ) & 3.84 ( @xmath103 t@xmath392 ) & t@xmath398 & t@xmath396/8 & t@xmath387 & 2 + 2 m 1217@xmath30311 & 1.71 ( t@xmath395/6 ) & 0.34 ( t@xmath396/8 ) & 0.92 ( t@xmath398 ) & 3.77 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath398 & [email protected] & 2 + gliese 570d & 1.48 ( t@xmath396/8 ) & 0.34 ( t@xmath396/8 ) & 0.87 ( t@xmath398 ) & 4.39 ( @xmath103 t@xmath392 ) & t@xmath397 & t@xmath398 & t@xmath388 & 1,2 +

let be a numerical semigroup and let be the ( locally finite ) poset induced by on the set of integers defined by if and only if for all integers and . in this paper , we investigate the _ mbius function _ associated to when is an _ arithmetic _ semigroup . * keywords : * mbius function , poset , numerical semigroup , arithmetic semigroup * msc 2010 : * 20m15 ; 05a99 ; 06a07 ; 11a25 ; 20m05

1209.1880

the _ mbius function _ is an important concept associated to ( _ locally finite _ ) posets . mbius function can be considered as a generalization of the classical mbius arithmetic function on the integers ( given by the mbius function of the poset obtained from the positive integers partially ordered by the divisibility ) . mbius function has been extremely useful to investigate many different problems . for instance , the _ inclusion - exclusion principle _ can be retrieved by considering the set of all subsets of a finite set partially ordered by inclusion . we refer the reader to @xcite for a large number of applications of the mbius function . in this paper , we investigate the mbius function associated to posets arising naturally from numerical semigroups as follows . let @xmath7 be @xmath8 relatively prime positive integers and let @xmath9 denote the _ numerical semigroup _ generated by @xmath7 , that is , @xmath10 throughout this paper , we consider the structure of the poset induced by @xmath0 on the set of integers @xmath2 , whose partial order @xmath11 is defined by @xmath12 for all integers @xmath5 and @xmath6 . this ( locally finite ) poset will be denoted by @xmath1 . we denote by @xmath13 the mbius function associated to @xmath1 . as far as we are aware , the only known result concerning @xmath13 is an old theorem due to deddens @xcite that determines the value of @xmath13 when @xmath0 has exactly two generators . here , we shall introduce and develop a new approach to investigate @xmath13 when @xmath0 is an _ arithmetic semigroup _ , that is , when @xmath14 for some integers @xmath15 and @xmath16 . this is a self - contained paper and it is organized as follows . in the next section , we review some classic notions of the mbius function and present some results needed for the rest of the paper . in section [ dedd ] , we give a new direct proof of deddens result , shorter than the original one ( based on a recursive case - by - case analysis ) . in section [ prelim ] , we discuss results about arithmetic semigroups , in particular , we prove the existence of _ unique _ representations . the latter is a key result that will be used , in section [ recurs ] , to give a recursive formula for @xmath13 when @xmath14 . finally , in section [ pair ] , we propose an explicit formula for @xmath13 ( based on the _ multiplicity function _ of a multiset ) in the case when @xmath17 and @xmath18 is even . background information on numerical semigroups can be found in the books @xcite . let @xmath19 be a partially ordered set , or _ poset _ for short . the _ strict partial order _ @xmath20 is the reduction of @xmath21 given by , @xmath22 if and only if @xmath23 and @xmath24 . for any @xmath18 and @xmath25 in the poset @xmath26 , the _ segments _ between @xmath18 and @xmath25 are defined by @xmath27}_{p } & \left\ { c\in p\ \middle|\ a\leq_p c\leq_p b \right\ } , & { \left]a , b\right]}_{p } & \left\ { c\in p\ \middle|\ a < _ p c\leq_p b \right\ } , \\[2ex ] { \left[a , b\right[}_{p } & \left\ { c\in p\ \middle|\ a\leq_p c < _ p b \right\ } , & { \left]a , b\right[}_{p } & \left\ { c\in p\ \middle|\ a < _ p c < _ p b \right\}. \\ \end{array}\ ] ] a poset is said to be _ locally finite _ if every segment has finite cardinality . in this paper , we only consider locally finite posets . let @xmath18 and @xmath25 be elements of the poset @xmath26 . a _ chain _ of length @xmath28 between @xmath18 and @xmath25 is a subset of @xmath29}_{p}$ ] containing @xmath18 and @xmath25 , with cardinality @xmath30 and totally ordered by @xmath31 , that is @xmath32}_{p}$ ] such that @xmath33 for any nonnegative integer @xmath34 , we denote by @xmath35 the set of all chains of length @xmath34 between @xmath18 and @xmath25 . the cardinality of @xmath35 is denoted by @xmath36 . this number is always finite because the poset @xmath26 is supposed to be locally finite . for instance , the number of chains @xmath37 , where the poset is the set @xmath38 partially ordered by divisibility , is equal to @xmath39 . indeed , there are exactly @xmath39 chains of length @xmath39 between @xmath39 and @xmath40 in @xmath41}_{{\mathbb{n}}}=\left\{2,4,6,12\right\}$ ] , which are @xmath42 and @xmath43 . for any locally finite poset @xmath26 , the _ mbius function _ @xmath44 is the integer - valued function on @xmath45 defined by @xmath46 for all elements @xmath18 and @xmath25 of the poset @xmath26 . one can remark that this sum is always finite because , for @xmath18 and @xmath25 given , there exists a maximal length of a possible chain between @xmath18 and @xmath25 since the segment @xmath29}_{p}$ ] has finite cardinality . the concept of mbius function for a locally finite poset @xmath47 was introduced by rota in @xcite as the inverse of the zeta function in the incidence algebra of a locally finite poset . let us see this with more detail . consider the set @xmath48 of all real - valued functions @xmath49 for which @xmath50 if @xmath51 . the sum @xmath52 and the multiplication by scalars @xmath53 are defined as usual in @xmath48 . the product of two functions @xmath54 and @xmath55 in @xmath48 is defined by @xmath56}_{p}}f(a , c)g(c , b),\ ] ] for all @xmath57 . then @xmath58 appears as an associative algebra over @xmath59 . this is the _ incidence algebra _ of @xmath26 . the _ kronecker delta function _ @xmath60 , defined by @xmath61 for all @xmath57 , is the identity element of @xmath48 . the _ zeta function _ @xmath62 is defined by @xmath63 for all @xmath57 . rota @xcite proved that the zeta function @xmath64 ( called the _ inverse function _ ) is invertible in @xmath48 and showed that @xmath44 is recursively defined as follows : for all @xmath57 , by @xmath65 let us see that both definitions of @xmath13 given by ( [ eq1 ] ) and by ( [ eq2 ] ) are equivalent . for , let @xmath18 and @xmath25 be two elements of the locally finite poset @xmath26 such that @xmath66 . then , @xmath67a , b\right]}_{p}}c_{l-1}(c , b),\ ] ] for all positive integers @xmath34 . indeed , every chain @xmath68 can be seen like an extension of a chain of @xmath69 or of @xmath70 . obviously , the identity @xmath71 directly comes from ( [ eq1 ] ) since @xmath72 and @xmath73 for all @xmath74 . by combining ( [ eq3 ] ) and ( [ eq1 ] ) , for all @xmath75 , we obtain that @xmath76 finally , since @xmath24 , it follows that @xmath77 and thus @xmath78 similarly , using the second identity of ( [ eq3 ] ) , we can also prove that , whenever @xmath75 , we have @xmath79a , b\right]}_{p}}\mu_p(c , b).\ ] ] therefore the two definitions of the mbius function ( for a locally finite posets ) are the same . all the results presented in this paper are derived from the recursive formula presented in ( [ eq2 ] ) . let @xmath0 be a numerical semigroup and @xmath1 its associated poset . observe that @xmath1 is a locally finite poset since @xmath80}_{({\mathbb{z}},\leq_s ) } \right| \leq y - x$ ] , for all @xmath81 . it is easy to see that @xmath13 can be considered as a univariable function of @xmath2 . indeed , for all @xmath81 and for all @xmath82 , we have @xmath83 the above follows since the set @xmath84 is in bijection with @xmath85 . indeed the map that assigns the chain @xmath86 to the chain @xmath87 is clearly a bijection . thus , by definition of @xmath13 and equality we obtain @xmath88 for all @xmath81 . in the sequel of this paper we shall only consider the reduced mbius function @xmath89 defined by @xmath90 this recursive formula given by ( [ eq2 ] ) can be more easily presented when the locally finite poset is @xmath1 . [ in ] let @xmath0 be a numerical semigroup and let @xmath91 . then , @xmath92 from ( [ eq2 ] ) , we deduce that @xmath93 the result follows since , by definition of @xmath13 , @xmath94 unless @xmath95 . in @xcite , deddens proved the following . @xcite[thmdedd ] let @xmath18 and @xmath25 be two relatively positive integers and let @xmath96 . then , for all @xmath97 , we have @xmath98 dedden s proof was based on a recursive argument and a case - by - case analysis . we may give the following direct proof of theorem [ thmdedd ] . _ proof of theorem [ thmdedd ] . _ we shall prove that @xmath99 for every @xmath100 . the result then follows since @xmath101 for all @xmath102 , @xmath103 , @xmath104 and @xmath105 . let us prove then equality . let @xmath106 and let @xmath107 . by proposition [ in ] we already know that @xmath108 since @xmath109 then @xmath110 moreover since @xmath111 then @xmath112 by applying for @xmath113 , that is , @xmath114 we obtain that @xmath115 by combining and , for @xmath116 , we obtain that @xmath117{l } = -\displaystyle\sum_{m_b=1}^{a-1}\mu_s(x - m_bb ) = -\mu_s(x - b)-\sum_{m_b=2}^{a-1}\mu_s(x - m_bb ) \\ = \displaystyle\sum_{m_b=2}^{a}\mu_s(x - m_bb)-\sum_{m_b=2}^{a-1}\mu_s(x - m_bb ) \\[3ex ] = \mu_s(x - ab ) \end{array}\ ] ] as desired . let @xmath0 be a numerical semigroup . the _ apry set _ of @xmath0 with respect with @xmath118 is defined as @xmath119 it is known that @xmath120 constitutes a complete set a of residues @xmath121 . roberts @xcite has proved that if @xmath122 with @xmath123 and @xmath124 then @xmath125 the following result gives a unique representation of elements in arithmetic semigroups . [ rep ] let @xmath126 with @xmath127 . then , there exists a unique triplet @xmath128 such that @xmath129 for some @xmath130 with @xmath131 . let @xmath132 and @xmath133 such that @xmath134 : let @xmath135 and @xmath136 be two integers such that @xmath137 . we notice that @xmath138 can be expressed as either @xmath139 or @xmath140 so , by repeatedly adding consecutive terms from the expression of @xmath5 in , we obtain that there exists a triplet @xmath141 such that @xmath142 with @xmath130 and @xmath143 . moreover , we may suppose that @xmath144 . otherwise , we use the following equality @xmath145{l } = \displaystyle\left ( \left\lfloor \frac{a}{k } \right\rfloor + 1 \right)a + \left ( \left\lfloor \frac{a}{k } \right\rfloor k + k \right)d \\[2ex ] = \displaystyle\left ( \left\lfloor \frac{a}{k } \right\rfloor + d + 1 \right)a + \left ( \left\lfloor \frac{a}{k } \right\rfloor k + k - a \right)d \\[2ex ] = \displaystyle\left ( \left\lfloor \frac{a}{k } \right\rfloor + d \right)a + \left ( a + \left ( \left\lfloor \frac{a}{k } \right\rfloor k + k - a \right)d \right ) , \end{array}\ ] ] where @xmath146 . finally , if @xmath147 then we consider the following representation @xmath148{l } = x_0a + x_i(a+id ) + x_k(a+kd ) \\[2ex ] = \displaystyle\left ( x_0+x_i+ x_k \right)a + \left ( i x_i+ k x_k\right)d \\[2ex ] = \displaystyle\left ( x_0 + x_i+ x_k+d \right)a + \left ( i x_i+ k x_k - a \right)d \\[2ex ] = \displaystyle\left ( x_0 + x_i + x_k+ d - 1 \right)a + \left ( a + \left ( i x_i+ k x_k- a \right)d \right ) , \end{array}\ ] ] where @xmath149 . obtaining the desired decomposition . : let us suppose that there exist two triplets of non - negative integers @xmath141 and @xmath150 such that @xmath151 with @xmath152 , @xmath153 , @xmath154 and @xmath155 . it follows that @xmath156 and since @xmath123 then @xmath157 moreover , since @xmath158 and @xmath159 , then @xmath160 we have four cases . case 1 ) if @xmath161 and @xmath162 then @xmath163 would be a multiple of @xmath164 which is impossible since @xmath165 case 2 ) if @xmath166 and @xmath167 then @xmath168 would be a multiple of @xmath164 which is impossible since @xmath169 case 3 ) if @xmath170 then @xmath171 would be a multiple of @xmath164 but since @xmath172 then @xmath173 implying that @xmath174 and thus @xmath175 . case 4 ) if @xmath176 then @xmath177 and since @xmath178 then @xmath174 and thus @xmath175 . let @xmath179 with @xmath130 and @xmath131 . we shall denote by @xmath180 $ ] the element in @xmath0 given by the representation of lemma [ rep ] . we shall now present a recursive formula for @xmath13 when @xmath122 . the following key remark led us to guess such recursion . if @xmath181 such that @xmath182 and @xmath183 then , @xmath184 [ thm ] let @xmath122 with @xmath123 and let @xmath185 with @xmath186 . let @xmath187 , then @xmath188 let @xmath107 . as for the proof of theorem [ thmdedd ] , we have @xmath117{l } = -\hspace{-10pt}\displaystyle\sum_{y\in s\setminus\{0\}}\mu_s(x - y ) \\[4ex ] = -\hspace{-8pt}\displaystyle\sum_{\stackrel{y\in s\setminus\{0\}}{y - a\in s}}\mu_s(x - y ) -\hspace{-8pt}\sum_{\stackrel{y\in s\setminus\{0\}}{y - a\notin s}}\mu_s(x - y ) \\[6ex ] = -\hspace{-2pt}\displaystyle\sum_{z\in s}\mu_s((x - a)-z ) -\hspace{-8pt}\sum_{\stackrel{y\in s\setminus\{0\}}{y - a\notin s}}\mu_s(x - y ) \\ = -\hspace{-8pt}\displaystyle\sum_{\stackrel{y\in s\setminus\{0\}}{y - a\notin s}}\mu_s(x - y)\quad\text{if}\quad x - a\neq 0 . \end{array}\ ] ] let us now determine the set @xmath189 for , we consider the set @xmath190 given by in function of the unique representation of lemma [ rep ] . we have three cases . case a ) if @xmath191 then @xmath192{l } \displaystyle\left\{y_k(a+kd)\ \middle|\ y_k\in\{1,\ldots , q-1\}\right\ } \\ \displaystyle\bigcup \left\ { ( a+id)+y_k(a+kd)\ \middle|\ \begin{array}{l } y_k\in\{0,\ldots , q-1\}\\ i\in\{1,\ldots , k-1\ } \\ \end{array } \right\ } \\ \end{array}\ ] ] case b ) if @xmath193 then @xmath192{l } \displaystyle\left\{y_k(a+kd)\ \middle|\ y_k\in\{1,\ldots , q\}\right\ } \\ \displaystyle\bigcup \left\ { ( a+id)+y_k(a+kd)\ \middle|\ \begin{array}{l } y_k\in\{0,\ldots , q-1\}\\ i\in\{r,\ldots , k-1\ } \\ \end{array } \right\ } \\ \end{array}\ ] ] case c ) if @xmath194 then @xmath192{l } \displaystyle\left\{y_k(a+kd)\ \middle|\ y_k\in\{1,\ldots , q\}\right\ } \\ \displaystyle\bigcup \left\ { ( a+id)+y_k(a+kd)\ \middle|\ \begin{array}{l } y_k\in\{0,\ldots , q\}\\ i\in\{1,\ldots , r-1\ } \\ \end{array } \right\ } \\ \displaystyle\bigcup \left\ { ( a+id)+y_k(a+kd)\ \middle|\ \begin{array}{l } y_k\in\{0,\ldots , q-1\}\\ i\in\{r,\ldots , k-1\ } \\ \end{array } \right\ } \\ \end{array}\ ] ] suppose that @xmath191 , i.e. @xmath195 . for @xmath196 , we have @xmath197 by applying to @xmath198 , that is , @xmath199 we obtain @xmath200 by combining and for @xmath201 , we obtain @xmath202 the cases when @xmath203 are similar to case a ( and it is left to the reader as an exercise ) . the _ multiplicity function _ of a multiset @xmath205 of @xmath38 is the function @xmath206 which assigns to each element @xmath207 its multiplicity , that is , the number of times that @xmath5 appears in the multiset @xmath205 . let @xmath208 and @xmath209 such that @xmath210 . for each @xmath211 , we consider the following multisets . @xmath212 b_i = \left\ { m(q+d)-nd+i \ \middle|\ m\in{\mathbb{n}}\setminus\{0,1\ } , n\in\{1,2,\ldots,\left\lfloor m/2\right\rfloor\ } \right\ } , \\[2ex ] c_i = a_i\bigcup b_i . \end{array}\ ] ] as we mentioned above , given a triple @xmath213 , we denote by @xmath214 $ ] the element in @xmath0 given by the representation in lemma [ rep ] . in the sequel of this section , we shall consider this representation for all @xmath215 , i.e. , @xmath216 = x_0a + x_1(a+d ) + x_2(a+2d)\ ] ] for all @xmath217 . in this case , it is clear that @xmath218\in s\longleftrightarrow x_0\in{\mathbb{n}}.\ ] ] [ thm1 ] let @xmath219 . let @xmath220 . then , @xmath221 ) = \left\{\begin{array}{ll } { ( -1)}^{x_1 } \left ( { \mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 ) & \text { if } x_2=0,\\ { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2 ) & \text { if } x_2\geq 1 . \end{array}\right.\ ] ] we notice that if @xmath222 is a constant then we should have the same value for @xmath223)$ ] . the latter is illustrated by the first values of @xmath223)$ ] , listed in table [ table ] given at the end of the section , for the case when @xmath224 and @xmath225 . indeed , we can see appearing diagonals ( corresponding to @xmath222 constant ) with the same value . [ prop1 ] let @xmath220 with @xmath226 . then , @xmath227\right ) = \begin{array}[t]{l } \mu_s\left([x_0-(q+d),x_1,0]\right ) \\[1.5ex ] + \mu_s\left([x_0-(q+d)-1,x_1,q-1]\right ) \\[1.5ex ] - \mu_s\left([x_0 - 2(q+d)-1,x_1,q-1]\right ) \\ \end{array}\ ] ] and @xmath228\right ) = \begin{array}[t]{l } \mu_s\left([x_0-(q+d),x_1,x_2]\right ) \\[1.5ex ] + \mu_s\left([x_0 - 1,x_1,x_2 - 1]\right ) \\[1.5ex ] - \mu_s\left([x_0-(q+d)-1,x_1,x_2 - 1]\right ) \\ \end{array}\ ] ] when @xmath229 . @xmath235\right ) \begin{array}[t]{l } = \begin{array}[t]{l } \mu_s\left ( [ x_0,0,0 ] - [ q+d,0,0 ] \right ) + \mu_s\left ( [ x_0,0,0 ] - [ q+d,1,0 ] \right ) \\ - \mu_s\left ( [ x_0,0,0 ] - [ 0,1,0 ] \right ) . \\ = \begin{array}[t]{l } \mu_s\left ( [ x_0-(q+d),0,0 ] \right ) + \mu_s\left ( [ x_0 - 2(q+d)-1,1,q-1 ] \right ) \\ - \mu_s\left ( [ x_0-(q+d)-1,1,q-1 ] \right ) . \end{array } \\ \end{array}\ ] ] by applying the recursive equality to @xmath236 \right)$ ] , we obtain @xmath237\right ) = \begin{array}[t]{l } \displaystyle\mu_s\left ( [ x_0 - 2(q+d)-1,1,q-1 ] \right ) \\[2ex ] + \displaystyle\mu_s\left ( [ x_0 - 2(q+d)-1,0,q-1]\right ) \\[2ex ] - \displaystyle\mu_s\left ( [ x_0-(q+d)-1,0,q-1]\right ) . \end{array}\ ] ] finally , by combining equations and we have @xmath238\right ) = \begin{array}[t]{l } \mu_s\left([x_0-(q+d),0,0]\right ) \\[1.5ex ] + \mu_s\left([x_0-(q+d)-1,0,q-1]\right ) \\[1.5ex ] - \mu_s\left([x_0 - 2(q+d)-1,0,q-1]\right ) . \\ \end{array}\ ] ] case b ) if @xmath233 and @xmath239 then , from we obtain @xmath240\right ) \begin{array}[t]{l } = \begin{array}[t]{l } \mu_s\left ( [ x_0,0,x_2 ] - x[q+d,0,0 ] \right ) + \mu_s\left ( [ x_0,0,x_2 ] - [ q+d,1,0 ] \right ) \\ - \mu_s\left ( [ x_0,0,x_2 ] - x[0,1,0 ] \right ) \end{array } \\ = \begin{array}[t]{l } \mu_s\left ( [ x_0-(q+d),0,x_2 ] \right ) + \mu_s\left ( [ x_0-(q+d)-1,1,x_2 - 1 ] \right ) \\ - \mu_s\left ( [ x_0 - 1,1,x_2 - 1 ] \right ) . \end{array } \\ \end{array}\ ] ] @xmath242\right ) = \begin{array}[t]{l } \displaystyle\mu_s\left ( [ x_0-(q+d)-1,1,x_2 - 1 ] \right ) \\[2ex ] + \displaystyle\mu_s\left ( [ x_0-(q+d)-1,0,x_2 - 1]\right ) \\[2ex ] - \displaystyle\mu_s\left ( [ x_0 - 1,0,x_2 - 1]\right ) . \end{array}\ ] ] finally , by combining equations and we have @xmath243\right ) = \begin{array}[t]{l } \mu_s\left([x_0-(q+d),0,x_2]\right ) \\[1.5ex ] + \mu_s\left([x_0 - 1,0,x_2 - 1]\right ) \\[1.5ex ] - \mu_s\left([x_0-(q+d)-1,0,x_2 - 1]\right ) . \\ \end{array}\ ] ] this concludes the proof for @xmath233 . the proof for the case @xmath244 is similar as the above case and it is left to the reader . by definition of the sets @xmath247 , for any integer @xmath248 such that @xmath249 , we have @xmath250 for any integer @xmath251 , we obtain @xmath252{ll } x\in a_i & \longleftrightarrow \ \text{there exists}\ m\in{\mathbb{n}}\ \text{with}\ \ x = m(q+d)+i \\ & \longleftrightarrow \ \text{there exists}\ m\in{\mathbb{n}}\ \text{with}\ \ x-(q+d)=(m-1)(q+d)+i \\ & \longleftrightarrow \ x-(q+d)\in a_i . \end{array}\ ] ] this completes the proof . by definition of the multisets @xmath254 and @xmath255 , for any integer @xmath256 , we have @xmath257 for any integer @xmath258 , we obtain @xmath252{ll } x\in b_i & \longleftrightarrow \ \text{there exists}\ m\in{\mathbb{n}}\setminus\{0,1\},\ 1\leq n\leq \left\lfloor m/2\right\rfloor \ \text{with}\ \ x = m(q+d)-nd+i \\ & \longleftrightarrow \ \text{there exists}\ m\in{\mathbb{n}}\setminus\{0,1\},\ 1\leq n\leq \left\lfloor m/2\right\rfloor \ \text{with}\ \ x-(2q+d)=\begin{array}[t]{l}(m-2)(q+d)\\ -(n-1)d+i \end{array}\\ & \longleftrightarrow \ \text{there exists}\ m\in{\mathbb{n}},\ 0\leq n\leq \left\lfloor m/2\right\rfloor \ \text{with}\ \ x-(2q+d)=m(q+d)-nd+i \\ & \longleftrightarrow \ x-(2q+d)\in c_i . \end{array}\ ] ] this completes the proof . for @xmath261 , since @xmath214\notin s$ ] and @xmath262 for all @xmath211 , it follows that @xmath263 ) = { ( -1)}^{x_1}({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_{1}})(x_0 ) = 0,\ ] ] and @xmath264 ) = { ( -1)}^{x_1}(2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_{1}})(x_0-x_2 ) = 0,\ ] ] for all @xmath265 . case a ) @xmath234 . + for @xmath268 , since @xmath269)=\mu_s(0)=1 $ ] and @xmath270)=\mu_s(a+d)=-1 $ ] , it follows , by definition of the multisets @xmath247 and @xmath254 , that @xmath271 ) = { ( -1)}^{x_1 } = { ( -1)}^{x_1}({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1})(0).\ ] ] for @xmath272 , since @xmath273)=\mu_s(a)=-1 $ ] and @xmath274)=\mu_s(a+(a+d))=c_2(0,2a+d)-c_1(0,2a+d)=2 - 1=1 $ ] , it follows , by definition of the multisets @xmath247 and @xmath254 , that @xmath275 ) = { ( -1)}^{x_1 + 1 } = { ( -1)}^{x_1}({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1})(1).\ ] ] suppose now that @xmath276 . from proposition [ prop1 ] , we have @xmath277 ) = \mu_s ( [ x_0-(q+d ) , x_1,0 ] ) + \mu_s ( [ x_0-(q+d)-1,x_1,q-1 ] ) \\[1.5ex ] - \mu_s ( ( x_0 - 2(q+d ) , x_1,q-1 ] ) .\ ] ] by induction hypothesis , we have @xmath278 ) \\ = { ( -1)}^{x_1 } \left ( { \mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - ( q+d)),\ ] ] @xmath279)\\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-(2q+d))\ ] ] and @xmath280 ) \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-(3q+2d)).\ ] ] by lemma [ lem2 ] , since @xmath281 , we already know that @xmath282 moreover , by lemma [ lem3 ] , we have @xmath283 for all @xmath211 . therefore , @xmath277 ) \begin{array}[t]{l } = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0-(q+d ) ) \right . \\ + \left(2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1}\right)(x_0-(q+d ) ) \\ + \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-(2q+d ) ) \\ \left . - \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-(3q+2d ) ) \right ] \end{array } \\ = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0 ) \right . \\ + \left(2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1}\right)(x_0-(q+d ) ) \\ + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 ) \\ \left . - \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0-(q+d ) ) \right ] \end{array } \\ = { ( -1)}^{x_1 } \left ( { \mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 ) . \end{array}\ ] ] case b ) @xmath284 . + for @xmath268 , since @xmath285)=\mu_s(a+2d)=-1 $ ] and @xmath286)=\mu_s((a+d)+(a+2d))=c_2(0,2a+3d)-c_1(0,2a+3d)=2 - 1=1 $ ] , it follows , by definition of the multisets @xmath255 , that @xmath287 ) = { ( -1)}^{x_1 + 1 } = { ( -1)}^{x_1}(2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1})(-1).\ ] ] for @xmath272 , since @xmath288)=\mu_s(a+(a+2d))=c_2(0,2a+2d)-c_1(0,2a+2d)=3 - 1=2 $ ] and @xmath289)=\mu_s(a+(a+d)+(a+2d))=-c_3(0,3a+3d)+c_2(0,3a+3d)-c_1(0,3a+3d)=-7 + 10 - 1=2 $ ] , it follows , by definition of the multisets @xmath255 , that @xmath290 ) = { ( -1)}^{x_1}2 = { ( -1)}^{x_1}(2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1})(0).\ ] ] suppose now that @xmath276 . from proposition [ prop1 ] , we have @xmath291 ) = \mu_s ( [ x_0-(q+d ) , x_1,1 ] ) + \mu_s ( [ x_0 - 1,x_1,0 ] ) - \mu_s ( [ x_0-(q+d)-1,x_1,0 ] ) .\ ] ] by using lemmas [ lem2 ] and [ lem3 ] and the induction hypothesis , we have @xmath292)\\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0 - ( q+d ) - 1),\ ] ] @xmath293 ) \\ = { ( -1)}^{x_1 } \left ( { \mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}+2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1)\ ] ] and @xmath294 first , since the multiset difference @xmath295 is equal to @xmath247 for all @xmath211 , it follows that @xmath296 therefore , @xmath297 ) \begin{array}[t]{l } = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0 - 1 ) - \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0-(q+d)-1 ) \right . \\ + \left(2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1}\right)(x_0-(q+d)-1 ) \\ - \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0-(q+d)-1 ) \\ \left . + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1 ) \right ] \end{array } \\ = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0 - 1 ) - \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0-(q+d)-1 ) \right . \\ + \left(2{\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_{-1}}-{\mathfrak{m}}_{a_1}\right)(x_0-(q+d)-1 ) \\ \left . + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1 ) \right]\\ \end{array } \\ = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0 - 1 ) + \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_{-1}}\right)(x_0-(q+d)-1 ) \right . + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1 ) \right ] . \end{array } \end{array}\ ] ] moreover , by lemma [ lem2 ] , since @xmath281 , we know that @xmath298 finally , in this case , we obtain @xmath297 ) \begin{array}[t]{l } = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_1}\right)(x_0 - 1 ) + \left({\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_{-1}}\right)(x_0-(q+d)-1 ) \right . + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1 ) \right ] . \end{array } \\ = { ( -1)}^{x_1 } \left [ \left(2{\mathfrak{m}}_{a_0}-{\mathfrak{m}}_{a_{-1}}-{\mathfrak{m}}_{a_{1}}\right)(x_0 - 1 ) + \left ( 2{\mathfrak{m}}_{b_0}-{\mathfrak{m}}_{b_{-1}}-{\mathfrak{m}}_{b_1 } \right)(x_0 - 1 ) \right ] \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0 - 1 ) . \end{array}\ ] ] case c ) @xmath299 . + from proposition [ prop1 ] , we have @xmath300 ) = \mu_s ( [ x_0-(q+d),x_1,x_2 ] ) \\[1.5ex ] + \mu_s ( [ x_0 - 1,x_1,x_2 - 1 ] ) - \mu_s ( [ x_0-(q+d)-1,x_1,x_2 - 1 ] ) .\ ] ] by induction , we have @xmath301 ) \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0 - x_2 - ( q+d)),\ ] ] @xmath302 ) \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2)\ ] ] and @xmath303 ) \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2-(q+d)).\ ] ] therefore , @xmath304 ) \begin{array}[t]{l } = { ( -1)}^{x_1 } \begin{array}[t]{l } \left [ \left(2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1}\right)(x_0-x_2-(q+d ) ) \right . \\ + \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2 ) \\ \left . - \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2-(q+d ) ) \right ] \end{array } \\ = { ( -1)}^{x_1 } \left ( 2{\mathfrak{m}}_{c_0}-{\mathfrak{m}}_{c_{-1}}-{\mathfrak{m}}_{c_1 } \right)(x_0-x_2 ) . \end{array}\ ] ] this completes the proof of theorem [ thm1 ] . @xmath305 & @xmath306 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath308 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath39 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 + @xmath310 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 + @xmath311 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 + @xmath312 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath313 & @xmath314 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath315 & @xmath306 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath316 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath317 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 + @xmath318 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 + @xmath319 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 + @xmath320 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath321 & @xmath314 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath322 & @xmath306 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath323 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 + @xmath324 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 + @xmath325 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 + @xmath326 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath307 + @xmath327 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath328 & @xmath314 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath329 & @xmath306 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 & @xmath305 + @xmath330 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath305 + @xmath331 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 + & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 + & @xmath305 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 & @xmath305 & @xmath305 & @xmath307 & @xmath309 & @xmath307 +

an extension of the nonlinear feedback ( nlf ) formalism to describe regimes of hyper- and high - inflation in economy is proposed in the present work . in the nlf model the consumer price index ( cpi ) exhibits a finite time singularity of the type , with , predicting a blow up of the economy at a critical time . however , this model fails in determining in the case of weak hyperinflation regimes like , e.g. , that occurred in israel . to overcome this trouble , the nlf model is extended by introducing a parameter , which multiplies all therms with past growth rate index ( gri ) . in this novel approach the solution for cpi is also analytic being proportional to the gaussian hypergeometric function , where is a function of , , and . for this hypergeometric function diverges leading to a finite time singularity , from which a value of can be determined . this singularity is also present in gri . it is shown that the interplay between parameters and may produce phenomena of multiple equilibria . an analysis of the severe hyperinflation occurred in hungary proves that the novel model is robust . when this model is used for examining data of israel a reasonable is got . high - inflation regimes in mexico and iceland , which exhibit weaker inflations than that of israel , are also successfully described .

1605.04945

when inflation surpasses moderate levels it damages real economic activities . for instance , it affects government s tax revenues because they are effectively received a period later after the declaration that fix them @xcite . the perception of relative prices changes becomes more difficult since it is not easy to distinguish whether some price grows as a consequence of a relative price change or it is part of general inflation ( the lucas problem @xcite ) . it produces inefficient changes of relative prices @xcite if the adjustment process is different for different kinds of goods inducing misleading allocation of resources @xcite . inflation also affects currency in its property as medium of exchange , store of value , and unit of account . the degree of perturbation is greater the higher the inflation . if consumption goods become relatively more expensive than leisure due to inflation , labor market may be negatively perturbed by reduction of working hours supply @xcite . unanticipated increase in prices reduces real wages and expand employment @xcite , although the positive employment effect could be lowered or reversed by the effects of falling investment . furthermore , investment demand may be especially affected because of the shorter planning time scope and growing uncertainty . in general , when the inflation is higher then the decision horizon is shorter . moreover , with no alternative allocation to money , savings decline and investment falls at expenses of actual consumption , lowering the capital stock growth . therefore inflation is not a pure nominal problem , m but it is linked to real economy in many non trivial ways @xcite . consequently , governments try to prevent high inflation , and to lower it when reach elevated levels . parameters can change once policy changes @xcite . the relation between sources of inflation and the evolution of parameters is complex and not direct . these issues are also analyzed in text books on econophysics @xcite . models of hyperinflation are especially suitable to emphasize that inflation implies bad `` states of nature '' in economy . wars , changes of social regimens , states bankruptcies are the characteristics of such regimens . these factors together with the influence of pure economical variables like expectations , money demand , velocity of circulation and quantity of money , give an increase of the consumer price index ( cpi ) larger than exponential as can be observed in the investigated cases . in turn , this behavior affects negatively the social network causing unpleasant situations . the model for hyperinflation available in the literature is based on a nonlinear feedback ( nlf ) characterized by an exponent @xmath1 of a power law . in such an approach the cpi exhibits a finite time singularity of the form @xmath8 , allowing a determination of a critical time @xmath2 at which the economy would crash . this model has been successfully applied to many cases . however , in the most recent paper @xcite , it is shown that for the episode of weak hyperinflation occurred in israel it is impossible to determine a value for @xmath2 within the nlf model because @xmath6 goes to zero . in this limit one gets the linear feedback approach which does not contain information on @xmath2 @xcite . this drawback was attributed to a permanent but partially successful efforts for stopping inflation in israel . in order to include in the nlf formalism information on efforts for stabilization like those observed in the case of israel , we developed an extension of this model introducing a parameter @xmath3 , which multiplies all the past growth rate index ( gri ) contributions changing the relative weight of the therm with the power law . the interplay between @xmath6 and @xmath3 leads to multiple equilibria phenomena for episodes of high inflation . the literature on this kind of behavior in models of economics is large and diverse @xcite , see also e.g. refs.@xcite . we use the well known multiple equilibria expression to indicate the existence of multiple trajectories compatible with data which lead to very different states of nature of the economy where the final outcome is stable or not . in the extended nlf approach the solution for cpi becomes proportional to the gaussian hypergeometric function @xmath4 , where @xmath5 is a function of @xmath6 , @xmath3 , and @xmath2 . since for @xmath7 this hypergeometric function diverges , a finite time singularity shows up allowing a determination of @xmath2 . it is important to notice that the gaussian hypergeometric appears in a variety of physical and mathematical problems . in quantum mechanics , the solution of the schrdinger equation for some potentials is expressed in terms of @xmath9 @xcite . moreover , the eigenfunctions of the angular momentum operators are sometimes written in terms of @xmath9 functions @xcite . in the present work we also investigated the effects of the parameter @xmath3 on the non linear dynamic evolution of real inflations . firstly , the robustness of the novel model is tested analyzing the severe hyperinflation occurred in hungary after world war ii . next , it is successfully applied for studying the weaker hyperinflation developed in israel and the high - inflation episodes observed in mexico and iceland . in order to understand better the evolution of prices in these countries , brief descriptions of historical events are provided . in these cases we found multiple equilibria . the paper is organized in the following way . in sec.[sec : theory ] the nlf model is outlined with some details because it is the starting point for the extension proposed in sec . [ sec : general ] . the limit @xmath10 is discussed . section [ sec : results ] is devoted to report and discuss the results obtained by applying the novel approach . finally , the main conclusions are summarized in sec . [ sec : summary ] . let us recall that the rate of inflation @xmath11 is defined as @xmath12 where @xmath13 is the cpi at time @xmath14 and @xmath15 is the period of the measurements . in the academic financial literature , the simplest and most robust way to account for inflation is to take logarithm . hence , the continuous rate of change in prices is defined as @xmath16 usually the derivative of eq . ( [ c_rate0 ] ) is expressed in a discrete way as @xmath17}{\delta t } \nonumber\\ & = & \frac{1}{\delta t}\,\ln \left [ \frac{p(t+\delta t)}{p(t ) } \right ] \ : . \label{c_rate1}\end{aligned}\ ] ] the gri over one period is defined as @xmath18 \nonumber\\ & = & \ln[1 + i(t+\delta t ) ] = p(t+\delta t ) - p(t ) \ : , \nonumber\\ \label{rate1}\end{aligned}\ ] ] where a widely utilized notation @xmath19 was introduced . it is straightforward to show that the accumulated cpi is given by @xmath20 } \;. \label{pt}\ ] ] in the fifties , cagan has proposed @xcite a model of inflation based on the mechanism of `` adaptive inflationary expectation '' with positive feedback between realized growth of the market price @xmath13 and the growth of people s averaged expectation price @xmath21 . these two prices are thought to evolve due to a positive feedback mechanism : an upward change of market price @xmath13 in a unit time @xmath15 induces a rise in the people s expectation price @xmath21 , and such an anticipation pushes on the market price . so , one may write @xmath22 actually @xmath23 indicates that the process induces a non exact proportional response of adaptation due to the fact that the expected inflation @xmath21 expands the response to the price level @xmath13 in order to forecast and meet the inflation of the next period . now , one may introduce the expected gri @xmath24 \ : . \label{rate2}\ ] ] sornette , takayasu , and zhou ( stz ) @xcite introduced a nonlinear feedback ( nlf ) process in the formalism suggesting that the people s expectation price @xmath21 obeys @xmath25 & = & \ln \left [ \frac{p(t)}{p(t-\delta t ) } \right ] \nonumber\\ & \times & \left ( 1 + 2\,a_p \biggr\ { \ln \left[\frac{p(t)}{p(t-\delta t)}\right]\biggr\}^\beta \right ) \ : . \nonumber\\ \label{rate00}\end{aligned}\ ] ] here @xmath26 is a positive dimensionless feedback s strength and @xmath27 is the exponent of the power law . this expression together with eq . ( [ cag1 ] ) leads to @xmath28^{1 + \beta } \;. \label{rate42}\ ] ] taking the continuous limit of this relation one obtains the following equation for the time evolution of @xmath29 @xmath30^{1+\beta } \;. \label{rate43}\ ] ] the solution for gri follows a power law exhibiting a singularity at finite - time @xmath2 @xcite @xmath31^{1/\beta } = r_0\,\left(\frac{t_c - t_0}{t_c - t}\right)^{1/\beta } \;. \label{r_time}\ ] ] the critical time @xmath2 being determined by the initial gri @xmath32 , the exponent @xmath6 , and the strength parameter @xmath26 @xmath33 the cpi is obtained by integrating @xmath34 according to eq.([pt ] ) . for @xmath35 , denoted as case ( i ) , it becomes @xmath36 = p(t ) - p_0 = \int^t_{t_0 } r(t ' ) \frac{dt'}{\delta t } \nonumber\\ & = & \frac{r_0^{1-\beta}}{(1-\beta)\,a_p } \biggr\ { \biggr[\frac{1}{1-\beta\,a_p\,r_0^\beta \left(\frac{t - t_0}{\delta t}\right ) } \biggr]^{\frac{1-\beta}{\beta } } - 1 \biggr\ } \;. \nonumber\\ \label{lptn}\end{aligned}\ ] ] ( i.a ) for @xmath37 one also gets a finite - time singularity in the @xmath38-cpi according to the power law @xmath39 \;. \label{price1}\ ] ] this expression has been used for the analysis of hyperinflation episodes reported in previous papers . ( i.b ) for @xmath40 one gets a finite - time singularity for @xmath34 , but the log - cpi evolves as @xmath41 \label{price3}\ ] ] converging to a finite value when time approaches the critical value @xmath2 @xmath42 \(ii ) for @xmath43 the integration in eq . ( [ pt ] ) yields @xmath44 = \int^t_{t_0 } r(t ' ) \frac{dt'}{\delta t } = \frac{1}{a_p } \ln \biggr [ \frac{1}{1-a_p\,r_0\,\left(\frac{t - t_0}{\delta t } \right ) } \biggr ] \;. \label{logdiv}\ ] ] upon introducing @xmath2 in this expression , the time dependence of @xmath45 exhibits a logarithmic divergence @xmath46 it was found that for severe episodes of hyperinflation one may get @xmath47 @xcite . in both regimes , ( i.a ) and ( ii ) , the cpi exhibits a finite - time singularity at the same critical value @xmath2 as gri . hence , these solutions correspond to a genuine divergence of @xmath48 . it is important to notice that for @xmath49 one gets the linear feedback ( lf ) model suggested previously by mizuno , takayasu , and takayasu ( mtt ) @xcite . in this limit one arrives at @xmath50 \;. \label{r_mizu}\ ] ] hence , the cpi grows as a function of @xmath14 following a double - exponential law @xcite @xmath51 - 1 \biggr\ } \;. \label{p_mizu}\ ] ] in the lf model no @xmath2 can be determined . furthermore , by setting now @xmath52 one gets @xmath53 and @xmath54 which is the cagan s original proposal . equations ( [ lptng3 ] ) and ( [ lptln1 ] ) have two domains of solutions , one for @xmath79 ( implying @xmath80 ) and the other for @xmath81 ( @xmath82 ) . in the appendix b we show that in the limiting case @xmath10 ( i.e. , @xmath83 ) from above , the generalized expressions for gri and cpi provided by the enlf model reduce to the forms reported previously in refs.@xcite and summarized in sec.[sec : theory ] . the finite time singularity occurs at the same value of critical @xmath91 for gri and for both sectors ( [ lptng4 ] ) [ with ( [ defxa ] ) ] and ( [ lptln2 ] ) of solutions for log - cpi , satisfying @xmath92 in turn , the critical time @xmath2 can be determined by equating this relation with @xmath93 given by eq.([deltaa ] ) @xmath94 \;. \label{singul_1}\end{aligned}\ ] ] the result is @xmath95 \ ; , \label{tc_g}\ ] ] providing the new expression for @xmath2 . let us now show that in the limit @xmath10 one retrieves for @xmath2 the expression given in sect . [ sec : feedback ] . expanding the logarithm in powers of @xmath96 one gets @xmath97 \biggr [ \frac{\gamma -1}{2\,a_p\,\gamma^{1+\beta}\ , r_0^\beta } \nonumber\\ & & -\,\frac{1}{2 } \biggr ( \frac{\gamma -1}{2\,a_p\,\gamma^{1 + \beta}\,r_0^\beta } \biggr)^2 + { \cal{o}}(h - o ) \biggr ] \nonumber\\ & \simeq & \frac{1}{\beta\,a_p\,\gamma^{1+\beta}\,r_0^\beta } \biggr ( 1 - \frac{\gamma -1}{4\,a_p\,\gamma^{1+\beta}\ , r_0^\beta } \biggr ) \;. \label{tc_gl}\end{aligned}\ ] ] then , for @xmath10 one arrives at @xmath98 recovering the relation of eq . ( [ c_time ] ) . upon introducing eq . ( [ singul_0 ] ) into eq . ( [ rate56 ] ) the gri becomes @xmath99^{1/\beta } \nonumber\\ & & ~~ \times \biggr[\biggr(1 + \frac{\gamma -1}{2 a_p \gamma^{1+\beta } r_0^\beta } \biggr ) \exp(-\delta ) - 1 \biggr]^{-1/\beta } \nonumber\\ & = & r_0 \biggr [ \frac{\exp(\delta_c)-1}{\exp(\delta_c-\delta)-1 } \biggr]^{1/\beta } = r_0 \biggr [ \frac{1-z_0}{\exp(-\delta)-z_0}\biggr]^{1/\beta } \ ; , \nonumber\\ \label{rate57}\end{aligned}\ ] ] with @xmath100 \;. \label{defxb0}\ ] ] lccccclc period & & model & @xmath101 + & @xmath2 & @xmath102 & @xmath6 & @xmath3 & @xmath103 & + + 1945:04:30 - 1946:07:15 & 1946:09:03 & 0.150 & 0.500 & & 3.82 & nlf@xmath104 & 1.168 + & 1946:07:30@xmath10501 & [email protected] & [email protected] & & 4.07 & nlf & 0.759 + & 1946:07:28@xmath10504 & [email protected] & [email protected] & & [email protected] & nlf & 0.704 + & 1946:07:30@xmath10501 & [email protected] & [email protected] & [email protected] & 4.07 & enlf & 0.761 + 1945:04:30 - 1946:07:31 & 1946:09:02@xmath10505 & [email protected] & [email protected] & & 4.07 & nlf & 1.123 + & 1946:09:01@xmath10503 & [email protected] & [email protected] & & [email protected] & nlf & 1.123 + & 1946:09:02@xmath10501 & [email protected] & [email protected] & [email protected] & 4.07 & enlf & 1.123 + @xmath104 the values of the parameters listed in this line were calculated using those reported by stz @xcite , see text . in a first step , we shall show that the enlf formalism is robust for episodes of severe hyperinflation , leading to values of @xmath3 very close to unity . next , cases of weaker inflations will be treated . let us begin the application of the novel approach by tackling an emblematic case . an important test for any formalism developed for describing regimes of high inflation is to verify whether it is able to account for the evolution of prices occurred in hungary right after world war ii @xcite . for this severe hyperinflation there is available a data series collected in table a.1 of ref . @xcite ( see also @xcite ) constructed on a basis of a biweekly frequency . the word `` biweekly '' is used to designate data available on the 15th and on the last day of each month . figure [ fig : hungary ] shows the cpi and gri during the period from april 30 , 1945 to july 31 , 1946 ( i.e. , from 1945:04:30 to 1946:07:31 , in the case of hungary the notation year : month : day will be used ) . notice that the cpi values are normalized to @xmath112 . several fits of these data were performed . an inspection to fig . [ fig : hungary](b ) suggests that the value at 1946:07:31 could be already affected by the stabilization policy adopted by the hungarian government @xcite . therefore , in a first step we performed a fit of cpi data up to 1946:07:15 with eq . ( [ price1 ] ) setting @xmath113 . the numerical task was accomplished by using a routine of the book by bevington @xcite cited as the first reference in chaps . 15.4 and 15.5 of the more recent _ numerical recipes _ @xcite . in practice , the applied procedure yields the uncertainty in each parameter directly from the minimization algorithm . the parameters yielded by these fits are listed in table [ tab : table3 ] together with the root - mean - square ( r.m.s . ) residue of the fit , @xmath101 . a simple inspection indicates a crash at the beginning of august 1946 . values of gri were evaluated with eq . ( [ r_time ] ) . the good quality of the fits is shown by solid curves in panels ( a ) and ( b ) of fig.[fig : hungary ] . table 1 in ref . @xcite indicates that stz have also analyzed data of hungary from 1945:04:30 to 1946:07:15 . in order to facilitate a quantitative comparison with the study reported by stz , we evaluated the parameters @xmath102 , @xmath6 , and @xmath103 utilized in the present work by inserting the values of @xmath2 , @xmath114 , @xmath115 , and @xmath116 listed in table 1 of ref.@xcite into the following relations @xmath117 these results are included in our table [ tab : table3 ] . a glance at this table indicates a shift between the present results and that of stz . for the sake of completeness , we also analyzed cpi data including the value at 1946:07:31 . this fit yielded the parameters quoted in table [ tab : table3 ] and the dashed curves in panels ( a ) and ( b ) of fig . [ fig : hungary ] show the adjustment . as expected , the new fit suggests a later date for the crash of the economy than the prediction provided by the shorter series , now the blow up would occur at the beginning of september 1946 . surprisingly , one may observe in table [ tab : table3 ] that the results of stz , including the @xmath101 , are almost the same as that obtained in the present work with the series ending at 1946:07:31 . for the sake of completeness , fits leaving @xmath103 free were also performed . looking at table [ tab : table3 ] one may realize that no significant changes are obtained . as planned the study of the hungarian hyperinflation is finished by applying the enlf formalism derived in sec . [ sec : general ] of the present paper . in so doing , both the shorter and larger series of cpi data were fitted to eq . ( [ lptng5 ] ) . the obtained parameters are also quoted in table [ tab : table3 ] , where one can observe that all the `` old '' parameters remain unchanged . predictions for gri were computed using eq.([rate57 ] ) . the match to eqs . ( [ rate57 ] ) and ( [ lptng5 ] ) and those corresponding to eqs . ( [ r_time ] ) and ( [ price1 ] ) are indistinguishable on the scales of fig.[fig : hungary ] . one can realize that in both cases @xmath3 stays equal to unity within the uncertainty . for the complete series up to 1946:07:31 the uncertainty is slightly larger , this feature could be attributed to the beginning of the stabilization process . all these results support the robustness of the extended model . [ cols="<,<,^,^,^,^,^,^ , < , < " , ] @xmath118 data from ref . @xmath119 @xmath120 . @xmath121 @xmath122 . @xmath123 the values of the parameters listed in this line were calculated using those reported by stz @xcite ( see text ) , in addition , @xmath103=1.04 . let us now examine the case of israel , which have been already studied in a previous work @xcite . figure[fig : israel_h](a ) shows the yearly data for the cpi in israel computed using data taken from a table of the international monetary fund ( imf ) @xcite . the upward curvature of the logarithm of cpi as a function of time indicates that from 1969 to 1985 the hyperinflation exhibits a faster than exponential growth . the inflation got triple - digit rates of about @xmath124 at their peak in the mid-1980 s . leiderman and liviatan @xcite attributed this response to the implicit preference for short - term considerations of avoiding unemployment over long - term monetary stability . in 1985 a new strategy was applied to stop the hyperinflation . the results for the parameters and the @xmath101 obtained in the previous work @xcite are reproduced in table [ tab : table4 ] . two time series of cpi were examined : one with data from 1969 until 1985 ( this series has been studied in refs.@xcite and @xcite ) and the other excluding the value of 1985 when the final stabilization started . from that analysis it was concluded that when applying the nlf model the fits indicate a strong correlation between @xmath6 and @xmath2 . in table [ tab : table4 ] two sets of parameters are provided , one corresponds to stopping minimization when the change of @xmath101 from the `` @xmath125 '' to the `` @xmath126''-iteration becomes less than @xmath127 and the other for @xmath128 . let us mention that for that study @xmath103 was set equal to zero . the fitting procedure showed that @xmath6 decreases approaching zero while @xmath2 increases , this occurs in such a way that the product @xmath129 converges to a constant yielding a well defined value of the parameter @xmath26 given by [ see eq . ( [ c_time ] ) ] @xmath130 which is also quoted in table [ tab : table4 ] . since for @xmath131 the nlf model converges towards the lf model of mtt @xmath132 = \frac{\delta t}{\beta\,(t_c - t_0 ) } \ ; , \label{a_p_ass}\ ] ] that study was completed by fitting the cpi data directly with eq . ( [ p_mizu ] ) . the obtained parameters are also quoted in table [ tab : table4 ] . figure [ fig : israel_h](a ) shows the good quality of the fit for the larger series . it is also worthwhile to mention that the present value for the parameter utilized in the original lf model , i.e. @xmath133 , is in good agreement with the result @xmath134 quoted by mtt in table 1 of ref . @xcite . for the sake of completeness , we plotted in fig . [ fig : israel_h](b ) the measured data of gri together with the theoretical values provided by eqs.([r_time ] ) and ( [ r_mizu ] ) , which are indistinguishable on the scale of the drawing . although the lf model provides a good fit , it does not predict any @xmath2 indicative for a possible crash of the economy . therefore , in order to estimate a @xmath2 , we shall apply the novel enlf model to analyze the evolution of prices in israel looking for multiple equilibria ( or trajectories ) . in practice , there are two strategies for treating the parameters of the enlf model . one is to adopt as free parameters @xmath102 , @xmath6 , @xmath3 , and @xmath2 like it was done for the analysis of data for hungary . the other , is to consider @xmath102 , @xmath6 , @xmath3 , and @xmath135 as free parameters . it is important to notice that the whole contribution of @xmath2 is carried by @xmath135 in the form of the product @xmath136 , which also determines @xmath137 as shown above . hence , it would be reasonable to expect a solution compatible with the lf model . therefore , in order to check this feature , the cpi data for the period 1969 - 1985 were , in a first step , fitted to eq . ( [ lptng5 ] ) expressed in therms of @xmath135 . the obtained parameters are listed in table [ tab : table4 ] , where one can observe that the @xmath101 is similar to that obtained for the long nlf run . the value @xmath138 is consistent with zero suggesting a lf with a renormalized strength given by eq . ( [ rate48 ] ) with @xmath49 @xmath139 in fact , the results for @xmath140 [ evaluated from eqs.([singul_1 ] ) and ( [ defxb0 ] ) ] and @xmath137 quoted in table [ tab : table4 ] satisfy this relation . this solution remains stable when the minimization procedure is continued . furthermore , the use of eq . ( [ defxb0 ] ) with the listed results for @xmath135 , @xmath6 , and @xmath3 yields @xmath141 , which is too large for a hyperinflation developing during the 1980 s . so , the prediction of this solution is not useful . the fit considering @xmath102 , @xmath6 , @xmath3 , and @xmath2 as free parameters yielded the set of results reported in table [ tab : table4 ] . the obtained values @xmath142 , @xmath143 , and @xmath144 are quite reasonable for a rather slow hyperinflation . if the fit is continued the parameters do not change , only the uncertainties diminish as can be seen in table [ tab : table4 ] . although the @xmath101 of this solution for @xmath2 is slightly larger than that determined by using @xmath135 , it is quite acceptable . the good quality of the present enlf(t@xmath145 ) description of measured data for both cpi and gri is shown in fig . [ fig : israel_h ] . the value of @xmath2 obtained by fitting @xmath48 with eq.([lptng5 ] ) is larger than that determined by stz * from a phenomenological fit of @xmath13 to eq . ( [ price1 ] ) . this feature is due to the fact that , for a regime with inflation , @xmath13 presents a steeper slope than @xmath48 @xmath146 the steeper is the curvature the earlier becomes @xmath2 [ see fig.[fig : israel_h](a ) ] . since the actions for stopping the inflation begun in 1985 @xcite , we also performed an analysis of the reduced period 1969 - 1984 . the values yielded by the new sequence of fits are also included in table [ tab : table4 ] . there one can observe small changes between the parameters corresponding to both series of data . in summary , we can state that in the case of the hyperinflation occurred in israel it is possible to predict a reasonable @xmath2 by applying the enlf model proposed in the present paper . in fact , the interaction between @xmath6 and @xmath3 leads to multiequilibria phenomena yielding solutions with @xmath49 ( no prediction for @xmath2 ) and @xmath27 ( @xmath2 is determined ) . as mentioned in sec . [ sec : introduction ] , the literature on multiple equilibria in economics is huge . this section is devoted to study regimes of high - inflation that are not catalog as episodes of hyperinflation . looking at tables of inflations one can find several examples of this type . if evolution of the cpi of such a sort of episodes is described by eq . ( [ price1 ] ) one finds similar features to that encountered in the case of israel . for illustrating this kind of behavior we selected the evolution of prices during periods of high - inflation occurred in mexico and iceland . the economy of mexico is rather large , the number of inhabitants during the period of high - inflation was close to 100 millions . the general conditions that gave rise to such an inflation have been based on domestic troubles and the management of petroleum business @xcite . political unrest escalated during the 1960 s . students of the autonomous national university of mexico began organizing large scale demonstrations in mexico city in 1968 . these protests were followed by violent episodes including the killing of demonstrators by members of the armed forces @xcite . at the beginning of the 70 s president luis echeverra lvarez trying to avoid an ongoing insurgency distributed land to a communal farming arrangement . after this step the production decreased substantially . in addition , he promoted extensive subsidies for public and private enterprises . for instance , one of the projects was a steel complex in michoacn . these programs could not be funded out of existing tax funds so the central bank printed money . one of the effects of these policies was a serious inflation accompanied by an economic crisis . at the end of the 70 s during the presidency of jos lpez portillo y pacheco new oil fields were discovered in the south of mexico . because of the existence of these new reserves and the rising international price of petroleum , foreign bankers were willing to lend mexico vast amount of money . by 1982 almost half of the petroleum exports earning were going to pay the interest and other scheduled payments for the foreign debt . so , these discoveries did not alleviate the problem of inflation , the annual rate of inflation hit 100% [ see fig . [ fig : mexico_high](b ) ] . in september 1982 the banks of mexico were nationalized . the price of petroleum began to fall in response to the increased quantity of petroleum being supplied . when miguel de la madrid hurtado came into office at the end of 1982 mexico s economic house was in a great state of disarray . there was a huge foreign debt requiring excessive foreign currency credit to service . he promulgated a program of economic austerity which included : increases in tax rates ; reduction of the federal government budget ; reduction of subsidies for some commodities ; postponement of many public projects ; increase of some interest rates ; and relaxation of capital transfer restrictions . the moves toward privatizations started during his administration . financial necessity forced the selling off of about 200 government enterprises . nevertheless , the inflation continuously grew , see fig . [ fig : mexico_high](b ) . in 1988 , carlos salinas de gotari , young technocrat with a ph.d . in economics from harvard university was elected president . the criterion was to find a person with expertize for dealing with mexico s financial and economic problems . he deepen successfully the stabilization program for stopping inflation began by de la madrid , see fig . [ fig : mexico_high](b ) . the cpi and gri were evaluated using data of inflation taken from ref . @xcite , these observables are plotted in panels ( a ) and ( b ) of fig . [ fig : mexico_high ] , respectively . the values from 1960 to 1988 , previous to stabilization , have been analyzed . for the study of the mexico s inflation the same procedure as that applied in the case of israel was adopted . so , in a first step , we fitted data of cpi with eq . ( [ price1 ] ) . the parameters together with the @xmath101 obtained by stopping the minimization procedure when the variation of @xmath101 between the `` @xmath126 '' and `` @xmath125 '' iterations was smaller than a standard choice @xmath147 are listed in table [ tab : table4 ] . a glance at this table shows a small exponent of the power law @xmath148 and a large critical time @xmath149 . if one allows the iterations to continue , the correlation between these parameters becomes clear . for instance , in table [ tab : table4 ] we quoted values obtained when the change of @xmath101 becomes less than @xmath150 . as in the case of israel , @xmath6 goes to @xmath151 and @xmath2 increases keeping @xmath26 given by eq . ( [ a_p_lim ] ) constant . these results indicate that the nlf model tends towards the lf one . therefore , the analysis was completed fitting the cpi data directly with lf s eq . ( [ p_mizu ] ) . the obtained parameters are included in table [ tab : table4 ] . notice the excellent agreement between the values of @xmath102 , @xmath26 , and @xmath101 yielded by the lf approach and those obtained from the `` long '' fit with eq . ( [ price1 ] ) of the nlf model . the quality of the fits is depicted in fig . [ fig : mexico_high](a ) . in addition , the theoretical gri was calculated using eqs . ( [ r_time ] ) and ( [ r_mizu ] ) and it is compared to data in fig . [ fig : mexico_high](b ) . no difference between lf and nlf can be observed . so , no estimation for @xmath2 can be achieved . next , the cpi data for the period 1960 - 1988 were fitted to eq.([lptng5 ] ) provided by the enlf model written in therms of @xmath102 , @xmath6 , @xmath3 , and @xmath135 . this procedure lead to the values listed in table [ tab : table4 ] . the value @xmath152 , although larger than in the case of israel , is also consistent with zero suggesting a lf . moreover , the result for @xmath140 also satisfies the relation ( [ a_p_g ] ) . finally , the same data of cpi were fitted with eq . ( [ lptng5 ] ) of the enlf model written in therms of @xmath102 , @xmath6 , @xmath3 , and @xmath2 . the obtained parameters are also included in table [ tab : table4 ] and the fit is displayed in fig.[fig : mexico_high](a ) . in this case , both parameters @xmath153 and @xmath154 are well defined and reasonable . although the @xmath101 is slightly larger than that obtained with the lf model a good match of theoretical cpi with measured data is got . for completeness , the gri was evaluated using eq . ( [ rate57 ] ) and plotted in fig . [ fig : mexico_high](b ) , where one may observe a good accordance with data . these results indicate that the enlf(@xmath2 ) model provides a satisfactory description of the episode occurred in mexico . hence , one can state that multiple equilibria phenomena are also present in this case . the solution with @xmath27 shows a trajectory towards the category of hyperinflation . it is also interesting to study the increase of prices occurred in iceland during the period from 1960 to 1983 @xcite right previously to the disinflation analyzed by mtt ( see table 2 in ref . let us mention that iceland is a nation with less than a half million inhabitants , so in contrast to mexico this is a rather small economy . iceland was under norwegian and danish kings along centuries . it is an independent nation since 1944 . during the examined period the government s fiscal policy was strictly keynesian , and their aim was to create the necessary industrial infrastructure for a prosperous developed country . it was considered essential to keep unemployment down to an absolute minimum and to protect the export of fishing industry through currency manipulation and other means . due to the country s dependence both on unreliable fish catches and foreign demand for fish products , iceland s economy remained very unstable well into the 1990 s , when the country s economy was greatly diversified . iceland then became a member of the european economic area in 1994 . economic stability increased and previously chronic inflation was drastically reduced . the gri and cpi for iceland were evaluated with data taken from @xcite and are plotted in panels ( a ) and ( b ) of fig . [ fig : iceland_high ] , respectively . one may observe that before stabilization the cpi reaches a value of about @xmath155 , being more than one order of magnitude smaller than the value @xmath156 corresponding to israel and slightly smaller than that of mexico . moreover , a further check of this relative strength of inflations can be done comparing fig . [ fig : iceland_high](b ) with figs . [ fig : israel_h](b ) and [ fig : mexico_high](b ) . hence , the inflation of iceland was studied in the same way as the episodes of that countries . however , in this case the analysis was begun by fitting data of cpi for the period 1960 - 1983 with eq . ( [ p_mizu ] ) of the lf model . the obtained parameters are listed in table [ tab : table4 ] . the quality of the adjustment is shown in fig . [ fig : iceland_high](a ) . the calculated @xmath157 is smaller than the values quoted in table 1 of ref . @xcite , indicating that in this case the inflation was less severe than for the examples examined there . next , a standard `` short '' fit of the same data for cpi with eq . ( [ price1 ] ) of the nlf model , similar to that performed in the cases treated previously , yielded the parameters quoted in table [ tab : table4 ] . one may realize that @xmath158 is consistent with zero and @xmath159 is rather undetermined . furthermore , a `` longer '' fit indicates that @xmath6 goes to 0 and @xmath2 increases presenting an even larger uncertainty , both these features can be seen in table [ tab : table4 ] . on the other hand , @xmath26 given by eq.([a_p_lim ] ) converges to a constant value , which coincides with that yielded by the fit with the lf model . the gri evaluated with eqs . ( [ r_time ] ) and ( [ r_mizu ] ) is displayed in fig.[fig : iceland_high](b ) , no difference between the lf and nlf approaches can be observed on the scale of the drawing . so , as expected , in this case the nlf model also converges towards the lf one . hence , no prediction for @xmath2 could be obtained . therefore , in order to get an estimation for @xmath2 , we also applied the enlf model for analyzing this episode . firstly , the same series of cpi data was fitted to eq . ( [ lptng5 ] ) written in therms of @xmath102 , @xmath6 , @xmath3 , and @xmath135 . this procedure lead to the values listed in table [ tab : table4 ] , where one can realize that the @xmath101 is equal to that obtained with the lf model . the value @xmath160 , although presents a very large uncertainty , it is consistent with zero suggesting a lf . on the other hand , the result for @xmath140 also satisfies the relation ( [ a_p_g ] ) . however , the critical time calculated with eq . ( [ defxb0 ] ) , @xmath161 , is too large making this prediction useless . finally , the enlf model written in therms of @xmath102 , @xmath6 , @xmath3 , and @xmath2 was applied for describing the examined data . the fit with eq . ( [ lptng5 ] ) yielded the parameters are included in table [ tab : table4 ] . in this case both parameters , @xmath162 and @xmath163 , are reasonable . if the minimization is continued the values of the parameters remain stable , while their uncertainties diminish . in addition , the @xmath101 is almost equal to that provided by the fit to the lf model . as depicted in fig . [ fig : iceland_high](a ) , the matching between theoretical cpi and measured data is quite good . for completeness , the gri was evaluated using eq . ( [ rate57 ] ) with the parameters of the enlf(@xmath2 ) model . the result is plotted in fig.[fig : iceland_high](b ) , where one may observe a good accordance with data . so , the enlf(@xmath2 ) model describes satisfactorily well the episode occurred in iceland providing an acceptable prediction for @xmath2 in case that this high inflation would become a hyperinflation . this is another example of multiple equilibria . in the present work we treated regimes of hyper- and high - inflation in economy . in a previous work @xcite it has been found that for a weak hyperinflation , like e.g. that developed in israel , was impossible to determine a value of @xmath2 within the frame of the nlf model . this model is based on a power law with an exponent @xmath1 , see eq . ( [ rate00 ] ) . the mentioned drawback has been attributed to a permanent but incomplete effort for stopping inflation . therefore , in the present work we suggested to include in the theory information on saturation by introducing a parameter @xmath3 , which multiplies all the past inflation growth rates . this parameter would account for the effort done by the government for coping inflation . in the extended approach , enlf , reported in the present paper the solutions for gri and cpi are also analytic as in the nlf model . in particular , the cpi is expressed in therms of the gaussian hypergeometric function @xmath4 , where @xmath5 is a function of @xmath3 , @xmath6 , @xmath102 , and @xmath2 , see eqs . ( [ defxb0])-([defxb ] ) . it is pertinent to notice that @xmath9 appears in a variety of mathematical and physical problems . for @xmath7 this hypergeometric function diverges leading to a finite time singularity , from which a value of @xmath2 can be determined . the same singularity is present in eq.([rate57 ] ) for gri . so , the enlf model proposed in the present work preserves the well - defined singularities yielded by the power law stemmed from a simple positive nonlinear feedback . this mechanism is important for understanding processes in financial crashes ( see ref . @xcite and references therein ) . for completeness , in the appendix it is shown that for the limit @xmath10 from above , one retrieves all the expressions of the nlf model . an analysis of the severe hyperinflation occurred in hungary after the world war ii proves that the novel enlf approach is robust . when it is used for examining data of israel there are two sorts of solutions . one yields @xmath6 consistent with zero ( lf model ) and the other one gives a well determined and reasonable @xmath2 . as a further application , high - inflation regimes exhibiting weaker inflations than that of israel were also analyzed . the episodes occurred in mexico and iceland are reported in the present work . data of both series of inflation can be described , as in the case of israel , with the lf model . however , the enlf(@xmath2 ) model also provides additional solutions forecasting possible blow up of the economies in case the high inflation regimes would become spirals of hyperinflation . the corresponding fits are very good and the predicted values of @xmath2 for crashes are acceptable . the phenomena of multiple equilibria , a known feature in models of economics , appears due to the fact that the introduction of @xmath3 enlarges the dimension of the @xmath101 hyper - surface , which now also presents minimums in domains where the parameters @xmath2 and @xmath6 are not strongly correlated . so , we can state that the parameter @xmath3 allows a more complete representation of inflationary processes . it is interesting to note that the exponent of the nonlinear feedback of inflation can be controlled by a parameter multiplying the growth rate . different combinations of parameters may be interpreted as forces acting with its own strength , producing dynamical paths that lead to very different outcomes . this work was supported in part by the ministry of science and technology of argentina through grants pip 0546/09 from conicet and pict 2011/01217 from anpcyt , and grant ubacyt 01/k156 from university of buenos aires . in this appendix we show that by imposing the limit @xmath164 , which is equivalent to @xmath83 , in the generalized expressions for gri and cpi the forms reported in ref.@xcite are recovered . so , starting from eq . ( [ rate55 ] ) for gri and keeping only linear terms of the expansion inpowers of @xmath84 one gets @xmath165}{s + q\ , r_0^\beta ( 1 - \exp[\beta\,s\,(t - t_0 ) ] ) } \biggr]^{1/\beta } \nonumber\\ & = & r_0 \biggr[\frac{s\,[1-\beta\,s\,(t - t_0)]}{(s + q\,r_0^\beta ) \,[1-\beta\,s\,(t - t_0 ) ] - q\,r_0^\beta}\,\biggr]^{1/\beta } \ ; , \label{b_1}\end{aligned}\ ] ] which in the limit @xmath83 , where @xmath166 $ ] , yields @xmath167^{1/\beta } \nonumber\\ & = & r_0 \biggr(\frac{t_c - t_0}{t_c - t}\biggr)^{1/\beta } \ ; , \label{b_2}\end{aligned}\ ] ] recovering eq . ( [ r_time ] ) . \(i ) starting from the general solution for cpi given by eq.([lptng3 ] ) written as @xmath73 & = & \frac{r_0}{\delta t } \biggr ( \frac{1}{s + q r_0^\beta } \biggr)^{1/\beta}\,s^{-1 + 1/\beta } \nonumber\\ & \times & \biggr[\ , _ 2f_1(1/\beta,1/\beta,1 + 1/\beta;z ) \ , \exp[s(t - t_0 ) ] \nonumber\\ & & - \ , _ 2f_1(1/\beta,1/\beta,1 + 1/\beta;z_0 ) \biggr ] \ ; , \label{b_3}\end{aligned}\ ] ] the limit @xmath83 is evaluated expanding the hypergeometric function for small values of @xmath84 . using the relation @xmath168 = ( 1-z)^{-1/\beta } \nonumber\\ & & -\ , _ 2f_1(1/\beta,1/\beta,1 + 1/\beta;z ) \biggr ] \ ; , \label{deriv}\end{aligned}\ ] ] and the on line mathematica one gets @xmath169 \nonumber\\ & & + \ , \frac{\gamma[1+\frac{1}{\beta } ] \gamma[-1+\frac{1 } { \beta}]}{(\gamma[\frac{1}{\beta}])^2 } \biggr ( \frac{1}{q\ , r^\beta_0 } - \beta\,(t - t_0 ) \biggr)^{1 - 1/\beta } \nonumber\\ & & + \ , { \cal{o}}(higher - order ) \;. \label{b_4}\end{aligned}\ ] ] upon introducing this result into eq . ( [ b_3 ] ) and using the property of @xmath170 functions @xmath171 \gamma[-1 + 1/\beta]}{(\gamma[1/\beta])^2 } = \frac{1}{1-\beta } \ ; , \label{gammas}\ ] ] the expansion of the exponential yields @xmath172 = \frac{r_0}{\delta t } \biggr ( \frac{1}{s + q r^\beta_0 } \biggr)^{1/\beta } \biggr\ { \biggr\ { s^{-1 + 1/\beta } \nonumber\\ & & \times\ , _ 2f_1(\frac{1}{\beta},\frac{1}{\beta},1+\frac{1 } { \beta};z ) \biggr|_{s=0 } \biggr [ 1 + \biggr ( \frac{1}{\beta q r^\beta_0 } - ( t - t_0 ) \biggr ) s \biggr ] \nonumber\\ & & + \frac{1}{1-\beta } \biggr ( \frac{1}{q\,r^\beta_0 } - \beta\,(t - t_0 ) \biggr)^{1 - 1/\beta } \biggr\ } [ 1+(t - t_0)\,s ] \nonumber\\ & & - s^{-1 + 1/\beta}\ , _ 2f_1(\frac{1}{\beta},\frac{1}{\beta } , 1+\frac{1}{\beta};z_0 ) \biggr|_{s=0 } \biggr [ 1 + \frac{s}{\beta q r^\beta_0 } \biggr ] \nonumber\\ & & - \frac{1}{1-\beta } \biggr ( \frac{1}{q\,r^\beta_0 } \biggr)^{1 - 1/\beta } \biggr\ } \ ; , \label{b_5}\end{aligned}\ ] ] when keeping in the cpi only the lowest order of @xmath84 one gets @xmath172 = \frac{r_0}{\delta t } \biggr ( \frac{1}{s + q r^\beta_0 } \biggr)^{1/\beta } \biggr\ { s^{-1 + 1/\beta } \nonumber\\ & & \times\ , _ 2f_1(\frac{1}{\beta},\frac{1}{\beta},1+\frac{1 } { \beta};z ) \biggr|_{s=0 } \biggr [ 1 + \frac{s}{\beta q r^\beta_0 } \biggr ] \nonumber\\ & & + \frac{1}{1-\beta } \biggr ( \frac{1}{q\,r^\beta_0 } - \beta\,(t - t_0 ) \biggr)^{1 - 1/\beta } [ 1+(t - t_0)\,s ] \nonumber\\ & & - s^{-1 + 1/\beta}\ , _ 2f_1(\frac{1}{\beta},\frac{1}{\beta } , 1+\frac{1}{\beta};z_0 ) \biggr|_{s=0 } \biggr [ 1 + \frac{s}{\beta q r^\beta_0 } \biggr ] \nonumber\\ & & - \frac{1}{1-\beta } \biggr ( \frac{1}{q\,r^\beta_0 } \biggr)^{1 - 1/\beta } \biggr\ } \;. \label{b_6}\end{aligned}\ ] ] furthermore , due to the fact that @xmath173 the remaining contributions of the gauss hypergeometric function cancel out leading to @xmath172 = \frac{r_0}{\delta t } \biggr ( \frac{1}{s + q r_0^\beta } \biggr)^{1/\beta } \frac{1}{1-\beta } \biggr ( \frac{1}{q\,r^\beta_0 } \biggr)^{1 - 1/\beta } \nonumber\\ & & \times \biggr [ \biggr ( 1 - \beta\,q\,r^\beta_0\,(t - t_0 ) \biggr)^{1 - 1/\beta } [ 1+(t - t_0)\,s ] - 1 \biggr ] \;. \nonumber\\ \label{b_7}\end{aligned}\ ] ] now , the limit @xmath83 , where @xmath166 $ ] , leads to @xmath73 & = & \frac{r_0}{(1-\beta ) \delta t } \frac{1}{q r^\beta_0 } \nonumber\\ & & ~~ \times \biggr [ \biggr(\frac{1}{1-\beta q r^\beta_0 ( t - t_0 ) } \biggr)^{\frac{1-\beta}{\beta } } - 1 \biggr ] \nonumber\\ & = & \frac{\beta\,\,r_0}{(1-\beta ) } \biggr(\frac{t_c - t_0}{\delta t } \biggr)\biggr [ \biggr(\frac{t_c - t_0}{t_c - t } \biggr)^{\frac{1- \beta}{\beta } } - 1 \biggr ] \ ; , \nonumber\\ \label{p_final}\end{aligned}\ ] ] in agreement with eq . ( [ price1 ] ) . \(ii ) in the case of @xmath43 , starting from the cpi given by eq . ( [ lptln1 ] ) @xmath44 = \frac{1}{q\,\delta t } \ln \biggr [ \frac{s}{s+q\,r_0\,(1-\exp [ s ( t - t_0 ) ] ) } \biggr ] \ ; , \label{b_15}\ ] ] for @xmath83 one gets @xmath73 & = & \frac{1}{q\,\delta t } \ln \biggr [ \frac{1}{1-q\,r_0\,(t - t_0 ) } \biggr ] \nonumber\\ & = & r_0 \biggr(\frac{t_c - t_0}{\delta t}\biggr ) \ln \biggr [ \frac{1}{1-(t - t_0)/(t_c - t_0 ) } \biggr ] \nonumber\\ & = & r_0 \biggr(\frac{t_c - t_0}{\delta t}\biggr ) \ln \biggr ( \frac{t_c - t_0}{t_c - t } \biggr ) \ ; , \label{b_16}\end{aligned}\ ] ] which coincides with eq . ( [ price2 ] ) . olivera , money , prices and fiscal lags : a note on the dynamics of inflation , banca nazionale del lavoro quarterly review * 20 * ( 1967 ) 258 - 267 . v. tanzi , inflation , lags in collection , and the real value of tax revenue , imf staff papers , * 24 * ( 1977 ) 154 - 167 ; and inflation , real tax revenue , and the case for inflationary finance : theory with an application to argentina , imf staff papers , * 25 * ( 1978 ) 417 - 451 . lucas jr , expectations and the neutrality of money , journal of economic theory * 4 * ( 2 ) ( 1972 ) 103 - 124 . m. king , twenty years of inflation targeting ( speech ) , the stamp memorial lecture , london school of economics 9 october 2012 . s. fischer , r.e . hall , and taylor , relative shocks , relative price variability , and inflation , brookings papers on economic activity 1981.2 ( 1981 ) 381 - 441 . cooley and g.d . hansen , the inflation tax in a real business cycle model , american economic review * 79 * ( 4 ) ( 1989 ) 733 - 748 . taylor , aggregate dynamics and staggered contracts , journal of political economy * 88 * ( 1 ) ( 1980 ) 1 - 23 . d. heymann and a. leijonhufvud , high inflation : the arne ryde memorial lectures , clarendon press , oxford ( england ) , 1995 . lucas jr , econometric policy evaluation : a critique , in : karl brunner and allan h. meltzer ( eds ) , the phillips curve and labor market vol 1 of carnegie - rochester conference series on public policy , north - holland , amsterdam , 1976 . r.n . mantegna and e. stanley , an introduction to econophysics : correlations and complexity in finance , cambridge university press , cambridge , england , 1999 . s. moss de oliveira , p.m.c . de oliveira , and d. stauffer , evolution , money , war and computers , teubner , stuttgart - leipzig , 1999 . d. sornette , why stock markets crash ( critical events in complex financial systems ) , princeton university press , princeton , 2003 . p. cagan , the monetary dynamics of hyperinflation , in : m. friedman(ed . ) , studies in the quantity theory of money , university of chicago press , chicago , 1956 . t. mizuno , m. takayasu , and h. takayasu , the mechanism of double - exponential growth in hyperinflation , physica a * 308 * ( 2002 ) 411 - 419 . d. sornette , h. takayasu , and w .- x . zhou , finite - time singularity signature of hyperinflation , physica a * 325 * ( 2003 ) 492 - 506 . m.a . szybisz and l. szybisz , finite - time singularity in the evolution of hyperinflation episodes , http://arxiv.org/abs/0802.3553 . m.a . szybisz and l. szybisz , finite - time singularities in the dynamics of hyperinflation in an economy , phys . b * 80 * ( 2009 ) 026116/1 - 11 . l. szybisz and m.a . szybisz , people s collective behavior in a hyperinflation , advances applic . stat . sciences * 2 * ( 2010 ) 315 - 331 . l. szybisz and m.a . szybisz , universality of the behavior at the final stage of hyperinflation episodes in economy , anales afa * 26 * ( 2015 ) 142 - 147 . szybisz and l. szybisz , hyperinflation in brazil , israel , and nicaragua revisited , http://arxiv.org/abs/1601.00092 . r. cooper , coordination games : complementarities and macroeconomics , cambridge university press , cambridge , 1999 . diamond , aggregate demand management in search equilibrium , journal of political economy * 90 * ( 1982 ) 881 - 894 . m. bruno and s. fisher , seigniorage , operating rules and the high inflation trap , national bureau of economic research cambridge , mass . , usa , 1987 . m. obstfeld , models of currency crises with self - fulfilling features , european economic review * 40 * ( 3 ) ( 1996 ) 1037 - 1047 . p. krugman , balance sheets , the transfer problem , and financial crises , in : international finance and financial crises , springer netherlands , 1999 . 31 - 55 ) s. morris and h.s . shin , rethinking multiple equilibria in macroeconomic modeling , in : nber macroeconomics annual , b.s . bernanke and k. rogoff ( eds . ) , mit press , massachusetts , 2000 . vol * 15 * pp . 139 - 181 . nadal , d. phan , m.b . gordon , and j. vannimenus , multiple equilibria in monopoly market with heterogeneous agents and externalities , quantitative finance * 5 * ( 2005 ) 557 - 568 . b. paal , measuring the inflation of parallel currencies : an empirical reevaluation of the second hungarian hyperinflation , stanford institute for economic policy research , discussion paper no . 00 - 01 , june 2000 . p.z . grossman and j. horvth , the dynamics of the hungarian hyperinflation , 1945 - 6 : a new perspective , scholarship and professional work - business , paper 29 , 2000 . bevington , data reduction and error analysis for the physical sciences , mcgraw hill , new york , 1969 . press , s.a . teukolsky , w.t . vetterling , and b.p . flannery , numerical recipes in fortran 77 , cambridge university press , cambridge , 1996 . table of the international monetary fund , http://www.imf.org/external/pubs/ft/weo/2002/01/data/index.htm . l. leiderman and n. liviatan , the 1985 stabilization from the perspective of the 1990 s , israel economic review * 1 * ( 1 ) ( 2003 ) 103 - 131 . l. rojas - surez , from the debt crisis toward economic stability : an analysis of the consistency of macroeconomic policies in mexico , international monetary fund , working paper no . 17 , 1992 . t. goodman , the economy , in : mexico a country study , t.l . merrill and r. miro ( eds . ) , federal research division , library of congress , 1997 . chapter 3 , pp . 141 - 228 . r. bergoeing , p.j . kehoe , t.j . kehoe , and r. soto , a decade lost and found : mexico and chile in the 1980 s , federal reserve bank of minneapolis , 2007 . federal reserve bank of st . louis , https://research.stlouisfed.org/fred2/series/ p.s . andersen and m. gumundsson , inflation and disinflation in iceland , bank for international settlements , working papers no . 52 , basel , switzerland , 1998 .

models involving large extra spatial dimension(s ) have interesting predictions on lepton flavor violating processes . we consider some 5d models which are related to neutrino mass generation or address the fermion masses hierarchy problem . we study the signatures in low energy experiments that can discriminate the different models . the focus is on muon - electron conversion in nuclei , and processes and their counterparts . their links with the active neutrino mass matrix are investigated . we show that in the models we discussed the branching ratio of like rare process is much smaller than the ones of like processes . this is in sharp contrast to most of the traditional wisdom based on four dimensional gauge models . moreover , some rare tau decays are more promising than the rare muon decays .

hep-ph0501161

in the standard model(sm ) with fifteen fermions per family neutrinos are strictly massless and the charged leptons weak eigenstates can be chosen to be their mass eigenstates . thus , each generation has a separately conserved lepton number . if one neglects the tiny effects from nonperturbative processes , there is no lepton flavor violating(lfv ) interaction in sm . however , recent neutrino experiments show strong evidence that neutrinos have none zero masses and the three active neutrinos mix@xcite . most physicists take this to be a harbinger of new physics beyond the sm . moreover , finite neutrino masses alone would imply the existence of lfv in charged lepton sector analogous to the quarks . if so we expect the glashow - iliopoulos - maiani(gim ) mechanism to be operative in the lepton sector then the rate of induced lfv processes will be proportional to the neutrino mass square difference , which is of the order of @xmath3 . hence , they will be hopelessly small for experimental verification . therefore , additional ingredients are essential for a detectable lfv signature . it is very common in model building to have the new physics that generate neutrino masses also given rise to lfv reactions . this link appears to be natural although there is no guarantee that this is case in nature . with this cautionary note we will focus attention to new physics that links the two phenomena . among the numerous beyond the sm models , lfv signatures are most intensely studied in supersymmetric ( susy ) ones . the connection with neutrino masses is established through the seesaw mechanism which is the orthodox way of getting a small mass for the active neutrinos . since the latter has a natural setting in grand unified theories ( gut ) the end result are rather bedecked supersymmetric seesaw models ; see e.g. @xcite . although the details are different the generic source of lfv lies in the mixing of various sfermions . the right - handed majorana neutrinos play a secondary role in this class of models . in general it is natural to expect @xmath4 in susy models . for non - supersymmetric models neutrino mass generation via the seesaw mechanism would required the right - handed neutrinos to be of the gut scale . in this simplest version all lfv are undetectable . attempts are now made to lower some right - handed neutrinos mandated by the seesaw mechanism to the tev scale so that the seesaw mechanism itself can be tested experimentally . if so then one can optimistically anticipate lfv signatures in the next round of experiments @xcite . independent of the details of the models one again expects @xmath4 to hold true . recently a new avenue has open up in the construction of models beyond the sm that exploits the possible existence of extra spatial dimensions . these theories are particularly interesting phenomenologically in the brane world context . it is fascinating that many long standing problems in the usual four dimensional ( 4d ) field theories can be overcome or take on new perspectives in these higher dimensional constructs . for example the hierarchy problem is solved by invoking large extra dimensions . in this note , we would like to draw the readers attention to the models which involve one or more flat extra spatial dimensions . furthermore , we focus on those that address the neutrino mass problem . in some cases , we predict a reversed pattern of @xmath5 compare to susy models . on the experimental side , it shall be interesting to see this . the current experimental limits on muon lfv have already put very stringent constraints on model building . on the other hand , the limits from tau lfv are rather loose . we shall constraint the extra dimension models by the muon rare processes data and place upper limits on the rare decays of the @xmath2 . to avoid any hadronic uncertainties we shall focus on purely leptonic processes . we shall also discuss the possible ways to discriminate different models and their connections to neutrino masses . in this brief review , we give few examples of extra dimension models which give potentially testable lfv signatures . these lfv processes are all directly or indirectly related to the generation of neutrino masses . we compare the lfv processes in a five dimension(5d ) @xmath6@xcite and @xmath7@xcite gut models where neutrino majorana masses are generated radiatively without a right - handed neutrino which is viable but less discussed alternative to the seesaw mechanism . a brief review of this construction is given in @xcite . also , we discuss the lfv processes in split fermion or multi - brane scenario . the following is our plan for the paper . in sec.ii , we will first review the general operator analysis for the lepton flavor violating processes . this will also set the notations for the rest of the discussions . iii examines lfv in a 5d @xmath8 model . new results of the one - loop calculations are given here . for details of the model and neutrino mass generation we refer to @xcite . in sec . iv , the lfv processes induced by 5d @xmath7 model will be discussed . the discussion here has not been presented before . the alternative way of studying the flavor problem using the split fermion model is examined in sec.v . calculations of the lfv processes in this scenario involves many new unknown parameters . the models lack predictive power even semi - quantitatively . however , very general generic trends for lfv can be discerned even in this early stages of development . our conclusions are given in sec . the necessary 5d gauge fixing details , which is crucial for loop calculations , are presented in an appendix . first of all , we collect the necessary general formulas for the study of lfv processes . the most important ones are the effective interactions of @xmath9 and @xmath10 where we use the notation @xmath11 to denote the heavier charged lepton which usually is either @xmath12 or @xmath2 and @xmath13 is the lighter daughter lepton which can be @xmath12 or @xmath14 . in lfv studies , the most important contribution comes from the effective @xmath9 vertex . the similar vertex where a virtual @xmath15 replaces the @xmath16 is subdominant in the class of models we are considering . for definiteness we will take @xmath17 and @xmath18 . the most general @xmath19 interaction amplitude allowed by lorentz and gauge invariance can be written as : = -e a^*_(q ) ( p_-q)\ { _ ( g^-q^q^q^2 ) + } u_(p _ ) with the convention @xmath20 used through out this paper and @xmath21 is the photon 4-momentum . for real photon emission , only @xmath22 and @xmath23 contribute . but if a off - shell photon is involved , then all 4 form factors contribute . after proper renormalization , the amplitude is finite as @xmath24 , so we must have @xmath25 . it is customary to factor out @xmath26 and rewrite the electric and magnetic form factors as [ eq : mu_e_form ] f_e0(q^2)=q^2m_^2_e0(q^2 ) , f_m0(q^2)=q^2m_^2_m0(q^2 ) and now @xmath27 and @xmath28 are finite at @xmath24 . using similar notations of @xcite , the most general effective lagrangian for @xmath1 and @xmath0 can be expressed as : -4 g_f & = & m_a_r ^_l f _ + m_a_l ^_r f _ + g_1 ( _ l)(e_l ) + g_2 ( _ r)(e_r ) + & + & g_3 ( ^_r)(_e_r ) + g_4 ( ^_l)(_e_l ) + & + & g_5 ( ^_r)(_e_l ) + g_6 ( ^_l)(_e_r ) + h.c . where a_r= - e 8 g_f m_^2 , a_l= - e 8 g_f m_^2 . also note the anapole form factors @xmath31 and @xmath32 have vector like effective contributions to @xmath33 : g_3 = g_5= e^2 4g_f m_^2 + g_4 = g_6= e^2 4g_f m_^2which shall be included in the @xmath34 . the above effective lagrangian leads to b(e)&= & 384 ^ 2 ( |a_l|^2+|a_r|^2 ) + [ eq : taudecay ] b(3e)&=&|g_1|^2+|g_2|^28 + 2(|g_3|^2+|g_4|^2)+|g_5|^2+|g_6|^2 + & & + 8e re + & & + 64e^2 \{-}(|a_r|^2+|a_l|^2 ) if electron mass is ignored . to carry out the calculation , it s convenient to define two dimensionless variables @xmath35 and @xmath36 . however , it is important to keep @xmath37 terms in the intermediate steps in order to properly extract the finite term in the last line of eq.([eq : taudecay ] ) . our result agrees with @xcite . the expressions of eq.([eq : taudecay ] ) , except the last line of dipole operators , can also apply to @xmath38 and @xmath39 processes . for @xmath40 processes , the part from dipole operators have double loop suppression from two flavor violation vertices and resulting in an insignificant branching ratios thus can be safely ignored . for @xmath2 decay , the branching ratios given above are normalized to @xmath41 . this holds for subsequent discussions of @xmath2 decays . to complete the story , we also give the expression for processes with no identical particles in the final state , namely , @xmath42 or @xmath43 . the above expression for the branching ratio is now modified to : [ eq : taudecay2 ] b(l_1 l_2 ) & = & |g_1|^2+|g_2|^24 + ( |g_3|^2+|g_4|^2 + |g_5|^2+|g_6|^2 ) + & & + 8e re + & & + 64e^2 \{-}(|a_r|^2+|a_l|^2 ) with trivial extension of @xmath44s . in arriving the last line of eq.([eq : taudecay2 ] ) , we have ignored the masses difference between @xmath45 and @xmath46 in phase space integration but keep the crucial mass singularity associated with the virtual photon . not surprisingly , the approximation agrees very well with the actual numerical integrations . if the photonic dipole operator is the only dominate lfv source , we have the following model independent prediction for b(e ) & = & b ( ) + b(e|e)b(e ) & = & \ { -32 } + b(e |)b(e ) & = & \ { -32 } to the accuracy of @xmath47 . to our knowledge , the last two relations have not been presented before . we can write the effective lfv lagrangian for @xmath48 conversion as : & = & |e(s -p^5 ) _ q |q(s_q - p_q^5)q + |e ^(v - a^5)_q |q _ ( v_q - a_q^5)q + & + & 12 |e(t_s + t_p ^5)^_q |q _ q + h.c . [ eq : lmueti ] with self explanatory notations . here , flavor changing terms in the quark sector are not included since they are not expected to be important here . the effective couplings are normalized to @xmath49 . for example , the sm @xmath15 boson has a vector coupling to quarks given by @xmath50 to calculate the conversion rate , we need to promote the interaction from quark level to the nucleon level by computing the matrix elements @xmath51 with @xmath52 denotes a nucleon and @xmath53 . since the coherent process is the important one only vector and scalar operators matter : p||q_q | p= g_v^q , p |p_p , n||q_q | n= g_v^q , n |n_n and p||q q | p= g_s^q , p|p p , n||q q | n= g_s^q , n|n n. by conserving of vector current , in the @xmath54 limit , one can determine that @xmath55 and @xmath56 . however , one has to rely on the nucleon model to evaluate the scalar operator . for qualitative estimation , we will use the result @xmath57 from full non - relativistic quark model but the reader should keep in mind that the uncertainty of nucleon model could be as large as few tens percent@xcite . following the approximations used in @xcite , the conversion rate , normalized to the normal muon capture rate @xmath58 , can be expressed as@xcite : [ eq : mu - e - barnchrate ] b_conv= p_e e_e g_f^2 f_p^2 m_^3 ^3 z_eff^4 2 ^2 z _ capt \ { |4 e a_l z + ( s - p)s_n + ( v - a)q_n|^2 + |4 e a_r z + ( s+p)s_n + ( v+a)q_n|^2 } by assuming that the proton and neutron density are equal and the muon wave function does not change very much in the nucleus , and @xmath59 is a form factor whose definition can be found in @xcite and @xmath60 is the electron momentum ( energy ) , @xmath61 . for @xmath62 , @xmath63 , @xmath64 , and @xmath65@xcite . where the coherent vector and scalar coupling strength of nuclei @xmath52 are defined as s_n & & s^u(2z+n)+s^d(2n+z ) , + q_n & & v^u(2z+n)+v^d(2n+z ) . if there are more than one gauge or scalar bosons mediate this process , the above expression can be trivially extended with modified couplings : ( sp)s_n _ i ( s^i p^i ) s^i_n , + ( va)q_n _ i ( v^i a^i ) q^i_n . note that the form factors @xmath66 and @xmath67 in eq.([eq : mu_e_form ] ) have extra contribution to the vector couplings : v= -2 e m_w g m__e0,a= -2 e m_w g m__m0,v_q= 2 e m_w g m_q_q , and if eq.([eq : mu_e_form ] ) is the only lfv source , then eq.([eq : mu - e - barnchrate ] ) reduces to the well - known formula given in @xcite b^_conv= 8 m_f_p^2 ^5 z_eff^4 z _ capt \{|f_m1+f_e0|^2 + @xmath48 conversion and the @xmath68 [ eq : weinberg ] b^_conv = m_^5 g_f^2 f_p^2 ^4 z_eff^4 z 12 ^ 3_capt ( brief review is sufficient for the phenomenological analysis we do . next , we will head for the extra - dimensional models and discuss their lfv signatures . it has been known for a long time that the sm lepton left - handed doublet and the right - handed singlet charged lepton in each family can beautifully form an @xmath6 fundamental representation , i.e. @xmath69 @xcite . this is implemented in an electroweak only unification in which @xmath70 is unified to @xmath6 . one of the attractive points of this unification model is the tree level prediction of @xmath71 . renormalization group considerations point to a relatively low scale of unification at @xmath72 few @xmath73 . we shall use @xmath74 to denote the @xmath75 gauge bosons which have sm quantum number @xmath76 . in 4d , the @xmath6 gut has a fundamental difficulty of embedding quarks into @xmath6 representations . this problem can be circumvented by promoting the model into five dimensional space time @xcite and @xcite . we give a brief summary of the model construction here . the extra spatial dimension , with coordinate denoted by @xmath77 , is compactified into an @xmath78 orbifold . where the circle @xmath79 of radius @xmath80 , or @xmath81 $ ] , is orbifolded by a @xmath82 which identifies points @xmath77 and @xmath83 . the resulting space is further divided by a second @xmath84 acting on @xmath85 to give the final geometry . we now have tow parity transformations @xmath86 and @xmath87 under which the bulk fields can be assigned either of the eigenvalues + or - . this freedom is used to break the bulk @xmath6 symmetry to @xmath70 . explicitly , one assigns the following properties to bulk gauge fields _ ( y ) = p a_(-y ) p^-1 , & & a_(y ) = p a_(-y ) p^-1 + a_5 ( y ) = -p a_5(-y)p^-1 , & & a_5(y ) = -p a_5(-y)p^-1 where the matrices @xmath88 and @xmath89 . now the @xmath90 parities of the sm gauge bosons and the @xmath91 gauge bosons are @xmath92 and @xmath93 respectively . it is easy to work out the fourier eigenmodes propagating in the bulk and see that only fields with @xmath92 parity have zero modes . in other words , only sm gauge bosons have zero modes . both the @xmath91 gauge bosons and all the @xmath94components are heavy kk excitation . note the second @xmath95 is necessary to avoid the presence of zero modes for both sm gauge boson and the exotic @xmath96 boson at the same time . the @xmath6 symmetry is explicitly broken to @xmath97 at the @xmath98 fixed point , where the 4d quarks field are forced to live on it . the extra degree of freedom in extra dimensional theories is the key to incorporate sm quarks into the @xmath6 symmetry . on the other hand , the lepton fields can be placed anywhere in the bulk or on either two fixed points . we choose to put the 4d lepton triplets at @xmath99 which is a @xmath6 symmetric fixed point so that they enjoy the @xmath6 symmetry . this also avoids possible proton decay contact interactions . one higgs triplet @xmath100 plus one higgs anti - sextet , denoted as @xmath101 , with parities _ 3(y)&= & p _ 3(-y ) , _ 3(y)= p _ 3(-y ) + _ 6(y)&= & p _ 6(-y)p^-1 , _ 6(y)=- p _ 6(-y ) p^-1 . is the minimal scalar set to give viable charged fermion masses ( see @xcite ) . another higgs triplet with parities @xmath93 is introduced to transmit lepton number violation essential for generating majorana neutrino mass through one - loop diagrams @xcite by a triple higgs interaction of the type of @xmath102 . this is a 5d realization of radiative neutrino mass generation first proposed in @xcite . the resulting mass matrix is necessarily of the majorana type . now we have all the ingredients to write down explicitly the 5d lagrangian density _ 5 & = & -12 + + & + & ( d_m_3)^(d^m_3 ) + ( d_m_3)^(d^m_3 ) + & + & ( y ) + & - & v_0(_6,_3,_3 ) - _ 3^t _ 6 _3 + h.c . + & + & l_gf + . [ 5dl ] where @xmath103 is the 5d field strength and @xmath104 is the 5d covariant derivative . the cutoff scale @xmath105 is introduced to make the coupling constants dimensionless . the other notations are self explanatory . the quark sector is not relevant now and will be left out . the complicated scalar potential is gauge invariant and orbifold symmetric and will not be specified since it is not needed here . to perform loop calculations , we need to specify the 5d gauge fixing term , @xmath106 , which will be exhibited later . the fields and their parities of this model are summarized below : 8^&= & _ b^ + _ a^ + _ ( u , v)^ + 8^y & = & _ b^y + _ a^y + _ ( u , v)^y + & = & _ h_w1 + _ h_s + & = & _ h_w1 + _ h_s + & = & _ h_t + _ h_w2+_h_s2where the sm quantum numbers are @xmath107 and the subscripts label the parities @xmath108 . then it is straightforward to obtain the 4d effective interaction by integrating over @xmath77 and the 4d effective gauge coupling can be identified as @xmath109 . the orbifold construction is engineered such that there is no tree level lfv in the sm gauge interactions . thus , the success of that model remains intact . but the tree level lfv interaction emerge in the @xmath91 gauge interaction which are heavy kk excitation and in the yukawa interactions . the lfv charged current is _ cc = g_2 _ n=1 ^p_l ( u_lep)_ij e^_r j u^-2_n , + h.c . + + g_2 _ n=1 ^p_l ( u_lep)_ij e^_rj v^-1_n , + h.c . where the subscripts @xmath11 and @xmath80 are kept for book keeping . the matrices @xmath110 are used to diagonalize the charged lepton mass matrix and @xmath111 is an extra ckm - like unitary mixing matrix for the lepton sector . the lfv yukawa interactions are given by _ y & = & 1 _ n=0 _ n + & + & 1 _ n=0_n + & + & 1 _ n=0 _ n + h.c . where @xmath112 and f^6_t= u_l^t f^6 u_l , f^6_s= u_r^t f^6 u_r , f^()3_s= u_l^t f^()3 u_l , f^()3_h= u_r^f^()3 u_l , f^6_h= u_r^f^6 u_l . note that in the new basis the symmetry of @xmath113 and @xmath114 are not changed . we begin the discussion by studying a special case that @xmath116 , such that @xmath117 also @xmath118 and @xmath119 are roughly diagonal . this hierarchical yukawa structure is also demanded to yield the observed charged lepton mass hierarchy . in other words , all the lfv sources are in the yukawa interaction of @xmath120 and @xmath121 . since @xmath122 has nothing to do with the charged lepton masses , we can further assume its lfv contribution is larger than @xmath120 , whose coupling is roughly @xmath123 , and @xmath124 . in general this class of decays proceeds via the one - loop diagrams . the ones involving the gauge boson @xmath125 and @xmath126 are suppressed by the gim mechanism . this leaves the singly charged and neutral scalars as the only possible contributors since they both carry two units of lepton charges in the usual scheme . we thus conclude that these decays are dominated by the scalar induced m1 and e1 operators only . therefore , they provide unique probes of the exotic scalar sector . later we will show that in contrast @xmath127 probes the gauge interactions of the model . in this case , the leading contribution loop diagrams are shown in fig.[fig : muetransition ] . and @xmath48 conversion in the case described in sec.[sec : su3case ] . the labels @xmath128 indicate the kk level.,scaledwidth=60.0% ] now , briefly discuss the gauge fixing in this model . because the orbifold parity for @xmath121 is chosen to be @xmath93 , it can not develop a vev and does nt participate the electroweak breaking . the goldstone bosons consist of the @xmath77-components of gauge bosons and the proper linear combinations of @xmath120 and @xmath101 . and the whole @xmath129 , @xmath130 and @xmath131 are physical higgs . so now it is straightforward to carry out loop calculation . for further details , see appendix . the @xmath132 form factors are calculated to be : [ eq : su3m1 ] f_m1^l l&= & m_^2 384 ^2 m_s0 ^ 2_i + [ eq : su3e1 ] f_e1^l l&= & m_^2 384 ^2 m_s0 ^ 2_i where @xmath133 , @xmath134 is the zero mode mass of @xmath135 , and @xmath136 . on arriving at the above expression , the contributions of all kk scalar excitation running in the loop have been summed . and if we drop the @xmath137-terms , the resulting branch ratio can be expressed as : b(ll+)&= & 96 ^ 3 g_f^2 m_^4 ( |f_e1^ll|^2+|f_m1^l l|^2 ) ~768 g_f^2 m^4_s0|_i = e , , f_li|^2 + & = & 2.7510 ^ -6 ( 300 m_s0)^4 | ( f^3_s , le)^*f^3_s , el+ ( f^3_s , l)^*f^3_s , l+ ( f^3_s , l)^*f^3_s , l |^2 because the yukawa couplings of triplet scalars are anti - symmetric , the @xmath115 processes have following forms : b(e+ ) = 2.7510 ^ -6 ( 300 m_s0)^4 | ( f^3_s , e)^*f^3_s,|^2 + b(e+ ) = 2.7510 ^ -6 ( 300 m_s0)^4 | ( f^3_s , e)^*f^3_s,|^2 + b(+ ) = 2.7510 ^ -6 ( 300 m_s0)^4 | ( f^3_s , e)^*f^3_s , e|^2 we have taken @xmath138gev as the reference point . if all of the yukawa couplings are real and none of them vanishes , their ratios can be further simplified to : b(e+):b(e+):b(+ ) = 1|f^3_s , e|^2:1|f^3_s , e|^2:1|f^3_s,|^2 . at this point one can use the data @xmath139@xcite to obtain the constrain @xmath140 . this is consistent with the expectation from the study of neutrino mass in this as given in @xcite . there it was found that the yukawa coupling @xmath141 has to be @xmath142 and the @xmath143 s exhibit the pattern @xmath144 . hence it reasonable @xmath145 to occur at a rate less than two orders of magnitude below current level . indeed in the @xmath6 model we can link the various @xmath146 transition branch ratios to the light neutrino mass matrix elements . assuming that the light neutrino mass are mostly coming from the one - loop quantum correction involving the zero modes of @xmath121 and @xmath101 , we have the prediction : b(e+):b(e+):b(+ ) ~(m_m_)^4 m_13m_23 : m_12m_23:m_12m_13 where @xmath147 is the @xmath148 entry of the light neutrino mass matrix . interestingly the model naturally accommodates an active neutrino mass matrix of the inverted hierarchy type as follows : m~ ( & 1&1 + 1 & & + 1&&^2 ) where @xmath149 . from the above equations , we see that @xmath150 is suppressed compared to the @xmath151 decays this is a striking feature of the model . the @xmath48 conversion in nuclei will be dominated by the virtual photon exchange . compared to @xmath68 it has additional contributions from the anapole terms . the corresponding photon @xmath152 form factors can be derived as : [ eq : su3e0 ] _ e0(-k^2)&= & m_^2 576 ^2 m_s0 ^ 2 _ i = e , , + [ eq : su3m0 ] _ m0(-k^2)&= & m_^2 576 ^2 m_s0 ^ 2 _ i = e,,where @xmath153 , @xmath154 , @xmath155 , and g ( , x)= -- x + 13 - 4x+(1 - 2x ) . as expected , the principal contribution is from the fig.[fig : muetransition](c ) with the @xmath135 zero mode running in the loop . the logarithmic enhancements in @xmath156 , is due to the exchange of neutral scalars , @xmath157 ( see fig.[fig : muetransition](a ) ) . although they are suppressed by the kk masses we find them to be compatible to the charged singlet contribution . in this process @xmath158 and @xmath159 has the following limits g_e , n ~-+13 , g_,n ~--1.515 , g_,n ~--6.978 the kk sum of these logarithmic enhancements are finite : _ n=1^ = ( m_r)^2 -0.8362 so the desired anapole form factors can be expressed as _ e0(m_^2)&= & m_^2 576 ^2 m_s0 ^ 2 _ i = e , , + _ m0(m_^2)&= & m_^2 576 ^2 m_s0 ^ 2 _ i = e,,@xmath160 . again , since @xmath161 is anti - symmetric , only @xmath162 can contributes . for simplicity we assume there is no new cp violation in the scalar sector ; then @xmath163 and the @xmath48 conversion rate in @xmath164 can be expressed as : [ eq : muegsu31 ] b^_conv ~0.01 b(e+ ) if taking @xmath165tev and @xmath166gev as a reference point . it is also possible to have extra contributions from kk photon and kk @xmath15 excitation , fig.[fig : muetransition](a - f ) . one will need to take care of the kk number conservation in the scalar - scalar - gauge boson vertices and sum over all the possible combinations . but generally speaking , their contributions are further suppressed by @xmath167 compared to the photon zero mode and can be safely ignored . the relation of eq.([eq : muegsu31 ] ) is based on the assumption that @xmath168 and @xmath121 is the dominate lfv source . however , we should point out that if @xmath169 is not so small the neutral scalar zero modes can make @xmath68 and @xmath170 compatible and deviates a lot from the pure photonic dipole prediction , eq.([eq : weinberg ] ) . again , this demonstrates that @xmath146 and @xmath48 conversion are very important for us to understand the yukawa structure in the @xmath6 model . the question now arises about the photonic dipole and anapole contribution to @xmath1 . the answer lies in eqs.([eq : su3m1],[eq : su3e1],[eq : su3e0 ] , [ eq : su3m0 ] ) . we estimated that b(3e ) < 0.04 b(e ) . this prediction is not very sensitive to what the yukawa pattern is . moreover , the decays @xmath171 have overwhelming contribution from other sources of new physics in the model to which we shall turn our attention to next . a characteristic of the model is the existence of double charged gauge bosons with lfv couplings . this will induce @xmath1 like processes for the @xmath2 . in addition there are also kk scalars @xmath172 and @xmath173 which has lfv yukawa couplings which are largely unknown . the feynman diagrams for the @xmath127 decays are depicted in fig.[fig : su3_tau3l ] . processes . , scaledwidth=80.0% ] since the yukawa coupling are totally unknown , we will postpone the discussion of the contributions from scalars and look at the branch ratios , normalized to @xmath174 , mediated by @xmath125 gauge boson alone first : b(3 ) & = & f ( |u_|^2 + |u_|^2 ) |u_|^2 , + b(3 e)&= & f ( |u_e|^2 + |u_e|^2 ) |u_ee|^2 , + b(| ee)&=&f ( |u_|^2 + |u_|^2 ) |u_e e|^2 , + b(|e)&=&f ( |u_e|^2 + |u_e|^2 ) |u_|^2 , + b(e|e)&= & 8 ( |u_e|^2 + |u_e|^2 ) ( |u_e|^2+|u_e|^2 ) , + b(e|)&= & 8 ( |u_|^2 + | u_|^2 ) ( |u_e|^2+|u_e|^2 ) . where @xmath175 . from the analysis given in sec.ii , we know all scalar operators give positive contribution . so even though we know nothing about the yukawa couplings , we can still derive an interesting lower bond from the unitarity of @xmath176 [ eq : tau3elowb ] b(3 e ) |u_ee|^2 ( 1-|u_ee |^2 ) for a given @xmath177 . if one wants to keep compactification scale @xmath177 low , say @xmath178 tev , then we would require @xmath179 to be either close to zero or one . furthermore , if we take the upper bound of @xmath180 tev derived from unification seriously we obtain b(3 e ) > 8.010 ^ -7 |u_ee|^2 ( 1-|u_ee |^2 ) . on the other hand , if we assume that the bilepton gauge boson exchange is the dominating fcnc source , another interesting upper bond can be derived : b(3 e ) < 4= 3.910 ^ -6 ( 2 1/r)^4 with @xmath181 in eq.([eq : tau3elowb ] ) . actually , if all the lfv yukawa couplings are associated with @xmath121 as discussed in the previous two subsections , the tree - level bilepton scalar contributions to @xmath182 vanish due to the antisymmetry of the yukawa couplings . however , the present experimental limit , @xmath183 @xcite will indicate that the compactification radius is closer to the upper limit of @xmath184 for this particular case . the orbifold @xmath6 model discussed above has many interesting and novel features ; however , the fact that quarks and leptons have to be treated differently is an obstacle towards complete unification . it s a natural attempt to further unify the quarks and leptons in a larger gut group . the simplest group for that is @xmath7 . now all fermions are on equal footing and can be clustered into 2 @xmath7 representations , i.e . @xmath185 . similar to the @xmath6 model , the model is embedded in the background geometry of @xmath186 orbifold . the bulk @xmath7 gauge symmetry is broken to the sm by orbifold parities , with parity matrices @xmath187 and @xmath188 for @xmath82 and @xmath95 transformations respectively . these are generalizations of the @xmath6 case . since no right - handed neutrinos are added , neutrino masses can be generated through quantum correction by using either @xmath189 or @xmath190 bulk scalars plus the @xmath191 interaction mandated by breaking to the sm gauge group . the orbifold parities of @xmath189 or @xmath190 bulk scalars are determined to be @xmath92 by considerations of proton decay . they split into following components : _ s(++ ) = p_15(6,1,-23)_+++t_15(1,3,1)_+++ c_15(3,2,16)_+- , + _ a(++ ) = p_10(|3,1,-23)_+++s_10(1,1,1)_+++ c_10(3,2,16)_+-.a careful analysis shows that by using @xmath192 the resultant neutrino mass matrix favor the normal(inverted ) hierarchy @xcite . it was also found that extra fine tuning efforts were needed to obtain phenomenologically acceptable neutrino mass patterns by using @xmath189 alone ; so we will only discuss the case which implements @xmath190 . the @xmath193 components induce tree - level @xmath194 mixing . to satisfy the experimental constraints , it is required that @xmath195 gev . on the other hand , the two bulk higgs in @xmath196 which are responsible for the sm electroweak symmetry breaking share the same @xmath92 parities as @xmath190 . the brane yukawa interaction term is easily constructed to be _ y = ( y ) , where @xmath197 are the @xmath7 symmetry indices . it can be seen that to contain the necessary lfv source to generate neutrino majorana masses.the neutrino mass matrix elements are proportional to @xmath198 where @xmath199 are the generation indices and @xmath200 is the mass of @xmath201-charged lepton running in the loop . the extra higgs doublet in the @xmath202 is good for gauge unification . by adding additional decaplet bulk fermion pair with @xmath93 parity and mass around @xmath203 tev , the unification is achieved at @xmath204 gev or equivalently @xmath205 gev . the high scale unification or tiny radius of extra dimension makes kk excitation decouple from most phenomenological studies and basically we only need to consider the zero modes . below unification scale or equivalently the low energy 4d effective theory is a two higgs doublets like model . in general the two yukawa patterns are not aligned which that can lead to severe tree level charged neutral flavor changing ( fcnc ) interaction . a @xmath82 symmetry is usually assumed to forbid such tree level fcnc @xcite . in this model , there is no such freedom since the yukawa patterns are determined by the geometrical setup . the @xmath206 of the first two generations are assigned to be bulk fields and the other fermion fields , @xmath207 and @xmath208 , are localized at the @xmath99 brane . in doing so , the salient @xmath7 prediction of @xmath209 ratio is preserved and give small hierarchy patterns in the yukawa couplings of both @xmath210 scalars , i.e. y_d ( ccc & & 1 + & & 1 + & & 1 ) , y_u ( ccc ^2 & ^2 & + ^2 & ^2 & + & & 1 ) . due to volume dilution factor we get @xmath211 which measures the amount of overlap between brane and bulk fields . the specific yukawa pattern above successfully generates mass and mixing hierarchy of charged fermions : m_b : m_s : m_d= m _ : m_:m_e ~1 : : ^2 + m_t : m_c : m_u ~1:^2 : ^4 , ( v_us , v_cb , v_ub)~ ( , , ^2 ) the rotation from weak to mass eigenbasis simultaneously diagonalizes the two higgs doublets yukawa couplings . thus , we do not have the fcnc problem due to mixing between two higgs doublets . instead , now tree level lfv processes can be mediated by the triplet component @xmath212 in @xmath190 . the only important ones are the @xmath1 like processes , see fig.[fig : su5_ktau ] . mixing and @xmath1 process . , scaledwidth=50.0% ] an explicit calculation give the branching ratio of @xmath1 : br(3e ) = 2|f^15_11 f^15_12|^2g_2 ^ 4(r m^*)^2 ( m_w m_t)^4 . the mass difference @xmath213 in @xmath214 mixing arises from @xmath215 can be used to eliminate the ambiguity of absolute strength of yukawa couplings . the ratio of yukawa couplings can be replaced by the ratio of the corresponding elements in @xmath216 . since only the @xmath190 higgs is used , we have br(3e)~3.0210 ^ -16 ( m^p_k m_k)^2 ( ) ^4 ( 2 m_11m_12 m_11m_22 + ( 2 m_12)^2)^2 . it is straightforward to extend the analysis to @xmath217 transitions . assuming that the hierarchy of the elements of neutrino mass matrix is smaller than factor 100 , this model predicts br(3e):br(3e):br(3 ) : br(e e):br(e ) + ~ : ( ) ^4 : ( ) ^4 : ( ) ^4 : ( ) ^4 . the photonic form factors due to @xmath218 scalar one loop diagrams can be obtained : f_m1^e = f_e1^e & = & -m_^2 16 ^2 m_t^2 5 f^15_i(f^15_ie)^*24 + ^e_m0= ^e_e0&=&m_^2 16 ^2 m_t^2 f^15_i(f^15_ie)^ * 6the chiral structure in the above result is easily understood because that only the lepton doublets interact with triplet scalar . due to the factor @xmath219 , the @xmath0 process is strongly suppressed . taking @xmath220gev , the @xmath221 conversion rate is estimated to be @xmath222 where we have kept only logarithmic terms which is sufficient for an order of magnitude estimate . the experimental bound is @xmath223 @xcite which is not stringent . on the other hand the recently proposed experiment aimed at a detecting a signal at the @xmath224 will be very encouraging . even a negative result will provide stringent constrain on the otherwise unknown yukawa couplings . an interesting scenario was introduced by @xcite to solve the charged fermion masses hierarchy problem . the basic idea is to postulate that fermions are bulk fields and they interact with a non - dynamic background scalar potential . in the 5d version the bulk fermions are vectorlike , but only one of the chiral zero modes will be localized at the zero of the background potential modulated by the 5d mass terms . the chirality of the zero mode is determined by the sign of slope of the background potential at the zero point . the fermion zero modes are given a gaussian profile in the fifth dimension and each has its own unique position in the extra dimension . the widths are controlled by the potential slope at the localized position . for simplicity , we will assume a universal width for all the sm fermions . the 5d fermion excitation are vectorlike and will be located at the same position of their zero modes . roughly speaking , the energy gap is @xmath225 and they decouple from the phenomenology we are interested in . since the sm fermions are scattered over the fifth dimension , to preserve the gauge interaction universality the sm gauge fields are forced to be bulk fields too . to illustrate the basic physics involved it suffices to build a model on an @xmath226 orbifold so that one can remove the unwanted @xmath77-components of gauge bosons zero modes which are identified with the sm gauge bosons . however , to break electroweak symmetry , a dynamic bulk higgs is necessary . to be more concrete we take the extra dimension to be the interval @xmath227 $ ] and the fermions are fixed in different positions @xmath228 in this interval . the usual kaluza - klein ansatz is invoked that any bulk field factorizes into a 4d field times a 5d wave function . thus , for the fermion field we have @xmath229 where @xmath230 is taken to be gaussian distribution in the fifth dimension : @xmath231\ ] ] where @xmath232 is the universal width of gaussian distribution . if @xmath233 , @xmath234 acts like the dirac delta function . effective 4d interactions are obtained by integrating out the @xmath77 direction . then any pair fermions get an exponential suppression @xmath235 g\left(\frac{z_1+z_2}{2},y\right)\ ] ] as a result of integrating over gaussian functions . therefore , the linear displacement between left - handed and right - handed fermions in the fifth dimension translates into exponential yukawa hierarchy in 4d theory . one set of solution for the quarks positions have been found numerically that can accommodate the mass hierarchy and the ckm mixing@xcite . to accommodate the cp violation , the overall yukawa coupling strength of up and down type quarks must be different@xcite . although a fundamental theory of where to place the fermions are located is lacking , we have at least one realistic solution for where the quarks are located in the extra dimension . for the lepton sector , we do nt have enough constraints to pin down the solution . as pointed out by@xcite , the fcnc interaction is induced geometrically where phenomenological constraint in quark sector have also been discussed . in fact , flavor violation is generic in any multi - position models . this comes from the fact that weak to mass eigenbasis rotations can only make the sm interactions ( zero modes ) flavor diagonal . the kk modes can not be simultaneously diagonalized . in general , the lfv can be discussed independently of the neutrino sector . the 4d effective lfv lagrangian can be expressed as ^lfv & = & _ n ( _ n g_2 ) |l_i ^l_j z_^n + & - & _ n ( _ n e ) |l_i ^l_j a_^n + & + & _ n ( _ n g_2 ) |l_i ^u^l_n(ij)_j w_^n- + h.c . where @xmath236 and @xmath237 are the n - th kk excitation of the photon , @xmath15 and @xmath238 bosons and @xmath239 . the matrices @xmath240 are a combination of the unitary transformations @xmath241 that take the lepton weak eigenstates to their mass eigenstates and the cosine weighting of the n - th kk modes . explicitly , they are u^l / r_n = v_l / r^ ( , , ) v_l / r for @xmath242 all the cosine factors become one and the interactions reduce to sm as expected . it is clear that the @xmath243 is no more diagonal nor unitary in general . in this model , the @xmath217 decay and @xmath48 conversion happen at tree - level , see fig.[fig : lfv_sf ] . the corresponding effective lfv couplings are : g_3&= & ( 2m_w r)^2 ( ^2_w+ g_r^2/^2_w ) _ n=1 u^r*_n , eiu^r_n , in^2 + g_4&= & ( 2m_w r)^2 ( ^2_w+ g_l^2/^2_w ) _ n=1 u^l*_n , eiu^l_n , in^2 + g_5&= & ( 2m_w r)^2 ( ^2_w+ g_r g_l/^2_w ) _ n=1 u^r*_n , eiu^l_n , in^2 + g_6&= & ( 2m_w r)^2 ( ^2_w+ g_l g_r/^2_w ) _ n=1 u^l*_n , eiu^r_n , in^2 and the @xmath244 and @xmath245 can be obtained in a similar way . also the lepton universality is broken due to flavor dependent couplings in the kk gauge interaction . we refer the reader to @xcite for a detailed analysis . the leading @xmath0 contribution comes from the one loop corrections . we need to fix the gauge before proceeding . the necessary details of 5d gauge fixing are collected in the appendix . after properly identifying the goldstone boson , the usual 4d @xmath246 gauge technique can be straightforwardly applied here . note that the yukawa couplings of the physical kk scalars are suppressed by the factor of @xmath247 . although there is residual gim cancellation in the kk gauge boson interaction , we expect the leading lfv are from kk gauge interaction . the lfv photonic form factors due to kk gauge boson and their goldstone boson can be calculated . the kk @xmath238 bosons contribution to the photonic form factors are given by : f^w_m1&=&f^w_e1=724g_2 ^ 216 ^ 2_n=1 ( m_r)^2n^2 u^l*_n , ieu^l_n , i , + & = & = 2372g_2 ^ 216 ^ 2_n=1 ( m_r)^2n^2 u^l*_n , ieu^l_n , i and for the kk @xmath15 bosons they are f^z_m1/e1&=&(m_r)^28 ^ 2 g_2 ^ 2 ^2_n=1_i 1n^2\ { -13 . + & & .+g_l g_r } , + _ e0/m0&= & -(m_r)^224 ^ 2 g_2 ^ 2 ^2_n=1_i 1n^2\ { ( g(_n , x_i)+ 12 ) } similar contributions from kk photons can be easily read from the above by replacing @xmath248 , @xmath249 , and @xmath250 . these photonic form factors give extra contribution to @xmath251 and @xmath48 conversion processes but ca nt compete with those tree - level kk gauge boson exchanging diagrams . however they are the sole sources of new physics for the @xmath115 process . in addition to the usual ignorance with regard to yukawa coupling there are more unknowns in the lepton locations and the gaussian widths . ad hoc simplifying assumptions have to be made . hence , this kind of model suffers from a lack of predictive power in lfv studies . more data such as the scale of neutrino mass and more complete knowledge of the neutrino mixing matrix will help greatly . however , we can extract some generic features for this kind of models as follow : 1 . @xmath252 and @xmath253(or @xmath254 ) will happen at tree - level from the exchange of kk scalars , photons and @xmath15 bosons . @xmath255 proceeds at the one - loop level and hence is expected to be suppressed compared to the previous modes 3 . violation of lepton universality will occur . the best signal will be to look for the violation in @xmath256 decays @xcite . unfortunately a more quantitative statement about the level of the effect eludes us for now . we have studied and reviewed lfv processes in 5d gauge models that are related to neutrino mass generation or address the flavor problem . specifically we focus on two complete models which generate neutrino masses radiatively . this allows us to see in detail how the two issues can be related . the models are based on @xmath6 and @xmath7 5d unification . they give rise to different neutrino mass patterns @xcite ; thus , it is not surprising that they give different prediction for lfv . the @xmath6 model has a unification scale at @xmath257 and makes essential use of bileptonic scalars . it also contains characteristic doubly charged gauge bosons . the @xmath7 model is a 5d orbifold version of the usual gut . the unification scale is much higher at @xmath258 gev . the important ingredient for lfv and neutrino masses is the @xmath259 higgs representation . the triplet higgs of this model plays the crucial role here . we found that for the @xmath6 model the rare @xmath2 decays are much more enhanced compare to their counterpart @xmath12 decays . among the @xmath38 decays the largest mode is the @xmath260 . even for this mode we expect it to be @xmath261 which is much lower then current experimental reach . the decay modes @xmath262 have a better chance of being observe . this stems from the fact that they are tree level processes induced by the bileptonic gauge bosons or scalars . since they are kk modes they have high masses controlled by the extra dimension compactification radius which is @xmath263 from consistency and unification considerations . an order of magnitude improvement on the current limit will be valuable information on the unknown yukawa couplings . for the orbifold 5d @xmath7 model the muon to electron conversion in nuclei can be within the experimental capability of the proposed experiment at brookhaven national laboratory@xcite . as in the previous model @xmath145 will not be observable . this is very different conventional 4d unificational models . the split fermion model also have the characteristic of @xmath127 and @xmath264 conversion dominating over @xmath146 . we can not be more quantitative due to proliferation of unknown parameters . this model have lepton universality violation which is not present in the previous two models . this can serve as a differentiating tool . it is clear that in order to unravel the physics behind the flavor problem all modes of lfv must be searched for . the usual 4d supersymmetric model will favor @xmath146 where as the 5d models prefer @xmath127 and/or @xmath265 conversion . to this we add lepton universality test as a probe . now we present the gauge fixing scheme used for the 5d electroweak interaction with one bulk higgs doublet . for simplicity the background geometry is @xmath226 . the fifth gamma matrix was chosen to be @xmath266 . the 5d lagrangian is _ 5 & = & -14 f^mnf_mn-14 g^(a),mng^(a)_mn + ( d_m ) ^(d^m ) + [ eqn : l5 ] where f_mn&= & _ m b_n - _ n b_m , + g^(a)_mn&= & _ m a^(a)_n - _ n a^(a)_m + _ 2 ^abca_m^(b ) a_n^(c ) + d_m & = & _ m -i _ 2 a_m^(a ) . - i _ 1 y b_m @xmath267 and @xmath268 stand for the @xmath269 hyper charge and @xmath270 gauge fields respectively . in this convention , we adopt the usual conventions : @xmath272 , @xmath273 and @xmath274 , or @xmath275 ) , where @xmath276 and @xmath277 . @xmath278 are introduced to simplify the notation . the symmetry breaking pattern is same as in the usual 4d sm . the bulk higgs doublet acquires a nonzero vev after ssb , =( c h^+ + + h^0 ) , h^0=^0+i^0 . the generalized linear @xmath246 gauge fixing is introduced @xcite(other schemes can be found in @xcite ) , _ gf= & - & ( _ p^+_y p^y ) ^2 + & - & |_w^+ , + |^2 + & - & ( _ z^+)^2 . to remove the mixing between gauge bosons and higgs . therefore , the goldstone bosons and the physical scalars can be easily identified : g^_n & = & p^y_n , m_^(n)= n / r + g^0_n & = & , s^0_n= + & & m_z^(n)= , s^z_n= m_z / m_z^(n ) + g^_n&= & , h^_n= + & & m_w^(n)= , s^z_n= m_w / m_w^(n ) + h^0_n & = & ^0_n , m_h^(n)= where @xmath279 and @xmath280 are the kk goldstone bosons , @xmath281 is the physical kk pseudo scalar , and @xmath282 are the physical kk scalars . the usual @xmath246 gauge can be extended to the 5d @xmath226 model with little modification , like @xmath283 and so on . the goldstone bosons are mainly constituted by the fifth gauge components with a small fraction of kk higgs bosons mixed interaction . on the other hand the goldstone bosons couple to brane fermions through their higgs components . in contrast the physical scalars are mainly composed of kk higgs plus small amount of the fifth components of gauge fields . this scheme can also be applied to the models built on the @xmath78 orbifold with little modification . the method can be easily extended to the cases with multi scalars . taking a 5d two higgs doublets model(2hdm ) as an example , with vevs @xmath284 , @xmath285 and @xmath286 , the physical charged scalars and pseudoscalars are h^= _ 1 ^ - _ 2^ , a^0 = im_1 ^ 0 - i m _ 2 ^ 0 just like the usual 4d 2hdm . the only difference is that the orthogonal linear combinations @xmath287 and @xmath288 will mix with the fifth components of gauge fields to form the real goldstone bosons : g^0_n & = & ^0_n v^0_n- ^0_n a^0_n , ^0_n= m_0 / , + g^_n & = & ^_n v^_n i ^_n g^_n , ^_n= m_v / . y. kuno and y. okada , rev . mod . phys . * 73 * , 151 ( 2001 ) [ arxiv : hep - ph/9909265 ] and references therein . s. t. petcov , phys . b * 68 * , 365 ( 1977 ) . k. huitu , j. maalampi , m. raidal and a. santamaria , phys . b * 430 * , 355 ( 1998 ) [ arxiv : hep - ph/9712249 ] . + t. s. kosmas , s. kovalenko and i. schmidt , phys . b * 511 * , 203 ( 2001 ) [ arxiv : hep - ph/0102101 ] . g. feinberg and s. weinberg , phys . lett . * 3 * , 527 ( 1959 ) . s. weinberg , phys . rev . d * 2 * , 1962 ( 1975 ) l.j . hall and y. normura , phys . b * 532 * 111 ( 2002 ) + t. li and w. liao , _ ibid _ * 545 * 147 ( 2002 ) + s. dimopoulos , d.e . kaplan , and n. weiner , _ ibid _ * 534 * 124 ( 2002 ) + i. antoniadis and k. benakli , _ ibid _ * 326 * 69 ( 1994 ) n. arkani - hamed and m. schmaltz , phys . d * 61 * , 033005 ( 2000 ) [ arxiv : hep - ph/9903417 ] ; + n. arkani - hamed , y. grossman and m. schmaltz , phys . d * 61 * , 115004 ( 2000 ) [ arxiv : hep - ph/9909411 ] . e. a. mirabelli and m. schmaltz , phys . d * 61 * , 113011 ( 2000 ) [ arxiv : hep - ph/9912265 ] . g. c. branco , a. de gouvea and m. n. rebelo , phys . b * 506 * , 115 ( 2001 ) [ arxiv : hep - ph/0012289 ] . w. f. chang and j. n. ng , jhep * 0212 * , 077 ( 2002 ) [ arxiv : hep - ph/0210414 ] . w. f. chang , i. l. ho and j. n. ng , phys . d * 66 * , 076004 ( 2002 ) [ arxiv : hep - ph/0203212 ] . j. l. popp [ meco collaboration ] , nucl . instrum . a * 472 * , 354 ( 2000 ) [ arxiv : hep - ex/0101017 ] . d. m. ghilencea , s. groot nibbelink and h. p. nilles , nucl . b * 619 * , 385 ( 2001 ) [ arxiv : hep - th/0108184 ] ; + a. muck , a. pilaftsis and r. ruckl , phys . d * 65 * , 085037 ( 2002 ) [ arxiv : hep - ph/0110391 ] . for example , k. r. dienes , e. dudas and t. gherghetta , nucl . b * 537 * , 47 ( 1999 ) [ arxiv : hep - ph/9806292 ] ; + j. papavassiliou and a. santamaria , phys . d * 63 * , 125014 ( 2001 ) [ arxiv : hep - ph/0102019 ] ;

three dimensional continuous and discrete fourier - like transforms , based on the three simple and four semisimple compact lie groups of rank 3 , are presented . for each simple lie group , there are three families of special functions (- ,- , and-functions ) on which the transforms are built . pertinent properties of the functions are described in detail , such as their orthogonality within each family , when integrated over a finite region of the 3-dimensional euclidean space ( continuous orthogonality ) , as well as when summed up over a lattice grid ( discrete orthogonality ) . the positive integer sets up the density of the lattice containing . the expansion of functions given either on or on is the paper s main focus . centre de recherches mathmatiques , universit de montral , c.p.6128-centre ville , montral , h3c3j7 , qubec , canada ; [email protected] + institute of mathematics of nas of ukraine , 3 , tereshchenkivska str , kyiv-4 , 04216 , ukraine ; [email protected]

0805.3731

new @xmath9-dimensional @xmath0- and @xmath1- and @xmath2-transforms were recently described in @xcite . each transform is based on a compact semisimple lie group of rank @xmath9 and comes in three versions : analogs of fourier series , fourier integrals , and fourier transforms on an @xmath9-dimensional lattice . they are named @xmath0- and @xmath1- and @xmath2-transforms @xcite in recognition of the fact that they can be understood as generalizations of the one dimensional cosine , sine , and exponential fourier transform . the aim of this paper is to set the grounds for the 3-dimensional exploitation of transforms described here as continuous transforms in a finite region @xmath3 of the 3-dimensional euclidean space @xmath10 , and also as discrete transforms of functions given on a lattice grid of points @xmath4 of any density in ready - to - use form . a positive integer @xmath5 specifies the density . in some cases , the density of the grid is more flexible as it is dictated by not one , but two or three positive intergers . the grid could thus be made denser along certain axes . the symmetry of the lattice is dictated by the shape of @xmath3 , or equivalently , by the choice of the lie group . there are seven compact semisimple lie groups of rank 3 : @xmath11 throughout the paper we identify these cases by symbols that are often used for their respective lie algebras : @xmath12 the immediate motivation for this paper is our anticipation of the extensive use of the transforms given the need for processing the rapidly increasing amount of 3d digital data gathered today . in 2d , our group transforms offered only in some cases more than a marginal advantage , having emerged when satisfactory practical methods had already been developed and adequately implemented . so far , practical use of the functions in 2d rested on the fact that the continuous extension of the transformed lattice data displayed remarkably smooth interpolation between lattice points @xcite ( see also references therein ) . special functions , which serve as the kernel of our transform ( we call them @xmath0- , @xmath1- , and @xmath2-functions or orbit functions ) , have simple symmetry property under the action of the corresponding affine weyl group . the affine group contains as a subgroup the group of translations in @xmath13 , which underlies the common fourier transform . this is the primary reason for the superior performance of our transforms , although detailed comparisons , rather than examples , will have to provide quantitative content to substantiate such a claim . other properties of the @xmath0- , @xmath1- , and @xmath2-functions are not less important . within each family , functions are described in a uniform way for semisimple lie groups of any type and rank . in this work , we illustrate this uniformity by considering all seven rank 3 group cases in parallel . the price to pay for the uniformity of methods is having to work with non - orthogonal bases which are not normalized . the functions are defined in @xmath13 and have continuous derivatives of all degrees . their orthogonality , when integrated over the finite region @xmath3 appropriate for each lie group , was shown in @xcite . the discrete orthogonality of @xmath0-functions in @xmath6 has already been described in @xcite and extensively used ( see for example @xcite and references therein ) . the completeness of these systems of functions directly follows from the completeness of the system of exponential functions . a laplace operator for each lie group is given in a different set of coordinates . the @xmath0- and @xmath1-functions are its eigenfunctions with known eigenvalues . on the boundary of @xmath3 , the @xmath0-functions have a vanishing normal derivative , while @xmath1-functions reach zero at the boundary . the functions have a number of other useful properties , which can be found in @xcite . for example , the decomposition of their products into sums , the splitting of functions into as many mutually exclusive congruence classes as is the order of the center of the lie group , etc@xmath14 . a different but valid viewpoint on some of the special functions presented here , namely , functions symmetrized by the summation of constituent functions over a finite group @xcite , may turn out to be rather useful . the finite group , in the case of @xmath0- and @xmath1-functions , is the weyl group of the corresponding semisimple lie group . in the case of @xmath2-functions , it is the even subgroup of the weyl group . the weyl group of @xmath15 is isomorphic to the group @xmath16 of the permutation of @xmath9 elements . this led to the recent implementations in @xcite , where instead of the weyl group of @xmath15 , the @xmath16 group is used , and variables are given relative to an orthonormal system of coordinates . furthermore , the even subgroup of @xmath16 is the alternating group . related transforms were introduced most recently in @xcite . the paper is organized as follows . in section [ sec_properties_definitions ] , necessary definitions and properties of lie groups and algebras are given and discussed . semisimple lie groups of rank 3 are considered in detail in section [ sec_algebras ] . for each of these groups , we lay down the information necessary to construct and use their orbit functions for 3d continuous and discrete transforms . section [ sec_orbit - funcs ] is devoted to @xmath0- , @xmath1- and @xmath2- orbit functions and to their pertinent properties . continuous and discrete orbit - function transforms are presented in section [ sec_c - s - e - transforms ] . some problems and possible applications arising in connection with orbit functions are formulated in the conclusion . an example of the application of orbit function transforms in the case of the group @xmath17 is given at the end of the paper . the notion of orbit function of @xmath9 variables depends essentially on the underlying semisimple lie group of rank @xmath9 . this section is intended to recall some of the standard properties of semisimple lie groups / lie algebras in general , and particularly those of rank 3 , as well as properties of related weyl groups . we also fix notation and terminology . additional information about such lie groups can be found for example in @xcite . let @xmath13 be the real euclidean space spanned by the simple roots of a simple lie group @xmath18 ( equivalently , lie algebra ) . the basis of the simple roots is hereafter referred to as the @xmath19-basis . an @xmath19-basis is not orthogonal and comprises simple roots of at most two different lengths . if a semisimple @xmath18 is not simple , the @xmath19-bases of its simple constituents are pairwise orthogonal . for important practical ( i.e. computational ) reasons , it is advantageous to introduce also the basis of fundamental weights , hereafter referred to as the @xmath20-basis . moreover , for lie groups with simple roots of two different lengths , it is useful to introduce bases dual to @xmath19- and @xmath20-bases , denoted here as @xmath21- and @xmath22-bases respectively . occasionally it is also useful to work with the orthonormal basis @xmath23 of @xmath13 . each subsection contains an explicit elaboration of these bases for the groups we consider . the cartan matrix @xmath0 of @xmath18 provides , in principle , all of the information needed about @xmath18 . it is an @xmath24-matrix . in particular , it provides the relation between @xmath19-and @xmath20-bases : @xmath25 where @xmath26 here @xmath27 denotes the inner product in @xmath13 . the length of the long simple roots is determined by an additional convention @xmath28 the dual bases are fixed by the relations @xmath29 where @xmath30 is the kronecker delta . the root lattice @xmath31 and the weight lattice @xmath32 of @xmath18 are formed by all integer linear combinations of the @xmath19-basis and @xmath20-basis , @xmath33 in general , @xmath34 , but in rank 3 lie groups the equalities do not occur . similarly , we can introduce the dual lattices @xmath35 and @xmath36 . in the weight lattice @xmath32 , we define the cone of dominant weights @xmath37 and its subset of strictly dominant weights @xmath38 @xmath39 for any simple lie group @xmath18 there is a unique highest root @xmath40 @xmath41 coefficients @xmath42 and @xmath43 , @xmath44 , are natural numbers , referred to as marks and comarks respectively . they are well known for all simple lie groups ( see for example @xcite ) . the weyl group @xmath45 of a semisimple lie group @xmath18 is the finite group generated in @xmath13 by reflections in @xmath46 dimensional mirrors ( hyperplanes ) orthogonal to the simple roots of @xmath18 and containing the origin of @xmath13 . for a simple root @xmath47 , @xmath44 the corresponding reflection @xmath48 is given by @xmath49 there is a general method for building weyl group orbits . in section [ sec_algebras ] , we limit ourselves to recording the result of its application for all lie groups of rank @xmath50 . it is assumed that we have fixed a semisimple lie group of rank 3 and that we consider its weight lattice @xmath32 . a @xmath51-orbit can be generated from any point @xmath52 , but in this paper we are almost always interested in @xmath51-orbits of points in @xmath32 . it is convenient to specify the orbit by its unique point @xmath53 with positive integer coordinates @xmath54 , @xmath55 , @xmath56 . we denote such a generic orbit by @xmath57 . every group contains the trivial one point orbit @xmath58 . the number of points of a generic orbit @xmath59 is equal to the order @xmath60 of the weyl group . consider the reflection @xmath61 with respect to the hyperplane containing the origin and orthogonal to the highest root @xmath40 , see ( [ highest_root ] ) @xmath62 we extend the set of @xmath9 reflections @xmath48 , given in ( [ reflection_alpha ] ) , generating the weyl group @xmath51 , by one reflection @xmath63 @xmath64 the resulting group transformations of @xmath13 , generated by @xmath65 reflections @xmath66 , is referred to as the affine weyl group @xmath67 . the order of @xmath67 is infinite . the fundamental region @xmath68 for any @xmath69 is the convex hull of the vertices @xmath70 , where @xmath43 , @xmath44 are comarks from ( [ highest_root ] ) , or equivalently , @xmath71 repeated reflections of @xmath72 in its @xmath46-dimensional sides results in tiling the entire space @xmath13 by copies of @xmath3 . we define grid @xmath4 , depending on an arbitrary natural number @xmath5 as given in ( [ highest_root ] ) , @xmath73 the number of points of the grid @xmath6 is denoted by @xmath74 , and the volume of the fundamental region @xmath3 with respect to the euclidian measure is denoted by @xmath75 . in case @xmath76 , where @xmath77 and @xmath78 are simple , the fundamental region of @xmath18 is the cartesian product of the fundamental regions of @xmath77 and @xmath78 . the same holds for grids @xmath6 , but for each simple constituent of @xmath18 , the numbers @xmath79 nd @xmath80 could be chosen independently . in the case of two different numbers @xmath79 and @xmath80 , such that , the corresponding grids are related as follows @xmath81 . elements of the subgroup @xmath83 are formed by an even number of reflections which generate @xmath51 . each @xmath84 and @xmath85function is built on a single weyl group orbit of point @xmath86 , and each @xmath87function is built on an orbit of the even subgroup of the weyl group , defined as follows . consider a subgroup of w generated by an even number of reflections @xmath88 @xmath89 is a normal subgroup of the index 2 of the weyl group , i.e. @xmath90 . let us denote the orbit of point @xmath86 with respect to the action of @xmath89 by @xmath91 , and the size of this orbit by @xmath92 , then @xmath93 and one of the following relations holds true @xmath94 , when @xmath95 or @xmath96 , when @xmath97 . note that each orbit of the even subgroup @xmath89 contains exactly one point from @xmath98 , @xmath99 . similarly , we define the even affine weyl subgroup @xmath100 and its fundamental region @xmath101 where @xmath3 is a fundamental region of @xmath67 ( [ fund_region_fm ] ) and @xmath102 . the same formula holds true for the grid on the fundamental region @xmath103 @xmath104 as it follows from ( [ fund_region_fem ] ) , the fundamental region of @xmath89 is not unique , hence it can be chosen in such a way as to be convenient for a given application . [ ex_a1 ] consider the rank one compact simple lie group @xmath105 ( the corresponding lie algebra is @xmath106 ) . the cartan matrix in this case is the @xmath107 matrix @xmath108 . the root system consists of two roots @xmath109 . the root and weight lattices are formed by integer multiples of the simple root and integer multiples of the fundamental weight @xmath20 . @xmath110 therefore @xmath111 . the weyl group @xmath51 of @xmath106 is of order 2 . it is a reflection group acting in @xmath112 . we have @xmath113 and @xmath114 . consequently , a weyl group orbit containing the point @xmath115 , also contains the point @xmath116 , but the orbit of the even subgroup of @xmath51 consists of a single point , either @xmath117 or @xmath116 . the fundamental region @xmath3 is the segment with endpoints @xmath118 . the grid @xmath4 is fixed by the positive integer @xmath5 and it consists of @xmath119 points @xmath120 in this section , we provide specific information about compact semisimple lie groups of rank 3 , namely @xmath17 , @xmath121 , @xmath122 , @xmath123 , and the simple groups @xmath124 , @xmath125 , and @xmath126 . for each of the groups , we lay down all of the information necessary to construct and use their orbit functions for 3d continuous and discrete transforms . note , that only the weight lattices of the lie groups @xmath17 , @xmath121 , @xmath125 and @xmath124 display cubic symmetries . below we present the dynkin diagram for each semisimple lie group of rank 3 . on these diagrams , long and short simple roots @xmath19 are respectively denoted by unfilled and filled circles and comarks are presented over the circles . of the lie algebra @xmath128 ; @xmath117 , @xmath129 and @xmath130 indicate respectively the orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 ; b ) the grid @xmath134 . ] this case is a straightforward concatenation of three copies of @xmath106 ( see example [ ex_a1 ] ) . the dynkin diagram and cartan matrix with its inverse are the following , @xmath135 all simple roots of the same length equal to @xmath136 . the bases of simple roots and fundamental weights are thus related by @xmath137 these bases are easily written in the orthonormal basis @xmath138 @xmath139 the highest roots for the simple subgroups @xmath106 are given by the formulas @xmath140 the fundamental region is a cube with vertices , see fig . [ fig_a1a1a1]a @xmath141 the volume of the fundamental region is given by the formula @xmath142 is fixed by the independent choice of three positive integers @xmath143 and @xmath144 , and consists of all the points @xmath145 the grid is cubic if @xmath146 , otherwise it is rectangular . the freedom to use unequal values of @xmath143 and @xmath144 may prove rather useful in the analysis of data given on rectangular , but not cubic grids . the number of points in the grid @xmath147 equals to @xmath148 consider the case @xmath149 , @xmath150 . there are 16 points of @xmath151 . explicitly , we have the following sets of integers @xmath152 $ ] and the corresponding grid points in the @xmath153-basis ( in this case the basis is orthogonal ) , @xmath154 : @xmath155=(0,0,0),&\ ; [ 0,0,1]=(0,0,\tfrac13),&\ ; [ 0,0,2]=(0,0,\tfrac23),&\ ; [ 0,0,3]=(0,0,1 ) , \\[1pt ] [ 0,1,0]=(0,1,0),&\ ; [ 0,1,1]=(0,1,\tfrac13),&\ ; [ 0,1,2]=(0,1,\tfrac23),&\ ; [ 0,1,3]=(0,1,1 ) , \\[1pt ] [ 1,0,0]=(1,0,0),&\ ; [ 1,0,1]=(1,0,\tfrac13),&\ ; [ 1,0,2]=(1,0,\tfrac23),&\ ; [ 1,0,3]=(1,0,1 ) , \\[1pt ] [ 1,1,0]=(1,1,0),&\ ; [ 1,1,1]=(1,1,\tfrac13),&\ ; [ 1,1,2]=(1,1,\tfrac23),&\ ; [ 1,1,3]=(1,1,1 ) . \end{array}\end{gathered}\ ] ] next , consider the case @xmath156 , @xmath157 , @xmath150 , see fig . [ fig_a1a1a1]b . the 16 points of @xmath151 form a part of @xmath134 . note that they correspond to another set of @xmath152 $ ] . more precisely , @xmath158 is the subset of @xmath159 with an even value of @xmath160 . thus @xmath134 contains an additional eight points : @xmath161=(\tfrac12,0,0),&\ ; [ 1,0,1]=(\tfrac12,0,\tfrac13),&\ ; [ 1,0,2]=(\tfrac12,0,\tfrac23),&\ ; [ 1,0,3]=(\tfrac12,0,1 ) , \\[1pt ] [ 1,1,0]=(\tfrac12,1,0),&\ ; [ 1,1,1]=(\tfrac12,1,\tfrac13),&\ ; [ 1,1,2]=(\tfrac12,1,\tfrac23),&\ ; [ 1,1,3]=(\tfrac12,1,1 ) . \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , always consists of the eight points @xmath164 orbit sizes for arbitrary points are given by the relations @xmath165 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath167 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 and the fundamental region of @xmath168 lies in the plane @xmath169 ; b ) the grid @xmath170 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath171 hence all the simple roots of the same length equal to @xmath136 . the bases of simple roots and fundamental weights are thus related by @xmath172 in order to visualize the implied geometry , it is useful to represent the @xmath20 and @xmath19 bases in the orthonormal basis . we have @xmath173 \alpha_2=(0,\ 1,\ -1)=e_2-e_3,&\quad \omega_2=(\tfrac13,-\tfrac13,-\tfrac23)=\tfrac13e_1+\tfrac13e_2-\tfrac23e_3={{\check\omega}}_2 , \\[1 pt ] \alpha_3=\tfrac{\sqrt{2}}{\sqrt{3}}(1,1,1)=\tfrac{\sqrt{2}}{\sqrt{3}}(e_1+e_2+e_3),&\quad \omega_3=\tfrac{1}{\sqrt6}(1,\ 1,\ 1)=\tfrac{1}{\sqrt6}(e_1+e_2+e_3)={{\check\omega}}_3 . \end{array}\end{gathered}\ ] ] the highest roots for the simple subgroups @xmath168 and @xmath106 are given by the formulas @xmath174 the fundamental region @xmath3 is a cylinder ( see fig . [ fig_a1a1a1]a ) with an equilateral triangle as its base and @xmath175 as its height . its vertices are @xmath176 the volume of the fundamental region is given by the formula @xmath177 the grid @xmath178 is fixed by the independent choice of the two positive integers @xmath5 and @xmath179 . it consists of all the points @xmath180 the number of points in the grid @xmath181 equals to @xmath182 consider the case @xmath156 , @xmath183 , see fig . [ fig_a2a1]b . there exist 24 points of @xmath184 @xmath185 $ ] and the corresponding grid points @xmath186 in @xmath153-basis : @xmath187=(0,1,0),&\ ; [ 0,2,3]=(0,1,1),&\ ; [ 0,2,2]=(0,1,\tfrac23),&\ ; [ 0,2,1]=(0,1,\tfrac13 ) , \\[1pt ] [ 2,0,0]=(1,0,0),&\ ; [ 2,0,3]=(1,0,1),&\ ; [ 2,0,2]=(1,0,\tfrac23),&\ ; [ 2,0,1]=(1,0,\tfrac13 ) , \\[1pt ] [ 0,0,0]=(0,0,0),&\ ; [ 0,0,3]=(0,0,1),&\ ; [ 0,0,2]=(0,0,\tfrac23),&\ ; [ 0,0,1]=(0,0,\tfrac13 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac12,0),&\ ; [ 0,1,3]=(0,\tfrac12,1),&\ ; [ 0,1,2]=(0,\tfrac12,\tfrac23),&\ ; [ 0,1,1]=(0,\tfrac12,\tfrac13 ) , \\[1pt ] [ 1,0,0]=(\tfrac12,0,0),&\ ; [ 1,0,3]=(\tfrac12,0,1),&\ ; [ 1,0,2]=(\tfrac12,0,\tfrac23),&\ ; [ 1,0,1]=(\tfrac12,0,\tfrac13 ) , \\[1pt ] [ 1,1,0]=(\tfrac12,\tfrac12,0),&\ ; [ 1,1,3]=(\tfrac12,\tfrac12,1),&\ ; [ 1,1,2]=(\tfrac12,\tfrac12,\tfrac23),&\ ; [ 1,1,1]=(\tfrac12,\tfrac12,\tfrac13 ) . \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of twelve points @xmath188 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath191 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 and the fundamental region of @xmath192 is the isosceles right - angled triangle in the plane @xmath193 , its right - angle is at the point @xmath194 ; b ) the grid @xmath195 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath196 hence @xmath197 is the shorter of the simple roots . the relative lengths of the simple roots are set as @xmath198 and @xmath199 . the bases of simple roots and fundamental weights are thus related by @xmath200 in the orthonormal basis these bases have the form @xmath201 the highest roots for the simple subgroups @xmath192 and @xmath106 are given by the formulas @xmath202 the fundamental region @xmath203 is a cylinder ( see fig . [ fig_c2a1]a ) with a triangular base . its vertices are @xmath204 the volume of the fundamental region is given by the formula @xmath205 the grid @xmath206 is fixed by the independent choice of the two positive integers @xmath5 and @xmath179 . it consists of all the points @xmath207 the number of points in the grid @xmath208 equals to @xmath209 + 1\right ) \left(m+1-\left[\tfrac{m}{2}\right]\right ) ( m'+1),\end{gathered}\ ] ] where @xmath210 $ ] denotes the integer part of a number . [ ex_grid_c2a1 ] consider the case @xmath211 , @xmath212 . there exist 18 points of @xmath195 , see fig . [ fig_c2a1]b @xmath185 $ ] and the corresponding grid points in the @xmath153-basis , @xmath186 : @xmath213=(\tfrac13,0,0),&\quad [ 1,0,1]=(\tfrac13,0,\tfrac12),&\quad [ 1,0,2]=(\tfrac13,0,2 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac13,0),&\quad [ 0,1,1]=(0,\tfrac13,\tfrac12),&\quad [ 0,1,2]=(0,\tfrac13,2 ) , \\[1pt ] [ 0,2,0]=(0,\tfrac23,0),&\quad [ 0,2,1]=(0,\tfrac23,\tfrac12),&\quad [ 0,2,2]=(0,\tfrac23,2 ) , \\[1pt ] [ 1,1,0]=(\tfrac13,\tfrac13,0),&\quad [ 1,1,1]=(\tfrac13,\tfrac13,\tfrac12),&\quad [ 1,1,2]=(\tfrac13,\tfrac13,2 ) , \\[1pt ] [ 0,0,0]=(0,0,0),&\quad [ 0,0,1]=(0,0,\tfrac12),&\quad [ 0,0,2]=(0,0,2 ) , \\[1pt ] [ 0,3,0]=(0,1,0),&\quad [ 0,3,1]=(0,1,\tfrac12),&\quad [ 0,3,2]=(0,1,2 ) . \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of the following set of points @xmath214 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath216 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 and the fundamental region of @xmath78 is the half of the equilateral triangle in the plane @xmath193 ; b ) the grid @xmath217 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath218 hence @xmath219 is the shorter of the simple roots . the relative lengths of the simple roots are set as @xmath220 and @xmath221 . the bases of simple roots and fundamental weights are thus related by @xmath222 relative to the orthonormal basis , we have @xmath223 the highest roots for the simple subgroups are given by the formulas @xmath224 the fundamental region @xmath3 is a cylinder ( see fig . [ fig_g2a1]a ) with a triangular base . its vertices are @xmath225 the volume of the fundamental region is given by the formula @xmath226\quad \text{denotes the vector product}.\end{gathered}\ ] ] the grid @xmath206 is fixed by the independent choice of two positive integers @xmath5 and @xmath179 . it consists of all the points @xmath227 the number of points in the grid @xmath228 equals to @xmath229 + 1+\sum_{i=0}^{\frac m3}\left[\tfrac{m-3i}{2}\right]\right)(m'+1),\end{gathered}\ ] ] where @xmath210 $ ] is the integer part of a number . [ ex_grid_g2a1 ] consider the case @xmath230 , @xmath183 . there are 16 points of @xmath231 ( see fig . [ fig_g2a1]b ) @xmath232 + 1+\sum_{i=0}^{\left[\frac{4}{3}\right]}\left[\tfrac{4 - 3i}{2}\right ] \right ) ( 3 + 1)=(1 + 1+(2 + 0))\cdot 4=16.\end{gathered}\ ] ] explicitly , we have the following sets of integers @xmath233 $ ] and the corresponding grid points in the @xmath153-basis , @xmath186 : @xmath155=(0,0,0),&\ ; [ 0,0,1]=(0,0,\tfrac13),&\ ; [ 0,0,2]=(0,0,\tfrac23),&\ ; [ 0,0,3]=(0,0,1 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac14,0),&\ ; [ 0,1,1]=(0,\tfrac14,\tfrac13),&\ ; [ 0,1,2]=(0,\tfrac14,\tfrac23),&\ ; [ 0,1,3]=(0,\tfrac14,1 ) , \\[1pt ] [ 1,0,0]=(\tfrac14,0,0),&\ ; [ 1,0,1]=(\tfrac14,0,\tfrac13),&\ ; [ 1,0,2]=(\tfrac14,0,\tfrac23),&\ ; [ 1,0,3]=(\tfrac14,0,1 ) , \\[1pt ] [ 2,0,0]=(\tfrac12,0,0),&\ ; [ 2,0,1]=(\tfrac12,0,\tfrac13),&\ ; [ 2,0,2]=(\tfrac12,0,\tfrac23),&\ ; [ 2,0,3]=(\tfrac12,0,1 ) . \end{array}\end{gathered}\ ] ] next , consider the case @xmath211 , @xmath212 . @xmath234 consists of the following points : @xmath235=(0,0,0),&\quad [ 0,0,1]=(0,0,\tfrac12),&\quad [ 0,0,2]=(0,0,1 ) , \\[1pt ] [ 1,0,0]=(\tfrac13,0,0),&\quad [ 1,0,1]=(\tfrac13,0,\tfrac12),&\quad [ 1,0,2]=(\tfrac13,0,1 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac13,0),&\quad [ 0,1,1]=(0,\tfrac13,\tfrac12),&\quad [ 0,1,2]=(0,\tfrac13,1 ) . \end{array}\end{gathered}\ ] ] in this case , none of the nine points of @xmath234 coincides with a point of @xmath217 . this is due to the fact that lattice densities @xmath211 and @xmath212 do not correspondingly divide the densities @xmath230 and @xmath183 . the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of 24 points @xmath236 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath238 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 , and the fundamental region of @xmath238 is the tetrahedron ; b ) the grid @xmath239 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath240 hence , all simple roots of the same length equal to @xmath136 . the bases of simple roots and fundamental weights are thus related by @xmath241 relative to the orthonormal basis , we have @xmath242 the highest root @xmath40 is given by the formula @xmath243 the fundamental region is a pyramid ( see fig . [ fig_a3]a ) with vertices @xmath244 the volume of the fundamental region is given by the formula @xmath245 where @xmath246 $ ] denotes the vector product . the grid @xmath247 is fixed by the choice of one positive integer @xmath5 . it consists of all the points @xmath248 the number of points in the grid @xmath249 equals to @xmath250 consider the case @xmath211 . there are 20 points of @xmath239 ( see fig . [ fig_a3]b ) . explicitly , we have the following sets of integers @xmath251 $ ] and the corresponding grid points in the @xmath153-basis @xmath252 : @xmath155=(0,0,0),&\ ; [ 0,0,3]=(0,0,1),&\ ; [ 0,3,0]=(0,1,0),&\ ; [ 3,0,0]=(1,0,0 ) , \\[1pt ] [ 2,0,1]=(\tfrac23,0,\tfrac13),&\ ; [ 2,1,0]=(\tfrac23,\tfrac13,0),&\ ; [ 1,0,2]=(\tfrac13,0,\tfrac23),&\ ; [ 0,1,2]=(0,\tfrac13,\tfrac23 ) , \\[1pt ] [ 1,2,0]=(\tfrac13,\tfrac23,0),&\ ; [ 0,2,1]=(0,\tfrac23,\tfrac13),&\ ; [ 1,0,0]=(\tfrac13,0,0),&\ ; [ 0,0,1]=(0,0,\tfrac13 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac13,0),&\ ; [ 1,1,0]=(\tfrac13,\tfrac13,0),&\ ; [ 1,0,1]=(\tfrac13,0,\tfrac13),&\ ; [ 0,1,1]=(0,\tfrac13,\tfrac13 ) , \\[1pt ] [ 1,1,1]=(\tfrac13,\tfrac13,\tfrac13),&\ ; [ 2,0,0]=(\tfrac23,0,0),&\ ; [ 0,0,2]=(0,0,\tfrac23),&\ ; [ 0,2,0]=(0,\tfrac23,0 ) . \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of 24 points @xmath253 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath255 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 and the face defined by vertices @xmath256 lies in the plane @xmath193 ; b ) the grid @xmath257 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath258 hence @xmath197 is the shorter of the simple roots . the relative lengths of the simple roots are set as @xmath198 and @xmath199 . the bases of simple roots and fundamental weights are thus related by @xmath259 relative to the orthonormal basis , we have @xmath260 the highest root @xmath40 is given by the formula @xmath261 the fundamental region is a pyramid ( see fig . [ fig_a1a1a1]a ) with vertices @xmath262 the volume of the fundamental region is given by the formula @xmath263\quad \text{denotes the vector product}.\end{gathered}\ ] ] the grid @xmath247 is fixed by the choice of an integer @xmath5 . it consists of all the points @xmath264 the number of points in the grid @xmath249 equals to @xmath209\left[\tfrac{m+1}{2}\right]+m+1 -\tfrac{m+2}{2}\left[\tfrac{m}{2}\right]\right ) + \sum_{i=0}^{\left[\frac{m}{2}\right]}i^2.\end{gathered}\ ] ] where @xmath210 $ ] is the integer part of a number . [ ex_grid_b3 ] consider the case @xmath230 , see fig . [ fig_b3]b . there are 14 points of @xmath265 @xmath266\left[\tfrac{4 + 1}{2}\right]+4 + 1 -\tfrac{4 + 2}{2}\left[\tfrac{4}{2}\right]\right ) + \sum_{i=0}^{\left[\frac{4}{2}\right]}i^2= 3(2\cdot2 + 4 + 1 - 3\cdot2)+0 + 1 + 4 = 14.\end{gathered}\ ] ] explicitly , we have the following sets of integers @xmath251 $ ] and the corresponding grid points in the @xmath153-basis , @xmath252 : @xmath155=(0,0,0),&\ ; [ 0,0,1]=(0,0,\tfrac14),&\ ; [ 0,0,2]=(0,0,\tfrac12),&\ ; [ 1,0,1]=(\tfrac14,0,\tfrac14 ) , \\[1pt ] [ 0,2,0]=(0,\tfrac12,0),&\ ; [ 0,1,1]=(0,\tfrac14,\tfrac14),&\ ; [ 0,1,2]=(0,\tfrac14 , \tfrac12),&\ ; [ 1,0,2]=(\tfrac14,0,\tfrac12 ) , \\[1pt ] [ 0,0,4]=(0,0,1),&\ ; [ 0,0,3]=(0,0,\tfrac34),&\ ; [ 1,1,0]=(\tfrac14,\tfrac14,0),&\ ; [ 2,0,0]=(\tfrac12,0,0 ) . \\[1pt ] [ 0,1,0]=(0,\tfrac14,0),&\ ; [ 1,0,0]=(\tfrac14,0,0 ) , \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of 48 points @xmath267 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] of the lie algebra @xmath269 ; @xmath117 , @xmath129 and @xmath130 indicate respectively orthogonal directions of the orthonormal basis @xmath131 , @xmath132 , @xmath133 and the face defined by vertices @xmath270 lies in the plane @xmath193 ; b ) the grid @xmath271 . ] the dynkin diagram and cartan matrix with its inverse are the following , @xmath272 hence @xmath273 is the longer of the simple roots . the relative lengths of the simple roots are set as and . the bases of simple roots and fundamental weights are thus related by @xmath274 in the orthonormal basis these bases have the form @xmath275 the highest root @xmath40 is given by the formula @xmath276 the fundamental region is a pyramid ( see fig . [ fig_a1a1a1]a ) with vertices @xmath277 the volume of the fundamental region is given by the formula @xmath278 where @xmath246 $ ] is the vector product . the grid @xmath247 is fixed by the choice of the integer @xmath5 . it consists of all the points @xmath279 the number of points in the grid @xmath249 equals to @xmath209\left[\tfrac{m+1}{2}\right]+m+1 -\tfrac{m+2}{2}\left[\tfrac{m}{2}\right]\right ) + \sum_{i=0}^{\left[\frac{m}{2}\right]}i^2.\end{gathered}\ ] ] where @xmath210 $ ] is the integer part of a number . [ ex_grid_c3 ] consider the case @xmath230 . there are 14 points of @xmath280 , see fig . [ fig_c3]b . @xmath266\left[\tfrac{4 + 1}{2}\right]+4 + 1 -\tfrac{4 + 2}{2}\left[\tfrac{4}{2}\right]\right ) + \sum_{i=0}^{\left[\frac{4}{2}\right]}i^2=14.\end{gathered}\ ] ] explicitly , we have the following sets of integers @xmath251 $ ] and the corresponding grid points in the @xmath153-basis , @xmath252 : @xmath155=(0,0,0),&\ ; [ 0,0,1]=(0,0,\tfrac14),&\ ; [ 0,0,2]=(0,0,\tfrac12),&\ ; [ 0,0,3]=(0,0,\tfrac34 ) , \\[1pt ] [ 0,1,0]=(0,\tfrac14,0),&\ ; [ 0,1,1]=(0,\tfrac14,\tfrac14),&\ ; [ 0,0,4]=(0,0,1),&\ ; [ 2,0,0]=(\tfrac12,0,0 ) , \\[1pt ] [ 1,0,0]=(\tfrac14,0,0),&\ ; [ 1,0,1]=(1,0,\tfrac13),&\ ; [ 1,0,2]=(1,0,\tfrac23),&\ ; [ 1,1,2]=(1,1,\tfrac23 ) . \\[1pt ] [ 1,1,0]=(1,1,0),&\ ; [ 1,1,1]=(1,1,\tfrac13),&\ ; \end{array}\end{gathered}\ ] ] the weyl group orbit of the generic point @xmath162 , @xmath163 , consists of 48 points @xmath281 orbit sizes for arbitrary points are given by the relations @xmath189 \end{array}\end{gathered}\ ] ] in this section , we define what we mean by @xmath0- , @xmath1- and @xmath2-functions , specified by a given point @xmath282 . we also show some of the properties inherent to those functions . namely , the following properties are of interest , their pairwise orthogonality ( using the appropriate scalar product , an integral over the fundamental region ) , their discrete orthogonality ( again , using a properly defined scalar product , a sum over the discrete grid ) , their product can be represented as a sum , and they are eigenfunctions of the laplace operator . we start with the @xmath0-functions . the @xmath0-function @xmath283 , @xmath284 is defined as @xmath285 where @xmath286 is the weyl group orbit generated from @xmath287 . if in ( [ def_c - function ] ) we restrict ourselves to the orbit of the even subgroup @xmath288 , then we define @xmath289 , @xmath290 @xmath291 the definition of an @xmath1-function @xmath292 , @xmath95 is almost identical , but the sign of each summand is determined by the number of reflections @xmath293 necessary to obtain @xmath294 from @xmath287 @xmath295 of course the same @xmath294 can be obtained by different successions of reflections , but all routes from @xmath287 to @xmath294 will have a length of the same parity , and thus the salient detail given by @xmath293 , in the context of an @xmath1-function , is meaningful and unchanging . for different families of orbit functions , the @xmath287 ( represented in the @xmath20-basis ) are taken from different sets , namely @xmath296 in particular , this implies that , for @xmath1-functions , the number of summands always equals to the size of the weyl group . in the case of @xmath297 ( the coordinates of @xmath117 are rational ) , the @xmath0- , @xmath1- and @xmath2-functions are formed by roots of unity . therefore , there can only be a finite number of possible orbit functions that can take distinct values on the points of @xmath6 , these functions are orthogonal on the grid . the number of pairwise orthogonal orbit functions on @xmath6 coincides with the size of @xmath6 , including the boundary in the case of @xmath0- and @xmath2-functions and excluding the boundary in the case of , @xmath1-functions . note that in the 1-dimensional case , @xmath0- , @xmath1- and @xmath2-functions are respectively a cosine , a sine and an exponential functions up to the constant . all three families of orbit functions are based on semisimple lie groups of finite order , the number of variables coincides with the rank of the corresponding lie algebra . in general , @xmath0- , @xmath1- and @xmath2-functions are the finite sums of exponential functions , therefore they are continuous and have continuous derivatives of all orders in @xmath13 . it is easy to prove that @xmath0- and @xmath1-functions are invariant with respect to the action of both weyl @xmath51 and affine weyl @xmath67 groups ( see e.g. @xcite ) and that @xmath2-functions are invariant with respect to the action of @xmath89 and @xmath298 groups ( see e.g. @xcite ) . therefore it is enough to consider them only on the fundamental domain of their affine weyl symmetry groups . the @xmath1-functions are antisymmetric with respect to @xmath46-dimensional boundary of @xmath3 . hence they are zero on the boundary of @xmath3 . the @xmath0-functions are symmetric with respect to @xmath46-dimensional boundary of @xmath3 . their normal derivative at the boundary is equal to zero ( because the normal derivative of a @xmath0-function is an @xmath1-function ) . a number of other properties of orbit functions are presented in @xcite . here we present the rules necessary for the calculation of the scalar products of the vectors given in the different bases . let vectors @xmath299 and @xmath300 are represented in the @xmath20-basis , then their scalar product is calculated as the matrix product @xmath301 here @xmath302 and @xmath0 is the cartan matrix . if vectors @xmath303 and @xmath304 are represented in the @xmath19-basis , then their scalar product is calculated as the matrix product @xmath305 here @xmath306 and @xmath0 is the cartan matrix . square lengths of the simple roots @xmath307 are always indicated in the dynkin diagrams of the semisimple lie groups , see e.g. @xcite . it is also useful to remind that @xmath308 consider the functions @xmath309 , @xmath310 and @xmath311 and suppose that the continuous variable @xmath117 is given relative to the orthogonal basis . in the case of lie algebra @xmath312 we use orthogonal coordinates @xmath313 and coordinates @xmath314 for @xmath315 , @xmath316 and @xmath317 ( the orthogonal bases for these algebras are well known and can be found e.g. in @xcite ) . the laplace operator in orthogonal coordinates has the form @xmath318 for the algebras @xmath312 , @xmath315 , @xmath316 and @xmath317 , the laplace operator gives the same eigenvalues on every exponential function summand of an orbit function with eigenvalue @xmath319 . hence , the functions @xmath309 , @xmath310 and @xmath311 are eigenfunctions of the laplace operator : @xmath320 now we consider the continuous variable @xmath117 given relative to the @xmath321-basis . let @xmath322 denote the laplace operator , where the differentiation @xmath323 is made with respect to the direction given by @xmath324 . @xmath325 it is known in lie theory that the matrix of scalar products of the simple roots is positive definite , moreover our definition makes matrix @xmath326 symmetric , hence it can be diagonalized and the laplace operator could be transformed to the sum of second derivatives by an appropriate change of variables . thereby , using the results of section [ sec_algebras ] we can write the explicit forms of the laplace operators given in the @xmath321-basis for all semisimple lie algebras of rank 3 @xmath327 for any two squared integrable functions @xmath328 and @xmath329 defined on the fundamental region @xmath330 , we define a continuous scalar product @xmath331 here , integration is carried out with respect to the euclidean measure , the bar means complex conjugation and @xmath332 , where @xmath333 is the fundamental region of either @xmath51 or @xmath89 . any pair of orbit functions from the same family is orthogonal on the corresponding fundamental region with respect to the introduced scalar product ( [ def_cont_scalar_product ] ) , namely @xmath334 where @xmath335 is the kronecker delta , @xmath60 is the size of weyl group , @xmath336 and @xmath337 are the sizes of weyl group orbits , and @xmath75 and @xmath338 are volumes of fundamental regions . all of the necessary information for each semisimple lie algebra of rank 3 can be found in section [ sec_algebras ] . in particular , @xmath339 and @xmath337 is defined by formula ( [ even_subgr_of_w ] ) . proof of the relations ( [ cont_orthog c funcs],[cont_orthog s funcs],[cont_orthog e funcs ] ) follows from the orthogonality of the usual exponential functions and from the fact that a given weight @xmath340 belongs to precisely one orbit function . therefore each family of orbit functions forms an orthogonal basis in the hilbert space of squared integrable functions @xmath341 . hence functions given on @xmath3 can be expanded in terms of linear combinations of @xmath0- , @xmath1- or @xmath2-functions . let us denote the discrete grid of the fundamental region as @xmath6 in the general case , even though in some cases it is determined by more than one positive integer @xmath342 a discrete scalar product of two functions @xmath328 and @xmath329 given on @xmath343 ( including @xmath0- , @xmath1- and @xmath2-functions ) is dependent on this grid and defined by the bilinear form @xmath344 here @xmath345 is the number of points conjugate to @xmath346 on the maximal torus of the lie group . the value of @xmath345 is given by the formula @xmath347 again , as in the continuous case , the @xmath0- , @xmath1- and @xmath2-functions are pairwise orthogonal , i.e. @xmath348 and @xmath349 denote the sizes of the orbits of the weyl group and its even subgroup . @xmath350 is the @xmath51-invariant abelian subgroup of the maximal torus @xmath351 of the simple compact group corresponding to @xmath51 @xmath352 . for different fixed @xmath353 the set of exponential functions @xmath354 , @xmath355 determines continuous and discrete fourier transforms on @xmath13 . in much the same way , the orbit functions ( which are a symmetrized version of exponential functions defined in section [ sec_orbit - funcs ] ) determine an analogue of the fourier transform . in this section , we introduce the essentials of the continuous and discrete @xmath0- , @xmath1- and @xmath2-transforms . the discrete transform can be used for the continuous interpolation of values of a function @xmath356 between its given values on a grid @xmath6 . each continuous function on the fundamental region with continuous derivatives can be expanded as the sum of @xmath0- , @xmath1- or @xmath2-functions . let @xmath356 be a function defined on @xmath3 ( or @xmath103 for @xmath2-functions ) , then it may be written that @xmath357 here @xmath358 denotes the continuous scalar product of ( [ def_cont_scalar_product ] ) . direct and inverse @xmath0- , @xmath1- and @xmath2-transforms of the function @xmath356 are in ( [ cont_trans_c ] ) , ( [ cont_trans_s ] ) and ( [ cont_trans_e ] ) respectively . let @xmath359 be the maximal set of points , such that for any two @xmath360 the condition of discrete orthogonality holds for any of the families of orbit functions in ( ( [ discr_ortog c ] ) , ( [ discr_ortog s ] ) or ( [ discr_ortog e ] ) ) . then we have the following discrete transforms for the function @xmath356 : @xmath361 here @xmath362 denotes the discrete scalar product given of ( [ def_discr_scalar_product ] ) . once the coefficients @xmath363 of the expansions ( [ discr_trans_c ] ) ( [ discr_trans_s ] ) , ( [ discr_trans_e ] ) are calculated , discrete variables @xmath346 in @xmath343 may be replaced by continuous variables @xmath117 in @xmath333 @xmath364 the function @xmath365 smoothly interpolates the values of @xmath366 , @xmath367 . at the points @xmath346 , we have the equality @xmath368 . if we calculate more than one discrete transform on the same grid @xmath343 and use the same set of orbit functions @xmath369 , @xmath370 , then it is reasonable to pre - compute and save the matrix @xmath371 this would save valuable computation time , especially for large @xmath5 , since the coefficients @xmath363 are easily calculated as a matrix product @xmath372 . moreover , the matrix @xmath373 does not depend on the function that is to be expanded into series , therefore it need be calculated only once for each @xmath5 and can be repeatedly used . * each of the transforms described here is based on a compact semisimple lie group of rank 3 . all seven types of such lie groups were considered . our goal was to provide the tools for the expansion of functions of 3 variables given on a bounded region @xmath374 of an euclidean space @xmath10 . the variables can be either continuous or discrete ( lattice grid points ) . the symmetry of the lattice is the weyl group of the lie group . the bounded region @xmath374 has to be scaled to fit into the fundamental region @xmath3 of the corresponding lie group . in case of functions given on a lattice grid , the scaling has to be accompanied with the matching density of the grid points in @xmath3 . fortunately , the formalism admits choosing any density one may need . the scaling resulting in the inclusion @xmath375 is not unique . various options may be considered for specific functions . + shows how additional frames could be added to a film . in much the same way , the continuous deformation of the picture can be proceed and a 3-dimensional image can be created from corresponding 2-dimensional layers or cuts . ] * the uncommon special functions of our transforms are defined for compact semisimple lie groups of any type and rank @xcite . their continuous and discrete orthogonality in @xmath3 is assured @xcite . unlike the translation symmetry required in traditional fourier expansions , the symmetry group of @xmath0- , @xmath1- , and @xmath2-functions is the appropriate affine weyl group , which contains the translations in @xmath10 as a subgroup . * the uniformity of our approach , as to the type of the rank 3 lie group , is illustrated here by considering the seven cases in parallel . the price paid for uniformity is the exploitation of non - orthogonal bases , @xmath19 , @xmath20-bases and their duals whenever necessary . the majority of practically useful digital data usually given on cubic / square lattices with the simplest symmetry group . only more costly experimental installations may use denser lattice arrangements of data collectors . + it should be pointed out that , at least for one type of transform , it is possible to avoid paying the price i.e. of having to work with orthonormal bases . the weyl group of @xmath15 is isomorphic to the permutation group @xmath16 of @xmath9 elements . recently introduced transforms @xcite , based on @xmath16 and on its alternating subgroup , exploit orthonormal bases in @xmath13 , although even there , the corresponding fundamental regions do not have orthogonal adjacent faces , in general . * there is an additional freedom of choice whenever the underlying lie group is not simple . suppose that group is a product of two simple lie groups , @xmath376 . the fundamental region is then the cartesian product of @xmath377 and @xmath378 . + for the expansion of class functions on the product group , we can combine @xmath0-functions on one with @xmath1-functions on the other . similarly , we can combine @xmath0- or @xmath1-functions with @xmath2-functions , enlarging appropriately the fundamental region of the @xmath2-functions . + in much the same way , discretization can proceed differently on @xmath377 then on @xmath378 . the corresponding integers @xmath79 , @xmath80 that fix it can be as different as one desires . thus the density of grid points in @xmath379 may be very different on the two orthogonal components . * a number of other properties of the orbit functions may prove to be useful ( see @xcite and references therein ) . let us point out that each of the three types of functions split into mutually exclusive congruence classes . for a given semisimple lie group , the number of congruence classes equals the order of the center of the lie group . * the possibility to introduce the @xmath0- , @xmath1- , and @xmath2-functions by summation over a finite noncrystallographic coxeter groups instead of the weyl group of a lie group appears to be rather interesting . in 3d there is just one such group @xmath380 , the icosahedral group of order 120 . most of the properties carry over to this case in a simple straightforward way . the exception is the orthogonality , continuous or discrete . there is an analog of the fundamental region , but no lattice . its role , perhaps , should be played by some quasicrystal ? as an example , we chose to interpolate a discretization of a known function , namely the gaussian function shown in equation ( [ eq_gauss ] ) @xmath381 where @xmath382 is a fixed point inside the fundamental region . the first test to be undertaken is to sample the function in the points of the grid @xmath6 for several values of @xmath5 . the continuous extension of @xmath383 , calculated from points of @xmath6 , is @xmath384 for different values of @xmath5 , fig . [ figerror]a represents the error , as defined by the integral @xmath385 the integral ( 24 ) was calculated using a simple monte carlo method with 10000 randomly chosen points . we made sure that the granularity of the randomly generated points was much higher than that of the grid @xmath6 . ) of the @xmath0-transform ( [ eq_c - transform ] ) as function of @xmath5 ; b ) the standard deviation of the difference between the gaussian function ( [ eq_gauss ] ) and its interpolation ( [ eq_c - transform ] ) as function of @xmath5 . ] fig . [ figerror]b shows the standard deviation of the sample set . it decreases in much the same way the error did in function of @xmath5 . as a means to show what takes place visually , we give a plot of @xmath386 as compared to @xmath387 on the parametric line given by the transformation @xmath388 fig . [ cut1d]a and fig . [ cut1d]b show the comparison between @xmath389 and @xmath390 for @xmath391 $ ] , with @xmath230 and @xmath392 , respectively . as seen when comparing figs . [ cut1d]a and [ cut1d]b , the degree of oscillation increases as @xmath5 increases , but the total error decreases . this is exactly as one would expect . a two dimensional cut from the graphs of @xmath387 is presented in fig . [ gauss3d ] . [ surface]a and [ surface]b are two dimensional cuts from the graphs of @xmath386 with @xmath230 and @xmath392 , respectively . in the case of @xmath396 ( as well as in the other cases , albeit less trivially ) , the set of pairwise orthogonal functions is not unique , i.e. one can scale all @xmath86 by integer multiplier @xmath397 with @xmath398 . this leads to a shifted system of orbit functions , that take the exact same values on the discrete grid @xmath6 , but do in fact differ in their continuous behavior . such a shifted system of functions can be understood as higher harmonics of the original functions on the grid @xmath6 . examples with original and shifted systems of orbit functions follow . [ harmonics]a shows @xmath387 and @xmath386 with @xmath399 , for @xmath386 computed using the fundamental set of @xmath0-functions and fig . [ harmonics]b shows the same graph , but for a @xmath386 that was computed using the set of @xmath0-functions shifted by a factor of @xmath400 . and @xmath389 ( [ eq_c - transform ] ) computed using the fundamental set of @xmath0-functions ( a ) or computed using the shifted by a factor @xmath400 set of @xmath0-functions ( b ) for @xmath391 $ ] ( [ eq_parametrization ] ) , with @xmath399 . ] the result is enlightening . the case using the shifted set of orbit functions exhibits a degree of error far above what could be tolerated , due to the fact that the higher harmonics of the @xmath0-functions , as expected , oscillate more . this derogatory result should not , however , dismiss the use of such higher harmonics . the gaussian function extrapolated here is very smooth , and should thus be extrapolated with a sum of smoother functions . but in real life applications , the data could be quite chaotic , thus the use of higher harmonics could be useful . it is also possible not to shift the entire set , but only a specific subset , and at that , not all should be shifted by the same factor . careful consideration must be taken , because the functions do not obey the simplistic rule that a shifted function is equal to its counterpart . in fact , the set of adjacent functions is paired according to a reflective symmetry , but not according to translation symmetry . work supported in part by the natural sciences and engineering research council of canada , the mind research institute , by mitacs , and by lockheed martin canada . we are grateful for the hospitality extended to us at the centre de recherches mathmatiques , universit de montral ( m.n . ) and at the aspen center for physics ( j.p . ) where most of this work was done . germain m. and patera j. , cosine transform generalized to lie groups @xmath401 and @xmath402 : application to textural image analysis _ ieee ccece _ 2006 , ottawa 2006 ; + germain m. and patera j. , multiresolution analysis of digital images using the continuous extension of discrete group transforms _ spie conf . on electronic imaging _ , san jose , 2006 , 6065 - 47 , s12 ; + grimm s. and patera j. , decomposition of tensor products of the fundamental representations of @xmath403 in _ advances in mathematical sciences crm s 25 years _ ed . l. vinet , crm proc . lecture notes , vol . 11 , amer . soc . , providence , ri , 1997 , pp . 329355 . kass s. , moody r.v . , patera j. , slansky r. , affine lie algebras , weight multiplicities , and branching rules , vol.1 and vol.2 los alamos series in basic and applied sciences , university of california press , berkeley , ca , 1990 . klimyk a u and patera j alternating group and multivariate exponential functions , in : `` groups and symmetries ; from the neolithic scots to john mckay '' , ams - crm proceedings and lectures notes series ( eds . j. harnad and p. winternitz ) , to appear ( 2008 ) . moody r.v and patera j. , voronoi domains and dual cells in the generalized kaleidoscope with applications to root and weight lattices ( dedicated to h. s. m. coxeter ) , _ can . j. math . _ , 1995 , * 47 * , 573605 .

2ex the thermoelectric properties of intermetallic compounds with ce or yb ions are explained by the single - impurity anderson model which takes into account the crystal - field splitting of the 4_f _ ground - state multiplet , and assumes a strong coulomb repulsion which restricts the number of _ f _ electrons or _ f _ holes to for ce and for yb ions . using the non - crossing approximation and imposing the charge neutrality constraint on the local scattering problem at each temperature and pressure , the excitation spectrum and the transport coefficients of the model are obtained . the thermopower calculated in such a way exhibits all the characteristic features observed in ce and yb intermetallics . calculating the effect of pressure on various characteristic energy scales of the model , we obtain the phase diagram which agrees with the experimental data on cerusi , cecusi , cepdsi , and similar compounds . the evolution of the thermopower and the electrical resistance as a function of temperature , pressure or doping is explained in terms of the crossovers between various fixed points of the model and the redistribution of the single - particle spectral weight within the fermi window .

cond-mat0501519

the thermoelectric power , @xmath5 , of intermetallic compounds with cerium and ytterbium ions exhibits some characteristic features which allow the classification of these compounds . into several distinct groups.@xcite in the case of cerium ions , the thermopower of the compounds belonging to the first group ( type ( a ) systems ) has a deep negative minimum at low temperatures @xcite and a high - temperature maximum , typically between 100 k and 300 k. at the maximum , @xmath5 could be either positive or negative , as shown in fig . [ fig : tepall ] . at very low temperatures , the type ( a ) systems order magnetically or become superconducting . the compounds of the second group ( type ( b ) systems ) have a negative low - temperature minimum and a positive high - temperature maximum but , in addition , the thermopower shows a smaller positive peak at lowest temperatures.@xcite this second peak is sometimes concealed by a low - temperature phase transition ; for example , in cecu@xmath4si@xmath4 it becomes visible only in an external magnetic field which suppresses the superconducting transition,@xcite and in ceru@xmath3ge@xmath4 it shows up when the external pressure suppresses the magnetic transition.@xcite the experimental evidence is now accumulating that the initial slope of the thermopower @xmath6 is positive for this class of ( heavy fermion ) materials , provided the measurements are performed at low enough temperature and with sufficient accuracy.@xcite in the third group ( type ( c ) systems ) , the low - temperature peak is well pronounced and shifted towards the high - temperature peak . the main difference with respect to the type ( b ) systems is that the sign - change of @xmath5 does not occur.@xcite finally , in some cases ( type ( d ) systems ) the thermopower grows monotonically towards the high - temperature maximum , and the low - temperature structure appears only as a shoulder on a broad peak , or is not resolved at all . @xcite the clue to these various types of behavior comes from the high - pressure@xcite and doping studies , @xcite which show that the thermopower of cerium compounds changes continuously from type ( a ) to type ( d ) . a typical example is provided by the @xmath5 of ceru@xmath3ge@xmath4 , which is plotted in fig.[fig : tepall ] as a function of temperature , for various pressures.@xcite at ambient pressure , ceru@xmath3ge@xmath4 is a type ( a ) system with a magnetic ground state and negative thermopower below 300 k. an increase of pressure leads to a thermopower with a small positive peak at low temperatures and an enhanced peak at high temperatures . a further increase of pressure enhances both peaks , shifts the low - temperature peak towards the high - temperature one , and makes the thermopower at intermediate temperatures less negative . for large enough pressure , the sign - change does not occur at all and for very high pressure the low - temperature peak merges with the high - temperature one , and transforms into a shoulder ( see inset to fig . [ fig : tepall ] ) . the high - temperature peak grows continuously but its position remains more or less constant , as @xmath5 changes from type ( a ) to ( c ) . eventually , for pressures above 10 gpa , the @xmath5 assumes the ( d ) shape . here , the initial slope of @xmath5 decreases and the position of the maximum shifts to higher temperatures , but its magnitude does not change as pressure increases . similar behavior is also seen in the high - pressure data of , cecu@xmath3si@xmath4,@xcite cecu@xmath4ge@xmath4,@xcite or cepd@xmath4si@xmath4.@xcite as regards doping , the substitutions which reduce the volume and make ce ions less magnetic , transform @xmath5 from type ( a ) to type ( b),@xcite from ( b ) to ( c),@xcite or from ( a ) to(c),@xcite while the substitutions which expand the volume and make the ce more magnetic , transform the thermopower from , say , type ( d ) to type ( c ) or from type ( c ) to type ( b).@xcite this variation of shape is an indication that the local environment plays an important role in determining the magnetic character of ce and yb ions . even at high temperatures , where each 4_f _ ion is an independent scatterer , the thermopower of a sample with a high concentration of 4_f _ ions can not be obtained by rescaling the low - concentration data . the ytterbium intermetallics can be classified using the mirror - image analogy with cerium systems . this holds because the yb ions fluctuate between 4_f_@xmath7 and 4_f_@xmath8 , while the ce ions fluctuate between 4_f_@xmath9 and 4 _ f_@xmath10 configurations , and the dynamics of a single _ f _ hole and a single _ f _ electron is the same . a well - defined local moment leads in yb systems to the type ( a ) behavior , such that the thermopower has a negative minimum at high temperatures and a positive maximum at low temperatures;@xcite the size of the minimum is about the same as the size of the maximum . the thermopower of ( b)-type yb systems@xcite mirrors the ( b)-type ce systems . here , one finds two negative minima separated by a small positive maximum . the type ( c ) yb systems have a nonmonotonic thermopower with a large ( negative ) minimum at high temperatures and a smaller one at low temperatures , but there is no sign - change.@xcite finally , the thermopower with a single negative peak centered around 100 k @xcite mirrors the type ( d ) behavior of ce systems . the reduction of volume by pressure or doping@xcite stabilizes the magnetic @xmath11 configuration of yb ions , and transforms @xmath5 from , say , type ( b ) to type ( a ) , from ( c ) to ( b ) , or from ( c ) to ( a ) . the experimental results show that 4_f _ systems with similar thermopowers exhibit similarities in other thermodynamic@xcite and transport@xcite properties , and there is an obvious correlation between the shape of @xmath5 and the magnetic character of the 4_f _ ions . the thermopower measurements provide a simple and sensitive tool for characterizing the magnetic state of a 4_f _ ion in a given metallic matrix : the shape of @xmath5 changes from the ( a)-type in the case of magnetic ce ( yb ) ions with stable f@xmath12 ( 4_f_@xmath7 ) configuration to the ( d)-type for non - magnetic ce ( yb ) ions which fluctuate between the 4_f_@xmath12 ( 4_f_@xmath7 ) and 4_f_@xmath13 ( 4_f_@xmath8 ) configurations . we explain the thermoelectric properties of ce and yb ions in terms of a single - impurity anderson model which takes into account the splitting of the 4_f _ states by the crystalline electric field ( cf ) , and assumes an infinitely large _ f f _ coulomb repulsion , which restricts the number of _ f _ electrons or _ f _ holes to @xmath0 for ce and @xmath1 for yb . we assume that pressure changes the coupling and the relative occupation of the _ f _ and conduction states , and impose the charge - neutrality constraint on the local scattering problem at each temperature and pressure . the total charge conservation provides a minimal self - consistency condition for a poor - man s treatment of pressure effects in stoichiometric compounds . the excitation spectrum of such a model in the vicinity of various fixed points , the crossovers induced by temperature and pressure , and the corresponding effects on @xmath5 , the number of _ f _ particles , @xmath14 , and the electrical resistance , @xmath15 , are calculated by the non - crossing approximation ( nca ) . the description of the stoichiometric compounds in terms of an impurity model is certainly inadequate at low temperatures where the _ f _ electrons become coherent . the errors due to such an approximation and the low - temperature errors inherent in the nca calculations are discussed in detail at the end of sec . [ theoretical description ] . our paper extends the long - standing theory of coqblin , @xcite which described the high - temperature properties of ce and yb intermetallics by the coqblin - schrieffer ( cs ) model with cf splitting , and its more recent version@xcite which improved the low - temperature calculations by rescaling the coupling constants . these previous theories explained the main features of the temperature dependence of the thermopower@xcite and the magnetic susceptibility@xcite but could not describe the pressure effects , because the cs model neglects charge fluctuations . furthermore , the approximations used to solve the effective high- and low - temperature models cease to be valid at temperatures at which @xmath5 changes sign,@xcite such that the shape of @xmath5 between the two maxima ( minima ) in ce ( yb ) systems could only be inferred from an interpolation . here , we consider both the local _ spin _ and _ charge _ fluctuations , and provide the full description of the impurity problem at various pressures and temperatures , from above the cf temperature to below the kondo temperature , including the intermediate regime where @xmath5 changes sign . our results explain the shapes ( a ) to ( d ) of the thermopower , which are found in the systems like ceal@xmath16,@xcite ceru@xmath3ge@xmath4 , @xcite cecu@xmath3si@xmath4,@xcite cecu@xmath4ge@xmath4,@xcite or cepd@xmath4si@xmath4 , @xcite and the mirror - image shapes found in systems like ybnisn , yninau@xmath4 , ybsi or ybcu@xmath4si@xmath4.@xcite we also explain the pressure data like on ceru@xmath3ge@xmath4@xcite or cecu@xmath3si@xmath4,@xcite and the chemical pressure data on ce@xmath17la@xmath18pdsn @xcite and ce@xmath17la@xmath18ru@xmath3si@xmath4@xcite or ybcu@xmath4si@xmath4.@xcite the paper is organized as follows . in sec . [ theoretical description ] we introduce the model , discuss its limitations , and describe the method of solution . in sec . [ results ] we provide the results for the transport coefficients of ce- and yb - based intermetallics . in sec . [ discussion ] we discuss the effects of temperature and pressure on the spectral function , analyze the fixed - point behavior , and relate the shapes of the thermopower to the properties of elementary excitations . [ conclusions ] , gives the summary and the conclusions . we model the intermetallic compounds by taking as many ce or yb ions per unit cell as required by the structure , but assuming that the scattering of conduction electrons on a given 4_f _ ion does not depend on other 4_f _ ions , except through the modification of the chemical potential . in other words , we consider an effective impurity model which treats the 4_f _ states as scattering resonances rather than bloch states but take into account the charge transfer due to the local scattering and adjust the chemical potential , @xmath19 , so as to maintain the overall charge neutrality of the compound . such a description of the lattice problem applies at temperatures at which the mean free path of the conduction electrons is short and the scattering is incoherent . we consider mainly the ce intermetallics and present only a few preliminary results for yb intermetallics . the cerium ions are allowed to fluctuate between the 4_f_@xmath13 and 4_f_@xmath12 configurations by exchanging electrons with the conduction band ; the ( average ) energy difference between the two configurations is @xmath20 and the hopping is characterized by the matrix element @xmath21 . the 4_f_@xmath22 configuration is excluded , i.e. , an infinitely strong coulomb repulsion @xmath23 between _ f _ electrons is assumed . the 4_f_@xmath12 configuration is represented by @xmath24 crystal field levels : there are @xmath25 excited states separated from the cf ground level by energies @xmath26 , where @xmath27 . the local symmetry is taken into account by specifying the respective degeneracies of these levels , @xmath28 . thus , the low - energy excitations of ce intermetallics are modeled by an effective single - impurity anderson hamiltonian,@xcite @xmath29 where @xmath30 describes the conduction band , @xmath31 describes the cf states , and @xmath32 describes the transfer of electrons between 4_f _ and conduction states . in the absence of mixing , the conduction band is described by a semielliptical density of states , @xmath33 , centered at @xmath34 and of half - width @xmath35 , and the unrenormalized _ f _ states are represented by a set of delta functions at @xmath36 and @xmath37 . the conduction states and the @xmath38 states have a common chemical potential , which is taken as the origin of the energy axis . the properties of the model depend in an essential way on the cf splittings and on the coupling constant @xmath39 , where @xmath40 measures the coupling strength between the _ f _ electrons and the conduction band , @xmath41 , and @xmath42 is the total degeneracy . we assume @xmath43 , @xmath44 , and @xmath45 limit of the anderson model is controlled by several fixed points which are well understood.@xcite in the case of a ce ion with two cf levels split by @xmath46 , the fixed point analysis can be summarized as follows . at small coupling , such that @xmath47 , we find @xmath48 and the model exhibits the kondo effect . that is , all the physical properties depend only on the kondo temperature , @xmath49 , which is uniquely determined by @xmath50 , @xmath46 , and the degeneracies of the cf states ( for the nca definition of @xmath49 see ref . @xcite and sec.[discussion ] below ) . the low - temperature behavior is characterized by the fermi liquid ( fl ) fixed point , which describes a singlet formed by an antiferromagnetically coupled _ f _ electron and conduction electron . an increase of temperature breaks the singlet and gives rise , at about @xmath49 , to a transition to the local moment ( lm ) fixed point , which describes a cf - split _ f _ state weakly coupled to the conduction band . the effective degeneracy of this _ f _ state is defined by the lowest cf level . for @xmath51 , there is a further crossover to another lm fixed point , which describes the scattering of conduction electrons on a fully degenerate local moment . at higher coupling , such that @xmath52 , the _ f _ charge is reduced to @xmath53 , and the impurity still behaves as a kondo ion , but the kondo scale is much higher than in the case @xmath54 . the two lm regimes are now close together and the crossover from the fl to the lm regime occurs at temperatures which are comparable to @xmath46 . at very high coupling , such that @xmath55 and @xmath56 , the _ f _ ions appear to be non - magnetic at all accessible temperatures due to the mixing of the 4_f_@xmath13 and 4_f_@xmath12 configurations . in this valence fluctuating ( vf ) regime , the behavior is non - universal and changes slightly , when the calculations are performed for different sets of parameters . away from @xmath57 , more than one energy scale is needed to fully characterize the model . other cf schemes pertinent to ce and yb ions in a different environment are characterized by similar fixed points . our calculations show that the functional form of the response functions changes at the crossover and that the @xmath58 limit of the anderson model captures all the main features of the experimental results on ce and yb intermetallics . to explain the pressure effects , which changes the thermopower of ce and yb systems in opposite ways , we assume that the exchange coupling @xmath50 increases in ce and diminishes in yb compounds , as pressure increases . this difference arises because ce fluctuates between 4_f_@xmath10 and 4_f_@xmath9 , while yb fluctuates between 4_f_@xmath8 and 4_f_@xmath7 configurations , so that the pressure - induced reduction of the number of electrons in the _ f_-shell makes ce ions less magnetic and yb ions more magnetic . in ce intermetallics , there is a substantial overlap between the _ f _ wave functions of ce and those of the neighboring atoms , and we associate the pressure - induced increase of @xmath59 with an enhancement of the hybridization @xmath60 . this enhances the kondo temperature , and pushes the system from the kondo to the vf limit . in stoichiometric compounds , the pressure - induced reduction of @xmath61 is accompanied by the increase of @xmath62 , because the total charge of a given compound , @xmath63 , is constant . the conservation of particles is enforced by adjusting @xmath19 , and since all the energies are measured with respect to @xmath19 , this amounts to shifting @xmath64 and @xmath65 by some amount @xmath66 . thus , we describe the pressure effects for a given ce compound by changing @xmath60 and keeping @xmath67 constant . the changes in the band - width and the cf splitting are neglected . in yb intermetallics , the _ f _ states are more localized than in ce systems , and we assume that the decrease in the radius of the 4_f _ shell as it loses charge at elevated pressure is sufficient to compensate for the increase in hybridization brought about by the reduction in unit cell volume . the reduction of @xmath59 in yb compounds is achieved through an enhancement of the hole binding energy @xmath64 as the neighbouring ions get closer to the rare earth , while @xmath60 remains essentially constant . this reduces the kondo temperature , and drives the system towards the kondo limit . since @xmath60 is treated as a material - specific constant , we model the pressure effects in yb systems by shifting @xmath64 and solving for @xmath65 , so as to preserve @xmath68 . this procedure shifts @xmath64 and @xmath65 by different amounts and makes the separation @xmath67 pressure dependent . however , when temperature is changed at constant pressure , the charge neutrality is enforced in the same way as for ce compounds , by shifting the chemical potential without changing the separation @xmath67 . the changes in the bandwidth and the cf splitting are neglected . describing the pressure effects in such a way , we can calculate the response functions of the model for any value of the external parameters , and study the transitions between various fixed points . the electrical resistivity and the thermopower of the single - impurity anderson model are obtained from the usual expressions,@xcite @xmath69 @xmath70 where @xmath71 and @xmath72 are given by the static limits of the current - current and current - heat current correlation functions , respectively . in the absence of nonresonant scattering the vertex corrections vanish and the transport integrals can be written as,@xcite @xmath73 where @xmath74 is material - specific constant , @xmath75 $ ] is the fermi function , @xmath76 is the conduction - electron scattering rate,@xcite @xmath77 @xmath78 is the _ f_-electron spectral function , @xmath79 is the green s function , and @xmath80 is the concentration of _ f _ ions . eqs.([eq : lij_final ] ) and ( [ eq : tau ] ) show clearly that the sign and the magnitude of @xmath5 are determined by the spectral weight within the fermi window ( f window ) , i.e. , by the shape of @xmath81 for @xmath82 . the sign of @xmath5 is positive if the f window shows more states above than below the chemical potential , and is negative in the opposite case . the difficult part is to find @xmath83 and , here , we solve this problem by the nca , following closely refs . @xcite , where all the technical details can be found . the main difference with respect to these nca calculations is that we take @xmath84 and enforce the overall charge neutrality . a detailed comparison with the experimental data shows that the transport coefficients obtained from the single - impurity anderson model have all the hallmarks of the experiments , but discrepancies appear at low temperatures . this indicates the limitations of our approach which should be considered before presenting the nca results . there are two main causes for the breakdown of the single - impurity model and the nca calculations . first , at temperatures much below @xmath49 the nca spectral function develops an unphysical spike , such that the resistivity and the thermopower become artificially enhanced . this error becomes particularly severe at high pressure , because the characteristic scale @xmath49 increases very rapidly with @xmath60 and the non - analytic nca spike appears at rather high temperatures . the unphysical enhancement of the low - frequency part of @xmath81 reduces the integral for @xmath85 , which is strongly underestimated at low - temperature . the integral @xmath86 is less affected by this pathology , because it has an additional @xmath87 factor which removes the states within the fermi window . thus , the overall shape of @xmath88 seems to be qualitatively correct , even though the low - temperature part of the curve has an unphysical enhancement . these difficulties are well known @xcite and relatively easy to resolve in the kondo limit , where the model has a unique low energy scale , @xmath49 . we can find @xmath49 in the lm regime , where the nca is reliable , and infer the low - temperature behavior from the universal power laws which hold in the fl regime . combining the nca results and fl theory , we can discuss the experimental data at all temperatures at which the single - ion approximation holds . a more serious problem is that , in stoichiometric compounds , the _ f _ electrons become coherent at low enough temperatures . this leads to a magnetic transition in ceru@xmath3ge@xmath4 , the formation of a heavy fl in ceru@xmath3si@xmath4 , superconductivity in cecu@xmath3si@xmath4 , or some more complicated ground states . at high pressure , the coherence sets in at very high temperatures , as revealed by low values of the electrical resistance . the onset of coherence ( like the nca pathology ) has its main impact on the low - energy states , giving @xmath89 , so that the impurity result badly overestimates the low - temperature electrical resistance . however , the considerations for the periodic anderson model@xcite , or other models with on - site correlation@xcite , show that the integral for @xmath86 also contains an additional @xmath87 factor which reduces the contribution of the low - energy coherent states to @xmath86 , like in the single impurity case . thus , our results for @xmath5 provide a qualitative description of the experimental data at temperatures well below the onset of coherence , but the calculated values of @xmath5 around @xmath49 are overestimated . we mention also that the analysis of the doping effects in terms of chemical pressure is not complete , because doping might give rise to a charge transfer or change the character of the ground state , and that the mirror - image analogy between ce and yb systems holds for the resonant scattering but may be lost in the presence of any additional scattering channel . despite these drawbacks , the nca solution of the anderson model provides a surprisingly accurate description for a large body of experimental data above the magnetic or superconducting transition temperature , and at low to moderate pressure . in this section we present the nca results for the transport coefficients , describe the thermopower and the electrical resistance due to ce ions in some detail , show the results for the _ f _ occupation , present a few preliminary results for the thermopower of yb , and compare our results with the experimental data . our strategy for ce intermetallics is to illustrate the behavior of one particular compound as a function of temperature at various pressures . the compound is characterized at ambient pressure and high temperature by an initial parameter set @xmath90 , where @xmath91 and @xmath92 are measured with respect to @xmath19 and the high - temperature limit is defined by temperature @xmath93 . for given values of @xmath60 and @xmath94 , we start the calculations at @xmath95 and find @xmath96 such that the total charge is the same as the one obtained for @xmath97 . ( in the absence of coupling we have @xmath98 and obtain @xmath62 by integrating the unperturbed density of states . ) at high temperatures , the _ f_-state is almost decoupled from the conduction band , the renormalization of the parameters is small , and the numerics converge very fast . we then reduce the temperature , find the new shift @xmath99 ensuring the charge conservation , and calculate the response functions for the resulting values of @xmath65 and @xmath64 . this process is continued until the nca equations break down at @xmath100 . to model the same system at different pressure , we change @xmath60 , find again @xmath96 , and repeat the same procedure as at ambient pressure for @xmath101 . note , @xmath102 and @xmath67 are conserved at all temperatures and pressures . as a numerical example we consider a semielliptical conduction band of half - width @xmath103 ev , centered at @xmath104 ev , and a _ f _ state split into a doublet and a quartet@xcite by the cf with @xmath105 ev . we take @xmath63=5.6301 electrons per ion ( 0.9383 electrons for each one of @xmath106 effective spin channels ) . the transport coefficients are calculated for the hybridization strength changing from 0.06 ev to 0.20 ev , i.e. , for @xmath60 varying from @xmath107 to @xmath108 . the single - particle excitation spectra corresponding to these parameters are discussed in sec . [ discussion ] . using the procedure outlined above we obtain for @xmath5 the results shown in figs . [ fig : theo_tep_low ] and [ fig : theo_tep_large ] . the calculated curves exhibit all the shapes ( a ) to ( d ) found in the experimentas and give @xmath5 of the right magnitude , except at low temperatures where the calculated peak is too large with respect to the experimental data . as discussed already , the reason for this discrepancy is that the nca overestimates the fermi - level scattering rate for @xmath100 , and that we neglected the coherent scattering , which sets in at temperatures of the order of @xmath49 . thus , our low - temperature result for @xmath85 is artificially reduced , which makes @xmath5 too large . the sign and the topology of the @xmath5 curves do not seem to be affected by this error . for @xmath109 , we have @xmath110 k and @xmath111 , and obtain @xmath5 with two well separated peaks , as shown by the @xmath60=60 and @xmath60=70 mev curves in fig . [ fig : theo_tep_low ] . the high - temperature peak is centered at @xmath112 and for our choice of parameters @xmath113 . the low - temperature maximum is at about @xmath114 and @xmath115 . the thermopower between the two maxima is mainly negative . since most of the type ( a ) and ( b ) systems order magnetically or become superconducting above @xmath49 , the low - temperature peak is not shown in fig . [ fig : theo_tep_low ] for @xmath116 mev . a small increase of @xmath60 ( due to , say , an increase of pressure ) reduces @xmath61 ( see fig . [ fig : nf ] ) , enhances @xmath49 and @xmath117 , and expands the temperature range in which @xmath5 is positive . such a behavior , which is typical of kondo systems with small @xmath49 , is in a qualitative agreement with the thermopower of the type ( a ) systems described in sec.[introduction ] , and with the data on ceru@xmath3ge@xmath4 @xcite at low pressures ( below 4 gpa ) and above the ordering temperature , as shown in fig.[fig : tepall ] . for @xmath118 , we have @xmath119150 k and @xmath120 0.8 , and still obtain @xmath5 with the two maxima . but @xmath117 is now positive , the value of @xmath121 is enhanced , the temperature interval in which @xmath122 is reduced , and @xmath5 at the minimum is less negative than for smaller @xmath60 . as we increase @xmath60 ( by increasing pressure ) , the sign - change of @xmath5 is removed , @xmath121 and @xmath117 are further enhanced , but @xmath123 is not changed . the two peaks are coming closer together and are merging eventually . these features are typical of kondo systems with moderate @xmath49 , say @xmath124 k , and are in a qualitative agreement with the data on the type ( b ) and ( c ) systems mentioned in sec . [ introduction ] . they are also shown by the ceru@xmath3ge@xmath4 data@xcite at intermediate pressures , ( see the curves in fig . [ fig : tepall ] for pressure above 3.4 gpa . ) at @xmath125 , the system enters the vf regime , @xmath5 acquires a single maximum at @xmath123 , with a shoulder on the low - temperature side . even this shoulder vanishes , when @xmath60 is increased further , as shown in fig . [ fig : theo_tep_large ] . for @xmath108 the thermopower is of the type ( d ) , with a single peak which is much steeper on the high- than on the low - temperature side . this peak shifts to higher temperatures with increasing @xmath60 ( pressure ) and @xmath117 saturates ( see inset to fig . [ fig : theo_tep_large ] ) ; the initial slope of @xmath5 decreases continuously with @xmath60 . such a behavior is in a qualitative agreement with the thermopower data on valence fluctuators@xcite and with the high - pressure data on ceru@xmath3ge@xmath4@xcite ( cf . inset to fig . [ fig : tepall ] ) and cecu@xmath3si@xmath4@xcite . however , a large discrepancy appears between theory and experiment at low temperatures , because @xmath81 is overestimated for @xmath126 , which makes the nca curves larger than the experimental ones . a possible correction of the initial @xmath6 values is discussed below . the _ f_-electron number @xmath61 , calculated for the parameters used in figs . [ fig : theo_tep_low ] and [ fig : theo_tep_large ] , is plotted in fig . [ fig : nf ] as a function of temperature . the overall temperature dependence is rather slow , but two different types of behavior can still be seen . for @xmath127 , we find that @xmath128 is nearly independent of temperature and close to 1 , which is typical of kondo systems.@xcite for @xmath129 , we find that @xmath128 is less than 0.8 and nearly constant at low temperatures but at about @xmath130 ( 271 k for @xmath46=0.07 ev ) there is an increase followed by the saturation at high temperatures . considered as a function of @xmath60 ( pressure ) , @xmath131 shows different behavior at high and low temperatures . at high temperatures @xmath61 decreases uniformly as @xmath60 increases . at low temperatures @xmath61 does not change much for @xmath132 and @xmath133 , but drops rapidly around @xmath134 , indicating the crossover from the kondo to the vf regime . the electrical resistivity , @xmath135 , obtained for small and intermediate values of @xmath60 , is shown in fig . [ fig : theo_rho ] . the interesting feature is the high - temperature maximum , which appears for @xmath107 , and correlates very well with the maximum in @xmath5 . for temperatures below the maximum , @xmath135 drops to a minimum and then rises logarithmically as @xmath49 is approached . this minimum and the subsequent low - temperature upturn are of a purely electronic origin and appear in systems with small @xmath49 and large cf splitting . in these systems , one can follow the evolution of the two peaks in @xmath135 , and observe the disappearance of the minimum with the application of pressure . for example , in ceru@xmath3ge@xmath4,@xcite cepd@xmath3ge@xmath4,@xcite cepd@xmath3si@xmath4,@xcite or cecu@xmath136au@xcite the minimum becomes more shallow , transforms into a shoulder , and vanishes at high enough pressure . as discussed already , the nca overestimates the low - temperature scattering and distorts the relative magnitude of the high- and low - temperature peaks . in addition , the single - ion approximation always gives @xmath135 which saturates at low temperature and can not explain the low - temperature reduction of @xmath135 , which is seen in stoichiometric compounds below the onset of coherence . the electrical resistance of ce - based kondo systems at very high pressure , and the ambient pressure data of valence fluctuators , can not be described by the nca solution . in these systems , the scattering on _ f _ ions remains coherent up to rather high temperatures and the nca solution is valid only above the high - temperature maximum and is not physically relevant . to illustrate the situation in yb intermetallics , we perform a generic calculation for a semielliptical ( hole ) conduction band and four cf doublets . the nca equations are solved for a single _ f _ hole ( @xmath137 ) . following the procedure outlined in sec . [ theoretical description ] , we start the calculations at ambient pressure and temperature @xmath138 , such that the _ f _ state is almost free , and calculate @xmath139 for the initial parameter set @xmath140 . at lower temperatures we shift @xmath65 and @xmath64 with respect to @xmath34 and @xmath141 by @xmath142 , so as to conserve @xmath68 , and calculate the response functions for this new parameter set ; this procedure does not change @xmath67 . at a higher pressure , we start again at @xmath143 , change @xmath64 so as to increase @xmath144 and find @xmath65 which conserves @xmath68 . since @xmath60 is not changed by this procedure , we now have @xmath145 . for temperatures below @xmath138 , the properties of the system are calculated by the same procedure as at ambient pressure , i.e. , @xmath68 is conserved by shifting @xmath65 and @xmath64 by the same amount @xmath142 . taking @xmath103 ev , @xmath146 ev , @xmath60=0.08 ev , @xmath147 ev , and three excited cf doublets at @xmath148=0.02 ev , @xmath149=0.04 ev , @xmath150=0.08 ev , respectively , we find @xmath151 at @xmath143 . the @xmath5 obtained for @xmath64 ranging from @xmath152 ev to @xmath153 ev , is shown in fig . [ yb_tep_nca ] . we recall , that an increase of pressure makes @xmath64 more negative . for @xmath152 ev , the thermopower shows a deep minimum , typical of yb ions in the vf state . for @xmath154 ev , @xmath5 develops a small maximum , which separates the high - temperature minimum at @xmath123 and the low - temperature minimum at @xmath49 . by making @xmath64 more negative we shift @xmath49 to lower values , as shown in fig . [ yb_tep_nca ] by the curves obtained for @xmath154 ev , @xmath155 ev , and @xmath156 ev , respectively . the low - temperature range in which @xmath5 is negative shrinks with pressure , in agreement with the experimental data.@xcite however , for @xmath64 much below @xmath19 , the nca calculations break down before this minimum is reached . as regards the value of @xmath5 at the maximum , it is negative at first but it becomes positive as pressure increases , i.e. , the thermopower changes from ( c)-type to ( b)-type . at very high pressures , such that the @xmath157 , the thermopower is dominated at low temperatures by a large positive peak and at high temperatures by a negative minimum , which is typical of yb - based systems with a small kondo scale . the shape of @xmath5 is directly related to the magnetic character of yb ions and our calculations explain the qualitative features of the thermopower of ybau@xmath4 , ybau@xmath16 , @xcite and ybsi , ynni@xmath4si@xmath4,@xcite which are of the ( a)-type , of yb@xmath4ir@xmath16al@xmath158,@xcite ybaucu@xmath159,@xcite and ybnisn,@xcite which are of the ( b)-type , of ybpdcu@xmath159,@xcite and yb@xmath4rh@xmath16al@xmath158,@xcite which are of the ( c)-type , and of ybagcu@xmath159,ybpd@xmath4cu@xmath4,@xcite and ybinau@xmath4@xcite which are ( d)-type . the pressure effects in ybsi,@xcite and the chemical pressure effects in ybcu@xmath4si@xmath4@xcite are also in a qualitative agreement with our results . an estimate of the low - temperature properties of the single - impurity model can be obtained by combining the nca results with the universal fl laws . the sommerfeld expansion of transport coefficients in eq.([eq : lij_final ] ) gives@xcite @xmath160 where @xmath161=283.5 @xmath19v / k and @xmath162 is the enhancement factor defined by the fermi - level derivative of the _ f_-electron self - energy , @xmath163_{\omega = t=0}$ ] . @xmath164 is related to the kondo scale by a factor of order 1 . setting @xmath165 , and using for @xmath166 the nca results of fig . [ fig : nf ] , we can estimate from eq.([resonant_siam ] ) the initial slope of the curves plotted in figs . [ fig : theo_tep_low ] and [ fig : theo_tep_large ] . the obtained values , which go from 32 @xmath19v / k@xmath167 for @xmath60=60 mev to 1 @xmath19v / k@xmath167 for @xmath60=130 mev and 0.1 @xmath19v / k@xmath167 for @xmath60=200 mev , are in the range reported recently for various ce - based heavy fermions and valence fluctuators ( see table 1 in ref . the initial slope of @xmath5 decreases as we move from the kondo to the vf limit , in agreement with pressure experiments . @xcite at higher temperatures , the non - linear corrections reduce @xmath5 and give rise to a maximum at @xmath49 . these non - linear corrections are non - universal , and a large slope does not necessary translate into large @xmath5 at @xmath49 . the corresponding calculations for the yb curves plotted in fig . [ yb_tep_nca ] show that an application of pressure enhances @xmath6 . the initial slope of @xmath5 can also be related to the @xmath168 coefficient of the specific heat , with the important result that the ratio @xmath169 is independent of @xmath162 . this expression and eq.([resonant_siam ] ) are valid in the single - impurity regime , and it is not obvious that they would produce the correct results for stoichiometric compounds . however , the characteristic energy scales of a coherent fl ( inferred from the experimental data on the initial slope of the thermopower , or the specific heat coefficient ) do not seem to be much different from the single - ion scale @xmath170 of the lm regime ( inferred , say , from the peak in the thermopower or the curie - weiss temperature ) . in this section we present the nca results for @xmath81 , study the low - energy spectral features in the vicinity of various fixed points , discuss the changes induced by the crossovers , and explain the behavior of @xmath5 in terms of the redistribution of spectral weight within the fermi window ( f window ) . only the ce case is considered and it is assumed that pressure gives rise to an increase of the hybridization width @xmath60 . the results obtained for @xmath171 are shown in fig . [ fig : spectrum_a ] , where @xmath81 is plotted as a function of frequency for several temperatures . at high temperatures , @xmath172 , the spectral function has a broad charge - excitation peak somewhat above @xmath64 and a narrower resonance of half - width @xmath46 , centered below @xmath19 . this low - energy resonance is a many - body effect due to the hybridization of the conduction states with the 4_f _ states and is typical of the exchange scattering on the full multiplet . in this temperature range , the f window shows more spectral weight below than above @xmath19 ( see the middle panel in fig . [ fig : spectrum_a ] ) and @xmath122 . the magnetic susceptibility@xcite is curie - weiss like , with a very small curie - weiss temperature and a curie constant which is close to the free ce ion value . the maximum of @xmath5 at about @xmath112 is here negative , @xmath173 , but a slight increase of @xmath60 would make @xmath117 positive . all these features are typical of the lm fixed point corresponding to a fully degenerate _ f_-state . , calculated for the hybridization strength @xmath60=0.06 ev and the cf splitting @xmath46=0.07 ev , plotted as a function of frequency for several temperatures . the solid , dashed , dashed - dotted and dotted curves correspond to @xmath1742 , 41 , 209 , and 670 k , respectively . the charge - excitation peak is visible in the upper panel . the middle panel shows the evolution of the cf and kondo resonances with temperature . for @xmath175 , the many - body resonance of half - width @xmath46 is centered well below @xmath19 . the f window has more states below than above @xmath19 and @xmath176 . the lower panel shows the position of the kondo resonance above @xmath19 . its center defines @xmath1771 k. for @xmath178 the f window has more states above than below @xmath19 and @xmath179 . , title="fig : " ] , calculated for the hybridization strength @xmath60=0.06 ev and the cf splitting @xmath46=0.07 ev , plotted as a function of frequency for several temperatures . the solid , dashed , dashed - dotted and dotted curves correspond to @xmath1742 , 41 , 209 , and 670 k , respectively . the charge - excitation peak is visible in the upper panel . the middle panel shows the evolution of the cf and kondo resonances with temperature . for @xmath175 , the many - body resonance of half - width @xmath46 is centered well below @xmath19 . the f window has more states below than above @xmath19 and @xmath176 . the lower panel shows the position of the kondo resonance above @xmath19 . its center defines @xmath1771 k. for @xmath178 the f window has more states above than below @xmath19 and @xmath179 . , title="fig : " ] at lower temperatures , @xmath180 , the cf splits the many - body resonance into two peaks . the larger one grows below @xmath19 and the smaller one above @xmath19 ( see the middle panel in fig . [ fig : spectrum_a ] ) . this asymmetry is enhanced as @xmath60 is reduced , which is typical of the anderson model with cf splittings ; @xcite the increase of the low - energy spectral weight below @xmath19 gives rise to a large negative thermopower . a further reduction of temperature leads , for @xmath181 , to a rapid growth of an additional peak very close to @xmath19 , such that @xmath81 acquires three pronounced low - energy peaks ( see fig . [ fig : spectrum_a ] ) . ( the physical origin of these many - body resonances is explained in detail in ref . @xcite ) the peak centered at @xmath182 is the kondo resonance and its appearance below @xmath183 40 k marks the onset of the lm regime , which is due to scattering of conduction electrons on the lowest cf level . the two cf peaks centered at about @xmath184 are outside the f window , and do not affect the low - temperature transport and thermodynamics . once the kondo peak peak appears , the f window shows more spectral weight above than below @xmath19 and @xmath5 is positive , which is just the opposite to what one finds for @xmath185 . the center of the kondo resonance saturates at low temperatures at the energy @xmath186 , which provides the nca definition of the kondo scale , @xmath187 . in the symmetric anderson model @xmath49 is related to the width of the kondo resonance but in the highly asymmetric case we are dealing with here , the current definition is more appropriate . the comparison with numerical renormalization group ( nrg ) calculations @xcite shows that @xmath188 gives a reliable estimate of the kondo temperature even for a doubly degenerate anderson model , and we assume that the nca definition of @xmath49 provides the correct kondo scale of the cf - split single - impurity anderson model as well . because the kondo resonance is asymmetric with respect to the @xmath189 line and has more states above than below @xmath19 , the reduction of temperature enhances @xmath5 unltil it reaches , at @xmath49 , the maximum value @xmath121 . a further temperature reduction leaves the top of the kondo resonance outside the f window , and the thermopower drops . however , most ce and yb system with very small @xmath49 have a phase transition above @xmath49 , and to discuss the normal - state properties of ( a)-type systems it is sufficient to consider the nca solution for @xmath190 . an increase of the coupling to @xmath191 has a drastic effect on @xmath81 , as illustrated in fig . [ fig : spectrum_b ] , where @xmath81 is plotted for @xmath192 ev . the charge - excitation peak is transformed into a broad background ( see the upper panel in fig . [ fig : spectrum_b ] ) and the only prominent feature at @xmath172 is the low - energy resonance of half - width @xmath46 centered above @xmath19 . this low - energy resonance is due to the exchange scattering of conduction electrons on the full cf multiplet , which gives rise to the maximum of @xmath5 in the lm regime . the f window ( see lower panel in fig . [ fig : spectrum_b ] ) shows more spectral weight above than below @xmath19 , so that @xmath193 . the reduction of temperature below @xmath123 removes some spectral weight above @xmath19 and brings additional spectral weight below @xmath19 , which reduces @xmath5 and leads to a minimum.@xcite a further reduction of temperature leads to the rapid growth of the kondo peak at @xmath188 , and the cf peak at @xmath194 , but the negative cf peak does not develop . that is , an increase of pressure removes the lower cf peak , and shifts the kondo and the upper cf peak to higher energies , without changing their separation @xmath46 . the f window shows more spectral weight above than below @xmath19 , so that @xmath5 is positive and grows as temperature is lowered . the maximum @xmath121 is reached at @xmath49 when the kondo resonance is fully developed . the characteristic energy scale is defined again by the position of the kondo peak , @xmath187 , which can now be quite large . for @xmath195 the f window becomes narrower than the kondo resonance and @xmath5 drops below @xmath121 . for @xmath100 , where the fl behavior is expected,@xcite the nca gives @xmath81 with an unphysical spike at @xmath19 , which makes @xmath135 and @xmath5 much larger than the exact result . however , once @xmath49 is obtained from the nca calculations , the low - temperature transport can be inferred from the universal power laws which hold in the fl regime , as discussed in the previous section . a further increase of @xmath60 shifts the kondo and the cf peaks to higher energies , and changes their relative spectral weight , as shown in fig . [ fig : spectrum_c ] , where the low - frequency part of @xmath81 is shown for @xmath196 . the kondo scale is still defined by the center of the kondo peak , even though it is now reduced to a hump on the low - energy side of a large peak centered at @xmath194 . the unphysical nca spike at @xmath189 can be seen at lowest temperatures . the thermopower is positive at all temperatures and has only a shoulder below @xmath123 . a quantitative comparison between @xmath49 , defined by the position of the kondo resonance , and the position of the kondo anomaly in @xmath5 becomes difficult . finally , for @xmath108 , we find @xmath81 with a single broad peak centred at @xmath197 , as shown in fig . [ fig : spectrum_d ] . the cf excitations are now absent , which is typical for the anderson model in the vicinity of the vf fixed point . the relevant energy scale at low temperature is defined as @xmath198 , and shows an almost linear dependence on @xmath199 . the thermopower is always positive and grows monotonically from small values at low temperatures towards a high - temperature maximum at @xmath123 . the unphysical spike at @xmath189 appears at higher temperatures and is more pronounced than for small @xmath60 , as illustrated in fig . [ fig : spectrum_d ] . the initial slope of @xmath5 obtained from the nca result for transport coefficients is very much overestimated with respect to the fl result based on eq.([resonant_siam ] ) . an increase of temperature above @xmath123 modifies the excitation spectrum on an energy scale of the order of @xmath200 and reduces @xmath5 . the values of @xmath200 and @xmath123 do not seem to be related in any simple way , and the temperature - induced breakdown of the local fl is not indicated by any fl scale . an increase of @xmath60 ( pressure ) changes @xmath81 and enhances @xmath123 but does not lead to further enhancement of @xmath117 . the pressure dependence of @xmath200 and @xmath123 looks different . in other words , the description of the vf regime requires more than one energy scale . the characteristic scales of the anderson model obtained for various values of @xmath60 provide the @xmath2 phase diagram of the system plotted in fig . [ t0-versus-gamma.eps ] . the @xmath49 line is defined by the position of the kondo peak and the @xmath123-line by @xmath117 . at small @xmath60 ( low pressure ) [ t0-versus-gamma.eps ] indicates two crossovers . the one around @xmath49 is between the fl and the lm regime defined by the lowest cf level . the one around @xmath123 is the crossover to the high - temperature lm regime , as defined by a full cf multiplet . when these lm regimes are well separated , the crossover between them is accompanied by a minimum and , possibly , a sign - change of @xmath5 . at intermediate pressures , the two lm regimes are too close for the sign - change to occur , and the crossover is indicated only by a shallow minimum or just a shoulder of @xmath5 . here , the relationship between the low - temperature maximum of @xmath5 and the center of the kondo peak can only be given as an order - of - magnitude estimate . at very high pressures , the system is in the vf regime and the crossover from a universal low - temperature fl phase to a non - universal high - temperature phase takes place at @xmath201 . this crossover is not defined by the fl scale @xmath202 , which is very large , but by a much smaller scale @xmath123 . we have applied the single - impurity anderson model to the investigation of the temperature and pressure dependence of the thermopower of ce and yb intermetallics , and found that the crossovers between various fixed points explain the seemingly complicated temperature dependence of @xmath5 . the basic assumption of our approach is that for a given concentration of rare - earth ions , and above some coherence temperature , the system is in an `` effective impurity '' regime . in that `` impurity limit '' , we treat the rare - earth ions as independent scattering centers and solve the ensuing single - impurity model by the nca . our calculations impose the charge neutrality constraint on the local scattering problem , which provides a minimal self - consistency condition for describing the pressure effects . the excitation spectrum obtained in such a way is very sensitive to the changes in the coupling constant @xmath203 , and an increase of the hybridization or a shift of the _ f _ state , which modifies the magnetic state of the _ f _ ion , has a huge effect on the spectral function . since the excitation spectrum is related to the transport coefficients by kubo formulas , the thermopower changes rapidly as a function of temperature , pressure or doping . the shapes ( a ) to ( d ) assumed by @xmath5 in various regions of the parameter space follow straightforwardly from the redistribution of the single - particle spectral weight within the fermi window . the nca solution of the single - impurity anderson model breaks down for @xmath204 . however , the excitations above the ground state are determined by the general fl laws , which can be used to find the initial slope of @xmath5 . estimating the kondo scale and the number of _ f _ electrons by the nca calculations , we find that an increase of pressure reduces ( enhances ) the low - temperature value of @xmath6 in ce ( yb ) compounds . combining the fl and the nca results , we can obtain @xmath5 in the full temperature range at any pressure . our results explain the thermopower of most intermetallic ce compounds mentioned in section [ introduction ] . the multi - peaked @xmath5 , which characterizes the type ( a ) and ( b ) systems , is obtained for kondo ions with small @xmath49 and well - defined cf resonances . a type ( c ) thermopower , which has weakly resolved peaks or just a broad hump , is obtained for kondo ions with large @xmath49 and partially overlapping cf resonances . a single - peaked @xmath5 of the ( d)-type is obtained for the vf ions which do not show any cf splitting of the single - particle excitations . if we assume that pressure increases @xmath60 , and reduces @xmath61 , our results account for the changes of @xmath49 , @xmath121 , @xmath6 , and @xmath117 , observed in pressure experiments on ce compounds . @xcite the strong doping dependance@xcite of @xmath5 is explained as a chemical pressure effect , which changes the _ f _ occupation . in our local model the `` effective concentration '' of _ f _ electrons is determined self - consistently , and the thermopower of concentrated and dilute systems can not be related by simple scaling . ( just like the high pressure data can not be described in terms of rescaled ambient pressure data . ) the relevant energy scales obtained for different values of @xmath60 agree with the @xmath2 phase diagram of ceru@xmath3ge@xmath4 and cecu@xmath3ge@xmath4 , inferred from the high - pressure transport and thermodynamic data.@xcite for yb compounds , we argue that pressure or chemical pressure mainly affect the position of the crystal field levels relative to the band center , while leaving the hybridization width unchanged . by shifting @xmath64 , so as to increase the number of holes , we obtain the thermopower of the type ( a ) to ( d ) , in agreement with the experimental data . the qualitative features seen in yb intermetallics at various pressures or chemical pressures are captured as well,@xcite but different compounds require different initial parameters , and a detailed analysis is yet to be done . in summary , the normal - state properties of stoichiometric compounds with ce and yb ions seem to be very well described by the local model which takes into account spin and charge fluctuations . the classification of the thermopower data follows straightforwardly from the fixed point analysis of the single - impurity anderson model with the cf splitting . a rich variety of shapes assumed by @xmath5 at various pressures or doping dramatically illustrates the effect of the local environment on the response of the single rare earth ion . the nature of the ground state , and the fact that the ground state of many ce and yb systems can be changed with pressure or doping , do not seem to affect the thermopower in the normal state . we take this as an indication that the normal state properties of ce and yb compounds are dominated by the local dynamics . the main problem with our poor man s treatment of ordered compounds is that it neglects the coherent scattering which sets in at low enough temperatures , and reduces the thermopower below the values predicted by the local fermi liquid theory . we can argue that these effects do not change the qualitative features of the thermopower but the proper answer can not be obtained without solving the lattice model . * acknowledgements * + we acknowledge useful comments and suggestions from b. coqblin , j. freericks , c. geibel , a. hewson , b. horvati , and h. wilhelm . this work has been supported by the ministry of science of croatia ( cro - us joint projects , grant number 1/2003 , by the national science foundation under grant number dmr-0210717 , and the swiss national science foundation under grant number 7krpj065554 - 01/1 . j. sakurai , h. takagi , s. taniguchi , t. kuwal , y. isikawa , and j .- l . tholance , j. phys . japan * 65 * , suppl . b , 49 ( 1996 ) . p. link , d. jaccard , and p. lejay , physica b * 225 * , 207 ( 1996 ) . v. zlati et al , phys . * b * 68 , 104432 ( 2003 ) . g. sparn , w. lieka , u. gottwick , f. steglich , and n. grewe , j. magn . mater . * 47 & 48 * , 521 ( 1985 ) . d. jaccard , j. m. mignot , b. bellarbi , a. benoit , h. f. braun , and j. sierro , j. magn . * 47 & 48 * , 23 ( 1985 ) . d. huo , t. kuwai , t. mizushima , y. isikawa , and j. sakurai , physica b * 312 & 313 * , 232 ( 2002 ) . d. huo , k. mori , t. kuwai , s. fukuda , y. isikawa , and j. sakurai , physica b * 281 & 282 * , 101 ( 2000 ) . j. sakurai , d. huo , d. kato , t. kuwai , y. isikawa , and k. mori , physica b * 281 & 282 * , 98 ( 2000 ) . p. b. van aken , h. j. van daal , and k. h. j. buschow , phys . a * 49 * , 201 ( 1974 ) . r. cibin , d. jaccard , and j. sierro , j. magn . magn . mater . * 108 * , 107 ( 1992 ) . h. wilhelm and d. jaccard , phys . b * 69 * , 214408 ( 2004 ) k. behnia , d. jaccard , and j. flouquet , j. phys . : cond matt . , * 16 * , 5187 ( 2004 ) f. steglich , festkrperprobleme xvii , 319 ( 1977 ) . j. sakurai , j. c. gomez sal , and j. rodriguez fernandez , j. magn . * 140 * - * 144 * , 1223 ( 1995 ) . a. amato , d. jaccard , j. sierro , p.haen , p. lejay , and j. flouquet , j. low . phys . * 77 * , 195 ( 1989 ) . y. nuki and t. komatsubara , j. magn . * 63 & 64 * , 281 ( 1987 ) . j. sakurai , t. ohyama , and y. komura , j. magn . mater . * 47 & 48 * , 320 ( 1985 ) . e. bauer , adv . * 40 * , 417 ( 1991 ) . c. fierz , d. jaccard , and j. sierro , j. appl . phys . * 63 * , 3899 ( 1988 ) . d. jaccard , k. behnia and j. sierro , phys . a * 163 * , 475 ( 1992 ) . d. jaccard , j. m. mignot , b. bellarbi , a. benoit , h. f. braun , and j. sierro , j. magn . mater . * 47&48 * , 23 ( 1985 ) . y. bando , j. sakurai , and e. v. sampathkumaran , physica b * 186 * - * 188 * , 525 ( 1993 ) . e. gratz , e. bauer , r. hauser , n. pillmayr , g. hilscher , h. mller , and b. barbara , j. magn . mater . * 76 & 77 * , 275 ( 1988 ) . j. sakurai , h. kamimura , and y. komura , j. magn . mater . * 76 & 77 * , 287 ( 1988 ) . d. huo , k. mori , t. kuwai , h. kondo y. isikawa , and j. sakurai , j. phys . japan * 68 * , 3377 ( 1999 ) . vol.68 no.10 , october , 1999 pp.3377 - 3382 m. oko , c. geibel , and f. steglich , phys . b * 64 * , 195107 - 1 ( 2001 ) . g. nakamoto , t. nobata , s. ueda , y. nakajima , t. fujioka , and m. kurisu , physica b * 259 * - * 261 * , 154 ( 1999 ) . k. alami - yadri , d. jaccard , and d. andreica , j. low temp . phys . * 114 * , 135 ( 1999 ) . r. casanova , d. jaccard , c. marcenat , n. hamdaoui , and m. j. besnus , j. magn . mater . * 90 & 91*,587 ( 1990 ) . o. trovarelli , c. geibel , b. buschinger , r. borth , s. mederle , m. grosche , g. sparn , and f. steglich , phys . b * 60 * , 1136 ( 1999 ) . d. andreica , k. alami - yadri , d. jaccard , a. amato , and a. schenk , physica b * 259 * - * 261 * , 144 ( 1999 ) . h. j. van daal , p. b. van aken , and k. h. j. buschow , phys . lett . a * 49 * , 246 ( 1974 ) . i. aviani , m. miljak , v. zlatic , k .- schotte , c. geibel , and f. steglich phys . rev . * b 64 * , 184438 ( 2001 ) h. wilhelm and d. jaccard , phys . rev . * b 66 * , 064428 ( 2002 ) a. demuer , a. t. holmes , and d. jaccard , j. phys . : condens . matter * 14 * l529-l535 ( 2002 ) h. wilhelm et al , science and technology at high pressure ( proc . airapt-17 ) m h manghnani , w j nellis and m f nicol ( hyderabad : universities press ) p 697 ( 2000 ) ( preprint cond - mat/9908442 ) h. wilhelm et al , preprint , cond - mat/0408280 for not too large values of @xmath60 , the thermopower at the minimum is negative ( see @xmath60=90 mev curve in fig . [ fig : theo_tep_low ] ) but if @xmath60 continues to increase the minimum of @xmath5 becomes more shallow and does not extend to negative values ( see curves in fig . [ fig : theo_tep_low ] obtained for @xmath60=100 - 120 mev ) .

we analyze quantal brownian motion in dimensions using the unified model for diffusion localization and dissipation , and feynman - vernon formalism . at high temperatures the propagator possess a markovian property and we can write down an equivalent master equation . unlike the case of the zwanzig - caldeira - leggett model , genuine quantum mechanical effects manifest themselves due to the disordered nature of the environment . using wigner picture of the dynamics we distinguish between two different mechanisms for destruction of coherence : scattering perturbative mechanism and smearing non - perturbative mechanism . the analysis of dephasing is extended to the low temperature regime by using a semiclassical strategy . various results are derived for ballistic , chaotic , diffusive , both ergodic and non - ergodic motion . we also analyze loss of coherence at the limit of zero temperature and clarify the limitations of the semiclassical approach . the condition for having coherent effect due to scattering by low - frequency fluctuations is also pointed out . it is interesting that the dephasing rate can be either larger or smaller than the dissipation rate , depending on the physical circumstances .

cond-mat9805023

classical brownian motion is described by the langevin equation @xmath1 where @xmath2 is the position of the particle , @xmath3 is the friction coefficient , and @xmath4 is a stochastic force . this equation is meaningful only in a statistical sense . the time evolution of a phase space distribution @xmath5 is obtained by solving ( [ e1 ] ) for various realizations of the stochastic potential , and then averaging over all these realizations . the stochastic potential is zero on the average and its correlations are @xmath6 typically these correlations are characterized by a ` short ' temporal scale @xmath7 and a ` microscopic ' spatial scale @xmath8 . usually , it is further assumed that higher moments are determined by gaussian statistics . the langevin description can be derived by considering a general hamiltonian of the form @xmath9 where @xmath2 and @xmath10 are the canonical variables that correspond to the distinguished degree of freedom , @xmath11 is the free motion hamiltonian , and @xmath12 are environmental degrees of freedom . the dynamical variable @xmath2 may represent the position of a large particle . in the one - dimensional version of brownian motion it may represent the position of a piston . the actual conditions for having the reduced langevin description turn out to be quite weak @xcite . the environment should consist of at least @xmath13 degrees of freedom with _ fast _ chaotic dynamics . _ fast _ implies that the classical motion is characterized by a continuous spectrum with high frequency cutoff , such that the motion of the environment can be treated adiabatically with respect to the _ slow _ motion of the particle . it is essential to assume that the following condition is fulfilled @xmath14 where @xmath7 and @xmath8 characterize the correlations of the stochastic potential which is experienced by the particle . eq.([e4 ] ) is the condition for using the white noise approximation ( wna ) . namely , the noise is characterized by its intensity @xmath15 and its temporal correlations are adequately described by the formal expression @xmath16 . without loss of generality we assume from now on the normalization @xmath17 . the correlations of the stochastic force satisfy @xmath18 in the general case , ( [ e6 ] ) is less informative than ( [ e2 ] ) . however , in case of a classical particle that experiences white noise , the additional information is not required at all ! it should be emphasized that this observation does not hold if ( [ e4 ] ) is not satisfied : if @xmath7 is larger than @xmath19 , then the particle will perform a stochastic motion that depends crucially on the `` topography '' of the stochastic potential . the classical analysis further reveals @xcite that associated with the stochastic potential there is also a dissipation effect . if the environment is characterized by either micro - canonical or canonical temperature @xmath20 , then the friction parameter will be @xmath21 thus , any generic environment , in the sense specified above , leads to the universal ohmic behavior . the motion of the particle is determined by the interplay between the friction and the noise . the friction leads to damping of the particle s velocity , while the noise pumps energy back into its motion . eventually we have diffusion with coefficient @xmath22 . if ( [ e4 ] ) is not satisfied , then we will have diffusion even in the absence of dissipation . the latter `` non - dissipative diffusion '' is characterized by the coefficient @xmath23 , where @xmath24 is the velocity of the particle . this latter type of diffusive behavior should not be confused with the generic `` dissipative diffusion '' that characterizes brownian motion . one wonders whether there is a well defined quantized version of the above langevin equation . is it possible , quantum - mechanically , to characterize a universal dynamical behavior that corresponds to classical brownian motion ? are stronger physical conditions required in order to guarantee generic behavior ? in the present paper we have no intention to give a full answer to all these questions . rather , as in previous publications @xcite , we follow the caldeira - leggett strategy @xcite . namely , we consider the motion of a particle under the influence of an effective ( non - chaotic ) bath that is composed of infinitely many oscillators . the proper model that corresponds ( classically ) to the langevin equation ( [ e1 ] ) with ( [ e2 ] ) is the dld model that has been introduced in @xcite . dld are the initials of `` diffusion localization and dissipation '' . these three effects comes out naturally from the quantum - mechanical solution of the dld model . quantum mechanically , it is convenient to use wigner function @xmath25 in order to represent the reduced probability density matrix . we can use the feynman - vernon ( fv ) formalism @xcite in order to find an expression for the propagator @xmath26 . this propagator is uniquely determined once the friction ( @xmath3 ) , the temperature of the bath ( @xmath20 ) and the spatial correlations @xmath27 are specified . the dependence of the propagator on the bath - temperature is via the appearance of a quantum - mechanical version of the noise kernel @xmath28 . the latter is related to the friction coefficient via a universal fluctuation - dissipation relation . at high temperatures the quantum - mechanical @xmath28 coincides with its classical expression . at the limit of zero temperature @xmath28 does not vanish , rather it develops a large negative tail . in the latter case , @xmath29 , the power spectrum of the noise , reflects the zero point fluctuations ( zpf ) of the environmental modes . the zwanzig - caldeira - leggett ( zcl ) model @xcite constitutes a special formal limit of the dld model . it is obtained by taking the limit @xmath30 . both the zcl model and the dld model will be generalized in this paper to allow the analysis of brownian motion in @xmath0-dimensions . a modified version of the the dld model incorporates the effect of long range ( power - law ) spatial correlations . on the other hand , we are not discussing the dld model in its full generality ( as in @xcite ) , but rather restricting ourselves to the particular circumstances that correspond in the classical limit to generic brownian motion . thus , we are considering the _ ohmic _ dld model , and further assume that ( [ e4 ] ) is satisfied . in case of the zcl model , the quantal propagator @xmath26 is a gaussian stochastic kernel . consequently the dynamics can be obtained by solving the langevin equation ( [ e1 ] ) with ( [ e6 ] ) . at low - temperatures the quantum mechanical version of @xmath28 should be used . at high temperatures @xmath28 becomes classical - like and consequently the zcl propagator coincides with its classical limit . these observations do not hold in case of the dld model @xcite . furthermore , the distinction between the quantal dld propagator and its classical limit persists even in the limit of high temperatures . it is important to define what is the meaning of ` high temperatures ' . as in the classical case , a relatively simple description of the dynamics is obtained if it possesses a markovian property . with a markovian property it is possible to obtain the long - time evolution by composing short - time evolution steps . it is also possible then to write down a corresponding master equation for the wigner function . the fv path - integral expression for the propagator contains the quantum - mechanical version of @xmath28 rather than the classical one . in order to have a markovian property one should argue that it is possible to use the wna , meaning to replace @xmath28 by the effective classical - like delta - function . still , in case of the dld model , the result of the path integration is not classical - like : the high - temperature markovian limit is not the same as the classical limit . it turns out that the quantum - mechanical condition for using the wna , thus having a markovian property , is more restrictive than ( [ e4 ] ) . for ballistic - like motion we shall see that the actual condition is , in most circumstances , @xmath31 a recent derivation of the high temperature ohmic dld model which is based on a synthetic rmt hamiltonian has been reported in @xcite . the existence of such a derivation is most encouraging since it further support the idea of having a universal description of quantal brownian motion , at least in the markovian limit . it has been speculated by the authors of the latter reference that a future extension of their formalism will lead to an agreement with the general result of @xcite . the high temperature ohmic dld model has been discussed in @xcite and has been further analyzed in @xcite . the physical picture of high - temperature dynamics will be further illuminated in the present paper . loss of coherence , or dephasing , is a central issue in the analysis of quantal brownian motion . wigner s picture of the dynamics leads to the distinction between two different mechanisms for loss of coherence . in case of zcl model one should consider the ` _ spreading mechanism _ ' . this classical - like non - perturbative mechanism is very effective in smearing away the interference pattern . in case of the dld model , coherence is much better maintained , and one should consider the perturbative ` _ scattering mechanism _ ' for dephasing . at the limit of zero temperature , the spreading mechanism is still effective in suppressing interference . one can use the langevin formalism in order to analyze the smearing of the interference pattern . it is important , however , to take into account the negative temporal - correlations of the effective noise @xcite . on the other hand , the analysis of low - temperature dephasing in case of the ` scattering mechanism ' is a quite subtle issue , that constitutes a main concern of the present paper . the langevin formalism is no - longer applicable , and the lack of a markovian property enforce a semiclassical approach . the semiclassical approach has a further advantage : it is possible to go beyond the analysis of a simple ballistic - like brownian motion and to analyze other types of transport . we shall distinguish between ballistic , diffusive and chaotic motions through cavities . we shall derive various results and contributions to the dephasing rate in the various temperature regimes and depending on the physical circumstances . some of these results coincide with similar computations that are related to electrons in metal @xcite . an important question is whether the scattering mechanism is still effective in suppressing interference at the limit of zero temperature @xcite . it turns out that our semiclassical approach , in spite of its other advantages , has a limited range of validity . it can be trusted if the kinetic energy of the particle is sufficiently large . for ballistic - like motion large energy means @xmath32 in analogy with ( [ e8 ] ) . the latter condition can be cast into the more suggestive form @xmath33 , where @xmath34 is the de - broglie wavelength of the particle . the large - energy condition is obviously satisfied in case of the zcl model ( @xmath30 ) . the large energy condition may not be satisfied in case of the dld model , and consequently the semiclassical result should be modified . in particular , in case of a low - temperature thermal motion , the contribution of the zero point fluctuations ( zpf ) to the dephasing rate should be excluded . we consider a bath that consists of infinitely many oscillators whose canonical coordinates are @xmath12 . the bath hamiltonian is @xmath35 in case of the dld model the interaction of the particle with the oscillators is described by @xmath36 where @xmath37 is the location of the @xmath38 oscillator , @xmath39 describes the interaction between the particle and the @xmath38 oscillator , and @xmath40 are coupling constants . it is assumed that the function @xmath41 depends only on @xmath42 . the range of the interaction will be denoted by @xmath8 . the oscillators are distributed uniformly all over space . locally , the distribution of their frequencies is ohmic . namely , @xmath43 it is useful to define the spatial auto - correlation function @xmath44 without loss of generality @xmath41 is normalized such that @xmath17 . for example , we may consider a gaussian @xmath41 for which @xmath45 the @xmath0-dimensional ft of @xmath46 will be denoted by @xmath47 . the mode - density ( after angular integration ) is @xmath48 . in general , we shall assume that @xmath49 where @xmath8 characterize the spatial scale of the correlations , and @xmath50 is a dimensionless constant . in case of the short range gaussian - type correlations ( [ e13 ] ) , the parameter @xmath51 equals simply to the dimensionality @xmath0 . for long range power - law interaction it may be less than @xmath0 , possibly negative . in a moment we shall argue that in order to have a well defined model we must have @xmath52 . therefore only @xmath53 is meaningful . the regime @xmath54 is well defined but it requires special treatment since @xmath55 diverges . the formal limit @xmath30 corresponds to the zcl model . the zcl model describes interaction of a particle with environmental modes whose wavelength is much larger compared with the distances that are explored by the particle . consequently , the interaction with the @xmath38 field - mode is approximated by a linear function . the interaction term in the hamiltonian that _ defines _ the zcl model is of the following form ( here generalized to @xmath0-dimensions ) : @xmath56 the unit vectors @xmath57 are assumed to be distributed isotropically . the distribution of the oscillators with respect to their frequencies is assumed to be ohmic , as in ( [ e11 ] ) with the @xmath58 omitted . the classical analysis of the @xmath0-dimensional dld model constitutes a trivial generalization of the one - dimensional dld model that has been considered in @xcite . equation ( 3.10 ) in the latter reference should be replaced by the following expression for the friction : @xmath59 where @xmath60 is the velocity of the particle . for ohmic bath @xmath61 can be replaced by @xmath62 . one obtains then @xmath63 . obviously , in order to obtain a finite result it is crucial to have @xmath52 . recall that by convention we use the normalization @xmath17 . thus , the generalized version ( [ e14 ] ) of the dld model is well - defined provided @xmath53 . the reduced motion of the particle will be described by the langevin equation ( [ e1 ] ) , where the noise satisfies ( [ e2 ] ) , with @xmath64 . for the realization of a classical trajectory , the global functional - form of @xmath46 is of no importance . only the force correlations ( [ e6 ] ) are important . the latter correlations are well defined as long as @xmath52 , and by our convention they are equal @xmath28 . the parameter @xmath8 is insignificant in the classical analysis of the ohmic dld model , and therefore the classical description of brownian motion is the same as in the case of the zcl model . quantum mechanically , the reduced dynamics of the system may be described by the propagator @xmath26 of the probability density matrix . for sake of comparison with the classical limit one uses wigner function @xmath25 in order to represent the latter . using feynman - vernon ( fv ) formalism @xcite , an exact path - integral expression for this propagator is obtained . the fv expression is a double sum @xmath65 over the path variables @xmath66 and @xmath67 . it is convenient to use new path variables @xmath68 and @xmath69 , and to transform the @xmath70 integral into the form @xcite @xmath71 \ \ \ \ , \end{aligned}\ ] ] where @xmath72 $ ] is a real functional , which is defined by the expression : @xmath73 = \int { \cal d}r \ \ \mbox{e}^{i\frac{1}{\hbar}(s_{free}+s_f ) } \ \ \mbox{e}^{-\frac{1}{\hbar^2}s_n } \ \ \ \ \ .\end{aligned}\ ] ] the @xmath74 integration is unrestricted at the endpoints , and the free action functional is @xmath75=-m\int_0^td\tau\ \ddot{r}r$ ] . the action functional @xmath76 $ ] corresponds to the friction , and the functional @xmath77 $ ] corresponds to the noise . the explicit expression for the friction action functional , assuming an ohmic bath , is @xmath78 \ = \ \eta \int_0^t d\tau \ \nabla w({\mathbf{r } } ) \cdot \dot{{\mathbf{r}}}\end{aligned}\ ] ] the general expression for the noise functional is @xmath79=\frac{1}{2 } \int_0^t\int_0^t dt_1 dt_2 \ \phi(t_2{-}t_1 ) \ \times \ \ \\ \ \ \ \ [ w({\mathbf{x}}_2''{-}{\mathbf{x}}_1'')+ w({\mathbf{x}}_2'{-}{\mathbf{x}}_1')-2w({\mathbf{x}}_2''{-}{\mathbf{x}}_1 ' ) ] \label{e20 } \end{aligned}\ ] ] where @xmath80 is a short notation for @xmath81 . the power spectrum of the noise @xmath29 is the ft of the noise kernel @xmath28 . for ohmic bath @xmath82 the zcl version for @xmath83 and @xmath84 is obtained by taking the limit @xmath30 which is equivalent to the formal substitution @xmath85 . the power spectrum ( [ e21 ] ) of the quantum mechanical noise is the same as that of a classical white noise in the restricted frequency regime @xmath86 , where @xmath87 . if the temperature is high enough , such that @xmath88 , then the power spectrum is essentially classical . at lower temperatures the power spectrum contains a non - trivial frequency zone @xmath89 , where one may use the approximation @xmath90 . this component of the power spectrum does not vanish in the limit of zero temperature . it corresponds to the `` zero point fluctuations '' ( zpf ) of the environmental modes . the temperature becomes effectively zero if @xmath91 . at intermediate temperatures , namely @xmath92 , it is convenient to write the power spectrum as a sum @xmath93 , where @xmath94 corresponds to the zpf and @xmath95 is the excess thermal noise : @xmath96 the propagator ( [ e17 ] ) possess a markovian property if it is legitimate to make the wna . the actual condition for having effectively white noise will be discussed in a later section . at low temperatures it is essential to consider the non - trivial nature the quantal noise in order to obtain the proper quantum - mechanical dynamical behavior @xcite . in the absence of noise and friction , the free - motion propagator of wigner function is the same as in the classical limit . namely , @xmath97 for simplicity , but without loss of generality , we shall consider whenever possible one - dimensional motion . in case of the zcl model , both @xmath83 and @xmath84 are quadratic forms , and therefore the propagator is a gaussian kernel . it follows that the dynamics of wigner function is purely stochastic , and it can be described by langevin equation ( [ e1 ] ) with ( [ e2 ] ) and ( [ e21 ] ) . in this sense , there are no genuine quantum mechanical effects that are associated with brownian motion , as long as the zcl model is used for its description . the situation is quite different in case of the dld model . here the propagator is non - gaussian , and , in general , it may have non - stochastic features . in the present section we shall restrict ourselves to the high temperature regime , where the white noise approximation ( wna ) , namely @xmath98 can be applied . whenever the wna applies , the zcl propagator becomes identical with the classical propagator for damped motion , namely @xmath99 in contrast , in case of the dld model , the high temperature limit of the propagator _ does not coincide _ with its classical limit . an explicit calculation , as in @xcite , gives the following expression : @xmath100 this expression is valid also for the generalized dld model provided @xmath101 . the precise value of the parameter @xmath8 is defined here via @xmath102 . above , @xmath103 is a smooth gaussian - like kernel that has unit - normalization . its spread in phase space is characterized by the momentum scale @xmath104 , and by an associated spatial scale . the symbol @xmath105 stands for convolution . thus , the classical propagator is smeared on a phase - space scale that correspond to @xmath106 and there is an additional un - scattered component that decay exponentially and eventually disappears . the structure of the propagator is illustrated in fig.[f_propg ] . the significance of this structure will be discussed in the next section . = 3.4 in to write down an explicit expression for the propagator is not very useful . rather , it is more illuminating to follow the standard procedure @xcite and to write an equivalent master equation . this is possible since at high temperatures the path - integral expression possess a markovian property . namely , since both @xmath83 and @xmath84 are local functionals of the paths , it is possible to regard the finite - time propagation as the convolution of ( infinitesimal ) short - time kernels . the derivation of the master equation is explained in appendix b. the final result is @xmath107 \ \rho\end{aligned}\ ] ] the friction kernel is defined as follows @xmath108 and the noise kernel is @xmath109 \label{e28.1}\end{aligned}\ ] ] both @xmath110 and @xmath111 are smooth gaussian - like scaling functions , properly normalized to unity . if wigner function does not possess fine details on the momentum scale @xmath104 , then the convolution with @xmath112 can be replaced by multiplication with @xmath113 , and the convolution with @xmath114 can be replaced by @xmath115 . these replacements are formally legitimate both in the classical limit @xmath116 , and in the zcl limit @xmath117 . one obtains then the classical fokker planck equation @xmath118 \ \rho\end{aligned}\ ] ] the same observation can be done by inspection of ( [ e22 ] ) . namely , if the propagator acts on a smooth wigner - function , ( no features on momentum scale @xmath104 ) , then the second term becomes vanishingly small , while the first term becomes effectively classical . wigner function may have some modulation on a fine scale due to an interference effect . the standard text - book example of a two slit experiment will be analyzed below , where the interference pattern has the momentum scale @xmath119 . see fig.[f_slits ] . @xmath120 is the distance between the slits . in this section we shall explain the possible mechanisms that lead to the disappearance of such interference pattern . in view of the semiclassical point of view of the next section , we use the term `` dephasing '' in order to refer to this disappearance effect . in the present section , as in the previous one , we are still limiting ourselves to the high temperature regime . = 3.4 in in case of the zcl model , the propagator is the same as the classical one , and therefore we may adopt a simple langevin picture in order to analyze the dephasing process . alternatively , we may regard the dephasing process as arising from gaussian smearing of the interference pattern by the propagator . in case of the dld model we should distinguish between two possible mechanisms for dephasing : * scattering ( perturbative ) mechanism . * spreading ( non - perturbative ) mechanism . actually , it is better to regard them as mechanisms to maintain coherence . the first mechanism to maintain coherence is simply not to be scattered by the environment . the second mechanism to maintain coherence is not to be smeared by the propagator . the first mechanism is absent in case of the zcl model . we shall see shortly that the first mechanism is much more effective in maintaining coherence . the notions perturbative and non - perturbative are used in order to suggest a connection with an earlier work @xcite . the smearing effect can be regarded as arising from accumulation of many small - momentum scattering events . a conventional perturbative approach can not be applied in order to analyze such dephasing process , even in the limit of very faint noise . having in mind the specific example of a two - slit experiment , we should distinguish between two cases . if the environment is characterized by a spatial correlation scale @xmath8 that is much larger than @xmath120 , then we can ` forget ' about the dld model and just use the zcl model . the relevant mechanism for destruction of coherence in this case is the spreading mechanism . see detailed analysis in the next paragraph . if the environment characterized by a spatial correlation scale @xmath121 then a totally different picture emerges . now , wigner function contains a modulation on a momentum scale much finer than @xmath104 . this modulation is not affected by the friction , but its intensity decays exponentially in time . this is based on inspection of either the propagator ( [ e22 ] ) , or the equivalent master equation ( [ e25 ] ) . in the latter case note that the convolution with @xmath114 can be replaced by multiplication with @xmath122 . the decay rate is @xmath123 it is a universal result for the dephasing rate due to the ` scattering mechanism ' , since it does not depend on details of the quantum - mechanical state involved . however , the validity of this result is restricted to the high temperature regime , where the white noise approximation ( wna ) can be applied . for completeness we turn back to the detailed analysis of dephasing due to spreading . this mechanism leads to non - universal results , since the calculation of the dephasing time depends on actual details of the interference pattern . for concreteness we consider the simplest case of a two - slit interference experiment . the wavepacket that emerges from the two slits is assumed to be a superposition @xmath124 , where @xmath125 is the momentum of the incident particle , and @xmath126 is a radial envelope function . the corresponding wigner function is @xmath127 the 1-slit wigner function is multiplied by @xmath128 , and therefore it is natural to regard the 2-slits wigner function as composed of partial - wavepackets , each characterized by a different transverse momentum . by definition the partial - wavepacket @xmath129 equals @xmath130 for @xmath131 and equals zero otherwise . each partial wavepacket represents the possibility that the particle , being scattered by the slits , had acquired a transverse momentum which is an integer multiplet of @xmath132 . note that the corresponding angular separation is @xmath133 , as expected . the associated spatial separation is @xmath134 where @xmath135 is the time up to the screen . it is important to distinguish between the `` preparation '' zone @xmath136 which is excluded from our considerations , and the far - field ( franhoufer ) zone @xmath137 where @xmath138 and consequently the individual partial - wavepackets can be resolved . due to the noise the interference pattern is smeared on a momentum scale @xmath139 , and on a corresponding spatial scale @xmath140 . the interference pattern disappears once @xmath141 or equivalently @xmath142 . this leads to the following expression for the dephasing time : @xmath143 comparing with ( [ e28 ] ) we observe that indeed the smearing mechanism is very effective in suppressing the interference pattern . the result is also non - universal since it depends on details of the interference pattern . in some other circumstances , where wigner function is characterized by a different type of interference pattern ( notably the case of `` vertical '' interference pattern ) , the dephasing time may be proportional to some fractional power of the noise intensity @xcite . in general , in the full analysis of the smearing process , one should take into account the effect of friction . in the above example this effect has been neglected . the analysis of dephasing in case of the zcl model is easily extended to the low - temperature regime . langevin formalism is still applicable , provided the appropriate @xmath28 is being used . the smearing of the interference pattern can be analyzed as in the previous paragraph . at the limit of zero temperature one should take into account the negative correlations of the noise @xcite . for the ballistic - like motion that has been considered above , the smearing scale @xmath144 is proportional , at zero temperature , to @xmath145 rather than to @xmath146 , leading to an anomalous ( non - exponential ) dephasing factor @xcite there is a totally different approach to analyze dephasing , that is based on the semiclassical point of view . the advantage of the semiclassical approach is the ability to pursue the study of dephasing to the low - temperature regime as well as to circumstance in which the simple ballistic - like brownian motion scenario is not applicable . specific examples that will be discussed later are the transmission via either chaotic or diffusive cavity . from now on we regard the two - slit experiment as a special case of transport problem . see fig.[f_trajc ] . the probability to cross the obstacle , or more generally to be transmitted via some cavity , can be written as a ( double ) sum over classical trajectories . namely , @xmath147{-}s[{\mathbf{x}}_b]}{\hbar } + i\frac{s_f[{\mathbf{x}}_a,{\mathbf{x}}_b]}{\hbar } -\frac{s_n[{\mathbf{x}}_a,{\mathbf{x}}_b]}{\hbar^2 } \right ) \end{aligned}\ ] ] where @xmath148 $ ] is the classical action for the classical trajectory @xmath149 , and @xmath150 is the corresponding classical amplitude . each pair @xmath151 of classical trajectories constitutes a stationary - phase point of the exact path - integral expression . however , the notation is somewhat misleading since once the friction functional is switched on , @xmath152 and @xmath153 no longer can be considered independent indexes . in particular , strictly speaking , the diagonal terms are no longer truly diagonal . however , here comes a very important observation . in case of the dld model , assuming that @xmath149 and @xmath154 are well separated with respect to the microscopic scale @xmath8 , one has @xmath155=0 $ ] . thus we conclude that friction has no effect on the interference part of the sum ( [ e31 ] ) , in consistency with our discussion of the scattering - mechanism in the previous section . on the other hand , friction will have an effect on the diagonal terms of the sum , as required by the correspondence principle . it should be emphasized that in case of the zcl model , friction may have a non - negligible effect also on the interference terms , again in consistency with our discussion of the smearing mechanism . = 1.4 in the suppression of an off - diagonal term in ( [ e31 ] ) is given by the dephasing factor @xmath156/\hbar^2)$ ] . using the definitions of appendix d it is possible to transform expression ( [ e20 ] ) into an integral over the fourier components of the motion involved . consequently , in most cases of interest , the dephasing factor can be cast into the form @xmath157 , where @xmath158 the domain of integration is illustrated in fig.[f_kwplan ] . the power spectrum @xmath159 of the motion is calculated in appendix d for ballistic , chaotic and diffusive motion . various results that can be derived by using the above expression will be presented in later sections . = 1.3 in one should be very careful in the physical interpretation of @xmath157 . if we have two unrelated trajectories @xmath151 , and we have also static disorder , then we will have `` statistical '' dephasing that reflects the effect of averaging over realizations of the disorder . consider for example a two - slit experiment : the interference pattern on the screen will be distorted for any particular realization of a static disorder , and will be washed away completely upon averaging over different realizations . this mechanism is non - effective if @xmath151 are related by time reversal , for which static disorder gives @xmath160=0 $ ] . still some suppression is expected also in the latter case , via the classical - like amplitudes @xmath161 . in case of dynamical environment the role of elastic scattering is taken by inelastic scattering events . therefore we have genuine dephasing , irrespective of whether @xmath151 are related by time reversal . in some typical circumstances , the dephasing factor can be re - interpreted as giving the probability to `` leave a trace '' in the environment ( see appendix c ) . we take now the limit of zero temperature . assuming for simplicity that @xmath151 are related by time reversal , it follows from the definition of the influence functional that @xmath156/\hbar^2)$ ] can be interpreted as the probability to excite at least one of the oscillators along the way . up to now we have encountered , depending on the physical circumstances , two possible interpretations for the dephasing factor . the `` statistical '' interpretations holds in case of static disorder , while the `` leaving - trace '' interpretation holds in typical cases of a dynamical environment . let us keep @xmath162 and take the limit @xmath163 , where @xmath164 is the available energy of the particle . in order to make the following argumentation more illuminating , one may assume that the particle has a small but finite kinetic energy @xmath164 , and that all the oscillators have relatively large frequencies , such that @xmath165 . consequently , all the scattering events are elastic . on the other hand @xmath166 has a non - vanishing value , which implies that the off - diagonals terms are being suppressed . obviously , the `` leaving - trace '' interpretation of the dephasing factor no longer holds . one wonders whether @xmath167 acquires , under such circumstances , a somewhat different physical meaning , as in the case of static disorder . maybe it gives now the probability to be scattered elastically ( rather than inelastically ) along the way , thus leading to dephasing of the `` statistical '' type . a perturbative treatment of the scattering process leads to the conclusion that the probability to be elastically scattered by a zero temperature oscillator is proportional to @xmath168 . this is because the elastic scattering off an oscillator involves a second - order process of virtual emission followed by absorption of a quanta @xmath169 . at the same time @xmath160 $ ] is proportional to @xmath170 to leading order . therefore , @xmath167 can not have the desired physical significance . we therefore must conclude that something goes wrong with the present semiclassical approach once low energy particle is concerned . we shall come back to this point later on . as a first step in the application of the semiclassical approach , it is interesting to recover results ( [ e28 ] ) and ( [ e30 ] ) for the dephasing rate . these results hold at high temperatures , such that the wna is applicable . the noise functional in such circumstances takes the following simple form @xmath171 \ = \ 2\eta k_bt \int_0^t \ [ w(0){-}w({\mathbf{r}}(t ' ) ) ] \ dt ' \end{aligned}\ ] ] where @xmath172 . if the two trajectories are well separated with respect to the microscopic scale @xmath8 , then one indeed recovers the universal result ( [ e28 ] ) . recall that @xmath173 . in the other extreme limit @xmath30 , that corresponds to the zcl model , one obtains @xmath174 \ = \ 2\eta k_bt \int_0^t \ { \mathbf{r}}(t')^2 \ dt ' \ \equiv \ 2\eta k_bt \ r_{\perp}^2 \cdot t\end{aligned}\ ] ] in case of the two - slit experiment @xmath175 , and one recovers ( [ e30 ] ) . obviously , it is not a universal result . for example , if the separation between the two trajectories grows linearly , then the dephasing time will be proportional to @xmath176 rather than to @xmath177 . the generalized dld model with @xmath178 , requires special care since @xmath55 diverges . we are still considering high temperatures , such that the wna is applicable . using the notations of appendix d expression ( [ e32 ] ) can be cast into the form @xmath179 \ = \ \nu t \int w({\mathbf{r } } ) p({\mathbf{r } } ) d{\mathbf{r } } \ = \ \nu t \int_0^{\infty } g(k)dk \ p(k ) \end{aligned}\ ] ] the latter expression converges for any @xmath53 . in the zcl limit , where @xmath8 is much larger than the average transverse distance between the trajectories , one may use the small-@xmath180 approximation of appendix d , and then ( [ e33 ] ) is recovered . in the dld case , where @xmath8 is a very small scale , the integration should be split into the domains @xmath181 and @xmath182 . the wavenumber @xmath183 is associate with the transverse distance @xmath184 between the two trajectories . namely , @xmath185 note that ergodic motion refers to either chaotic or diffusive trajectories with @xmath186 . again we use the notations of appendix d. for @xmath101 the integration is dominated by the upper cutoff @xmath187 . one recovers then the standard result ( [ e28 ] ) . on the other hand , for @xmath178 , the integration is dominated by the @xmath188 modes . the wavelength of these modes is like the transverse distance between the trajectories . one obtains then @xmath189 \ = \ 2\eta k_b t \ \ell^2 \left ( \frac{\ell}{r_{\perp}}\right)^{\sigma } \cdot t \ \ \ \ \ \mbox{wna for $ -2<\sigma<0$}\end{aligned}\ ] ] for non - ergodic diffusive motion @xmath184 depends on @xmath146 and therefore the dephasing factor is of the type @xmath190 . in all other cases we have a simple exponential dephasing factor . obviously ( [ e34 ] ) is a special case of the general formula ( [ e37 ] ) . note that ( [ e37 ] ) should be multiplied by @xmath191 in order to become a proper expression for @xmath84 . by comparing ( [ e37 ] ) to ( [ e34 ] ) one can reveal the actual condition for the validity of the wna . within the interval @xmath192 , the power spectrum of the motion occupies a frequency range @xmath193 , where the effective cutoff is @xmath194 as long as @xmath195 one can use the wna in estimating the integral in ( [ e37 ] ) . one observes that under such circumstances ( [ e37 ] ) reduces , after multiplication by @xmath191 , to ( [ e34 ] ) . as the temperature becomes low , such that @xmath196 , the dephasing rate becomes larger than the value which is predicted by the wna . this is due to the `` zero point fluctuations '' ( zpf ) in the frequency zone @xmath197 . the temperature becomes effectively zero if @xmath198 . time longer than @xmath146 is required in order to resolve such low temperatures . one can use then the zero temperature limit of @xmath28 for which @xmath199 . this power spectrum corresponds to the zpf of the environmental modes . for either ballistic or chaotic motion , the integral ( [ e37 ] ) is dominated by @xmath200 modes that are concentrated below the curve @xmath201 . one obtains @xmath202 similarly , for diffusive motion the integral is dominated by @xmath200 modes that are concentrated below the curve @xmath203 . one obtains @xmath204 in both cases most of the contribution comes from modes with large wavenumber , namely @xmath205 . the validity of the present semiclassical approach , which is based on the stationary - phase approximation is limited . common wisdom @xcite is that the applicability of the semiclassical approach is restricted to circumstances such that the energy transfer between the particle and the environment is much smaller than the particle s available energy . technically , it is equivalent to the assumption of uncoupled wave - equations @xcite . the _ coupled _ wave - equations for the particle - oscillators system can be _ uncoupled _ provided certain conditions are satisfied . then we can treat the particle as moving with constant velocity @xmath24 and solve for the oscillators . it turns out that this reduction requires the assumption of small energy transfer . therefore , one should anticipate problems once oscillators with @xmath206 larger than @xmath164 are involved . in the latter case , there is no justification to think of the particle as decoupled from the bath , moving with some constant velocity , capable of exciting oscillators along the way . therefore the corresponding factor @xmath167 loses its physical significance . from the above considerations it follows that a reasonable condition for the validity of our semiclassical approach is @xmath207 on physical grounds this is the condition for being able to leave a trace `` along the way '' . the semiclassical significance of this condition can be further illuminated . for ballistic - like motion it is equivalent to the condition of small momentum transfer @xmath208 . this is precisely the condition for the applicability of the semiclassical methods for the scattering of the particle by the oscillators . more generally , one can define a quantum mechanical version of @xmath159 as the ft of the correlator of the heisenberg - picture operator @xmath209 . the quantal @xmath159 coincides with the classical @xmath159 only for @xmath210 . larger classical frequencies are not supported by the effectively banded energy spectrum . our practical conclusion is that the contribution of zpf should be included if non - thermal motion with large energy is concerned . it should be excluded in case of thermal motion with @xmath211 . the possible contribution of zpf to the dephasing has been discussed in the previous section . now we are interested in the the thermal noise contribution ( tnc ) . the tnc can be calculated using ( [ e37 ] ) with @xmath29 replaced by @xmath95 . see ( [ e42 ] ) for the definition of the latter . as the temperatures is being raised various regimes are encountered : @xmath212 the time scale @xmath213 corresponds to the transverse distance @xmath184 of ( [ e35 ] ) . it equals @xmath214 for ballistic motion , and @xmath215 for ergodic motion . in case of non - ergodic diffusive motion , where @xmath216 , the extremely low temperature regime is absent . at high temperatures we can use the wna as discussed in previous sections . the low and the extremely low temperature regimes are discussed in the next two paragraphs . when considering _ extremely low _ temperatures it is useful to define the critical exponent @xmath217 for ballistic motion , and @xmath218 for diffusive motion . if @xmath219 the integration in ( [ e37 ] ) is dominated by @xmath220 and one obtains the result @xmath221 for @xmath222 the integration is dominated by @xmath223 and one obtains the result @xmath224 in the formulas above @xmath225 is a numerical factor of order unity . note that in both cases , disregarding possible contribution of zpf , the dephasing rate is proportional to @xmath226 . this should be contrasted with the high temperature behavior where , disregarding the special case of non - ergodic diffusive motion , the dephasing rate is proportional to @xmath20 . at _ low _ ( but not extremely low ) temperatures it is useful to define a temperature dependent wavenumber as follows : @xmath227 for @xmath228 the integration in ( [ e37 ] ) is dominated by @xmath220 , and one can use the wna result ( [ e36 ] ) . thus , for @xmath228 , in the absence of zpf contribution , we can trust the wna at both high and low temperatures , and an actual crossover is expected only when extremely low temperatures are involved . for @xmath229 the integration is dominated by @xmath230 , and one obtains the result @xmath231 if @xmath222 the integration is dominated by @xmath223 and one obtains again ( [ e46 ] ) . thus , for @xmath229 , there is a non - trivial low - temperature regime where the tnc to the dephasing rate is proportional to a non - universal power of @xmath20 . below this intermediate temperature - regime the tnc is proportional to @xmath226 . above this intermediate temperature - regime we can trust the wna and the dephasing rate is proportional to @xmath20 . we shall consider in this section quantal brownian motion in @xmath232 dimensions , that is described by the the dld model with short range spatial correlations . for concreteness let us assume that the fluctuations of the effective stochastic potential satisfy : @xmath233 if we are interested in the dynamics over a _ finite _ time interval @xmath146 , then obviously all the fourier components in the frequency regime @xmath234 will have the same effect as static disorder . let us denote by @xmath235 the `` static '' component of the effective stochastic potential . the variance of @xmath236 is determined by the product @xmath237 . recall that @xmath238 . consequently we have @xmath239 where @xmath240 . we would like to find out whether this effective static disorder will manifest itself . similar question has been raised in @xcite . one should not take for granted that the effect of low - frequency fluctuations is completely masked by the the incoherent effect of the high frequency modes . the first obvious step is to calculate the `` statistical '' dephasing rate due to the presence of the ( effective ) disorder . we can use ( [ e37 ] ) with the formal substitution @xmath241 , thus obtaining @xmath242 for ballistic - like motion one obtains @xmath243 in order to have a non - vanishing effect we should have @xmath244 . this condition can be cast into the form @xmath245 where @xmath246 is the mean free path . still another way to express this condition is @xmath247 , where @xmath248 is given by ( [ e28 ] ) . on the over hand , the actual dephasing time should satisfy @xmath249 , else coherent effects due to the scattering by the ( effective ) disordered potential will not manifest themselves . thus , we come to the conclusion that the following condition should be satisfied in order to have manifestation of coherent effects due to scattering by the effective static disorder : @xmath250 obviously , this condition can be satisfied only at the low temperature regime . it is quite striking that the friction coefficient @xmath3 is not affected by quantum mechanical effects . having the boltzmann picture in mind it is not anticipated that all the quantum mechanical scattering events will conspire to give the classical result . still , this conclusion follows from the fv - formalism quite easily . in order to have a clear physical picture , let us consider the time evolution of a gaussian wavepacket using either the propagator ( [ e24 ] ) or the equivalent master equation ( [ e25 ] ) . clearly , the dissipation rate @xmath251 is the same as in the classical picture , with small transient corrections . thus , also in the quantum - mechanical picture , the damping process is characterized by the time constant @xmath252 . consider a ballistic - like motion with non - thermal energy , such that @xmath253 . consequently the equivalent condition ( [ e41 ] ) is satisfied , and we can trust semiclassical considerations as far as dephasing is concerned . we can cast the universal wna result ( [ e28 ] ) and the zpf result ( [ e39 ] ) into the form @xmath254 where @xmath255 is the thermal wavelength , and @xmath256 is the de - broglie wavelength . we see that for the above _ non - thermal motion _ the dephasing time is always shorter than the damping time . the dephasing rate is linear in @xmath20 at high temperature , and saturates at low temperatures . for _ thermal motion _ the latter statement is no - longer true . at the low temperature regime the zpf contribution should be excluded , and therefore the dephasing rate goes to zero , while the damping time remains finite . one should distinguish between various physical circumstances . at extremely low temperatures , assuming ergodic motion , the dephasing rate is proportional to @xmath226 . at higher temperatures we can trust the wna provided @xmath228 . otherwise , there is an intermediate low - temperature regime where the dephasing rate is proportional to a non - trivial power of @xmath20 . in this paper we have introduced a systematic derivation of a general formula for the dephasing rate , eq.([e37 ] ) , that holds in all physical circumstances , including the zcl limit . expressions similar to eq.([e37 ] ) , that incorporate integration over the @xmath200 environmental modes , are encountered frequently in the literature , starting from the well known work in @xcite . whenever diffusive motion of electrons is concerned ( see appendix e ) , our expressions agree with well know results @xcite . however , most publications avoid a straightforward application of the fv formalism , and introduce in some stage heuristic considerations in order to obtain a convergent result . there is a `` zoo '' of cutoffs that are introduced in performing the @xmath200 integration , some of them are questionable . for example , it is customary to take @xmath166 itself as a lower cutoff for the @xmath257 integration . the upper cutoff is sometimes @xmath258 , sometimes fermi energy , sometimes the kinetic energy that corresponds to the drift velocity , sometimes the inelastic scattering rate , and so on . the role of fermi statistics in the determination of the various cutoffs is usually left unclear . there are similar ambiguities in the determination of the the proper cutoffs for the @xmath180 integration . our derivation has led to a proper definition for the power - spectrum of the motion , and has made it unnecessary to introduce ad - hoc cutoffs into the calculations . furthermore , in the analysis of dephasing at the limit of zero temperature , we were successful in reducing our considerations to the level of `` one particle physics '' , thus avoiding a complicated discussion of the role played by pauli exclusion principle . we first cite the well known identity @xmath259 \delta(x - x_0 ) \end{aligned}\ ] ] both sides of this identity should be interpreted as kernels of operators . when applied to wave - functions , the left hand side corresponds to the free - motion propagator , and the second term in the right hand side corresponds to the free - motion hamiltonian . we shall explain now how to derive the following related identity : @xmath260 \delta(r - r_0)\delta(r - r_0 ) \end{aligned}\ ] ] both sides of the latter identity should be interpreted as kernels of operators that operate on phase - space functions . we start the derivation by writing an equality that follows from the first identity via simple replacements : @xmath261 \delta(x_{\pm}\mp{\scriptstyle\frac{1}{2}\displaystyle}\eta\delta t ) \end{aligned}\ ] ] upon multiplication it follows that @xmath262 \left(\delta(x_{+})-\frac{1}{2}\eta\delta t\delta'(x_{+})\right ) \cdot \left[1-\frac{i}{2}\delta t \partial_{-}^2\right ] \left(\delta(x_{-})+\frac{1}{2}\eta\delta t\delta'(x_{+})\right)\end{aligned}\ ] ] simplification of the right hand side gives @xmath263 \delta(x_{+ } ) \delta(x_{- } ) \end{aligned}\ ] ] the replacements @xmath264 and @xmath265 , followed by the transformation to the variables @xmath266 and @xmath267 , gives the desired result . note that the derivation holds also in the case where @xmath3 is replaced by some function of @xmath268 . wigner function @xmath25 is the fourier transform of the reduced probability density matrix @xmath269 in the variable @xmath270 . the path integral expression for the kernel @xmath271 is the type @xmath70 with an obvious endpoint conditions , and its integrand is essentially the same as in ( [ e18 ] ) . see @xcite for more details . the infinitesimal - time kernel equals simply to the integrand of the path integral expression , namely @xmath272 \delta(r - r_0)\delta(r - r_0 ) \end{aligned}\ ] ] where the second line is obtained by employing the identity of appendix a. thus , to leading order , the kernel @xmath273 has the same effect as operating with a differential operator of the type @xmath274 . consequently , we find that @xmath269 satisfies a master equation of the form @xmath275 . this master equation is easily transformed into the wigner representation , where any derivative @xmath276 is replaced by multiplication with @xmath277 . similarly , any multiplication by a ( real symmetric ) function @xmath278 is transformed into a convolution with a ( real symmetric ) kernel @xmath279 . for the convolution @xmath280 we use the notation @xmath281 . a kernel @xmath279 and a function @xmath278 are related by a fourier transform with the convention @xmath282 in the above expression we have modified the standard ft convention , by including the factor @xmath283 . this has been done in order to have properly normalized kernels , with the measure @xmath284 rather than @xmath285 . note also that the ft of @xmath286 equals to the @xmath287 derivative of the kernel @xmath112 that corresponds to the real symmetric function @xmath288 . consider the interference contribution of two trajectories @xmath151 that are related by time reversal . assuming short range interaction with the environmental modes we can prove that dephasing is related to the probability for leaving a trace in the environment . this statement is true at any temperature . `` short range interaction '' means that @xmath101 and that @xmath8 is a small scale . `` leaving a trace '' means that at least one of the oscillators has changed its quantum - mechanical state . it follows that under such circumstances dephasing - rate is equal to the inelastic scattering rate . recall the definition of the influence functional . for simplicity we consider first a zero - temperature bath , meaning that all the bath - oscillators are initially in the ground state . these oscillators are driven by the motion of the particle . the excited states of the bath will be denoted by @xmath289 . the evolution operator of a driven oscillator will be denoted by @xmath290 $ ] . @xmath291 \ \equiv \ \sum _ { \{n_{\alpha}\ } } \ \left\langle \{n_{\alpha}\ } \left| \prod_{\alpha } u_{\alpha}[{\mathbf{x}}_b ] \right| \{0\ } \right\rangle \left\langle \{n_{\alpha}\ } \left| \prod_{\alpha } u_{\alpha}[{\mathbf{x}}_a ] \right| \{0\ } \right\rangle^{\star } \end{aligned}\ ] ] if @xmath151 are related by time reversal , then one can consider only those oscillators that are located `` along the loop '' . each oscillator along the loop is characterized by its natural frequency @xmath206 and by the time @xmath292 at which @xmath293 . consequently we have the following expression for the influence functional : @xmath294 \ = \ \sum_{\{n_\alpha\ } } \ \mbox{e } ^{-i\sum_{\alpha}n_{\alpha}\omega_{\alpha}\cdot(2t_{\alpha}{-}t ) } \ \ p(\{n_\alpha\}|\{0\})\end{aligned}\ ] ] where the excitation probability is @xmath295=p(\{0\}|\{0\})$ ] . indeed , the summation in ( [ e201 ] ) contains excitations for which @xmath296 , excitations for which @xmath297 and so on . let us consider all the excitations that involves @xmath206 in the range @xmath298 $ ] . each of these excitations contributes the same @xmath299 , but with a different phase factor @xmath300 . by construction of the dld model ( eq.([e11 ] ) ) , the summation over the phase factors will lead to a zero contribution . the meaning of `` zero contribution '' is as follows : either we average over realizations of @xmath301 , or else we recall that the influence functional appears inside a path - integral expression . the summation over paths will have the same effect like an averaging procedure , and therefore we will have indeed a `` zero contribution '' in the sense . similar argumentation applies to other subsets of excitations . if the bath is initially in thermal equilibrium , rather than in zero temperature , than the relation @xmath302=p(\{0\}|\{0\})$ ] can be generalized into @xmath303 \ = \ \exp(-s_n[{\mathbf{x}}_a,{\mathbf{x}}_b]/\hbar^2 ) \ = \ p(\mbox{leaving no trace } ) \ \ \ .\end{aligned}\ ] ] in order to prove the latter statement one should use the same procedure as above with @xmath304 replaced by some arbitrary initial preparation @xmath305 . then the result should be thermally averaged . having a pair @xmath151 of trajectories , we can define the function @xmath306 where the average is over @xmath307 within the time interval @xmath308 $ ] which is considered . we shall use the notations @xmath309 and @xmath310 for @xmath311 . it is also useful to define the functions @xmath312 the fourier transform of @xmath313 will be denoted by @xmath314 , and the double fourier transform of @xmath315 will be denoted by @xmath316 . the above functions all appear within integrals where `` isotropic '' integration over @xmath317 or @xmath318 is being performed . therefore it is convenient to average all these functions over all orientations , and so to have functions that depend on either @xmath319 or @xmath320 respectively . a useful notation is @xmath321 , where the average is over the orientation of the unit vector @xmath322 . the function @xmath323 depends only on @xmath42 . it equals @xmath324 in 1-d , regular bessel function @xmath325 in 2-d , and @xmath326 in 3-d . we shall distinguish now between _ ballistic _ trajectories as in fig.[f_trajc](a ) , _ chaotic _ trajectories as in fig.[f_trajc](b ) , and _ diffusive _ trajectories as in fig.[f_trajc](c ) . we can also add to this list the case of diffusive trajectories that cover ergodically the whole available space . in case of a ballistic trajectory @xmath327 , leading to @xmath328 , where @xmath329 . similar expression hold for @xmath330 , where @xmath331 , and @xmath120 is the transverse distance between the slits . for a chaotic trajectory @xmath332 starts as in the ballistic case , while @xmath333 is uniform in @xmath317 . the ergodic time is given essentially by the ballistic time , namely @xmath334 , where @xmath335 is the linear dimension of the cavity . for @xmath336 both @xmath332 and @xmath333 become uniform in @xmath317 . for diffusive motion @xmath337 is as for the ballistic case as long as @xmath338 where @xmath339 . on larger times it becomes a gaussian , and consequently @xmath340 , where @xmath341 is the diffusion coefficient . the latter expression as well as the approximation @xmath342 hold for non - ergodic diffusive motion . after the ergodic time @xmath343 , both @xmath332 and @xmath333 become uniform in @xmath317 . in the following paragraphs we discuss the various @xmath180 regimes of the function @xmath159 . we shall distinguish between the small-@xmath180 regime and the large-@xmath180 regime . in case of diffusive trajectories the large-@xmath180 regime will be further divided into a ballistic regime ( @xmath344 ) and a diffusive regime ( @xmath345 ) , where @xmath346 is the inverse of the mean free path . we turn first to define the notions of small and large @xmath180 . the transverse distance between two trajectories has a distribution @xmath347 , and we can define a typical value @xmath348 . see eq.([e35 ] ) . the associated wavenumber is @xmath349 . large @xmath180 means from now on @xmath350 . for large @xmath180 we may use the approximation @xmath351 for either ballistic , diffusive or chaotic trajectories we have @xmath352 where @xmath353 is the ft of the @xmath354 function . this rectangular - like scaling - function has a unit width and a unit normalization . in case of diffusive motion we also have @xmath355 diffusive regime means here any large @xmath180 that satisfies @xmath345 . the validity of the latter approximation is further restricted by the condition @xmath356 . the collision frequency @xmath357 should be used as a cutoff to the slow @xmath358 power - law decay . in our calculations we shall assume that @xmath359 , meaning that the motion is diffusive also on the spatial scale @xmath8 . for small @xmath180 , meaning @xmath360 , the transverse term can not be ignored , and we no - longer can use the approximation ( [ e102 ] ) . we can define a time scale @xmath213 that correspond to @xmath361 . it equals @xmath214 for ballistic motion , @xmath146 for diffusive motion , and @xmath215 for ergodic motion . for a given small @xmath180 , the function @xmath362 is concentrated within @xmath363 , since @xmath364 and @xmath330 are not identical there . consequently @xmath365 the function @xmath366 is the ft of @xmath367 , and therefore we have for @xmath368 the following small @xmath180 approximation @xmath369 for large @xmath180 we have @xmath370 , as implied by ( [ e102 ] ) and the definition ( [ e101 ] ) . the effect of electron - electron coulomb interaction can be analyzed by considering the motion of a single electron under the influence of a fluctuating electrostatic potential @xmath371 that is created by the other electrons . thus the electron experience an interaction with a fluctuating field that is characterized by @xmath372 . it is well known , using fluctuation - dissipation theorem , that for diffusive electrons the corresponding fluctuation spectrum is @xmath373 the ohmic behavior is cut - off by the drude collision frequency @xmath374 . the elastic mean free path is @xmath375 , where @xmath24 is the fermi velocity . for @xmath376 the power spectrum @xmath159 of the diffusive motion is concentrated below the drude cutoff frequency . the ballistic regime @xmath377 is of no interest because its contribution is suppressed due to the drude cutoff . one observes that these fluctuations corresponds to the generalized dld model ( [ e14 ] ) with @xmath378 . having observed that the mean free path is the physical ( effective ) cutoff for the spatial fluctuations ( this statement is true for @xmath379 ) , it follows from the convention @xmath380 that the friction constant ( up to dimensionless factor of order unity ) is given by the expression @xmath381 here @xmath382 is the charge of an electron , and @xmath383 is the conductivity defined for @xmath384 dimensions . the last equality is obtained by using einstein relation in order to express @xmath383 in terms of @xmath341 . the notation @xmath385 stand for the mean level spacing within a cube whose volume is @xmath386 .

the generalized function matching ( gfm ) problem has been intensively studied starting with [ ehrenfeucht and rozenberg , 1979 ] . given a pattern p and a text t , the goal is to find a mapping from the letters of p to non - empty substrings of t , such that applying the mapping to p results in t. very recently , the problem has been investigated within the framework of parameterized complexity [ fernau , schmid , and villanger , 2013 ] . in this paper we study the parameterized complexity of the optimization variant of gfm ( called max - gfm ) , which has been introduced in [ amir and nor , 2007 ] . here , one is allowed to replace some of the pattern letters with some special symbols `` ? '' , termed wildcards or do nt cares , which can be mapped to an arbitrary substring of the text . the goal is to minimize the number of wildcards used . we give a complete classification of the parameterized complexity of max - gfm and its variants under a wide range of parameterizations , such as , the number of occurrences of a letter in the text , the size of the text alphabet , the number of occurrences of a letter in the pattern , the size of the pattern alphabet , the maximum length of a string matched to any pattern letter , the number of wildcards and the maximum size of a string that a wildcard can be mapped to .

1409.2398

in the generalized function matching problem one is given a text @xmath0 and a pattern @xmath1 and the goal is to decide whether there is a match between @xmath1 and @xmath0 , where a single letter of the pattern is allowed to match multiple letters of the text ( we say that @xmath1 gf - matches @xmath0 ) . for example , if the text is @xmath2 and the pattern is @xmath3 , then a generalized function match ( on short , gf - match ) is @xmath4 , but if @xmath5 and @xmath6 , then there is no gf - match . if , moreover , the matching is required to be injective , then we term the problem generalized parameterzied matching ( gpm ) . in @xcite , amir and nor describe applications of gfm in various areas such as software engineering , image searching , dna analysis , poetry and music analysis , or author validation . gfm is also related to areas such as ( un-)avoidable patterns @xcite , word equations @xcite and the ambiguity of morphisms @xcite . gfm has a long history starting from 1979 . ehrenfeucht and rozenberg @xcite show that gfm is np - complete . independently , angluin @xcite studies a more general variant of gfm where the pattern may contain also letters of the text alphabet . angluin s paper received a lot of attention , especially in the learning theory community @xcite ( see @xcite for a survey ) but also in many other areas . recently , a systematic study of the classical complexity of a number of variants of gfm and gpm under various restrictions has been carried out @xcite . it was shown that gfm and gpm remain complete for many natural restrictions . moreover , the study of gfm and its variants within the framework of parameterized complexity has recently been initiated @xcite . in this paper we study the parameterized complexity of the optimization variant of gfm ( called max - gfm ) and its variants , where one is allowed to replace some of the pattern letters with some special symbols `` ? '' , termed wildcards or do nt cares , which can be mapped to an arbitrary substring of the text . the goal is to minimize the number of wildcards used . the problem was first introduced to the pattern matching community by amir and nor @xcite . they show that if the pattern alphabet has constant size , then a polynomial algorithm can be found , but that the problem is complete otherwise . then , in @xcite , it is shown the hardness of the gfm ( without wildcards ) and the hardness of the gfm when the function @xmath7 is required to be an injection ( named gpm ) . more specifically , gfm is hard even if the text alphabet is binary and each letter of the pattern is allowed to map to at most two letters of the text @xcite . in the same paper it is given a @xmath8 approximation algorithm for the optimization variant of gfm where the goal is to search for a pattern @xmath9 that gf - matches @xmath0 and has the smallest hamming distance to @xmath1 . in @xcite the optimization versions of gfm and gpm are proved to be * apx*-hard . [ [ our - results ] ] our results + + + + + + + + + + + before we discuss our results , we give formal definitions of the problems . in the following let @xmath0 be a text over an alphabet @xmath10 and let @xmath11 be a pattern over an alphabet @xmath12 . we say that @xmath1 _ gf - matches _ @xmath0 if there is a function @xmath13 such that @xmath14 . to improve the presentation we will sometimes abuse notation by writing @xmath15 instead of @xmath16 . let @xmath17 be a natural number . we say that a pattern @xmath1 _ @xmath17-gf - matches _ @xmath0 if there is a text @xmath9 over alphabet @xmath18 of hamming distance at most @xmath17 from @xmath1 such that @xmath9 gf - matches @xmath0 . given a text @xmath0 , a pattern @xmath1 , and an integer @xmath17 , decide whether @xmath1 @xmath17-gf - matches @xmath0 . the max - gfm can be seen as the optimization variant of gfm in which we want to replace some of the pattern letters with special wildcard symbols , i.e. , the symbols @xmath19 , which can be mapped to any non - empty substring of the text . we also study the max - gpm problem . the only difference between max - gpm and max - gfm is that for max - gpm the function @xmath7 is required to be injective . the notions of gp - matching and @xmath17-gp - matching are defined in the natural way , e.g. , we say a pattern @xmath1 _ gp - matches _ a text @xmath0 if @xmath1 _ gf - matches _ @xmath0 using an injective function . in this paper we study the parameterized complexity of the two problems using a wide range of parameters : maximum number of occurrences of a letter in the text @xmath20 , maximum number of occurrences of a letter in the pattern @xmath21 , size of the text alphabet @xmath22 , size of the pattern alphabet @xmath23 , the maximum length of a substring of the text that a letter of the pattern alphabet can be mapped to ( i.e. , @xmath24 ) , the number of wildcard letters @xmath25 , and the maximum length of a substring of the text that a wildcard can be mapped to , denoted by @xmath26 . our results are summarized in table [ table : tab2 ] . we verified the completeness of our results using a simple computer program . in particular , the program checks for every of the @xmath27 possible combinations of parameters @xmath28 that the table contains either : i ) a superset of @xmath28 under which max - gfm / gpm is hard ( and thus , max - gfm / gpm is hard if parameterized by @xmath28 ) ; or ii ) a subset of @xmath28 for which max - gfm / gpm is fpt ( and then we have an fpt result for the set of parameters @xmath28 ) . since some of our results do not hold for both max - gfm and max - gpm , we carried out two separate checks , one for max - gfm and one for max - gpm . .parameterized complexity of max - gfm and max - gpm . [ cols="^,^,^,^,^,^,^,^",options="header " , ] the paper is organized as follows . in section [ sec : prel ] we give preliminaries , in section [ sec : fpt ] we present our fixed - parameter algorithms and in section [ sec : hardness ] we show our hardness results . we define the basic notions of parameterized complexity and refer to other sources @xcite for an in - depth treatment . parameterized problem _ is a set of pairs @xmath29 , the _ instances _ , where @xmath30 is the main part and @xmath17 the _ parameter_. the parameter is usually a non - negative integer . a parameterized problem is _ fixed - parameter tractable ( fpt ) _ if there exists an algorithm that solves any instance @xmath29 of size @xmath31 in time @xmath32 where @xmath7 is an arbitrary computable function and @xmath33 is a constant independent of both @xmath31 and @xmath17 . is the class of all fixed - parameter tractable decision problems . because we focus on fixed - parameter tractability of a problem we will sometimes use the notation @xmath34 to suppress exact polynomial dependencies , i.e. , a problem with input size @xmath31 and parameter @xmath17 can be solved in time @xmath35 if it can be solved in time @xmath36 for some constant @xmath33 . parameterized complexity offers a completeness theory , similar to the theory of np - completeness , that allows the accumulation of strong theoretical evidence that some parameterized problems are not fixed - parameter tractable . this theory is based on a hierarchy of complexity classes @xmath37 } } } \subseteq { { { { { { \mbox{\textnormal{\textbf{w}}}}}}}}{{[2 ] } } } \subseteq { { { { { { \mbox{\textnormal{\textbf{w}}}}}}}}{{[3 ] } } } \subseteq \cdots$ ] where all inclusions are believed to be strict . fpt - reduction _ from a parameterized problem @xmath38 to a parameterized problem @xmath39 is a mapping @xmath40 from instances of @xmath38 to instances of @xmath39 such that ( i ) @xmath29 is a yes - instance of @xmath38 if and only if @xmath41 is a yes - instance of @xmath39 , ( ii ) there is a computable function @xmath42 such that @xmath43 , and ( iii ) there is a computable function @xmath7 and a constant @xmath33 such that @xmath40 can be computed in time @xmath44 , where @xmath31 denotes the size of @xmath29 . for our hardness results we will often reduce from the following problem , which is well - known to be [ 1]complete @xcite . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ multicolored clique + _ instance : _ a @xmath17-partite graph @xmath45 with a partition @xmath46 of @xmath47 . + _ parameter : _ the integer @xmath17 . + _ question : _ are there nodes @xmath48 such that @xmath49 and @xmath50 for all @xmath51 and @xmath52 with @xmath53 ( i.e. the subgraph of @xmath54 induced by @xmath55 is a clique of size @xmath17 ) ? _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ for our hardness proofs we will often make the additional assumptions that ( 1 ) @xmath56 for every @xmath51 and @xmath52 with @xmath57 and ( 2 ) @xmath58 for every @xmath51 , @xmath52 , @xmath59 , and @xmath60 with @xmath61 and @xmath62 , where @xmath63 for every @xmath51 and @xmath52 as before . to see that multicolored clique remains @xmath64}}}$]hard under these additional restrictions we can reduce from multicolored clique to its more restricted version using a simple padding construction as follows . given an instance @xmath65 of multicolored clique we construct an instance of its more restricted version by adding edges ( whose endpoints are new vertices ) between parts ( i.e. @xmath66 ) that do not already have the maximum number of edges between them and then adding isolated vertices to parts that do not already have the maximum number of vertices . even stronger evidence that a parameterized problem is not fixed - parameter tractable can be obtained by showing that the problem remains complete even if the parameter is a constant . the class of these problems is called . a _ square _ is a string consisting of two copies of the same ( non - empty ) string . we say that a string is _ square - free _ if it does not contain a square as a substring . in this section we show our fixed - parameter tractability results for max - gfm and max - gpm . in particular , we show that max - gfm and max - gpm are fixed - parameter tractable parameterized by @xmath22 , @xmath23 , and @xmath67 , and also parameterized by @xmath25 , @xmath68 , @xmath23 , and @xmath67 . we start by showing fixed - parameter tractability for the parameters @xmath22 , @xmath23 , and @xmath67 . we need the following lemma . [ lem : dyn ] given a pattern @xmath11 over an alphabet @xmath12 , a text @xmath69 over an alphabet @xmath10 , a natural number @xmath70 , and a function @xmath71 , then there is a polynomial time algorithm deciding whether @xmath1 @xmath70-gf / gp - matches @xmath0 using the function @xmath7 . if we are asked whether @xmath1 @xmath70-gp - matches @xmath0 and @xmath7 is not injective , then we obviously provide a negative answer . otherwise , we use a dynamic programming algorithm that is similar in spirit to an algorithm in @xcite . let @xmath72 . for every @xmath73 , we define the function @xmath74 to be the hamming gfm / gpm - similarity ( i.e. , @xmath75 minus the minimum number of wildcards needed ) between @xmath76 and @xmath77 . then , we obtain the hamming gfm / gpm - similarity between @xmath1 and @xmath0 as @xmath78 . consequently , if @xmath79 , we return no , otherwise we return yes . we now show how to recursively compute @xmath74 . if @xmath80 , we set @xmath81 and if @xmath82 , we set : @xmath83 where @xmath84 is @xmath85 if the strings @xmath86 , and @xmath87 are the same , and @xmath88 otherwise . we must first show that the dynamic programming procedure computes the right function and then that it runs in polynomial time . we can see immediately that @xmath89 for all @xmath51 because in this case the pattern is empty . the recursion step of @xmath74 has two cases : if @xmath90 , then it is possible to map @xmath91 to @xmath92 , and we can increase the number of mapped letters by one . otherwise , we can not increase the hamming gfm / gpm - similarity . however , we know that @xmath91 has to be set to a wildcard and therefore we find the maximum of the previous results for different length substrings that the wildcard maps to . it is straightforward to check that @xmath74 can be computed in cubic time . [ the : fpt - max - gfm - text ] max - gfm and max - gpm parameterized by @xmath22 , @xmath23 , and @xmath67 are fixed - parameter tractable . let @xmath1 , @xmath0 , and @xmath70 be an instance of max - gfm or max - gpm , respectively . the pattern @xmath1 @xmath70-gf / gp - matches @xmath0 if and only if there is a function @xmath71 such that @xmath1 @xmath70-gf / gp - matches @xmath0 using @xmath7 . hence , to solve max - gfm / max - gpm , it is sufficient to apply the algorithm from lemma [ lem : dyn ] to every function @xmath71 that could possible constitute to a @xmath70-gf / gp - matching from @xmath1 to @xmath0 . because there are at most @xmath93 such functions @xmath7 and the algorithm from lemma [ lem : dyn ] runs in polynomial time , the running time of this algorithm is @xmath94 , and hence fixed - parameter tractable in @xmath22 , @xmath23 , and @xmath67 . because in the case of max - gpm it holds that if @xmath22 and @xmath67 is bounded then also @xmath12 is bounded by @xmath95 , we obtain the following corollary . [ cor : fpt - max - gpm ] max - gpm parameterized by @xmath22 and @xmath67 is fixed - parameter tractable . we continue by showing our second tractability result for the parameters @xmath23 , @xmath67 , @xmath25 , and @xmath68 . [ the : fpt - max - gfm - wild ] max - gfm and max - gpm parameterized by @xmath23 , @xmath96 , @xmath25 , @xmath68 , are fixed - parameter tractable . let @xmath1 , @xmath0 , and @xmath70 be an instance of max - gfm or max - gpm , respectively . observe that if we could go over all possible functions @xmath97 that could possible constitute to a @xmath70-gf / gp - matching from @xmath1 to @xmath0 , then we could again apply lemma [ lem : dyn ] as we did in the proof of theorem [ the : fpt - max - gfm - text ] . unfortunately , because @xmath22 is not a parameter , the number of these functions can not be bounded as easily any more . however , as we will show next it is still possible to bound the number of possible functions solely in terms of the parameters . in particular , we will show that the number of possible substrings of @xmath0 that any letter of the pattern alphabet can be mapped to is bounded by a function of the parameters . because also @xmath23 is a parameter this immediately implies a bound ( only in terms of the given parameters ) on the total number of these functions . let @xmath98 and consider any @xmath70-gf / gp - matching from @xmath1 to @xmath0 , i.e. , a text @xmath99 of hamming distance at most @xmath70 to @xmath1 and a function @xmath100 such that @xmath101 . then either @xmath33 does not occur in @xmath9 or @xmath33 occurs in @xmath9 . in the first case we can assign to @xmath33 any non - empty substring over the alphabet @xmath10 ( in the case of max - gpm one additionally has to ensure that the non - empty substrings over @xmath10 that one chooses for distinct letters in @xmath12 are distinct ) . in the second case let @xmath102 for some @xmath51 with @xmath103 be the first occurrence of @xmath33 in @xmath9 , let @xmath104 , and let @xmath105 . furthermore , for every @xmath106 and @xmath107 , we denote by @xmath108 the number of times @xmath109 occurs in @xmath110 . then @xmath111 where @xmath112 , which implies that the value of @xmath113 is fully determined by @xmath114 and @xmath115 . because the number of possible values for @xmath115 is trivially bounded by the parameters ( it is bounded by @xmath96 ) , it remains to show that also @xmath114 is bounded by the given parameters . because @xmath116 , we obtain that the value of @xmath114 is fully determined by the values of @xmath117 and @xmath118 for every @xmath119 . for every @xmath120 there are at most @xmath121 possible values for @xmath122 ( namely @xmath88 and @xmath85 ) and there are at most @xmath123 possible values for @xmath124 . similarly , for every @xmath125 there are at most @xmath126 possible values for @xmath117 ( the values @xmath127 ) and there are at most @xmath96 possible values for @xmath118 . hence , the number of possible values for @xmath114 is bounded in terms of the parameters , as required . since @xmath23 and @xmath128 together bound @xmath25 , we obtain the following corollary . [ cor : fpt - max - gfm - wild ] max - gfm and max - gpm parameterized by @xmath128 , @xmath23 , @xmath96 , and @xmath68 are fixed - parameter tractable . furthermore , because all considered parameters can be bounded in terms of the parameters @xmath129 and @xmath22 , we obtain the following corollary as a consequence of any of our above fpt - results . [ cor : fpt - occt - sigt ] max - gfm and max - gpm parameterized by @xmath129 and @xmath22 are fixed - parameter tractable . in this subsection we give our hardness results for max - gfm and max - gpm . we will show the theorem by a parameterized reduction from multicolored clique . to simplify the proof we will reduce to the variant of max - gfm and max - gpm , where we are allowed to map wildcards to the empty string . it is however straightforward to adapt the proof to the original versions of max - gfm and max - gpm . hence , in the following , whenever we refer to max - gfm and max - gpm , we mean the version of max - gfm and max - gpm , where wildcards can be mapped to the empty string . let @xmath132 be a @xmath17-partite graph with partition @xmath46 of @xmath47 . let @xmath133 for every @xmath51 and @xmath52 with @xmath134 . again , as we stated in the preliminaries we can assume that @xmath135 and @xmath136 for every @xmath51 and @xmath52 with @xmath61 . let @xmath137 and @xmath138 . we construct a text @xmath0 and a pattern @xmath1 from @xmath54 and @xmath17 such that @xmath1 @xmath59-gf / gp - matches @xmath0 with @xmath139 if and only if @xmath54 has a @xmath17-clique . we set @xmath140 and @xmath141 . for an edge @xmath142 between @xmath143 and @xmath144 where @xmath145 and @xmath146 , we write @xmath147 to denote the text @xmath148 . for @xmath149 and @xmath150 we write @xmath151 to denote the text consisting of repeating the letter @xmath152 exactly @xmath51 times . we first define a preliminary text @xmath153 as follows . _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ @xmath163 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ this completes the construction of @xmath0 and @xmath1 . clearly , @xmath0 and @xmath1 can be constructed from @xmath54 and @xmath17 in fpt - time ( even polynomial time ) . furthermore , @xmath164 , as required . it remains to show that @xmath54 has a @xmath17-clique if and only if @xmath1 @xmath59-gf / gp - matches @xmath0 . we put @xmath59 wildcards on the last @xmath59 occurrences of @xmath175 in @xmath1 , i.e. , every occurrence of @xmath175 that corresponds to an occurrence in @xmath9 . then length of the text the wildcards are mapped to is determined as follows . for an edge @xmath176 look at the `` block '' in @xmath1 that corresponds to the edge , i.e. , the block : the first @xmath178 occurrences of @xmath175 ( in this block ) are replaced with a wildcard which is mapped to a text of length @xmath88 , the last @xmath178 occurrences of @xmath175 are replaced with a wildcard which is mapped to a text of length @xmath121 , and all other occurrences of @xmath175 are replaced with a wildcard that is mapped to a text of length @xmath85 . it is straightforward to check that @xmath7 together with the mapping of the wildcards maps the pattern @xmath1 to the text @xmath0 . we show that @xmath169 since the remaining cases are similar . because the pattern @xmath1 starts with @xmath179 repetitions of the letter @xmath175 , it follows that at least @xmath85 of these occurrences of @xmath175 is not replaced with a wildcard . because every letter of @xmath1 is replaced by at most @xmath121 letters of the text the first occurrence of @xmath175 that is not replaced by a wildcard is mapped to a letter of the text at position at most @xmath180 , i.e. , a @xmath175 . this concludes the proof of the claim . it follows from the previous claim that @xmath169 for any function that @xmath59-gf / gp - matches @xmath1 to @xmath0 . because every letter of @xmath1 that is not replaced with a wildcard is replaced with exactly @xmath85 letter from the text , it follows that the first occurrence of @xmath175 in @xmath1 that corresponds to an occurrence of @xmath175 in @xmath9 is mapped to ( if it is not replaced with a wildcard ) to a letter of the text at position at least @xmath181 . however , since the text @xmath0 does not contain the letter @xmath175 after position @xmath179 , this occurrence of @xmath175 in @xmath1 ( and all other occurrences of @xmath175 in @xmath1 that follow ) has to be replaced with a wildcard . since @xmath9 contains exactly @xmath59 occurrences of @xmath175 the only letters of @xmath1 that are replaced with wildcards are these occurrences of @xmath175 . let @xmath7 be a function that @xmath59-gf / gp - matches @xmath1 to @xmath0 . because of claim [ clm : gfm - questionmarksize - hard - s21 ] , it holds that @xmath169 , @xmath170 , @xmath171 , and @xmath172 . furthermore , because of claim [ clm : gfm - questionmarksize - hard - s21 ] the only letters in @xmath1 that are replaced with wildcards are the last @xmath59 occurrences of @xmath175 in @xmath1 . because the number of occurrences of the letter @xmath182 is the same in @xmath0 and @xmath1 each occurrence of @xmath182 in @xmath1 has to be mapped to its corresponding occurrence in @xmath0 . it follows that for every @xmath51 and @xmath52 with @xmath61 the `` block '' we will show the theorem by a parameterized reduction from multicolored clique . let @xmath132 be a @xmath17-partite graph with partition @xmath46 of @xmath47 . let @xmath133 for every @xmath51 and @xmath52 with @xmath134 . again , as we stated in the preliminaries we can assume that @xmath135 and @xmath136 for every @xmath51 and @xmath52 with @xmath61 . let @xmath137 and @xmath138 , and @xmath189 . we construct a text @xmath0 over alphabet @xmath10 and a pattern @xmath1 over alphabet @xmath12 from @xmath54 and @xmath17 such that @xmath1 @xmath190-gf / gp - matches @xmath0 using a function @xmath7 with @xmath191 if and only if @xmath54 has a @xmath17-clique . the alphabet @xmath10 consists of : * the letter @xmath182 ( used as a separator ) ; * the letter @xmath192 ( used to forced the wildcards ) ; * one letter @xmath193 for every @xmath142 ( representing the edges of @xmath54 ) ; * one letter @xmath194 for every @xmath51 with @xmath195 ( used as special separators that group edges from the same vertex ) ; * the letters @xmath196 , @xmath197 , @xmath198 , @xmath199 for every @xmath51 and @xmath52 with @xmath61 ( used as dummy letters to ensure injectivity for gpm ) ; * the letter @xmath200 and @xmath201 for every @xmath142 and @xmath202 with @xmath203 ( used as dummy letters to ensure injectivity for gpm ) . for a vertex @xmath205 and @xmath52 with @xmath206 we denote by @xmath207 the set of edges of @xmath54 that are incident to @xmath208 and whose other endpoint is in @xmath209 . furthermore , for a vertex @xmath202 , we write @xmath210 to denote the text @xmath211 , where @xmath212 , for vertex @xmath208 and a set @xmath213 of edges with @xmath214 , is the text @xmath215 . let @xmath243 . for @xmath149 and @xmath150 we write @xmath151 to denote the text consisting of repeating the letter @xmath152 exactly @xmath51 times . we also define @xmath244 to be the text @xmath245 and @xmath246 to be the pattern @xmath247 . then , @xmath0 is the concatenation of @xmath244 , @xmath216 and @xmath220 and @xmath1 is a concatenation of @xmath246 , @xmath223 and @xmath241 . this completes the construction of @xmath0 and @xmath1 . clearly , @xmath0 and @xmath1 can be constructed from @xmath54 and @xmath17 in fpt - time ( even polynomial time ) . furthermore , @xmath248 , @xmath249 , @xmath250 and hence bounded by @xmath17 , as required . it remains to show that @xmath54 has a @xmath17-clique if and only if @xmath1 @xmath190-gf / gp - matches @xmath0 using a function @xmath7 with @xmath251 . we put @xmath190 wildcards on the last @xmath190 occurrences of @xmath254 in @xmath1 . the mapping of these wildcards is defined very similar to the mapping of the letters @xmath255 , @xmath256 , @xmath257 , @xmath258 , and @xmath259 in the proof of lemma [ lem : gfm - mobile-1-hard-1 ] and will not be repeated here . using this mapping ensures that every wildcard is mapped to an non - empty substring of @xmath0 and no two wildcards are mapped to the same substring of @xmath0 . we define the function @xmath7 that @xmath190-gf / gp - matches @xmath1 to @xmath0 as follows : we set @xmath172 and @xmath260 . moreover , for every @xmath51 and @xmath52 with @xmath61 , we set @xmath261 and @xmath262 . [ clm : gfm - mobile-2-hard-1 ] let @xmath7 be a function that @xmath190-gf / gp - matches @xmath1 to @xmath0 with @xmath251 , then : @xmath172 and @xmath260 . moreover , all wildcards have to be placed on all the @xmath190 occurrences of @xmath254 in @xmath9 . we first show that @xmath260 . observe that the only squares in the string @xmath0 are contained in @xmath244 ( recall the definition of square - free from section [ sec : prel ] ) . it follows that every two consecutive occurrences of pattern letters in @xmath246 have to be mapped to a substring of @xmath244 . because there are @xmath263 occurrences of @xmath254 in @xmath246 it follows that at least two consecutive occurrences of @xmath254 in @xmath246 are not replaced with wildcards and hence @xmath254 has to be mapped to a substring of @xmath244 . furthermore , since all occurrences of @xmath254 are at the end of @xmath246 , we obtain that @xmath254 has to be mapped to @xmath192 , as required . because all occurrences of @xmath254 in @xmath9 have to be mapped to substrings of the concatenation of @xmath216 and @xmath220 , but these strings do not contain the letter @xmath192 , it follows that all the @xmath190 occurrences of @xmath254 in @xmath223 and @xmath241 have to be replaced by wildcards . since we are only allowed to use at most @xmath190 wildcards , this shows the second statement of the claim . since no wildcards are used to replace letters in @xmath246 it now easily follows that @xmath172 . let @xmath7 be a function that @xmath190-gf / gp - matches @xmath1 to @xmath0 with @xmath251 . because of claim [ clm : gfm - mobile-2-hard-1 ] , we know that @xmath172 and that no occurrence of @xmath182 in @xmath1 is replaced by a wildcard . because @xmath0 and @xmath1 have the same number of occurrences of @xmath182 , it follows that the @xmath51-th occurrences of @xmath182 in @xmath1 has to be mapped to the @xmath51-th occurrence of @xmath182 in @xmath0 . we obtain that : 1 . for every @xmath51 , @xmath52 with @xmath61 , the substring @xmath265 of @xmath1 has to be mapped to the substring @xmath266 of @xmath0 . 2 . for every @xmath51 with @xmath174 , the substring @xmath267 of @xmath1 has to be mapped to the substring @xmath268 of @xmath0 . because for every @xmath51 with @xmath174 the letters @xmath269 are the only letters that occur more than once in the substring @xmath268 of @xmath0 , we obtain from ( 2 ) that @xmath270 has to be mapped to @xmath269 for some @xmath52 with @xmath271 . consequently : it follows from ( 1 ) that for every @xmath51 , @xmath52 with @xmath274 , @xmath275 is mapped to an edge between @xmath276 and @xmath209 . furthermore , because of ( 3 ) it follows that for every @xmath51 with @xmath174 , it holds that the edges mapped to any @xmath277 with @xmath278 such that @xmath279 or @xmath280 have the same endpoint in @xmath276 . hence , the set of edges mapped to the letters @xmath229 for @xmath281 form a @xmath17-clique of @xmath54 . we will show the above theorem by a parameterized reduction from multicolored clique . let @xmath132 be a @xmath17-partite graph with partition @xmath46 of @xmath47 . let @xmath133 for every @xmath51 and @xmath52 with @xmath134 . again , as we stated in the preliminaries we can assume that @xmath135 and @xmath136 for every @xmath51 and @xmath52 with @xmath61 . let @xmath137 and @xmath138 . for a vertex @xmath282 and @xmath52 with @xmath283 we denote by @xmath207 the set of edges of @xmath54 that are incident to @xmath208 and whose other endpoint is in @xmath209 . 1 . the parameters @xmath20 and @xmath21 are bounded by @xmath17 ( note the parameters @xmath25 and @xmath123 are bounded since we consider gfm and gpm ) . @xmath1 gf / gp - matches @xmath0 using a function @xmath7 with @xmath284 if and only if @xmath54 has a @xmath17-clique . let @xmath285 . the alphabet @xmath10 consists of ( 1 ) the letter @xmath182 , ( 2 ) the letters @xmath286 and @xmath287 for every @xmath288 and @xmath289 , ( 3 ) the letters @xmath290 and @xmath291 for every @xmath282 , @xmath283 , and @xmath292 , where @xmath174 and @xmath293 , ( 4 ) the letters @xmath294 and @xmath295 for every @xmath174 and @xmath296 , ( 5 ) the letter @xmath297 for every @xmath61 and @xmath187 , and ( 6 ) the letter @xmath194 for every @xmath195 . the alphabet @xmath12 consists of ( 1 ) the letter @xmath182 , ( 2 ) the letters @xmath298 and @xmath299 for every @xmath288 and @xmath289 , ( 3 ) the letters @xmath300 and @xmath301 for every @xmath302 with @xmath303 , and @xmath292 , ( 4 ) the letters @xmath304 and @xmath305 for every @xmath174 and @xmath296 , ( 5 ) the letter @xmath229 for every @xmath61 , and ( 6 ) the letter @xmath270 for every @xmath195 . furthermore , for a vertex @xmath202 and @xmath51 with @xmath309 , we write @xmath310 to denote the text @xmath311 , where @xmath312 ( for a set of edges @xmath213 ) is a list of all the letters in @xmath10 that correspond to the edges in @xmath213 . for a vertex @xmath282 , and @xmath52 with @xmath283 , we denote by @xmath318 the text @xmath319 if @xmath320 and the empty text if @xmath321 . furthermore , we denote by @xmath322 the text @xmath323 . let @xmath220 be the text : for @xmath51 , @xmath52 with @xmath227 , let @xmath228 be the letter @xmath229 if @xmath230 , the letter @xmath231 if @xmath232 and the empty string if @xmath233 . furthermore , let @xmath331 be the pattern @xmath332 if @xmath303 and the empty pattern otherwise . let @xmath241 be the pattern : we also define @xmath244 to be the text @xmath335 and @xmath246 to be the pattern @xmath335 . then , @xmath0 is the concatenation of @xmath244 , @xmath216 and @xmath220 and @xmath1 is a concatenation of @xmath246 , @xmath223 and @xmath241 . this completes the construction of @xmath0 and @xmath1 . clearly , @xmath0 and @xmath1 can be constructed from @xmath54 and @xmath17 in fpt - time ( even polynomial time ) . furthermore , because @xmath336 , @xmath337 , condition ( c1 ) is satisfied . to show the remaining condition ( c2 ) we need the following intermediate lemmas and claims . we first give the gf / gp - matching function @xmath7 for the letters in @xmath12 that occur more than once in @xmath1 as follows : we set @xmath172 , @xmath339 , and @xmath340 , for every @xmath51 and @xmath52 with @xmath61 . informally , we will map the remaining letters in @xmath12 to substrings of @xmath0 of length between @xmath85 and @xmath121 in such a way that the occurrences of the letters @xmath182 , @xmath229 , and @xmath270 are placed over the right positions in the text @xmath0 . more formally , we define @xmath7 for the remaining letters in @xmath12 as follows : * for every @xmath61 , we define @xmath341 in such a way that @xmath342 for every @xmath343 and @xmath344 for every @xmath345 . * for every @xmath61 , we define @xmath346 in such a way that @xmath347 for every @xmath348 and @xmath342 for every @xmath349 . * for every @xmath227 with @xmath303 , we define @xmath350 in such a way that @xmath351 for every @xmath352 , where @xmath60 is the position of @xmath176 in @xmath353 and @xmath354 for every @xmath355 . * for every @xmath227 with @xmath303 , we define @xmath356 in such a way that @xmath357 for every @xmath358 , where @xmath60 is the position of @xmath176 in @xmath353 and @xmath357 for every @xmath359 . * for every @xmath174 , we define @xmath360 in such a way that @xmath361 for every @xmath352 , where @xmath60 is position of @xmath362 in the substring @xmath363 of @xmath0 and @xmath344 for every @xmath364 . * for every @xmath174 , we define @xmath365 in such a way that @xmath366 for every @xmath358 , where @xmath60 is position of @xmath362 in the substring @xmath363 of @xmath0 and @xmath367 for every @xmath368 . [ clm : gfm - occt - max-1 ] for any function @xmath7 that @xmath190-gf / gp - matches @xmath1 to @xmath0 it holds that : @xmath170 , @xmath172 , and @xmath369 . moreover , all wildcards have to be placed on all the @xmath190 occurrences of @xmath39 in @xmath1 . we first show that @xmath369 . observe that the concatenation of the strings @xmath216 and @xmath220 is square - free ( recall the definition of square - free from section [ sec : prel ] ) . it follows that every two consecutive occurrences of pattern letters in @xmath246 have to be mapped to a substring of @xmath244 . because there are @xmath263 occurrences of @xmath39 in @xmath246 it follows that at least two consecutive occurrences of @xmath39 in @xmath246 are not replaced with wildcards and hence @xmath39 has to be mapped to a substring of @xmath244 . furthermore , since all occurrences of @xmath39 are at the end of @xmath246 , we obtain that @xmath39 has to be mapped to @xmath192 , as required . because all occurrences of @xmath39 in @xmath223 and @xmath241 have to be mapped to substrings of the concatenation of @xmath216 and @xmath220 but neither @xmath216 nor @xmath220 contain the letter @xmath192 , it follows that all the @xmath190 occurrences of @xmath39 in @xmath223 and @xmath241 have to be replaced by wildcards . since we are only allowed to use at most @xmath190 wildcards , this shows the second statement of the claim . since no wildcards are used to replace letters in @xmath246 it now also follows that @xmath170 and @xmath172 . let @xmath7 be the function that gf / gp - matches @xmath1 to @xmath0 with @xmath338 . we first show that @xmath172 . suppose for a contradiction that @xmath371 because @xmath0 and @xmath1 start with @xmath335 it follows that @xmath372 is a string that starts with @xmath335 . however , @xmath0 does not contain any other occurrence of the string @xmath335 and hence the remaining occurrences of @xmath182 in @xmath1 can not be matched by @xmath7 . because @xmath0 and @xmath1 have the same number of occurrences of @xmath182 , it follows that the @xmath51-th occurrences of @xmath182 in @xmath1 has to be mapped to the @xmath51-th occurrence of @xmath182 in @xmath0 . we obtain that : 1 . for every @xmath51 , @xmath52 with @xmath61 , the substring @xmath326 of @xmath1 has to be mapped to the substring @xmath313 of @xmath0 . 2 . for every @xmath51 with @xmath174 , the substring @xmath373 of @xmath1 has to be mapped to the substring @xmath374 of @xmath0 . because for every @xmath51 with @xmath174 the letters @xmath269 are the only letters that occur more than once in the substring @xmath375 of @xmath0 , we obtain that @xmath270 has to be mapped to @xmath269 for some @xmath52 with @xmath271 . consequently : it follows from ( 1 ) that for every @xmath51 , @xmath52 with @xmath274 , the function @xmath7 maps @xmath229 to edges between @xmath276 and @xmath209 . we can assume that @xmath229 is mapped to exactly one such edge because mapping it to many edges makes it only harder to map the following occurrences of @xmath229 in @xmath1 . because of ( 3 ) it follows that for every @xmath51 with @xmath174 , it holds that the edges mapped to any @xmath277 with @xmath278 such that @xmath279 or @xmath280 have the same endpoint in @xmath276 . hence , the set of edges mapped to all the letters @xmath229 for @xmath281 form a @xmath17-clique of @xmath54 . we will show the above theorem by a parameterized reduction from multicolored clique . let @xmath132 be a @xmath17-partite graph with partition @xmath46 of @xmath47 . let @xmath133 for every @xmath51 and @xmath52 with @xmath134 . as we stated in the preliminaries we can assume that @xmath135 and @xmath136 for every @xmath51 and @xmath52 with @xmath61 . 1 . the parameters @xmath21 , @xmath23 , and @xmath25 can be bounded as a function of @xmath17 . 2 . @xmath1 @xmath190-gf / gp - matches @xmath0 using a function @xmath7 with @xmath251 if and only if @xmath54 has a @xmath17-clique . for an edge @xmath142 between @xmath143 and @xmath382 where @xmath145 and @xmath383 , we write @xmath147 to denote the text @xmath384 . for @xmath149 and @xmath150 we write @xmath151 to denote the text consisting of repeating the letter @xmath152 exactly @xmath51 times . we first define a preliminary text @xmath153 as follows . this completes the construction of @xmath0 and @xmath1 . clearly , @xmath0 and @xmath1 can be constructed from @xmath54 and @xmath17 in fpt - time ( even polynomial time ) . furthermore , because @xmath395 , @xmath164 , and @xmath396 , condition ( c1 ) above is satisfied . to show the remaining condition ( c2 ) , we need the following intermediate lemmas . we put @xmath190 wildcards on the last @xmath190 occurrences of @xmath254 in @xmath1 . informally , these wildcards are mapped in such a way that for every @xmath61 the substring @xmath398 of the pattern @xmath1 is mapped to the substring @xmath399 of the text @xmath0 . more formally , for @xmath51 and @xmath52 with @xmath61 let @xmath400 . we map the wildcard on the @xmath401-th occurrence of the letter @xmath254 in @xmath9 with the text @xmath402 and similarly we map the wildcard on the @xmath403-th occurrence of the letter @xmath254 in @xmath9 with the text @xmath404 . note that in this way every wildcard is mapped to a non - empty substring of @xmath0 and no two wildcards are mapped to the same substring of @xmath0 , as required . we then define the @xmath190-gf / gp - matching function @xmath7 as follows : @xmath170 , @xmath171 , @xmath172 , @xmath405 , @xmath260 , for every @xmath51 and @xmath406 with @xmath174 and @xmath407 . it is straightforward to check that @xmath7 together with the mapping for the wildcards maps the pattern @xmath1 to the text @xmath0 . [ clm : gfm - questionmarksize - hard - s21 ] let @xmath7 be a function that @xmath190-gf / gp - matches @xmath1 to @xmath0 with @xmath251 , then : @xmath170 , @xmath171 , @xmath172 , and @xmath260 . moreover , all wildcards have to be placed on all the @xmath190 occurrences of @xmath254 in @xmath9 . we first show that @xmath260 . observe that the string @xmath153 is square - free ( recall the definition of square - free from section [ sec : prel ] ) . it follows that every two consecutive occurrences of pattern letters in @xmath393 have to be mapped to a substring of @xmath391 . because there are @xmath263 occurrences of @xmath254 in @xmath393 it follows that at least two consecutive occurrences of @xmath254 in @xmath393 are not replaced with wildcards and hence @xmath254 has to be mapped to a substring of @xmath391 . furthermore , since all occurrences of @xmath254 are at the end of @xmath393 , we obtain that @xmath254 has to be mapped to @xmath192 , as required . because all occurrences of @xmath254 in @xmath9 have to be mapped to substrings of @xmath153 and @xmath153 does not contain the letter @xmath192 , it follows that all the @xmath190 occurrences of @xmath254 in @xmath9 have to be replaced by wildcards . since we are only allowed to use at most @xmath190 wildcards , this shows the second statement of the lemma . since no wildcards are used to replace letters in @xmath393 it now easily follows that @xmath170 , @xmath171 and @xmath172 . let @xmath7 be a function that @xmath190-gf / gp - matches @xmath1 to @xmath0 such that @xmath251 . we claim that the set @xmath188 is a @xmath17-clique of @xmath54 . because of lemma [ clm : gfm - questionmarksize - hard - s21 ] , we know that @xmath172 and that no occurrence of @xmath182 in @xmath1 is replaced by a wildcard . since the number of occurrences of @xmath182 in @xmath0 is equal to the number of occurrences of @xmath182 in @xmath1 , we obtain that the @xmath51-th occurrence of @xmath182 in @xmath1 is mapped to the @xmath51-th occurrence of @xmath182 in @xmath0 . consequently , for every @xmath51 and @xmath52 with @xmath61 , we obtain that the substring @xmath408 is mapped to a substring of the string @xmath409 in @xmath0 . again , using lemma [ clm : gfm - questionmarksize - hard - s21 ] and the fact that @xmath251 , we obtain that both @xmath276 and @xmath209 are mapped to some letter @xmath410 and @xmath382 for some @xmath152 and @xmath60 with @xmath411 such that @xmath412 . hence , @xmath188 is a @xmath17-clique of @xmath54 . we will show the theorem by a parameterized reduction from multicolored clique . let @xmath132 be a @xmath17-partite graph with partition @xmath46 of @xmath47 . let @xmath133 for every @xmath51 and @xmath52 with @xmath134 . again , as we stated in the preliminaries we can assume that @xmath135 and @xmath136 for every @xmath51 and @xmath52 with @xmath61 . let @xmath137 and @xmath138 . we construct a text @xmath0 and a pattern @xmath1 from @xmath54 and @xmath17 such that @xmath1 gf / gp - matches @xmath0 if and only if @xmath54 has a @xmath17-clique . the alphabet @xmath10 consists of : * the letter @xmath182 ( used as a separator ) ; * one letter @xmath193 for every @xmath142 ( representing the edges of @xmath54 ) ; * one letter @xmath194 for every @xmath51 with @xmath195 ( used as special separators that group edges from the same vertex ) ; * the letters @xmath196 , @xmath197 , @xmath198 , @xmath199 for every @xmath51 and @xmath52 with @xmath61 ( used as dummy letters to ensure injectivity for gpm ) ; * the letter @xmath200 and @xmath201 for every @xmath142 and @xmath202 with @xmath203 ( used as dummy letters to ensure injectivity for gpm ) . for a vertex @xmath205 and @xmath52 with @xmath206 we denote by @xmath207 the set of edges of @xmath54 that are incident to @xmath208 and whose other endpoint is in @xmath209 . furthermore , for a vertex @xmath202 , we write @xmath210 to denote the text @xmath211 , where @xmath212 , for vertex @xmath208 and a set @xmath213 of edges with @xmath214 , is the text @xmath414 . we also define @xmath244 to be the text @xmath335 and @xmath246 to be the pattern @xmath335 . then , @xmath0 is the concatenation of @xmath244 , @xmath216 and @xmath220 and @xmath1 is a concatenation of @xmath246 , @xmath223 and @xmath241 . this completes the construction of @xmath0 and @xmath1 . clearly , @xmath0 and @xmath1 can be constructed from @xmath54 and @xmath17 in fpt - time ( even polynomial time ) . furthermore , @xmath424 , @xmath425 , @xmath426 and hence bounded by @xmath17 , as required . it remains to show that @xmath54 has a @xmath17-clique if and only if @xmath1 gf / gp - matches @xmath0 . we define the function @xmath7 that gf / gp - matches @xmath1 to @xmath0 as follows : we set @xmath172 and @xmath170 . moreover , for every @xmath51 and @xmath52 with @xmath61 , we set @xmath261 , @xmath262 , @xmath427 , @xmath428 , @xmath429 , and @xmath430 . * for every @xmath51 and @xmath52 with @xmath434 , @xmath435 , @xmath303 , and @xmath436 , we map @xmath437 to the substring of @xmath438 in between the occurrences ( and not including these occurrences ) of the letters @xmath439 and @xmath431 . * we map @xmath440 to be the prefix of @xmath441 ending before the letter @xmath442 . * for every @xmath51 with @xmath443 , we map @xmath444 to the prefix of @xmath438 ending before the letter @xmath445 . * for every @xmath51 with @xmath446 , we map @xmath447 to the suffix of @xmath438 starting after the letter @xmath448 . * we map @xmath449 to be the suffix of @xmath441 starting after the letter @xmath450 . let @xmath7 be a function that gf / gp - matches @xmath1 to @xmath0 . we first show that @xmath172 . suppose for a contradiction that @xmath371 . because @xmath0 and @xmath1 start with @xmath335 it follows that @xmath372 is a string that starts with @xmath335 . however , @xmath0 does not contain any other occurrence of the string @xmath335 and hence the remaining occurrences of @xmath182 in @xmath1 can not be matched by @xmath7 . because @xmath0 and @xmath1 have the same number of occurrences of @xmath182 , it follows that the @xmath51-th occurrences of @xmath182 in @xmath1 has to be mapped to the @xmath51-th occurrence of @xmath182 in @xmath0 . we obtain that : 1 . for every @xmath51 , @xmath52 with @xmath61 , the substring @xmath451 of @xmath1 has to be mapped to the substring @xmath266 of @xmath0 . 2 . for every @xmath51 with @xmath174 , the substring @xmath267 of @xmath1 has to be mapped to the substring @xmath268 of @xmath0 . because for every @xmath51 with @xmath174 the letters @xmath269 are the only letters that occur more than once in the substring @xmath268 of @xmath0 , we obtain from ( 2 ) that @xmath270 has to be mapped to @xmath269 for some @xmath52 with @xmath271 . consequently : it follows from ( 1 ) that for every @xmath51 , @xmath52 with @xmath453 , @xmath275 is mapped to some edges between @xmath276 and @xmath209 . we can assume that @xmath275 is mapped to exactly one edge between @xmath276 and @xmath209 , because mapping it to more than one edge would make the matching of the latter occurrences of @xmath229 in @xmath1 even harder , i.e. , whenever a latter occurrence of @xmath229 in @xmath1 can be mapped to more than one edge it also can be mapped to any of these edges . furthermore , because of ( 3 ) it follows that for every @xmath51 with @xmath174 , it holds that the edges mapped to any @xmath277 with @xmath278 such that @xmath279 or @xmath280 have the same endpoint in @xmath276 . hence , the set of edges mapped to the letters @xmath229 for @xmath281 form a @xmath17-clique of @xmath54 . dana angluin . finding patterns common to a set of strings ( extended abstract ) . in _ proceedings of the 11h annual acm symposium on theory of computing , april 30 - may 2 , 1979 , atlanta , georgia , usa _ , pages 130141 , 1979 . raphal clifford , aram w. harrow , alexandru popa , and benjamin sach . generalised matching . in jussi karlgren , jorma tarhio , and heikki hyyr , editors , _ string processing and information retrieval , 16th international symposium , spire 2009 , saariselk , finland , august 25 - 27 , 2009 , proceedings _ , volume 5721 of _ lecture notes in computer science _ , pages 295301 . springer , 2009 . raphal clifford and alexandru popa . ( in)approximability results for pattern matching problems . in jan holub and jan zdrek , editors , _ proceedings of the prague stringology conference 2010 , prague , czech republic , august 30 - september 1 , 2010 _ , pages 5262 . prague stringology club , department of theoretical computer science , faculty of information technology , czech technical university in prague , 2010 . henning fernau and markus l. schmid . pattern matching with variables : a multivariate complexity analysis . in johannes fischer and peter sanders , editors , _ combinatorial pattern matching , 24th annual symposium , cpm 2013 , bad herrenalb , germany , june 17 - 19 , 2013 . proceedings _ , volume 7922 of _ lecture notes in computer science _ , pages 8394 . springer , 2013 . henning fernau , markus l. schmid , and yngve villanger . on the parameterised complexity of string morphism problems . in anil seth and nisheeth k. vishnoi , editors , _ iarcs annual conference on foundations of software technology and theoretical computer science , fsttcs 2013 , december 12 - 14 , 2013 , guwahati , india _ , volume 24 of _ lipics _ , pages 5566 . schloss dagstuhl - leibniz - zentrum fuer informatik , 2013 . dominik d. freydenberger , daniel reidenbach , and johannes c. schneider . unambiguous morphic images of strings . in clelia felice and antonio restivo , editors , _ developments in language theory _ , volume 3572 of _ lecture notes in computer science _ , pages 248259 . springer berlin heidelberg , 2005 . yen kaow ng and takeshi shinohara . developments from enquiries into the learnability of the pattern languages from positive data . , 397(13):150 165 , 2008 . forty years of inductive inference : dedicated to the 60th birthday of rolf wiehagen . takeshi shinohara . polynomial time inference of extended regular pattern languages . in eiichi goto , koichi furukawa , reiji nakajima , ikuo nakata , and akinori yonezawa , editors , _ rims symposia on software science and engineering _ , volume 147 of _ lecture notes in computer science _ , pages 115127 . springer berlin heidelberg , 1983 .

we perform a general parton level analysis for the search of heavy resonant states in the production of pairs at the lhc with an integrated luminosity of 30 fb . we assume the existence of resonances that only couple to quarks and propose kinematic cuts in order to increase the amount of events produced through quark - annihilation . we study the interplay between different variables and their impact on the purity of the selected sample . we make focus on the longitudinal ( ) and transverse ( ) momentum of the pair , and the scattering angle ( ) in the center of mass reference frame . we observe that is replaced by as a suitable discriminating variable of quark - annihilation processes for invariant masses above 1 tev . finally , we illustrate the analysis with a gluon resonance of 1.5 tev and show the improvement in the sensitivity of the signal when cuts on are imposed . * top - antitop resonance searches beyond 1 tev * + + _ conicet , ifiba and departamento de fsica , fceyn , universidad de buenos aires , + ciudad universitaria , pab.1 , ( 1428 ) buenos aires , argentina _ + _ iflp , conicet - dpto . de fsica , universidad nacional de la plata , + c.c . 67 , 1900 la plata , argentina _ + e - mail : [email protected] , [email protected] , [email protected]

1212.5606

with what seems to be the recent discovery of the higgs boson @xcite , the high energy physics enters into the solitude of a path that no one knows where or how it may found new physics ( np ) . although a light higgs requires either , fine - tuning , or np at the tev scale , the experimental results do not show any hint of new phenomena insofar . as a matter of fact , _ natural _ supersymmetry , which is one of the most popular theories to avoid fine - tuning and to simultaneously explain many other phenomena , is being highly constrained by direct searches at the lhc @xcite . from a general theoretical point of view , one may expect to have np effects in the detailed study of the heavier particles properties . in particular , the top quark is the heaviest known particle and its properties have not been explored in depth insofar . in fact , last years results from tevatron @xcite seemed to point to np in the top forward - backward asymmetry . however , recent results from the top charge asymmetry at lhc @xcite , have contradicted most of the np proposals except for very few models @xcite that have survived . regardless of these recent results in top physics , in this article we consider the inclusive @xmath7 process where we study the @xmath8 invariant mass ( @xmath9 ) spectrum as a sensitive observable to np resonant phenomena . this is due to a potential enhancement in the coupling of the top quark to the np . in this work , we address the question of how to increase the sensitivity of this observable in the case of np that couples to quarks and not to gluons . the reason for this study is that at the lhc most ( @xmath10 at @xmath11 tev and increasing with energy ) of the @xmath8 events have a gluon in the initial state partons and , therefore , a possible np as the stated above would be per - se diluted just because of the initial state partons of the event . for the sake of brevity , from this point forward we call gluon - fusion any event with at least one gluon in the initial state . the goal of this article is to propose kinematic cuts which enrich the quark - annihilation fraction ( @xmath12 ) of a selected @xmath8 sample in order to increase the sensitivity to the stated np in the @xmath9-spectrum . moreover , in light of the upcoming experimental results , we focus our work to the @xmath13 tev region of energy . this range of energies has two special features which determine the results in this work : _ i ) _ we propose to use as one of the variables to discriminate quark - annihilation events , the center of mass scattering angle , which is useful for this purpose at high @xmath9 ; and _ ii ) _ the lack of statistics at high @xmath9 requires a special selection criteria that with small cuts makes considerable increases in the sensitivity . this latter is in contrast to the low energy regime , where strong cuts that yield a very large increase in the sensitivity are sought . the search for kinematic cuts that increase @xmath12 in a given @xmath8 sample has been studied in the last years @xcite mainly to increase the sensitivity in the top charge - asymmetry at the lhc and very few in the @xmath9-spectrum @xcite . these articles deal mainly with two different features in the production mechanisms that allow to distinguish @xmath14 production in @xmath7 . the first one is that valence quarks inside the colliding protons tend to have larger fraction of the proton momentum than gluons and anti - quarks . therefore , top pairs produced through quark - annihilation tend to be boosted along the pipe line . this characteristic can be measured through the kinematic variable = , [ beta ] which ranges from @xmath15 for not boosted pairs to @xmath16 for maximum boosted pairs . notice for future purposes that the origin of this feature is in the proton parton distribution function ( pdf ) . the second feature which allows to isolate top pairs produced through quark - annihilation is that initial state gluons tend to produce more initial state radiation ( isr ) than quarks . this is an effect due to a larger numeric factor in the three gluon vertex in the qcd lagrangian . this characteristic can be measured through the transverse momentum of the top pair , p_t=| p_t(t ) + p_t(|t ) | . [ pt ] the larger is @xmath3 , the more probable is to have a gluon in the initial state . notice that , in this case , the origin of this feature is pure qcd and is independent not only from the pdf , but also from the dynamics in the @xmath8 production process . the usage of these two variables to increase the sensitivity to np in the @xmath9-spectrum has been exploited in refs . @xcite . there is a third variable which may serve to distinguish quark - annihilation processes . this is the center of mass scattering angle ( @xmath4 ) of the top quark direction relative to the right moving parton . this variable has been previously studied in refs . since , as we show in the article , to take profit of this variable requires large @xmath9 , we use it to study the upcoming experimental results where the @xmath9 region above @xmath16 tev will be better analyzed . this variable is related to the dynamics of the @xmath17 process and the spin of the particles . the key feature in using the angular distribution is that for large @xmath9 the quark - annihilation processes have a smooth angular distribution , whereas gluon - fusion processes accumulate most of the events in the forward @xmath18 region . or , to be more precise , the @xmath19-channel @xmath20 amplitude is the one that peaks the production in the forward region for relativistic tops . therefore , the angular distribution may also be understood as a discriminator of @xmath21- from @xmath19-channel contributions , as in dijet resonance searches @xcite . the aim of this work is to analyze simultaneously cuts in all three variables ( @xmath22 and @xmath4 ) to enhance the sensitivity to resonant np in the @xmath9-spectrum by increasing @xmath12 . notice that , although all three variables have different origins , they have some degree of indirect correlation through the pdf s . in any case due to their different origin none of them can be expressed as a function of the other two . the price to pay for increasing @xmath12 is to reduce the fraction of the original sample ( @xmath23 ) that is selected for the analysis . this reduction , which always yields an increase in the statistical uncertainty , may not worth the selection . the critical point up to where it is convenient to cut the sample depends on how @xmath12 and the total uncertainty behave as a function of @xmath23 . henceforth , it is crucial at this point also to take into account systematic uncertainties in the analysis in order to know if the selection cut is useful or not . however , since a full realistic simulation of the systematic uncertainty would require a detailed simulation of the detector , which is beyond the scope of this work , we use a simpler model which is enough to lead us to the desired interplay between @xmath12 , @xmath23 and statistic and systematic uncertainties in order to decide whether the selection is suitable or not . under this scenario , for the setup of the problem and its solution , it will be enough to consider @xmath17 processes up to parton level including isr . hadronization , detector simulation and reconstruction would be useful only if one could perform a fully realistic detector simulation including its systematic uncertainties . this is left for the experimental groups in case they consider to follow the guidelines in this work . the purpose of this article is to show that exists an interplay between a selection cut on the three variables @xmath2 , @xmath3 and @xmath4 and the statistic and systematic uncertainties , that yields an optimal cut which enhances the sensitivity of the @xmath9-spectrum to np that couples to quarks . we show along the article that this interplay , and therefore also the optimal cut , is strongly dependent on the energy of the process and the accumulated luminosity . for instance , we show that for the 2012 lhc data , as the energy increases beyond @xmath24 tev , strong cuts in @xmath2 should be gradually replaced by mild cuts in @xmath4 . this article is divided as follows . in next section we present the analytic results for the angular distributions in the standard model ( sm ) for quark - annihilation and @xmath25-fusion . we also study angular distribution of np models which couple to quarks and not to gluons . in section 3 , we study the parton level variables @xmath2 , @xmath3 and @xmath4 and their interplay with @xmath12 and @xmath23 , as well as the relationships of these two with the uncertainties . in section 4 , we present an example of a resonant np in @xmath8 production and show how the progressive cuts in the studied variables leads to the visibility of the resonant bump . section 5 contains a discussion on the results and previous works , and section 6 the conclusions . in this section we study analytically the leading order @xmath8 production through the processes @xmath26 and @xmath20 within the sm and for different possible resonant np models . at high energy , events initiated by quarks have different angular distributions than those initiated by gluons . this is due to the @xmath25-fusion amplitude where a top is exchanged in the @xmath19-channel producing @xmath8 pairs mainly in the forward region . therefore , one may impose kinematic cuts to the relevant angular variables in order to disentangle contributions from the two different production modes . we begin analyzing the angular distribution of the two production mechanisms within sm . the expressions for the production squared amplitudes are given in the helicity basis in the initial parton center of mass frame . a sum over spin and color for the initial and final states as well as proper spin and color averaging factors which have not been included are assumed all along this section . for the quark - annihilation initial state , we have itself . ] @xcite _ ll , rr m ( q |q t |t ) ^2 & = & 8 g^4 ( 1-_t^2 ) ^2 , [ qqll ] + _ lr , rl m ( q |q t |t ) ^2 & = & 8 g^4 ( 1+^2 ) . [ qqlr ] where @xmath27(@xmath28 ) stands for left(right ) helicity of the @xmath19 or @xmath29 , @xmath4 is defined as the angle between the direction of motion of the top and the initial parton moving to the right , and @xmath30 is the speed of the top quark . as it is apparent from these equations , the like - helicity ( @xmath31 and @xmath32 ) @xmath26 production is suppressed as @xmath33 . this behavior is a consequence of the helicity conservation . for the @xmath25 initial state we have @xcite _ ll , rr m ( gg t _ lr , rl m ( gg t |t ) ^2 & = & g^4 _ t^2 ^2 ( 1 + ^2 ) . [ gglr ] for @xmath30 close to @xmath16 , the common factor to both equations governs the growth of @xmath8 production through @xmath25-fusion as @xmath34 . furthermore , this growth becomes stronger as @xmath33 . hence , we may expect that the angle @xmath4 becomes a more useful variable to distinguish quark - annihilation from @xmath25 events for increasing values of the invariant mass of the @xmath8 pair . again , like - helicity contribution is suppressed in the limit @xmath33 as for quark - annihilation production , except for @xmath35 where both production mechanisms contribute . this is understood because the @xmath19-channel that arises in this limit allows to flip the helicity of the tops . notice that at threshold ( @xmath36 ) both production mechanisms have an isotropic angular distribution . then , no angular distribution based discrimination between the two production mechanisms is expected close to the threshold . + from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] .5 cm + from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] from left to right : normalized unlike- , like - helicity and total angular distributions for @xmath26 ( top ) and @xmath37 ( bottom ) for three values of @xmath38 ( solid blue ) , @xmath39 ( dashed red ) , @xmath40 ( dot - dashed green ) . these correspond to @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively . for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 ( notice that this only takes place when the angular distributions are normalized).,scaledwidth=90.0% ] in fig . [ analytic ] we show the normalized like- ( left ) and unlike- ( center ) helicity angular distributions from eqs . ( [ qqll])-([gglr ] ) for three values of @xmath44 ( @xmath41 gev , @xmath42 gev and @xmath43 gev , respectively ) . also , we plot the total angular distributions for the sum of all final helicity states ( right ) . notice that for quark - annihilation production the normalized like- and unlike - helicity angular distribution are independent of @xmath30 since in eq.([qqll ] ) @xmath30 dependence enters just as a global factor and in eq.([qqlr ] ) there is no dependence on @xmath30 at all . we can see that @xmath26 production in an unlike - helicity state is more likely to be produced in the forward region whereas @xmath26 production in a like - helicity state is mostly central . the suppression in the like - helicity @xmath26 production for increasing values of @xmath9 results in a slight population of the forward region as we can see from the total @xmath26 angular distribution . on the other hand , for @xmath45 the unlike - helicity production for @xmath20 is turned on and both like- and unlike - helicity production begin to populate the forward region as @xmath30 increases . hence , although both mechanisms tend to populate those regions for increasing values of @xmath9 , @xmath20 production makes it further significantly due to the particular behavior of its total angular distribution as @xmath33 . the comparison between @xmath26 and @xmath20 productions in this limit is shown in the right - hand side column of fig . [ analytic ] . these differences between the angular distribution of the two production mechanisms for large @xmath9 will prove to be useful to distinguish them . we may expect from these plots that as @xmath33 a loose upper cut in @xmath46 results in a slight decrease of the total number of events together with a sudden increment of the quark - annihilation fraction of the sample . this is in opposition to @xmath2 where one usually imposes tight cuts that end up removing a large amount of events from the sample in order to get an enhancement in the quark - annihilation fraction . + [ color online ] angular distribution of the @xmath47 np events through spin-0 ( a ) , spin-1 vectorial or axial ( b ) or spin-2 ( c ) massive resonances . the different curves correspond to different @xmath30 as stated in fig . [ analytic].,title="fig:",scaledwidth=90.0% ] ( a ) + [ color online ] angular distribution of the @xmath47 np events through spin-0 ( a ) , spin-1 vectorial or axial ( b ) or spin-2 ( c ) massive resonances . the different curves correspond to different @xmath30 as stated in fig . [ analytic].,title="fig:",scaledwidth=90.0% ] ( b ) + [ color online ] angular distribution of the @xmath47 np events through spin-0 ( a ) , spin-1 vectorial or axial ( b ) or spin-2 ( c ) massive resonances . the different curves correspond to different @xmath30 as stated in fig . [ analytic].,title="fig:",scaledwidth=90.0% ] ( c ) in the presence of np that couples only to quarks , the angular distribution of the @xmath25-fusion events remains the same as in the sm . therefore , the fraction of these events that are removed with a given cut on @xmath4 does not change with respect to the sm . for instance , we obtain from this simple parton level analysis that making a forward angular cut @xmath48 for @xmath49 , only @xmath50 of the @xmath25-fusion initiated events pass the cut . this has to be compared with the fraction of sm quark - annihilation events that are left after the same selection , which is @xmath51 . moreover , it is easy to verify that this selection cut is also useful for np propagating in the @xmath21-channel . we test this with three benchmark cases where the np corresponds to spin-0 , spin-1 or spin-2 massive resonances as is shown in fig . [ np ] . since the spin-0 colorless resonance ( fig . [ np]a ) does not interfere with the quark - annihilation qcd production and has no angular dependence , the same cut yields that @xmath52 of the np quark - annihilation events are selected . for the spin-1 resonance ( fig . [ np]b ) we choose a color octet @xmath53 with vectorial couplings . the amplitude squared in this case is _ m(q |q g / g t |t ) ^2 = a(s ) ( 2-_t^2 + _ t^2 ^2 ) . [ m_np ] the same angular dependence that the sm is obtained with @xmath54 being a factor that depends on the resonance mass and width . this is consequence of the equal angular dependence of the qcd and np amplitudes . thus , we also obtain that @xmath55 of this np quark - annihilation events are left after the selection . moreover , we find no difference with this and the case of an axial spin-1 color octet . ( notice that in this case , if we had chosen @xmath4 as the angle between the top and the quark instead of a fixed beam direction , as is usual in forward - backward asymmetry studies , we would have obtained an asymmetric angular distribution for the @xmath56 process , but the symmetric cut @xmath48 would have had the same effect . ) finally , in fig . [ np]c we illustrate as an example a spin-2 colorless resonance @xcite which we restrict its couplings to quarks . in this case we find that @xmath57 of the np events pass the selection . thus , provided that the np propagates in the @xmath21-channel , the cut in @xmath4 could be used to improve the quark - annihilation fraction and the sensitivity to resonant np . in fact , although we find some slight differences in the efficiency of a given cut on @xmath46 to keep quark - annihilation events from np , we always obtain the same amelioration in reducing the dilution coming from @xmath25-fusion events . we now discuss results from simulations of @xmath8 events within the sm in order to find phase space regions where quark - annihilation is favored over gluon - fusion as the dominant production mechanism . we consider each kinematic variable separately to analyze them as potential filters of quark - annihilation processes . finally , we perform random scans of cuts on these variables and show that to impose restrictions on @xmath4 is more efficient to isolate quark - annihilation production than to force limits on @xmath2 for the high @xmath9 region . we have simulated sm inclusive @xmath8 production at the lhc@8tev up to one extra jet with madgraph5 @xcite and then showered it with pythia @xcite . to avoid double counting , we have matched the matrix element to the parton shower through the mlm scheme @xcite implemented in madgraph5 . this generation procedure allows to include nontrivial @xmath3 distributions . it is beyond the scope of this work to deal with final states , therefore we have assumed a simplified selection cut @xmath58 ( for future purposes , it is worth noticing that @xmath59 corresponds to @xmath60 , for the case where both laboratory and center of mass frames coincide ) . the prediction for the inclusive production cross section has been obtained with mcfm lhc@8tev at nlo as 225.2 pb @xcite and an overall selection efficiency for semileptonic @xmath8 detection of @xmath61 @xcite has been used in the analysis . this strategy has ended up with a total of 400.000 @xmath8 events at the 2012 s 30 fb@xmath1 . + ( 94200 events ) + ( 2200 events ) + ( 120 events ) differential @xmath8 distribution for quark - annihilation ( red dashed ) and gluon - fusion ( black solid ) events for each one of the kinematic variables @xmath2 , @xmath3 and @xmath4 in three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . all the distributions have been normalized to one.,scaledwidth=100.0% ] differential @xmath8 distribution for quark - annihilation ( red dashed ) and gluon - fusion ( black solid ) events for each one of the kinematic variables @xmath2 , @xmath3 and @xmath4 in three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . all the distributions have been normalized to one.,scaledwidth=100.0% ] differential @xmath8 distribution for quark - annihilation ( red dashed ) and gluon - fusion ( black solid ) events for each one of the kinematic variables @xmath2 , @xmath3 and @xmath4 in three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . all the distributions have been normalized to one.,scaledwidth=100.0% ] we show in fig . [ histograms1 ] the differential distribution of quark - annihilation ( red dashed ) and gluon - fusion ( black continuum ) for each one of the kinematic variables @xmath2 , @xmath3 and @xmath4 for three different bins of @xmath9 ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) that will be used from now on all along the analysis . notice also that the central values of these bins correspond to the @xmath9 values selected in the previous section as referential cases . since all the distributions have been normalized to one , the plots are not useful to compare the number of events initiated by quark - annihilation or gluon - fusion . instead , they allow to evaluate whether a given cut is suitable for increasing the quark - annihilation fraction . thus , we can see from the first row of the figure that quark - annihilation events tend to populate the high @xmath2 region more than gluon - fusion events . in particular , the peak in the distribution of quark - annihilation events at large @xmath2 in the 400 - 450 gev can be understood according to the following . for large @xmath62 , due to the suppression of the pdf s , it is easier the energy to be shared by the initial quark and anti - quark than having a configuration with large momentum difference between them . this situation is relaxed as @xmath9 decreases and it is more likely to find a quark with a large momentum . we conclude from the plots in the first row that a lower cut in @xmath2 always increases @xmath12 . by contrast , from the second row , we expect that an upper cut in @xmath3 may favor the quark - annihilation production . finally , the angular distributions for both production mechanisms is depicted in the third row . as we have discussed in the previous section , both quark - annihilation and gluon - fusion events have essentially the same almost isotropic distribution in the 400 - 450 gev bin . for increasing @xmath9 values , the angular distributions of both mechanisms starts to differentiate from each other . for the 1000 - 1100 gev bin and even more remarkably for the 1600 - 1800 gev bin , gluon - fusion events produce @xmath19 and @xmath29 around the incoming direction of the initial partons whereas the events initiated by quark - annihilation do not have such a strong @xmath4 dependence . we notice that the efficiency in performing cuts on @xmath2 and @xmath3 to disentangle quark - annihilation from gluon - fusion events is slightly sensitive to @xmath9 . on the contrary , from the angular distributions we observe that gluon - fusion events populate the forward direction for increasing @xmath62 values and @xmath4 becomes a more convenient variable to discriminate quark - annihilation events . in order to go into greater detail on this matter , we have plotted in fig . [ fq_vs_fs ] the quark - annihilation fraction ( @xmath12 ) vs. the fraction of events ( @xmath23 ) that remain after a given cut for each variable . for the 1600 - 1800 gev bin , a larger quark - annihilation fraction with a less loss of events in the sample is achieved by setting cuts on @xmath4 . therefore , it is expected that cuts on @xmath4 to be more suitable to small samples of events as it is the case in resonance searches beyond @xmath16 tev . in summary , cuts on @xmath4 may lead to larger values of @xmath23 in the relevant invariant mass region . @xmath12 ( vertical axes ) vs. @xmath23 as the sample is cut in @xmath63 ( red dotted ) , @xmath64 ( blue dashed ) or @xmath65 ( magenta solid ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . the range of the cuts are @xmath66 $ ] , @xmath67 $ ] gev and @xmath68 $ ] , scaledwidth=100.0% ] @xmath12 ( vertical axes ) vs. @xmath23 as the sample is cut in @xmath63 ( red dotted ) , @xmath64 ( blue dashed ) or @xmath65 ( magenta solid ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . the range of the cuts are @xmath66 $ ] , @xmath67 $ ] gev and @xmath68 $ ] , scaledwidth=100.0% ] @xmath12 ( vertical axes ) vs. @xmath23 as the sample is cut in @xmath63 ( red dotted ) , @xmath64 ( blue dashed ) or @xmath65 ( magenta solid ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev ) . the range of the cuts are @xmath66 $ ] , @xmath67 $ ] gev and @xmath68 $ ] , scaledwidth=100.0% ] so far we have considered each kinematic variable separately . now , we combine cuts in the three variables with the aim to improve the quark - annihilation fraction . we perform then simultaneously a lower random cut @xmath69 and upper random cuts @xmath70 and @xmath71 . we have set @xmath72 $ ] , @xmath67 $ ] gev and @xmath68 $ ] . a cut on @xmath2 stronger than @xmath73 may suffer from large systematic uncertainties @xcite and @xmath74 gev @xcite corresponds to the minimal experimental sensitivity . in fig . [ fq_vs_fs.scanning ] we show @xmath12 vs. @xmath23 for a scan on the three variables ( vertical lines area ) , and only in @xmath2 and @xmath3 ( shaded area ) ( namely , no cut on @xmath4 is demanded ) . we have plotted for the three invariant mass bins 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev . as expected from the discussion given in the previous section , an enhancement in @xmath12 is obtained at large invariant masses when we combine cuts on all the three variables instead of only imposing cuts on @xmath2 and @xmath3 . moreover , we have also checked that , for the 1600 - 1800 gev bin , the maximum @xmath12 for a fixed @xmath23 is reached without any cut on @xmath2 . the independence on the cut on @xmath2 is consistent with the behavior observed in fig . [ fq_vs_fs ] within the same @xmath9 range . therefore , in order to end up with @xmath23 as large as possible for a given value of @xmath12 , we conclude that performing cuts on @xmath4 is more efficient than restricting @xmath2 . this is an important observation since @xmath2 is usually thought as a central kinematic variable in phenomenological analyses that aim at achieving quark - annihilation and gluon - fusion discrimination @xcite . these sm results can be used to search for new physics in the @xmath8 invariant mass spectrum . for the 1600 - 1800 gev bin , simulated np distributions follow the same behavior as in right plots in fig . [ histograms1 ] . in particular , to impose upper cuts on @xmath46 is still an efficient procedure to remove a considerable fraction of events initiated by gluon - fusion while keeping a large @xmath23 , as discussed in previous section . therefore , the results of this section provide us with a guideline to look for new physics in the @xmath9-spectrum . @xmath12 ( vertical axes ) vs. @xmath23 for a scan on the cuts on @xmath2 , @xmath3 and @xmath4 ( vertical lines area ) , and on @xmath2 and @xmath3 ( shaded area ) ( no cut on @xmath4 ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev).,scaledwidth=100.0% ] @xmath12 ( vertical axes ) vs. @xmath23 for a scan on the cuts on @xmath2 , @xmath3 and @xmath4 ( vertical lines area ) , and on @xmath2 and @xmath3 ( shaded area ) ( no cut on @xmath4 ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev).,scaledwidth=100.0% ] @xmath12 ( vertical axes ) vs. @xmath23 for a scan on the cuts on @xmath2 , @xmath3 and @xmath4 ( vertical lines area ) , and on @xmath2 and @xmath3 ( shaded area ) ( no cut on @xmath4 ) for three invariant mass bins ( 400 - 450 gev , 1000 - 1100 gev and 1600 - 1800 gev).,scaledwidth=100.0% ] we consider now a possible matter of concern since , as the value of @xmath12 becomes larger , the number of events in the sample decreases with a logical increment in the statistical uncertainty . this may spoil the potential power of the whole analysis and , for resonance searches , we do not only need a large @xmath12 but also to have uncertainties under control . from fig . [ fq_vs_fs.scanning ] we can see that , as @xmath23 gets reduced , there is always a set of cuts that improves @xmath12 . if the systematic uncertainties dominate over the statistical ones , any cut enhancing @xmath12 always improves the sensitivity to new physics . however , if we are searching for resonances in the region beyond @xmath16 tev , a reduction on the amount of events is expected and further cuts may lead immediately to a regime where both systematic and statistical uncertainties are competitive . even if @xmath12 is large , the statistical uncertainties can spoil the sensitivity to np @xcite . therefore , a regime where systematic and statistical uncertainties are of the same order has to be considered as a minimal necessary condition to be fulfilled . it is beyond the scope of this work to perform an exhaustive analysis of the uncertainties and , for the sake of the discussion , we simply assume that the relative systematic uncertainty is constant . this is a good approximation as far as we do not push the cuts to extreme values . we model the systematic uncertainty in each bin as @xmath75 , where @xmath76 is the number of events in the bin and @xmath77 is the relative systematic error in corresponding bin . a typical value for @xmath77 in a @xmath8 -spectrum measurement without jet substructure top tagging is about @xmath78 ( @xmath79 ) @xcite . ( this number , and the selection efficiency , are expected to change if top tagging through jet substructure @xcite is used in the analysis . ) thus , we get that the systematic and statistical uncertainties are of the same order provided that @xmath80 . it is worth noticing at this point that if the systematic uncertainties are reduced ( @xmath77 decreases ) then the optimal number of events for a given bin increases in order to reduce the statistical uncertainties . therefore , if the systematic uncertainties are reduced then the optimal selection cuts should be relaxed . we illustrate the above discussion with an example . at @xmath81 tev with @xmath82 fb@xmath1 , we expect in the sm 120 events in the bin of 1600 - 1800 gev of invariant mass . as we have mentioned , the systematic and statistical uncertainties will be comparable for @xmath83 , i.e. @xmath84 . from fig . [ fq_vs_fs.scanning ] we observe that for @xmath84 there exists a set of cuts on @xmath2 ( which turns out to be needless , @xmath85 ) , @xmath3 and @xmath4 that achieves @xmath86 . whereas for cuts only on @xmath2 and @xmath3 we would need @xmath87 to achieve the same @xmath12 ( see the dashed line in fig . [ fq_vs_fs.scanning]c ) . accordingly , we would obtain the same significance with a cut on @xmath2 , @xmath3 and @xmath4 at 30 fb@xmath1 as with a cut only on @xmath2 and @xmath3 for 60 fb@xmath1 . for this reference example , the set of cuts is @xmath88 , @xmath89 gev and @xmath90 for the former case , and @xmath91 and @xmath92 gev for the latter . this case shows that to include @xmath4 as a discrimination variable is equivalent to double the luminosity . in this section we apply the previously discussed cuts to a specific np model in order to explicitly show the amelioration in the invariant mass spectrum sensitivity to a new resonance . we study the effects of including cuts in different sets of @xmath2 , @xmath3 and in @xmath4 to visualize the improvement in sensitivity as more variables are included in the cuts . @xmath8 invariant mass spectrum for the @xmath82 fb@xmath1 expected lhc sample . the ( black ) histogram with error bars corresponds to only sm , and the ( red ) histogram corresponds to sm+np . in panel ( a ) we plot both spectrum without applying any cut on the sample , in ( b ) we apply the best cut that can be achieved in using @xmath2 and @xmath3 , in ( c ) the same as in ( b ) but using @xmath2 , @xmath3 and @xmath4 , and finally ( d ) corresponds to an ideal cut , as explained in the text . in each panel we determine the significance of the deviation according to a @xmath93 test ( see text ) . , title="fig:",scaledwidth=90.0% ] ( a ) @xmath8 invariant mass spectrum for the @xmath82 fb@xmath1 expected lhc sample . the ( black ) histogram with error bars corresponds to only sm , and the ( red ) histogram corresponds to sm+np . in panel ( a ) we plot both spectrum without applying any cut on the sample , in ( b ) we apply the best cut that can be achieved in using @xmath2 and @xmath3 , in ( c ) the same as in ( b ) but using @xmath2 , @xmath3 and @xmath4 , and finally ( d ) corresponds to an ideal cut , as explained in the text . in each panel we determine the significance of the deviation according to a @xmath93 test ( see text ) . , title="fig:",scaledwidth=90.0% ] ( b ) @xmath8 invariant mass spectrum for the @xmath82 fb@xmath1 expected lhc sample . the ( black ) histogram with error bars corresponds to only sm , and the ( red ) histogram corresponds to sm+np . in panel ( a ) we plot both spectrum without applying any cut on the sample , in ( b ) we apply the best cut that can be achieved in using @xmath2 and @xmath3 , in ( c ) the same as in ( b ) but using @xmath2 , @xmath3 and @xmath4 , and finally ( d ) corresponds to an ideal cut , as explained in the text . in each panel we determine the significance of the deviation according to a @xmath93 test ( see text ) . , title="fig:",scaledwidth=90.0% ] ( c ) @xmath8 invariant mass spectrum for the @xmath82 fb@xmath1 expected lhc sample . the ( black ) histogram with error bars corresponds to only sm , and the ( red ) histogram corresponds to sm+np . in panel ( a ) we plot both spectrum without applying any cut on the sample , in ( b ) we apply the best cut that can be achieved in using @xmath2 and @xmath3 , in ( c ) the same as in ( b ) but using @xmath2 , @xmath3 and @xmath4 , and finally ( d ) corresponds to an ideal cut , as explained in the text . in each panel we determine the significance of the deviation according to a @xmath93 test ( see text ) . , title="fig:",scaledwidth=90.0% ] ( d ) we take as an example a benchmark model with a gluon resonance of @xmath94 tev reminiscent of strongly coupled / composite models . a characteristic feature of this kind of models is that this massive gluon couples stronger to the heavy than to the light quarks , and it can not be created from a fusion of two qcd gluons . thus , this kind of models can illustrate and test the studied cuts . the interaction with sm particles is : l_g = g_s _ q_i f_q |q g q where @xmath95 stands for all quarks . we take @xmath96 , @xmath97 and all other couplings are set to zero . with this choice of parameters the width of @xmath53 yields @xmath98 gev , and the resonance would not be visible with the lhc running at @xmath81 tev and an accumulated luminosity of @xmath82 fb@xmath1 without any cut to improve the @xmath12 fraction . with the same computational tools used in the previous section , we have simulated a @xmath82 fb@xmath1 sample of @xmath17 and assumed a reference selection efficiency of @xmath61 @xcite . we have simulated only sm and jointly sm+np , to take into account the interference coming from the large width effects . in fig . [ histograms ] we show the comparison in the region of interest of the expected @xmath9-spectrum of only sm ( histogram with error bars , in black ) and sm+np ( histogram , in red ) for : ( @xmath99 ) no cuts at all ; ( @xmath100 ) cuts in @xmath2 and @xmath3 ; ( @xmath77 ) cuts in @xmath2 , @xmath3 and @xmath4 ; and @xmath101 an ideal cut where only the quark - annihilation events in the sample are left . the error bars correspond to statistical and systematic uncertainties , as discussed in the previous section . we choose the cuts in plots @xmath102 and @xmath103 from a scanning that looks for the cuts that maximize the significance of a @xmath93 test to 3 bins of @xmath104 gev width from @xmath105 gev to @xmath106 gev . we have checked that if we had chosen the _ a priori _ cuts from the previous section sm analysis to the @xmath107 gev bin , then we would have obtained an analogous , but not optimal , improvement . .[table][color online ] set of cuts used to differentiate sm+np from sm using different set of variables . the first two columns indicate the plot in fig . [ histograms ] and the performed cuts , the third and fourth columns show their effect on the sample ( @xmath23 and @xmath12 ) , and the following columns the expected differentiation in computing a @xmath93 test , as explained in the text . [ cols="^,^,<,<,<,<",options="header " , ] to quantify the gradual improvement in resolving the resonance in fig . [ histograms ] , we show in table [ table ] the @xmath23 , @xmath12 , @xmath108-values for the @xmath93 tests , and the gaussian - equivalent significance ( @xmath109 ) of all the set of cuts in the figure . we also show the value of the cuts for fig . [ histograms]b and fig . [ histograms]c . as we can determine from the table ( second and third row ) , a cut in @xmath2 is not too tight at these energies and luminosities . moreover , in passing from second to third row , when the cut in @xmath4 is included , we see that the cut in @xmath2 is practically replaced by a cut in @xmath4 . we see in this example that this exchange of cuts increases @xmath23 by @xmath110 and @xmath12 by @xmath111 , which yields an improvement in the @xmath108-value from @xmath112 to @xmath113 . we have repeated the same computations as in this section for other models of resonant np and obtained similar results . for the best cuts in the analogous to fig . [ histograms]c we always obtain @xmath114 , @xmath115 and @xmath116 gev . this is in agreement with the previous section analysis . it could be valuable to stress at this point that the results in this section have the only purpose of showing the qualitative effect of these cuts in the increment in the sensitivity through a specific example . the aim of this section is to relate the present analysis with previous works . besides @xmath3 and @xmath2 , the rapidity of the top and spin correlations of its decays also have been used to separate np and sm @xmath26 production from the @xmath20 background . first , we discuss in this section some differences among the present analysis that makes focus on @xmath4 as a suitable kinematic variable and previous studies where the rapidity @xmath117 is used to impose tight selection cuts . and finally , we discuss briefly about the different regimes where @xmath4 cuts and spin correlation are useful to improve the @xmath12 fraction . it is well known that there is a correlation between the center of mass angle @xmath4 and the difference of rapidity @xmath118 , y = , where @xmath119 . the partonic energy @xmath120 entering this relation may dilute the correlation between this two variables . only if @xmath30 is fixed , there is a one - to - one correspondence between @xmath4 and @xmath121 . however , if we look at a bin of invariant mass where @xmath30 has large variations then there could be a lower correlation between them . for instance , if we consider the @xmath9 bin of 1600 to 1800 gev , @xmath121 varies up to @xmath122 when @xmath4 is fixed . therefore , for a high @xmath9 bin , cuts on @xmath4 and @xmath121 are roughly equivalent . moreover , if we take several bins in the high @xmath9 region , for the same cut on @xmath4 , we will have a different ( but equivalent ) cut on @xmath121 in each bin . on the other hand , close to the threshold , @xmath30 varies greatly and the correlation is more diluted . thus , within the 400 - 425 gev bin , @xmath121 varies up to @xmath78 for fixed values of @xmath4 . cuts on @xmath123 have already been considered in the literature @xcite . however , they have been applied at low energies and for certain class of models . for instance , in @xmath124 flavor violating models , @xmath8 are produced through the exchange of the @xmath124 in the @xmath19-channel and then it produces @xmath8 events mostly in the forward region . this situation , which does not correspond to resonant np , is quite the opposite to the present analysis . for such models , it has been shown that a lower cut on @xmath123 enhances the sensitivity to the new interactions . since these interactions arise at low @xmath9 , there is not a considerable increment of the gluon - fusion fraction that may spoil the enhancement generated by the new interactions . in cases where the new physics contributions are significant at higher @xmath9 , we have to deal with an interplay between increasing the np @xmath19-channel contribution and decreasing the sensitivity with the increment of the gluon - fusion @xmath8 production . the rapidity has also been used in the context of the charge asymmetry in @xmath8 production at the lhc @xcite . in those cases , the basic idea is also to discriminate quark - annihilation from @xmath25-fusion events by imposing different cuts on @xmath125 and @xmath126 . the procedure traces the discrimination back to the correlation between @xmath117 and @xmath2 . however , the angular variable @xmath4 ( or @xmath121 ) plays no role in that method . finally , for many of the models that have been proposed to accommodate the @xmath127 , there exist a high @xmath9 resonance that induces an interference at low @xmath9 , for instance , a heavy axigluon that has been integrated out @xcite . therefore , the new physics contribution arise at low @xmath9 and this may seem to be the reason why @xmath4 has never been taken as a relevant kinematic variable in previous studies . spin correlations in the production and decay of tops have been extensively discussed @xcite . the aim of those studies is to compare predictions made within the sm or its extensions with the measured angular distributions of the decay products of @xmath19 and @xmath29 . in general , different models give place to distinct angular distribution patterns and the measurement of spin correlations may provide additional and relevant information on the structure of possible new interactions . such spin correlations are optimized when the asymmetry between like- and unlike - helicities reaches certain maximum value . usually , for a given axis from which the spin projection is defined , one seeks to perform kinematic cuts that increase this asymmetry . this in turn allows to discriminate like- and unlike - helicities . from eqs . ( [ qqll])-([gglr ] ) and their corresponding discussion in section 2 we have learned that angular distribution for both quark - annihilation and @xmath25-fusion production mechanisms are essentially flat close to the threshold . therefore , @xmath4 cuts can not be used to improve the quark - annihilation fraction . on the contrary , a distinction based on spin correlation is envisaged since only like - helicity @xmath20 production is possible whereas the two spin production mechanism are present for @xmath26 as we can see from the eqs . ( [ qqll])-([gglr ] ) . the opposite situation we encounter at large @xmath9 . there , only unlike - helicity @xmath26 and @xmath20 production mechanisms are possible , being hardly expected that spin correlation being useful to distinguish quark - annihilation from @xmath25-fusion production contrary to @xmath4 cuts that has been proven to be suitable to increase the quark - annihilation fraction . we have studied the collider phenomenology of @xmath8 resonance searches above 1 tev at 30 fb@xmath1 . we have focused our analysis on resonances that couple exclusively to quarks which , at the lhc , would be naturally hidden behind the large gluon - fusion @xmath0 production . the usual variables to enlarge the quark - annihilation fraction @xmath12 of a @xmath8 sample are the longitudinal and transverse momentum of the @xmath8 pair , @xmath2 and @xmath3 , respectively . in addition to these variables , we exploit the fact that for masses above 1 tev the large boost of the tops in the center of mass system yields a correlation between their center of mass scattered angle @xmath4 and the initial partons of the event ( this correlation is suppressed for invariant masses below 1 tev ) . since the origin of the relationship between the initial state partons and @xmath2 is the proton pdf , whereas for @xmath3 is the initial state radiation , and for @xmath4 the dynamics of the partonic collision , there are not a priori reasons to expect these variables to be considerably correlated , and all three should be used together to texture the initial state partons of an event . we have simulated @xmath8 production in the large invariant mass regime and scanned over different set of cuts on the variables mentioned above to increase the quark - annihilation fraction of the sample . given a set of invariant mass bins where the np resonance could be expected , the optimal cut is the one that enhances the most the significance of a resonance search over the sm background . this is achieved when a set of cuts increases @xmath12 while keeping the selected fraction @xmath23 still large since , for resonances beyond 1 tev , a critical reduction in the amount of events could spoil the signal due to an increase of the statistical uncertainties . our results show that @xmath2 is not as useful as expected in this resonance search . the reason for this is that this variable requires tight cuts in order to yield significant increases in @xmath12 , and there are not too many events in the large invariant mass bins in the 2012 data . on the other hand , @xmath4 does not reach large values of @xmath12 , but it has an important increase as one begins cutting the sample . hence , the reasons why @xmath4 is useful in the large invariant mass bins are two - folded : the variable requires relativistic tops to distinguish the initial state partons , and in this region the cuts should be moderate . at last , the cut in @xmath3 is always useful for any invariant mass bin . we have studied a reference case of a @xmath94 tev resonance and found that the optimal cuts in these three parton level variables is in agreement with the previous sections analyses . we have also shown how the significance of the resonance search behaves as one gradually includes all variables in the selection cuts . summarizing , when we include all three variables , we found that @xmath2 is not relevant , and @xmath3 and @xmath4 provide the most useful cuts to enhance the significance of the resonance search . the results in this article are a guideline to point the searches of np resonances above 1 tev for 30 fb@xmath1 at a center of mass energy of @xmath81 tev . it is important to stress the perspectives for upcoming luminosity and energy increases . for np resonances coupled to quarks and masses above 1 tev , a rise in the luminosity will make @xmath4 remain as a useful variable , whereas @xmath2 will improve as a quark - annihilation discriminator too . on the other hand , a rise of the center of mass energy , will allow the exploration of heavier resonances within this framework . we thank l. da rold for useful conversations . this work has been partially supported by anpcyt ( argentina ) under grants pict - prh 2009 - 0054 ( a.s ) and pict 2011 - 0359 ( a.e . and j.i.s.v . ) and by conicet ( argentina ) . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * , 1 ( 2012 ) [ arxiv:1207.7214 [ hep - ex ] ] ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * , 30 ( 2012 ) [ arxiv:1207.7235 [ hep - ex ] ] . https://twiki.cern.ch/twiki/bin/view/atlaspublic/supersymmetrypublicresults ; https://twiki.cern.ch/twiki/bin/view/cmspublic/physicsresultssus . the cdf collaboration , conf . note 10807 , http://www-cdf.fnal.gov/physics/new/top/ 2012/lepjet_afb_winter2012/cdf10807.pdf ; a. lister [ cdf and d0 collaboration ] , arxiv:0810.3350 [ hep - ex ] ; t. aaltonen _ et al . _ [ cdf collaboration ] , phys . * 101 * , 202001 ( 2008 ) [ arxiv:0806.2472 [ hep - ex ] ] ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . rev . d * 84 * , 112005 ( 2011 ) [ arxiv:1107.4995 [ hep - ex ] ] ; t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 83 * , 112003 ( 2011 ) [ arxiv:1101.0034 [ hep - ex ] ] . arxiv:1203.4211 [ hep - ex ] ; s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 709 * , 28 ( 2012 ) [ arxiv:1112.5100 [ hep - ex ] ] ; g. aad _ et al . _ [ atlas collaboration ] , eur . j. c * 72 * , 2039 ( 2012 ) [ arxiv:1203.4211 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 717 * , 129 ( 2012 ) [ arxiv:1207.0065 [ hep - ex ] ] . r. barcelo , a. carmona , m. masip and j. santiago , phys . b * 707 * , 88 ( 2012 ) [ arxiv:1106.4054 [ hep - ph ] ] ; e. alvarez , l. da rold , j. i. s. vietto and a. szynkman , jhep * 1109 * , 007 ( 2011 ) [ arxiv:1107.1473 [ hep - ph ] ] ; g. marques tavares and m. schmaltz , phys . d * 84 * , 054008 ( 2011 ) [ arxiv:1107.0978 [ hep - ph ] ] ; e. alvarez and e. c. leskow , arxiv:1209.4354 [ hep - ph ] ; j. drobnak , a. l. kagan , j. f. kamenik , g. perez and j. zupan , arxiv:1209.4872 [ hep - ph ] . y. -k . wang , b. xiao and s. -h . zhu , phys . d * 82 * , 094011 ( 2010 ) [ arxiv:1008.2685 [ hep - ph ] ] ; y. -k . wang , b. xiao and s. -h . zhu , phys . d * 83 * , 015002 ( 2011 ) [ arxiv:1011.1428 [ hep - ph ] ] ; b. xiao , y. -k . wang , z. -q . zhou and s. -h . zhu , phys . d * 83 * , 057503 ( 2011 ) [ arxiv:1101.2507 [ hep - ph ] ] ; j. -f . arguin , m. freytsis and z. ligeti , phys . d * 84 * , 071504 ( 2011 ) [ arxiv:1107.4090 [ hep - ph ] ] . j. l. hewett , j. shelton , m. spannowsky , t. m. p. tait and m. takeuchi , phys . d * 84 * , 054005 ( 2011 ) [ arxiv:1103.4618 [ hep - ph ] ] . y. bai and z. han , jhep * 1202 * , 135 ( 2012 ) [ arxiv:1106.5071 [ hep - ph ] ] . j. a. aguilar - saavedra , a. juste and f. rubbo , phys . lett . b * 707 * , 92 ( 2012 ) [ arxiv:1109.3710 [ hep - ph ] ] . j. h. kuhn and g. rodrigo , jhep * 1201 * , 063 ( 2012 ) [ arxiv:1109.6830 [ hep - ph ] ] . e. alvarez , phys . d * 85 * , 094026 ( 2012 ) [ arxiv:1202.6622 [ hep - ph ] ] . k. agashe , a. belyaev , t. krupovnickas , g. perez and j. virzi , phys . d * 77 * , 015003 ( 2008 ) [ hep - ph/0612015 ] . e. alvarez , phys . d * 86 * , 037501 ( 2012 ) [ arxiv:1205.5267 [ hep - ph ] ] . v. barger , t. han and d. g. e. walker , phys . lett . * 100 * , 031801 ( 2008 ) [ hep - ph/0612016 ] . r. frederix and f. maltoni , jhep * 0901 * , 047 ( 2009 ) [ arxiv:0712.2355 [ hep - ph ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 704 * , 123 ( 2011 ) [ arxiv:1107.4771 [ hep - ex ] ] ; [ atlas collaboration ] , atlas - conf-2012 - 038 . g. mahlon and s. j. parke , phys . d * 53 * , 4886 ( 1996 ) [ hep - ph/9512264 ] . b. grinstein , c. w. murphy , d. pirtskhalava and p. uttayarat , jhep * 1208 * , 073 ( 2012 ) [ arxiv:1203.2183 [ hep - ph ] ] . j. alwall , p. demin , s. de visscher , r. frederix , m. herquet , f. maltoni , t. plehn , d. l. rainwater _ et al . _ , jhep * 0709 * ( 2007 ) 028 , [ arxiv:0706.2334 [ hep - ph ] ] . t. sjostrand , s. mrenna and p. z. skands , jhep * 0605 * ( 2006 ) 026 [ arxiv : hep - ph/0603175 ] . j. alwall , s. hoche , f. krauss , n. lavesson , l. lonnblad , f. maltoni , m. l. mangano and m. moretti _ et al . _ , eur . j. c * 53 * , 473 ( 2008 ) [ arxiv:0706.2569 [ hep - ph ] ] . [ cms collaboration ] , cms - pas - top-12 - 007 . j. m. campbell and r. k. ellis , nucl . proc . suppl . * 205 - 206 * , 10 ( 2010 ) [ arxiv:1007.3492 [ hep - ph ] ] . [ cms collaboration ] , cms - pas - top-11 - 009 . m. backovic and j. juknevich , arxiv:1212.2978 [ hep - ph ] ; j. gallicchio and m. d. schwartz , arxiv:1211.7038 [ hep - ph ] , and reference therein . s. berge and s. westhoff , arxiv:1208.4104 [ hep - ph ] . c. degrande , j. -m . gerard , c. grojean , f. maltoni and g. servant , jhep * 1103 * , 125 ( 2011 ) [ arxiv:1010.6304 [ hep - ph ] ] . g. mahlon and s. j. parke , phys . b * 411 * , 173 ( 1997 ) [ hep - ph/9706304 ] ; t. stelzer and s. willenbrock , phys . b * 374 * , 169 ( 1996 ) [ hep - ph/9512292 ] ; and many others .

this introductory review discusses the main problems facing the attempt to build quantum information processing systems ( like quantum computers ) from spin - based qubits . we emphasize bottom - up attempts using methods from chemistry . the essentials of quantum computing are explained , along with a description of the qubits and their interactions in terms of physical spin qubits . the main problem to be overcome in this whole field is decoherence - it must be considered in any design for qubits . we give an overview of how decoherence works , and then describe some of the practical ways to suppress contributions to decoherence from spin bath and oscillator bath environments , and from dipolar interactions . dipolar interactions create special problems of their own because of their long range . finally , taking into account the problems raised by decoherence , by dipolar interactions , and by architectural constraints , we discuss various strategies for making chemistry - based spin qubits , using both magnetic molecules and magnetic ions .

0807.1986

the basic question addressed in this short review is : how might one build a spin - based quantum computer using the bottom - up methods of chemistry ? before even starting on this topic , the reader might suppose that we should first answer two other questions , viz . ( i ) can one build a quantum computer at all ? ( ii ) why would this be a subject for chemists ? the answer to the first question is not known . although one can easily imagine various schemes for doing quantum information processing , to actually build a quantum information processing system " ( qips ) one has to deal with a really quite fundamental problem in nature , usually called the decoherence problem. quantum computing involves manipulating multiply entangled wave - functions , involving many different sub - systems in the qips - in these terribly complex quantum states , distributed over the whole qips , the individual wave - functions of the individual components lose all meaning and only the wave function of the entire system is physically meaningful . the problem is that such states are extremely delicate , and can be destroyed by very weak interactions with their surroundings - and yet in order to use them we need to be able to probe and manipulate them . physicists and chemists working in nanoscience have come up with many designs , and there has been considerable success in building qips involving just a few entangled qubits ( the simplest kind of sub - system , involving only two quantum states ) . this success has simply made the importance of the decoherence problem more obvious - solving it will be the key that unlocks the qips door ( and doubtless many other doors as well ) . concerning the second question - so far the most successful designs for qips have been made from ion traps@xcite , but it is not yet clear how one might scale these up to the larger systems , involving many entangled qubits , that are needed for quantum computation . the most likely designs may be hybrids between solid - state circuits of qubits , with communication effected by photons - nobody really knows yet . there are then 2 good reasons for concentrating on spin - based systems , made using bottom - up methods . the first is that a device which only uses spin degrees of freedom avoids moving charges around - this is good because moving charges invariably causes decoherence . the second is that if the basic sub - units are made by nature ( in the form of molecules or ions ) then quantum mechanics guarantees they will all be identical , provided we can eliminate impurities , defects , etc . this of course takes us into the realm of chemistry . in writing this review we have assumed a reader who may not be too familiar with this subject , but who is interested in understanding the main ideas and problems , and who is also looking for suggestions for future work in this area . we therefore begin , in section 2 , by describing briefly what a qips is , and how one would describe its physical components if they were indeed made from spin qubits using chemistry . this description also includes a discussion of the physical mechanisms responsible for decoherence in such systems . given the importance of decoherence , we have devoted all of section 3 to how it works . the emphasis is on simple pictures , which often convey a better feeling for what is going on than complex mathematics . then , in section 4 , we outline some features of the chemistry , and a few desiderata for any future architecture . this allows us to make some suggestions for future work . this last section is necessarily open - ended - at present nobody knows which designs may ultimately succeed . physicists and chemists on the one hand , and computer scientists on the other , have quite different ways of describing a system whose ultimate aim is to do computing ( quantum or classical ) . each kind of description is valuable , but the differences often cause misunderstanding . in this section we describe and then connect the two kinds of description ; we then discuss an effective hamiltonian for a set of qubits made from real materials . computer scientists like to reduce physical computational systems to `` turing machines''@xcite ; a clear physical discussion of this was given by feynman@xcite . in a turing machine , a `` tape '' moves forward or backwards past a `` head '' ; the head has the job of writing in or reading out particular discrete states ( labelled by different symbols ) on the tape ( and it may also erase states ) . there is also a table of instructions which tell the tape and head what to do ( the `` program '' ) , and a `` stable register '' or memory , which stores the state of the machine at any given time . ( turing thought of this as being analogous to the state of mind of a sentient being ) . the net result is shown in fig . [ turing ] . note that the state of a turing machine at any time can be understood in binary code as a string of 1 s and 0 s ( or a sequence of on / off states , or up / down , or heads / tails , etc . ) . even though real computers may not be constructed like a turing machine ( indeed turing machines provide a very clumsy model for a realistic computing architecture ) , nevertheless any computer can be mapped , in its logical and operational structure , to a turing machine . moreover , turing showed that a rather simply constructed machine of this kind ( since made even simpler ) could do any discrete computation . thus all digital computers can be described in terms of such `` universal ( classical ) turing machines '' . this idea of classical turing machines can be enlarged to cover quantum turing machines , in which the notion of a state is generalised to include the usual quantum superpositions of states , in the head , tape , register and program@xcite . this generalisation seems straightforward - however the way one describes such a system has led to much misunderstanding and a severe divergence in the literature . computer scientists like to talk in terms of states , and logical operations on them - in quantum computing these operations are called `` unitaries '' . a `` _ quantum computation _ '' is then a sequence of unitary transformations @xmath0 performed on some initial state @xmath1 , to produce the final state @xmath2 . the initial state is then the input to the computer , and the final state the output . these states are a quantum mechanical generalisation of the binary strings mentioned above - however they are much more complicated . consider , e.g. , a classical pair of binaries , which can be in one of four states ( ie , ( 11 ) , ( 10 ) , ( 01 ) , or ( 00 ) ) . in quantum mechanics , systems that have 2 states are called two - level systems ( tls ) , or qubits. now however a single qubit can exist in an arbitrary superposition of states @xmath3 , where the coefficients @xmath4 are real , @xmath5 , the states @xmath6 and @xmath7 correspond to the original classical states ( 0 ) and ( 1 ) , and @xmath8 are phases . pair _ of qubits has the general wave - function @xmath9 and we see that if we have @xmath10 qubits then we will end up with @xmath11 coefficients and @xmath11 relative phases ( of which one is redundant ) . the great power of quantum computation resides in all the information that can be stored in these phases , allowing one to superpose many computations at once@xcite . a key feature of these states is that typically the @xmath10 qubits are _entangled_ ; one can not write them as a product over independent wave - functions for each individual qubit . as first noted by einstein@xcite , this leads to very counter - intuitive features - it is simply not correct to assign a quantum state to an individual qubit , and only the @xmath10-qubit wave - function , existing with all its relative phases in a vast @xmath11-dimensional hilbert space , is physically meaningful . the above description is very abstract . to get a more intuitive understanding , it is helpful to map the qips onto a `` quantum walk '' system , in which a single particle hops around a graph . each node of the graph represents a state in the hilbert space of the qips , and the links between the nodes represent the transition matrix elements ( in general time - dependent ) between them . thus the time evolution of a qips can be represented as a `` quantum walk '' in information space@xcite ; and it is always possible to make this mapping for a qips composed of qubits . to see how this works one needs to specify a hamiltonian , to give the system dynamics . the simplest quantum walk hamiltonians take the form@xcite : @xmath12+\sum_{\mu } \epsilon_\mu(t)c^{\dagger}_\mu c_\mu \label{qw}\ ] ] where the operators @xmath13 , @xmath14 , create or destroy a particle at the @xmath15-th node , @xmath16 is the @xmath15-th node energy , and @xmath17 the non - diagonal hopping matrix element between nodes @xmath15 and @xmath18 . this hamiltonian must be generalized to describe `` composite quantum walks '' ( where discrete internal `` spin '' variables at each node can also couple to the walker ) if we are to fully describe a qips@xcite . the great advantage of this representation , at least for physicists and chemists , is that we have a good intuitive understanding of how particles move around such graphs . one also clearly sees where things can get complicated , if the @xmath19 and @xmath20 are strongly disordered or if their time - dependence is non - trivial . how this works in practice can be found in the literature@xcite . we now turn to a description of a qips of the kind favoured by physicists and chemists , one which will allow them to design and make one , written in terms of the actual physical components of the system . so let us now start from a much more physical model , using a hamiltonian written in terms of the qubits themselves : @xmath21 here the pauli operator @xmath22 operates on the j - th qubit , and both the local fields @xmath23 and the interaction @xmath24 can be time - dependent . now qubits are real physical objects , and many different physical systems have been proposed to make them . solid - state examples which have achieved some experimental success include spins in semiconductors and quantum dots@xcite , various designs based on superconductors@xcite , vacancy centers in diamond@xcite , single molecule magnets ( smms)@xcite , and rare earth spins@xcite . to focus the discussion here , let us consider an example . o@xmath25(oh)@xmath26(tacn)@xmath27@xmath28 single molecule magnet , showing the octahedral iron ions , the oxo and hydroxo bridges and the 1,4,7-triazacyclononane ( tacn ) ligands completing the coordination spheres.,scaledwidth=35.0% ] _ an smm example _ : a `` single molecule magnet '' ( smm ) typically has a magnetic core of transition metal ( tm ) ions , surrounded by a protective cage of organic ligands ; the inter - ion couplings are usually antiferromagnetic . a useful ( because well - understood ) example is the `` fe@xmath29 '' molecule@xcite , shown in fig.[fe8 ] , in which a core of eight fe@xmath30 , spin-@xmath31 ions forms a total spin-10 system . the simplest ( and very commonly employed ) model of a system like fe@xmath29 starts by reducing the spin hamiltonian @xmath32 written in terms of the individual tm spins ( for simplicity we ignore here terms like dzyaloshinski - moriya interactions ) to a single `` giant spin '' model of form @xmath33 where we assume all low - energy states have the same total spin @xmath34 . it is very common in the literature to drop many of the higher terms in this - for example , in the case of fe@xmath29 one usually writes the simple quadratic biaxial form @xmath35 which gives a 2-well crystal field potential acting on @xmath36 . here @xmath37 k , @xmath38 k , @xmath39 k , @xmath40 , and @xmath41 is the isotropic @xmath42-factor of the spin-10 moment . all the higher terms @xmath43 in ( [ gs ] ) are thus dropped . this effective hamiltonian for fe@xmath29 is used below roughly @xmath44 ; above this one certainly has to worry about other states , contained in the more general 10-spin hamiltonian ( [ ss ] ) . by application a real transverse field @xmath45 at different angles in the fe@xmath29 system.,scaledwidth=45.0% ] at still lower energies one can truncate ( [ gs ] ) to its lowest singlet or doublet - in the latter case we have a qubit hamiltonian of form @xmath46 where the effective field @xmath47 depends both in the crystal field potential and the applied field @xmath48 . a very important special case of this arises for the 2-well potential , where there is a natural basis for the states , in which @xmath49 represents the state localised in one well and @xmath50 the state localised in the other . one then often writes @xmath51 so that @xmath52 ; here @xmath45 , @xmath53 are the applied fields perpendicular and parallel to the fe@xmath29 easy axis . we can think of @xmath54 as an `` effective transverse field '' , driving transitions between the @xmath55 and @xmath50 states of the qubit . this effective field can be varied by applying the real external transverse field @xmath45 - how it does this is shown for fe@xmath29 in fig [ graphfe8 ] . the 2-well system possesses a real advantage - it is protected against external transverse field fluctuations , which have a small effect on @xmath54 . * connection between quantum walk and qubit formulations * : we have seen that a quantum computer can be thought of as a particle walking in information space , described by a hamiltonian like ( [ qw ] ) , or as a set of interaction qubits with hamiltonian ( [ qb ] ) . so how do we connect these two ? the answer is shown in fig . [ qbit - lattice ] for a set of 3 interacting qubits - in general the @xmath11 states of an @xmath10-qubit system map to the corners sites of an @xmath10-dimensional hypercube , and the quantum walker hops from one site to another on this hypercube . how it hops depends on the exact details of the qubit hamiltonian . although it is too lengthy to give here , the explicit mapping between the quantum walk hamiltonian @xmath56 in ( [ qb ] ) and the qubit hamiltonian @xmath57 in ( [ qw ] ) , for general qubit couplings , is fairly straightforward - it was given in ref.@xcite . another commonly used representation is in terms of `` quantum gates''@xcite , which can also be mapped to quantum walks@xcite , and thus also allows a connection to the quantum turing machine language . we have seen above how to describe a qips in terms of interacting qubits , of a quantum walker , and of a quantum turing machine ( and there are other descriptions - for example , we have for reasons of space ignored adiabatic quantum computation entirely ) . but how well do such toy models match the real world in which physicists and chemists work ? we now address this question . obviously both the general models ( [ qw ] ) , ( [ qb ] ) , as well as the specific models ( [ ss ] ) , ( [ gs ] ) , and ( [ qb0arrow ] ) for magnetic molecules , are oversimplified . many papers have appeared discussing both the failures of general models like ( [ qw ] ) , ( [ qb ] ) , or of the simple qubit forms ( [ qb0 ] ) and ( [ qb0arrow ] ) , as well as more specific corrections to the giant spin model ( [ gs ] ) . let us begin by enumerating some of the problems of such models as applied to smm s , and then discuss the more general case . the principal shortcomings are as follows : 1 . the giant spin model does not include many other discrete electronic degrees of freedom ( spin and charge ) ; if these lie at low enough energies , they couple to @xmath58 ( and of course to @xmath36 ) . because they are discrete , they behave like `` loose spins '' coupling to @xmath58 ( see refs.@xcite ) . this happens , for example , if one has structural transformations in the smm ( e.g. jahn - teller distortions ) , or if some exchange couplings @xmath59 are much weaker than others . note that in a multi - smm system , there will also be impurities , and `` rogue molecules '' which are different from the usual smm because of some internal distorsion , or because of stray local fields created by , e.g. , dislocations . these will also behave like `` loose spin '' degrees of freedom coupling to @xmath60 . 2 . in reality @xmath58 will couple to many other environmental modes in the vicinity of the smm . this includes any nuclear spins in the smm , coupled via hyperfine interactions , nuclear spins outside the smm , coupled via magnetic dipolar or possibly transfer hyperfine interactions@xcite . also @xmath58 will couple to photons ( via magnetic dipole and electric dipole couplings ) and both extended phonon modes ( i.e. , magnetoacoustic couplings@xcite ) and local phonons ( e.g. , vibrons , or impurity - induced local phonons ) . in a conducting system , @xmath58 can also couple to extended electronic excitations in various ways@xcite . 3 . finally , an ensemble of smms , as in a qips , will have various stray couplings between the @xmath61 , mediated by all the degrees of freedom discussed above , and not included in the bare interactions @xmath24 which already appear in ( [ qb ] ) above . these include slow couplings like those mediated by phonons , in which retardation effects can be important . actually there is nothing specific to smm s about problems like these . quite generally , in describing a qips , we need to include ( i ) coupling of the qubits to stray environmental degrees of freedom , and ( ii ) stray couplings between the qubits from these environmental degrees of freedom . so what kind of model should we really be using for a set of coupled smm qubits , instead of ( [ qb ] ) ? the answer proceeds in 2 stages . * ( i ) environmental modes * : we can divide all the environmental modes into 2 classes@xcite . first , for all the discrete _ localized _ modes , like defects , nuclear spins , `` loose '' spins , localized phonons , etc , we can use a model which couples the qubits to a `` spin bath '' environment@xcite of localized modes . in its simplest form this describes a set @xmath62 of two - level systems , and we can generalize the bare qubit hamiltonian in ( [ qb ] ) to @xmath63 in which the pauli matrices @xmath64 act on the 2-level systems , and where the spin bath has its own hamiltonian @xmath65 in which the local fields @xmath66 typically dominate over the rather weak interspin coupling @xmath67 ; and finally the qubit couples to the spin bath via the term @xmath68 with couplings @xmath69 . more generally we have to couple to a spin bath in which the individual modes have more than 2 levels for example , to a set of nuclear spins with spin @xmath70 , having @xmath71 levels each . this description may look rather complicated , but note that we can also write @xmath56 in the more transparent form @xmath72 where we define the time - dependent vectors @xmath73 and @xmath74 , having components @xmath75 equation ( [ h - qb - red ] ) simply says that the @xmath76-th bath spin moves in a dynamic field coming from both the qubit ( incorporated in @xmath77(t ) ) and the other bath spins ( incorporated in @xmath78(t ) ) . note that @xmath77(t ) is the sum of a static field @xmath79 and a dynamic field coming from the qubit . of course the motion of the @xmath76-th bath spin ( and all the other bath spins ) will also react back on the qubit via the interaction @xmath80 . . however each qubit is also interactiong with a `` spin bath '' and an `` oscillator bath '' ( for details see text).,scaledwidth=50.0% ] as an example consider the fe@xmath29 molecule again , in which the core of eight fe@xmath30 spin-5/2 ions couples to over 200 nuclear spins in each molecule ( these being a mixture of @xmath81h , @xmath82h , and possibly @xmath83o and @xmath84fe spin 1/2 nuclear spins , as well as some higher spin br and n nuclear spins ) . in this system , analysed in detail in refs@xcite , the @xmath85 are weak internuclear couplings ( @xmath86 even for nearby nuclei ) , the @xmath87 are hyperfine couplings ( with @xmath883 - 4@xmath89 for @xmath84fe nuclei or nearby proton spins ) and the @xmath66 include both the weak nuclear zeeman couplings , and a term arising when the fe@xmath29 giant spin is not flipping between antiparallel orientations when it tunnels . what about _ delocalized _ modes , like phonons or photons ? these can be handled in a standard way , by coupling the qubit to a bath of oscillators@xcite , which now adds an oscillator term @xmath90 plus a coupling @xmath91 to the qubit . in a smm system , @xmath92 would be e.g. the coordinate of a phonon with momentum @xmath93 , @xmath94 its energy , and @xmath95 is related to the magnetoacoustic coupling@xcite between the smm and this phonon . this coupling takes the form @xmath96 where @xmath97 is an operator acting on the giant spin @xmath36 , @xmath98 an operator acting on the phonons , and @xmath99 a coupling function ; typically @xmath100 for a smm . when we then truncate the giant spin hilbert space down to the qubit space , we get a coupling of the form in ( [ h - obint ] ) . * stray interactions * : now consider the interaction @xmath24 between the qubits . essentially this incorporates all interactions mediated by fields which operate so fast that they can be treated as instantaneous on the timescale of the qips . this includes all photon - mediated interactions ( including dipolar interactions and exchange interactions ) . however it does _ not _ include phonon - mediated interactions , which are slow - if the time @xmath101 taken for phonons to travel across a qip of size @xmath102 at velocity @xmath103 is not much less than the timescale of switching or operation of qubits , then we must treat these interactions as retarded , and this can actually lead to decoherence . we ignore this problem in what follows but it is important . this spin - phonon interaction typically has a complicated form - however the main part will be of dipolar form , of strength @xmath104 , where @xmath105 is the @xmath106 component of @xmath99 above . it is essential to distinguish between the short - range and long - range parts of @xmath24 ; we write @xmath107 where @xmath108 is the short - range interaction ( typically exchange or superexchange ) , with strength @xmath109 for transition metal spins and @xmath110 for rare earths ; and @xmath111 is the magnetic dipolar interaction , of strength @xmath112 , where @xmath113 , where @xmath114 is a microscopic lattice length . the key point is that @xmath108 is only appreciable for nearest - neighbour systems - it falls off exponentially and very fast with distance . decoherence is fundamental problem blocking the manufacture of any qips . it is considered by many to be an insuperable problem ( an opinion we do not share ! ) , as well as being at the very heart of our understanding of quantum mechanics@xcite . for anyone who wants to make a qips , decoherence creates a very practical problem - since a qips will not work if there is too much decoherence , then in designing one , we need to know how big the decoherence will be . this question has a chequered history . early calculations of decoherence were far too optimistic - they used oscillator bath models , but at low temperatures most decoherence comes from spin bath environments of localised modes . our purpose in this section is ( i ) to give readers an intuitive feeling for how decoherence works in practise in a system of qubits , and ( i ) give some results for decoherence rates which we hope will help to better design a qips . quite generally , decoherence in the dynamics of some system a is caused by its entanglement with its environment@xcite . a key result of quantum mechanics says that if we do not what the environment is doing , then we must average over all of its possible states when calculating the dynamics of a. but if the environment is entangled with a , then this will `` smear out '' the phase dynamics of a. another way of putting this is that if the the environment is reacting to the dynamics of a , then it is actually in effect measuring it - and this actually destroys phase coherence in this dynamics . to see how decoherence works , it is best to divide the discussion between the 3 main mechanisms . acting on it , a sum of a static field @xmath115 , and a part which varies rapidly from @xmath116 to @xmath117 when @xmath73 moves between @xmath118 and @xmath119 ; this happens when the qubit flips from @xmath120 to @xmath121 ( a typical path for @xmath73 during this flip is shown ) . , scaledwidth=35.0% ] to understand how spin bath decoherence works we need first to see how the bath spins move in the presence of the qubit . recall that the @xmath76-th bath spin moves in a dynamic field @xmath122 ( see equations ( [ h - qb - red ] ) and ( [ gam - xi ] ) ) . since @xmath123 is very small compared to @xmath77 , let us begin with @xmath77 , the net field on the k - th bath spin @xmath124 ; recall it is the sum of the static field @xmath79 and the time - varying field from the qubit . now @xmath77 spends almost all of its time in the quiescent states @xmath119 and @xmath118 , associated with the qubit states @xmath121 and @xmath120 , and given by @xmath125 where @xmath126 corresponds to @xmath121 , @xmath120 and the vector @xmath127 is defined by its components @xmath128 ( see fig . [ gammaf ] ) . however every time the qubit flips ( taking a very short `` bounce time '' @xmath129 to do so ) , the vector @xmath77 moves rapidly between these two quiescent states . the vector @xmath78 defining the field from the other bath spins also varies in time , but quite differently . typically @xmath78 will look much like a small and slowly - varying random noise source , adding a small `` jittering '' perturbation on @xmath77 , to give the total time - dependent field acting on @xmath130 . how does @xmath130 itself move in response to this , as @xmath77 `` jerks '' back and forth between @xmath119 and @xmath118 ? the answer is shown in fig [ precessf ] . at @xmath131 , @xmath130 will begin to precess about the initial field @xmath132 ( which is , say , @xmath119 ) . however as soon as @xmath119 jumps to @xmath118 , then @xmath133 must begin to precess around the new field . the resulting motion of @xmath130 depends on the exact time sequence of flips of @xmath77 . actually there are two small corrections to this picture . first , there is a small extra `` wobble '' caused by @xmath74 . second , we have described the motion of @xmath118 as though the jumps between @xmath119 and @xmath118 were instantaneous , so that @xmath130 can not follow them . however this is not true although they are sudden , there is still a small perturbation on the dynamics of @xmath130 . now notice a crucial consequence of this picture - it is that the path followed by @xmath130 is _ conditional _ on that followed by @xmath77 . this means that the 2 systems are entangled - indeed rather strongly so , since the path followed by @xmath130 depends more or less entirely on the time evolution of @xmath77 ( apart from the small perturbation coming from @xmath78 ) . now we can see how a spin bath will cause decoherence in the dynamics of a qubit . if we can not follow the coupled dynamics of the qubit and each individual bath spin , we must average over the different possible time evolutions of them - a definite time dynamics then gets converted to a time - evolving probability distribution . thus this entanglement between qubit and spin bath dynamics causes decoherence in the qubit dynamics . in fact we can classify three different decoherence mechanisms here , as follows : \(i ) : the precessional motion of @xmath130 around the time - varying @xmath77 , shown in fig [ precessf ] , gives the strongest decoherence from a spin bath . essentially the spin bath is absorbing from the qubit , and it turns out that the rate at which the qubit phase dynamics is smeared is roughly proportional to the average solid angle swept out by all of the @xmath124 together in their precessional motion . a detailed calculation@xcite gives simple results for the decoherence rate @xmath134 in two limiting cases : @xmath135 where @xmath136 in ( [ ksigma ] ) , and @xmath137 . note that this `` precessional decoherence '' ( originally called `` orthogonality blocking''@xcite ) no energy is exchanged between qubit and bath , only precessional phase . thus this decoherence mechanism , by far the most important one coming from the spin bath , is not even dissipative ! the simplest and best - studied example of precessional decoherence is the case of a spin qubit coupled to nuclear spins . consider the example of fe@xmath29 , already discussed above , where the main contribution to @xmath79 is just the zeeman coupling to the external field , i.e. , @xmath138 , where @xmath139 and @xmath140 are the nuclear g - factor and magnetic moment . then from ( [ gammaphip ] ) we see that when @xmath141 , the decoherence rate @xmath142 ( this is because @xmath143 flips then through 180@xmath144 ; such a flip has no effect on the dynamics of @xmath130 ) . decoherence then rises rapidly until reaching a maximum when @xmath145 , where @xmath146 is the hyperfine coupling as before - at this point the bath spins are precessing wildly as the field @xmath143 flips between orientations separated by angles @xmath147 . further increase in @xmath148 then @xmath149 , and in the high - field limit @xmath149 again becomes very small ( in this case @xmath150 and @xmath151 are almost parallel , so their flipping of @xmath152 again hardly affects the dynamics ) . the quantitative details for fe@xmath29 were worked out in ref.@xcite . \(ii ) : the above discussion concerned the bath spin dynamics when the qubit was sitting quietly in either @xmath55 or @xmath50 . but what about when the qubit is actually flipping , and @xmath73 is moving rapidly between @xmath118 and @xmath119 ? if this flip were instantaneous , there would be no effect on the bath spins - they simply could nt follow the sudden change . however the time @xmath153 taken for @xmath77 to flip from @xmath150 to @xmath151 is actually finite ( though still very short ) . then elementary time - dependent perturbation theory tells us that the amplitude @xmath154 for the spin @xmath124 to make an inelastic transition under this sudden change , ie . , during the qubit flip itself , is @xmath155 . by the same arguments as before this bath transition must cause decoherence in the qubit dynamics the bath is actually the change in the qubit state ( it is in effect performing a measurement ) , and this transition actually adds a random topological phase to the qubit dynamics@xcite . the resulting `` topological decoherence '' contribution @xmath156 to the decoherence rate is found to be@xcite : @xmath157 for most systems this will be very small , since @xmath158 as a rule . thus , for fe@xmath29 , @xmath159k in low applied fields , whereas most of the @xmath160mk ; even after summing over c. 200 nuclear spins , one still finds a very small contribution to decoherence , changing slowly with field as @xmath161 decreases@xcite . \(iii ) : the best - known kind of decoherence is that coming from some unknown noisy vector field @xmath162 coupling to @xmath58 . chemists are very familiar with this in nmr and epr - it leads to the usual `` @xmath163 phenomenology '' , in which the noise causes incoherent relaxation of the qubit energy over a time @xmath164 and the qubit phase over a timescale @xmath165 . a spin bath also gives rise to this noise decoherence ; the weak interactions between the bath spins cause a slowly fluctuating field to act on the qubit , causing @xmath164 and @xmath165 relaxation in the standard way . notice an interesting consequence of this , which is seen very nicely in experiments on smms . since most of the spin bath dynamics is being driven by the qubit itself ( see again fig [ precessf ] ) , then when the qubit is frozen by an external longitudinal bias field @xmath166 , the bath spins must also largely freeze without the time - varying field @xmath167 to drive them , only the very weak @xmath168 are left to give the spin bath any independent dynamics ( via the small slowly - varying @xmath74 ) . thus we expect @xmath164 and @xmath165 to drastically increase when the qubit is frozen . this was seen in very pretty experiments by the leiden group on the smm mn@xmath26 ; they observed roughly a hundredfold increase in @xmath164 and @xmath165 when they pushed the tunneling mn@xmath26 molecules off resonance@xcite . : an obvious questions which arises here is the following : to what extent are all these decoherence results already incorporated into the @xmath163 phenomenology used in experimental chemistry and physics in the last few decades ? we note first that the common practise in the literature ( particularly the quantum information literature ) of associating decoherence with some `` experimental noise '' fluctuation source , and then applying standard ideas such as the fluctuation - dissipation theorem@xcite to associate this noise with a dissipation , is quite wrong for a spin bath . as noted above , the most important part of spin bath decoherence , i.e. , precessional decoherence , simply involves precessional motion there is no transfer of energy between qubit and bath spins , and therefore no dissipation . thus there is no classical analogue to decoherence from a spin bath , and no connection between decoherence and dissipation@xcite . now the `` @xmath163 '' phenomenology in nmr is based on the classical bloch equations . it throws away all the internal phase information in the nuclear spin bath dynamics ( including relative phases between spins , entanglement , etc ) , leaving only the self - consistent description of a single nucleus in the fluctuating mean field from all the other spins ( plus any fluctuating fields coming from electron spins , fluctuating external fields , motional narrowing , etc ) . , the absorption probability described in the text , for a qubit coupled to a spin bath . without decoherence this would be a sharp line at @xmath169 . the shaded part shows contributions from processes where the spin bath polarisation is unchanged.,scaledwidth=45.0% ] however this @xmath163 phenomenology , although it is commonly used in the qubit literature ( in discussions of , e.g. , rabi oscillation , or spin echo for a qubit ) , it is not appropriate to the qubit dynamics@xcite . to give an idea of what can really happen , in fig . [ chif ] we show the imaginary part of the fourier transform @xmath170 of the spin correlation function @xmath171 for a qubit coupled to a spin bath . this function is the probability that a qubit in state @xmath121 at the time @xmath131 will be found there at time @xmath172 . thus @xmath170 is the absorption intensity at frequency @xmath173 in a qubit resonance experiment . this looks nothing like the lorentzian expected in the @xmath163 phenomenology , and it shows singular peaks which come from the unusual coupled dynamics of qubit and spin bath . these features actually result from decoherence and are not associated with dissipation . this point is important , because it means that eventually experimentalists working on qubits are going to have to abandon the @xmath163 phenomenology in describing their experiments , when spin bath decoherence is important . the phenomenology that will be required needs a whole paper on its own . this has been well covered in other articles@xcite and books@xcite . the theoretical problem is simply formulated one now couples a qubit to a bath of oscillators , and tries to calculate the dynamics . this `` spin - boson '' model is well understood@xcite . in contrast to the spin bath , there is a quantum - classical correspondence for oscillator baths@xcite , and a fluctuation - dissipation theorem relating both the dissipative dynamics and the decoherence of the quantum system to the fluctuational `` quantum noise '' acting on it , coming from the oscillator bath . thus we can understand decoherence in the spin - boson model in a fairly simple way . the oscillator bath has two main effects . first , it slows down the qubit , increasing its `` inertia '' , so that @xmath174 , the qubit flip rate , is `` renormalized '' ( ie . , @xmath175 with @xmath176 ) . second , it causes dissipation in the `` quantum relaxation '' of the qubit dynamics , either because the qubit spontaneously emits an excitation ( i.e. , excites an oscillator from its ground state ) or scatters an existing excitation ( i.e. , stimulated emission an excited oscillator changes its state , and possibly another one is excited ) . all the complexity in the qubit dynamics comes from repeated absorption and emission of oscillator excitations@xcite . nevertheless the phenomenology is very familiar one defines a `` spectral function''@xcite of form @xmath177 ( where @xmath178 ) , and the entire dynamics can be calculated in terms of this function , which is familiar from second order perturbation theory or from fermi s golden rule . it has the form @xmath179(density of states ) , where @xmath180 is a coupling constant and @xmath173 the energy denominator , and the sum over @xmath181 measures the density of oscillator states at @xmath173 . most of the theoretical work on the spin - boson system has been done for the simple @xmath182-independent `` ohmic '' form @xmath183 . however most experiments , certainly on qubit systems , are described by an ohmic model , for two reasons . first , as already discussed , spin bath decoherence is not at all described by such a model . second , even without the spin bath , the oscillator bath is often not ohmic . it certainly will not be if the system is insulating , since for phonons @xmath184 depends strongly on t and is a higher than linear power of @xmath173 . only for simple conductors is the ohmic form always valid . for a proper discussion of the different forms of @xmath185 , see refs.@xcite . as an example let us again take the smm . the principal source of oscillator bath decoherence is then the magnetoacoustic coupling to acoustic phonons ( for which @xmath186 . both quantum relaxation rates@xcite and decoherence rates@xcite have been calculated for this interaction . the usual form for the decoherence rate in a transverse field is @xmath187 where @xmath188 is the spin - phonon matrix element discussed in section 2 , @xmath189 the density , and @xmath103 the sound velocity . this expression illustrates nicely the points made above ; we have a ( matrix element)@xmath82 , multiplied by a density of states @xmath190 , divided by an energy denominator @xmath54 , multiplied by a thermal factor @xmath191 . in other words , just the golden rule . a more sophisticated calculation hardly changes this result , simply renormalizing @xmath54 . an ohmic bath would give a completely different result@xcite . actually , decoherence from an ohmic bath is typically very large , which is why in making a qips we want to avoid having moving electrons ( ie . , one should use insulating systems ) . of course , this still leaves decohrence from the spin bath . we now turn to what may in the long run be the most insidious kind of decoherence - that caused by unwanted long - range interactions between the qubits . there are two kinds of these , viz . ( i ) interactions like dipolar or ( in the case of current - carrying devices ) unscreened inductive interactions , which can be treated as instantaneous ; and ( ii ) long - range interactions mediated by slow particles like phonons , which may have a retarded character . the reason such interactions are so dangerous is that they affect many qubits at a time . only a few theoretical studies have so far appeared . aharonov et al.@xcite argue that concatenated error correction codes can handle long - range interactions , but only just . it is however likely that the model in this paper is over - simplified , since detailed calculations for both insulating systems@xcite(where dipolar inter - qubit interactions are important ) and conducting systems@xcite ( where inter - qubit interactions mediated by the electron bath are important ) give rather different results . thus the problem is still open , and we do not give any detailed results for decoherence rates here . we can nevertheless give a simplified description of what is going on ( see fig . [ dipdec ] ) . with long - range ( @xmath192 ) interactions , it is _ always _ possible for a given qubit to find resonance with many others in 3 dimensions , whether we want it or not , provided the strength of these interactions exceeds a critical threshold ( this result basically goes back to anderson@xcite ) . typically for dipolar interactions the resonances will occur over rather long distances . there are really only two ways to get round this problem - first , to use error correction ( a topic whose details@xcite we have eschewed in this article ) ; and ( ii ) use a lower dimensional geometry , which drastically reduces the number of qubits that are far apart ( compared to the 3d case ) . ideally one uses both - in section 4 we discuss certain architectures which can reduce the effective dimensionality of the qips . let us now use the results given in the previous section to discuss the best way to _ make _ a qips . to recapitulate - we want a system of qubits whose mutual interactions , and the local fields acting on them , can be accurately controlled ( or , if they are static in time , are very accurately fixed and can be tuned ) . we want to be able to write quantum information onto the states of these qubits , do a computation , and then read out the answer . we do not want stray interactions or fields interfering with this , and we certainly want to minimize decoherence caused by interactions with other uncontrolled degrees of freedom . so how can we best achieve all these desiderata ? as advertised in the introduction , we concentrate here on designs for spin qips using materials chemistry , ie . , bottom - up approaches . the main advantages of such an approach ( as opposed to top - down nanofabrication approaches ) , are : a1 ) reproducibility - chemistry uses quantum mechanics to produce absolutely identical molecules or other kinds of spin complex ; a2 ) size - the qubits and other parts of the architecture are really small , so that few degrees of freedom are involved , and few defects ; hence decoherence is reduced and timescales become short ; a3 ) no moving electrons - moving electrons always cause decoherence . with insulating molecules or other spin complexes one can make a true spintronics system , which only involves spin degrees of freedom . there are also disadvantages : the main ones are d1 ) size - molecular and ionic spin moments are currently too small to be either controlled or examined individually , except in very small numbers . in most qip algorithms , we need to be able to control , address , and read out the states of individual qubits . d2 ) dipolar interactions - as discussed above these are potentially fatal to qips - one needs architectures that suppress them . our task is thus to find designs that use the advantages ( a1)-(a3 ) to their fullest extent , and overcome the disadvantages ( d1 ) and ( d2 ) as much as possible . in bottom - up designs the main weapons we have are ( i ) the methods of chemistry , and ( ii ) a huge variety of potential geometries , ie . , possible qips architectures. we deal with these in turn . at the level of individual ions or molecules , materials chemists are well - placed to deal with the problems of reproducibility and decoherence , both at the nuclear and electronic scale . * ( i ) nuclear chemistry * : unless one wishes to use nuclear spins as a qips resource ( for which see below ) , then their enormous contribution to decoherence means we simply want to get rid of them . now while all `` odd '' nuclear isotopes ( ie . , with an odd number of protons ) have @xmath193 , at least one isotope of every even element has @xmath194 . indeed , in natural mixtures of even - numbered elements , the @xmath195 isotopes are the most common . thus we should aim for systems with even - numbered elements only . of course , if one aims strictly for even nuclei , the chemical playground is dramatically reduced , and isotopic purification is still typically required . obvious transition metal choices are fe ( with 2@xmath196 @xmath84fe ) or ni ( with 1@xmath196 @xmath197ni ) . there are also a number of rare earth candidates . these are either even - numbered ( like nd , dy , er , yb ) with a complex mixture of isotopes and the possibility of having only @xmath195 through costly isotopic purification , or odd - numbered ( like tb , ho ) , where @xmath198 but only a single isotope is naturally prevalent . note however that in this case we deal not with a single rare earth ionic doublet at low energies , but an electronuclear pair of spins - the nuclear spin hyperfine coupling is so strong that one can not rotate without the other@xcite . an entirely new range of possibilities opens up if we use nuclear spins as part of the qips ( e.g. , in a quantum register ) . we then must be able to switch on and off the interaction ( and hence the information flow ) between the nuclear and electronic spins . one way to do this is dynamical - one tunes the electronic and nuclear spin splittings ( e.g. , with external fields ) so that resonant hyperfine - mediated interactions can be switched on and off . concrete practical designs are needed here - the big stumbling block will be decoherence from dipolar interactions between nuclear spins and remote electronic spins ( even if these latter do not partake in the resonant interactions , their dipolar field adds extra dynamic phase to the nuclear dynamics , and can throw them off resonance ) . another possibility is to switch the hyperfine interactions on and off . in spin transition systems the electron spins can be switched on and off without a change in the redox state , using temperature , pressure or light ( the liesst effect,@xcite or haxiesst@xcite / soxiesst@xcite for hard / soft x - rays ) . if a design based on this effect could be made to work it would be very powerful - one would simply switch off the qips when it was not needed , storing information in the nuclear register . \(a ) ( b ) * ( ii ) electronic chemistry * : undoubtedly one area where chemists can play a big role is in the design of interactions ( ie . , engineering of effective hamiltonians ) , and the preparation of uniform & reproducible qubits . the first problem then is is the design of individual spin qubits . let us first note that seemingly obvious candidates such as molecular kondo systems or electrically - controlled quantum dots@xcite , while promising , have one very serious drawback - they have an intrinsically high decoherence , as their operation is based on moving electrons . so let us now consider insulating smms and rare earth ions as qubit candidates . a question often asked by molecular chemists is - what kind of crystal field is needed for the spin qubit ? as noted earlier , designs in which there is a natural basis for the qubit , created by a 2-well potential , are usually good ones . many tunneling smms and rare - earth ion molecular systems now use this design - the question at issue is how big to make the spin anisotropy energy barrier @xmath199 . one wishes to operate the qips at energy scales @xmath200 , where @xmath201 is the crossover energy to the qubit regime ( above @xmath202 , the spin dynamics proceeds by incoherent thermal activation ) . however at the same time we want to suppress decoherence from nuclear spins and phonons . this imposes competing restrictions : to reduce nuclear spin decoherence one wants @xmath203 , where the @xmath204 are the couplings to the spin bath ( nuclear spins in this case ) , but to reduce phonon decoherence one wants @xmath54 as small as possible compared to typical phonon energies ( whose scale is determined by the debye energy @xmath205 ) . thus there is a coherence window@xcite , defined by @xmath206 , defining a desirable size for @xmath54 . noting that @xmath54 is certainly less than @xmath207 , we can put these various restrictions together as a requirement that @xmath208 . thus we would like to have a large magnetic anisotropy and weak hyperfine couplings - and a system that can go to very low temperatures . certainly the most important requirement here is weak hyperfine couplings . it also helps if @xmath205 is big and if the spin - phonon couplings are weak ( which will happen if the molecule is , eg . , nearly spherically symmetric ) . if one uses transition ion smms based on even nuclei ( so as to eliminate nuclear spin decoherence ) , severe restrictions are placed on the chemical bridging between the spins - one has to discard all the halogens and almost every organic molecule because of their hydrogen and nitrogen nuclear spins . for coordination and short - range bridging , obvious choices are then oxo , tio or oxalate anions , but there are others , like carbonyl , thiocarbonyl or carbide . alternatively one can go for rare earth - based molecules . single lanthanide cations in adequate coordination environments have shown smm behaviour@xcite ; and recently , polyoxometalates ( poms ) composed completely of w , o and si have shown crystal fields capable of giving qubit dynamics@xcite . this offers exciting new possibilities , as poms are chemically quite different from the usual smms ( they have high negative charge , and are stable against oxidation ) . two advantages of any rare earth design are ( i ) the relevant magnetic parameters can be calculated theoretically from their structures and/or extracted from epr and optical data ; and ( ii ) their strong magnetic signals and hyperfine coupling ( when using nuclei with @xmath209 ) facilitate their detection and characterization . another key requirement is the reduction of disorder to very low levels - this means eliminating defects , dislocations , etc . , from the qips , and using molecules or other spin complexes that only come in one species ( ie . , not using systems that have active structural degrees of freedom , various kinds of isomerism , or free rotation about single bonds , or solvents / counterions that can take different positions which are close in energy to each other ) . let us turn now to the design of inter - qubit interactions . these need to be precisely controlled ( whether they are static or dynamic ) and inter - qubit dipolar interactions and spin - phonon interactions need to be suppressed . since both are @xmath210 the qubit spin moment , one possibility is to have antiferromagnetically ordered spin qubits . the inter - qubit interactions would then vanish much more quickly at long range ( for an antiferromagnetic pair , they are @xmath211 , where @xmath212 is the separation between the spins ) . if the spins in each pair are not equivalent ( eg . , in an antiferromagnetic transition metal dimer where the two metals are either different or have different chemical environments ) , an external stimulus ( e.g. an epr pulse ) will differentiate the @xmath213 and @xmath214 configurations . this way of suppressing dipolar interactions also suppresses the spin - phonon interaction . important work that has already been done in this direction with dimeric molecular nanomagnets@xcite , including molecules where there is a built - in switch@xcite . we would suggest making analogues to these systems in which the dipolar magnetic moments cancel completely while a strong quadrupolar moment remains . alternatively , heterometallic antiferromagnetic wheels@xcite , with @xmath215 , are clearly promising building blocks , and qubit designs for these systems have already been explored@xcite . the most obvious way to implement inter - qubit interactions is to use exchange or superexchange interactions between neighbouring spins . transition metal - based systems have interactions @xmath216 , making them better than rare earths , where these interactions are typically @xmath217 . if controllable long - range inter - qubit interactions are desired , one can imagine propagating these through reduced carbon nanotubes@xcite or poms@xcite . both can in principle mediate relatively long - range indirect exchange through the delocalized electrons , compared to superexchange , and both can be prepared using only @xmath195 isotopes . however , they involve moving electrons , a serious serious source of decoherence ; a better way would be to propagate spin signals down a chain of strongly exchange - coupled insulating spins . the actual spatial arrangement of qubits , and the read in / read out probes which couple to them , is quite crucial to the design of a qips . there are 3 main issues , viz . ( i ) the small size of spin qubits based on molecules or other spin complexes means that with current detection / control systems , it is hard to manipulate , read in , or read out the state of a single qubit ; ( ii ) the much larger size of these detection / control systems means that they can not be crowded in to interact with more than a few qubits at a time ; and ( iii ) the geometrical design of a qips will greatly influence decoherence , both from stray fields ( e.g. , from substrates ) and from long - range dipolar interactions . * ( i ) geometry and architecture * : current probes ( microsquids , stm and mfm systems , hall probes , optical detection and control systems , etc . ) are partly dogged by the uncertainty principle - if one is not careful , attempts to probe or control at a length scale of order a qubit size ( ie . , a few nm ) will have far too destructive an effect on the qubits and their quantum states ( thus if photons were used , we would be dealing with soft @xmath218-rays ! ) . stm and mfm probes can have a very weak effect , even though their tips are very small - but the rest of them is very big , and there is no obvious way to address many qubits at a time with this technology . the most obvious way to get round this problem is to use fixed probes of _ nm _ size , connected remotely to the outside world along one - dimensional connecters . conventionally one thinks here of wires and moving electrons - conceivably some future nanosquid array could do the job . however it seems better to us to use a genuinely spintronic arrangement of interacting spins ( e.g. , using strongly - coupled chains of spins , or possibly a set of nanotubes ) to couple into the qubit array . these can themselves be connected to much larger control and read in / read out systems outside the qips . as yet there are no concrete designs in this area - it would be useful to have some . another possibility , at least in the near future , is to make larger spin - based qubits - a large ( e.g. , 50 @xmath219 50 ) spin array ( square , hexagonal , linear ) of microscopic spins or smms , coupled antiferromagnetically . these would interact via a few nearest neighbour spins . a further advantage of this design would be obtained if each qubit had no net spin , thereby suppressing dipolar errors . promising advances here are being made with the synthesis and study of mesoscopic antiferromagnetic grids@xcite . another question which we find interesting is the way in which both the read in / read out and decoherence problems might be alleviated by using a sparse architecture for the qubits@xcite . by this we mean an arrangement of qubits on , e.g. , a planar substrate , whose total number does not increase linearly with the area of the system , but more slowly . this allows easier access for probes , and also strongly reduces errors caused by dipolar interactions . many geometries are possible here , ranging from coupled lines or nanorods of various shapes to fractal patterns on surfaces . the chemistry and physical properties of the substrates will be important in limiting these designs . for example , while dendrimers@xcite could be a promising scalable support for sparse architectures , steric hindrance usually induces disorder in their structures , which would cause decoherence . finally , let us mention two other interesting alternative designs which could alleviate the architectural problem . the first involves using specific algorithms that avoid single - qubit addressing altogether @xcite ; only the entire qips is addressed , but on multiple occasions after a series of pulses . the other uses molecular cellular automata @xcite which only need addressability at edges , thus reducing the addressability problem by one dimension . both of theses designs suffer because they require a long time to carry out a computation - they can only work if decoherence times are really long . * ( ii ) fabrication * : this brings us back to the problem of fabrication of these qips arrays . the range of possibilities is enormous ; we mention just a few . as far as smms are concerned , we gave the main desiderata above - the interesting question is what new possibilities would be useful to explore . one is poms , which have a rich chemistry offering different topologies , sizes , and the ability to encapsulate magnetic metal ions in arbitrary ligand fields and/or host delocalized electrons if desired@xcite . as noted above , they can be built entirely using even elements ( mainly w , mo and o ) . thus , one could explore `` giant '' pom wheels@xcite , currently about 4 nm in radius . it may be possible to build a wheel which ( i ) is even bigger , ( ii ) is free from nuclear spins ( current designs contain many water molecules ) , and ( iii ) has the magnetic network we want inside . alternatively , with an array of stms , one could chemically prepare a self - assembled monolayer of poms or fullerenes , and inject a single electron on each molecule directly under each tip . the chemistry would be quite easy , and avoids nuclear spins , since fullerenes and poms are based on even - numbered elements . the main problems would be ( i ) electron delocalisation in the molecules introduces new paths to decoherence , and ( ii ) this scheme would shift part of the difficulty from the chemistry to the nanoengineering . we also note that a way to prepare antiferromagnetic 2-d lattices which are poor in nuclear spins and have two distinct sites is to use bimetallic honeycomb oxalate layers@xcite ; see fig . [ chemsystems ] . one of the metals will always carry a nuclear spin , but the other can be clean if one so chooses . they have no water , but do require a nearby cationic layer , to compensate their own anionic charge . this cationic layer could be ca@xmath220-based . finally , we emphasize the care required in designing substrates . they will cause strong decoherence if there are any spin bath defects ( dangling bonds / free radicals , or charge defects , or paramagnetic spins , or nuclear spins ) which can couple to the qubits . solving this will not be easy - getting rid of nuclear spins in the substrate may be the hardest task of all . poms , among a vast variety of molecules , have been organized bidimensionally in different ways , including langmuir - blodgett films and through covalent modification of the surface . of course , usually these ways make extensive use of elements with an odd number of protons , so potentially the cleanest way would be to use self - assembled monolayers ( sams ) on an adequate substrate . highly ordered pyrolytic graphite@xcite and silicon@xcite have been demonstrated to support sams of - among other chemical species - different poms , and are all free from nuclear spins . quartz would be a good insulating substrate , but the preparation of pom sams on quartz usually involves organic molecules to provide for a positive charge@xcite , which defeats the purpose of choosing a nuclear - spin - free substrate . the international effort to make a qips has assumed very large proportions , but building a qips will obviously not be easy - the main problem , as we have seen , is to suppress decoherence , particularly from nuclear spins and dipolar interactions . the use of insulating atomic or molecular scale spin qubits offers important advantages - reduced decoherence , and with appropriate care , high reproducibility . we have explored herein a large number of design strategies for building a qips . it seems likely to us that experimental efforts to gain control of decoherence and make systems of multiply - entangled spin qubits will succeed in the next few years . at this point it should be possible to devise realistic architectures for large scale quantum information processing , and we anticipate that chemical considerations will play a large role in these efforts . t. fujisawa et al . , _ nature _ , 2002 * 419 * , 278 ; t hayashi et al , _ phys . _ , 2003 , * 91 * , 226804 ; k ono et al . , _ science _ , 2002 , * 297 * , 1313 ; j.r . petta et al . , _ science _ , 2005 , * 309 * , 2180 ; f.h.l . koppens et al . _ nature _ , 2006 , * 442 * , 776 j brooke et al . , _ science _ , 1999 , * 284 * , 779 ; r. giraud et al . lett . _ , 2001 , * 87 * , 057203 , and _ ibid _ , 2003 , * 91 * , 257204 ; s. bertaina et al . , _ nature nanotech _ , 2007 , * 2 * , 39 ; s nettutla et al . , _ phys . lett . _ , 2007 , * 99 * , 137601 for introductory discussionos of decoherence , see r. omnes , `` _ the interpretation of quantum mechanics _ '' , princeton univ . press ( 1994 ) ; and p.c.e . stamp , studies hist . phys . * 37 * , 467 ( 2006 ) j. lehmann , a. gaita - ario , e. coronado and d. loss , _ nature nanotech . _ , 2007 * 2 * 312 - 317 ; j. lehmann , a. gaita - ario , e. coronado and d. loss , _ j. mat . chem . _ , 2008 ( in this same special issue , insert volume and pages ) . f. k. larsen , e. j. l. mcinnes , h. el mkami , j. overgaard , s. piligkos , g. rajaraman , e. rentschler , a. a. smith , g. m. smith , v. boote , n. jennings , g. a. timco and r. e. p. winpenny , _ angew . eng . _ , 2003 , * 115 * 105109

we compare the h line strengths of star - forming galaxies observed with the near - ir grism of the _ hubble space telescope _ with ground - based measurements of ly from the hetdex pilot survey and narrow - band imaging . by examining the line ratios of 73 galaxies , we show that most star - forming systems at this epoch have a ly escape fraction below . we confirm this result by using stellar reddening to estimate the effective logarithmic extinction of the h emission line ( ) and measuring both the h and ly luminosity functions in a mpc volume of space . we show that in our redshift window , the volumetric ly escape fraction is at most , with an additional systematic uncertainty associated with our estimate of extinction . finally , we demonstrate that the bulk of the epoch s star - forming galaxies have ly emission line optical depths that are significantly greater than that for the underlying uv continuum . in our predominantly [ o iii ] -selected sample of galaxies , resonant scattering must be important for the escape of ly photons .

1409.8304

ly@xmath2 is the most common electronic transition in the universe . most often , it is a product of the photo - ionizing photons emitted by young stars : as recombining electrons cascade through the energy levels , they are funneled into hydrogen s @xmath10 state by the high optical depth of the interstellar medium to lyman series transitions . the result is that strong ly@xmath2 is a signature of star formation , and indeed , @xcite noted that this feature may be our best probe for identifying galaxies in the act of formation . due to the resonant nature of the line , a typical ly@xmath2 photon must undergo tens or even hundreds of absorptions and re - emissions before escaping into intergalactic space . consequently , the radiative transfer of this line is quite complex , and even a small amount of dust can break the chain of interactions which is necessary for its escape . this fact is reflected in the observed redshift evolution of ly@xmath2 emitting galaxies ( laes ) : in the nearby universe , such objects are quite rare , but between @xmath11 @xcite and @xmath12 there is a strong increase in both the number density of ly@xmath2 emitters and their characteristic luminosity @xcite . three - dimensional radiative transfer models have demonstrated that the ly@xmath2 emission line can contain a great deal of information about the distribution of a galaxy s ism , its surrounding circum - galactic medium , and the physics of its on - going star formation ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? however , to extract this information , one needs accurate measurements of the fraction of ly@xmath2 photons escaping the galaxy ( ) , and the profile of the emission line . over the past decade , there have been numerous attempts to estimate in the normal ( non - agn ) galaxies of the distant ( @xmath13 ) universe , mostly by comparing ly@xmath2 to measurements of galactic emission in the rest - frame uv ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , the far - infrared @xcite , or the x - ray @xcite . the premise behind these measurements is straightforward : like ly@xmath2 , the strength of a galaxy s uv , far - ir , and x - ray emission all depend in some way on the existence of young stars and star formation . consequently , the ratio of ly@xmath2 to these quantities should yield a measure how efficiently ly@xmath2 is escaping its environment . using a compilation of such measurements , @xcite determined that , over time , the `` volumetric '' ly@xmath2 escape fraction of the universe has declined monotonically , from @xmath14 at @xmath15 to @xmath16 locally . this evolution is consistent with models in which ly@xmath2 is quenched by dust , which slowly builds up as the universe ages . there is , however , one difficulty with this analysis : all the star - formation rate tracers listed above are somewhat indirect and rely on empirical calibrations derived from galaxies in the local @xmath17 universe . for example , not only is the observed uv luminosity of a galaxy extremely sensitive to the effects of internal extinction , which may depend on such factors as star formation rate , inclination , and redshift ( e.g. , * ? ? ? * ; * ? ? ? * ) , but it also arises from a stellar population that is slightly different from that which is producing the ionizing photons . although both the uv continuum and ly@xmath2 are generated by the flux from hot , young stars , ly@xmath2 is excited by the far - uv emission of stars with @xmath18 , whereas the uv light originates in the atmospheres of @xmath19 objects ( e.g. , * ? ? ? thus , the ratio of the two quantities is subject to shifts in the initial mass function , metallicity , extinction law , and the timescale over which star formation is occurring . indeed , @xcite has examined these effects and has shown that , even if both the star - formation rate ( sfr ) and extinction are well determined , measurements of the rest - frame uv in galaxies at @xmath20 will underestimate the flux of ionizing photons by almost a factor of two . to overcome the need for empirical calibrations , one requires a more direct probe of the ionizing flux from hot , young stars . since ly@xmath2 is produced by transitions out of the @xmath10 state of hydrogen , the best possible tracer of its intrinsic strength is one which measures the preceding transitions into the @xmath10 state . to date , only one such investigation of this type has been made in the @xmath13 universe . by performing narrow - band surveys for @xmath21 galaxies in both h@xmath2 and ly@xmath2 , @xcite was able to fix the epoch s volumetric ly@xmath2 escape fraction at @xmath22% . however , the precision of this measurement was limited by the survey s small volume ( @xmath23 mpc@xmath6 ) , and the dearth of galaxies brighter than @xmath24 . to improve upon this situation , we have combined the data from four recent surveys : 3d - hst and aghast @xcite , the pilot survey for hetdex , the hobby - eberly telescope dark energy experiment ( hps ; * ? ? ? * ) , and the 3727 narrow - band observations of the chandra deep field south ( cdf - s ; * ? ? ? * ; * ? ? ? the first two of these studies unambiguously measures total h@xmath0 fluxes in the redshift range @xmath1 via wfc3 grism observations with the _ hubble space telescope ; _ the latter two provide ly@xmath2 measurements ( or upper limits ) for many of these same galaxies via integral field spectroscopy and narrow - band imaging . by comparing their data products , we can place constraints on via statistically complete samples of star - forming galaxies in the goods - n , goods - s , and cosmos fields . in section 2 , we describe the observational data and detail the procedures used for identifying and measuring h@xmath0 and ly@xmath2 in our target fields . in section 3 , we use these data to measure ( or place limits on ) the ly@xmath2/h@xmath0 ratio of 73 galaxies in the redshift range @xmath25 . by converting stellar reddenings into nebular extinctions via the empirical @xcite obscuration law , we show that the typical ly@xmath2 escape fraction of these galaxies is just a few percent . in section 4 , we consider the epoch s volumetric ly@xmath2 escape fraction by deriving the h@xmath0 and ly@xmath2 luminosity functions for a @xmath5 mpc@xmath26 volume of space . after correcting for nebular h@xmath0 extinction , we demonstrate that , at most , only @xmath7 of the ly@xmath2 photons escape their galaxies , and argue that any systematic error associated with this measurement must be less than @xmath8 . we conclude by discussing this measurement , and the implications it has for the evolution of galaxies . for this paper , we assume a @xmath27cdm cosmology , with @xmath28 , @xmath29 and @xmath30 km s@xmath31 mpc@xmath31 @xcite . we begin our analysis with a sample of @xmath20 galaxies observed with the g141 near - ir grism of the _ hubble space telescope s _ wide field camera 3 ( go programs 11600 , 12177 , and 12328 ) . this dataset , which is the product of the 3d - hst @xcite and aghast @xcite surveys , consists of @xmath32 slitless spectroscopy over the wavelength range @xmath33 m , and records total emission line fluxes over 625 arcmin@xmath34 of sky . tens of thousands of spectra are observable on these images , but of special interest are those produced by galaxies in the redshift range @xmath1 , where the emission lines of [ o ii ] @xmath35 , h@xmath0 , and the distinctively - shaped [ o iii ] blended doublet @xmath36 are simultaneously present in the bandpass . for these objects , redshift determinations are unambiguous , and total h@xmath0 fluxes can be measured to a 50% completeness flux limit of @xmath37 ergs @xmath38 s@xmath31 , independent of redshift . @xcite has used these data to measure the h@xmath0 fluxes of 260 [ o ii ] and [ o iii ] selected galaxies in the goods - s , goods - n @xcite , and cosmos @xcite fields , while @xcite has derived metallicities and masses for these systems . comparisons with the deep x - ray surveys of the regions @xcite confirm that the vast majority of these objects are normal galaxies with star - formation rates between @xmath39 and @xmath40 yr@xmath31 and no evidence of agn activity @xcite . any h@xmath0 source projected within @xmath41 of a cataloged x - ray position has been excluded from our analysis . our ly@xmath2 measurements ( and upper limits ) come principally from hps , a blind integral - field spectroscopic study of four areas of sky , including cosmos and goods - n . a full description of this survey and its data products is given by @xcite , but in brief , a square 246-fiber array mounted on the harlan j. smith 2.7-m telescope at mcdonald observatory was coupled to the @xmath42 george and cynthia mitchell spectrograph , a proto - type of the visible integral - field replicable unit spectrograph ( virus - p ) designed for hetdex @xcite . at the focal plane , each fiber subtends an angle @xmath43 in diameter , enabling simultaneous spectroscopy of @xmath44 arcmin@xmath34 of sky between the wavelengths 3550 and 5800 . at @xmath21 , 50% of the survey s pointings reach a @xmath45 monochromatic flux limit of @xmath46 ergs @xmath38 s@xmath31 ( or @xmath47 ergs @xmath38 s@xmath31 ) and 90% reach @xmath48 ergs @xmath38 s@xmath31 ( @xmath49 ergs @xmath38 s@xmath31 ) . above these flux limits , the hps s recovery fraction of emission lines is greater than 95% for emission line equivalent widths greater than 5 , and better than 90% for equivalent widths as small as 1 @xcite . moreover , because the virus - p spectrograph s response increases rapidly towards the red , the luminosity limits of this survey are roughly constant throughout our redshift range of our observations @xcite . in total , the hps survey covered 169 arcmin@xmath34 , with 107 arcmin@xmath34 in the cosmos and goods - n regions . roughly 76 arcmin@xmath34 of this area overlaps the fields studied by _ hst , _ with @xmath50 of the overlap region useable for science ( see section [ subsec : hbeta_lf ] ) . thus , the intersection of two surveys encompasses a co - moving volume of @xmath51 mpc@xmath6 between @xmath52 . to compare the hps ly@xmath2 measurements with the h@xmath0 fluxes from the _ hst _ grism , we began by examining the virus - p spectra at the location of every galaxy in the _ hst _ emission - line selected sample . since the original hps survey ignored all detections below @xmath53 confidence , we re - measured these spectra , mimicking the procedures of @xcite using a @xmath54 radius aperture centered on the position of each h@xmath0 source . this spectrophotometry did present some challenges . because the dither pattern of the hps placed roughly two of the @xmath55 diameter fibers at any given position , we had to precisely compute the fraction of each fiber falling within any given aperture . this calculation was done by summing the total fiber area within the designated aperture and then normalizing the flux by this aperture area . in addition , although we did have prior knowledge of the approximate redshift of each source , the limited spectral resolution of the _ hst _ grism ( @xmath56 at @xmath20 ; * ? ? ? * ; * ? ? ? * ) prevented us from knowing the exact wavelength of the corresponding ly@xmath2 line . we therefore searched a spectral window about the h@xmath0-defined wavelength of ly@xmath2 that was six times the wavelength resolution of the grism , ( @xmath57 ) , summing up the putative ly@xmath2 flux in a series of 4.2 bandpasses ( i.e. , twice the spectral resolution of the instrument ) . the total flux and noise were then corrected for flux losses due to the fixed spatial and spectral aperture by assuming an effective psf through the @xmath43 fibers of @xmath54 fwhm and adopting an 8 fwhm for the line profile . finally , we searched for sources with a signal - to - noise detection greater than 3 within the spectral bandpass . this procedure recovered the fluxes of all the @xmath53 ly@xmath2 detections found by @xcite in our h@xmath0 redshift window , and identified one additional h@xmath0 counterpart with a ly@xmath2 @xmath58 of @xmath59 . for the remaining h@xmath0 sources we used the @xmath60 ly@xmath2 limits in our analysis . to supplement the hps data , we also used a second source of ly@xmath2 measurements : the deep , narrow - band observations of the extended chandra deep field south . @xcite imaged this field with the ctio 4-m and mosaic camera for 28.17 hours through a 50 wide interference filter centered at 3727 , and obtained a sample of over 200 ly@xmath2 emitting objects in the redshift range @xmath61 . [ see @xcite for more details on this dataset . ] a subset of these sources fall in the goods - s region surveyed by 3d - hst , thus allowing us to increase the number of ly@xmath2 emitters with reliable h@xmath0 constraints . because these narrow - band observations sample only a small redshift slice of the universe , the actual volume covered by the goods - s data is rather small , just @xmath62 mpc@xmath6 , or @xmath63 that of the overlap region between hps and the fields surveyed by the _ hst _ grism . also , due to the redshift uncertainty associated with the _ hst _ grism measurements , the ly@xmath2 emission from any given h@xmath0 source with @xmath64 may , or may not fall within the narrow - band filter s bandpass . this ambiguity , plus the flux errors introduced by the filter s gaussian - shaped transmission curve , complicates the interpretation of these photometric measurements . nevertheless , the narrow - band ly@xmath2 data do reach significantly deeper than the hps spectra ( to a 90% completeness limit of @xmath65 ergs @xmath38 s@xmath31 ) and are useful , for both probing the faint end of the lae luminosity function , and performing the inverse experiment of measuring for targets where ly@xmath2 is already detected . finally , to complete our sample , we performed a reverse search and examined the _ grism frames for evidence of h@xmath0 emission at the position of known ly@xmath2 sources . for several objects , this proved to be impossible due to contamination from overlapping spectra . however , in a few cases , we were able to extract the h@xmath0 fluxes for the lae candidate , thereby reconfirming its existence . our final sample in the area of hps / hst overlap therefore consists of 54 h@xmath0 emitting galaxies and 13 ly@xmath2 emitters , with four objects detected in both surveys . the goods - s region contains 24 galaxies with grism - based redshifts between @xmath66 , though only 13 have redshifts that fall within the effective volume surveyed by the gaussian - shaped narrow - band filter ( defined by the filter s full - width at two - thirds maximum ; see * ? ? ? * for the details of this calculation ) . of the 10 narrow - band selected ly@xmath2 emitters falling within the 3d - hst survey area , 8 have h@xmath0 detections . a list of these objects appears in table [ tab : galaxies ] . the lower panel of figure [ fig : ratio ] compares our _ hst _ h@xmath0 fluxes to ly@xmath2 measurements and limits made via hps spectroscopy . the measurements are plotted against @xmath0 , the slope of the rest - frame uv continuum between 1250 and 2600 , as derived from the deep multicolor photometry compiled and homogenized by @xcite . also plotted are the laes found in the goods - s field . from the figure , it is immediately obvious that for most galaxies , ly@xmath2 is below the detection limit of the ground - based surveys . as stated above , only four of the 54 h@xmath0 emitting galaxies were detected with the hps spectroscopy ; this agrees with the results of @xcite , who found little overlap in the blind samples of ly@xmath2 and h@xmath2 emitters at @xmath67 . the fraction of ly@xmath2 recoveries is greater for the deeper goods - s data , but since it is not known exactly where these objects fall on the narrow - band filter s transmission curve , the uncertainties associated with their ly@xmath2 fluxes are generally large . nevertheless , these objects do suggest the existence of an upper limit to the ly@xmath2/h@xmath0 ratios of galaxies . of course , the values shown in the lower panel of figure [ fig : ratio ] do not reflect the true ratio of these emission lines . before we can use these limits to infer ly@xmath2 escape fractions , we must correct h@xmath0 for the effects of internal extinction . for the dataset under consideration , this is not straightforward . the _ hst _ grism spectra do not extend to @xmath20 h@xmath2 , and h@xmath68 is generally too faint ( and too close in wavelength to h@xmath0 ) to constrain the balmer decrement . consequently , we have no direct measure of the extinction affecting the objects recombination lines . we do , however , have access to a measure of _ stellar _ reddening , as each field has deep multicolor photometry that extends throughout the rest - frame uv , from 1250 to 2600 @xcite . @xcite has shown that over this spectral range , the intrinsic slope of the stellar continuum of star - forming galaxies is very nearly a power law , with @xmath69 . for steady - state star - formation extending over @xmath70 gyr , @xmath71 , while for extremely young starbursts , @xmath72 may be as steep as @xmath73 . if the rest - frame uv slope of a star - forming galaxy is observed to be flatter than this value , the most likely explanation is reddening due to dust . we next need to know how the observed slope of the uv continuum translates into nebular extinction . as has been noted many times in the literature , the distribution of stars within a galaxy is generally wider than that of the dust , and the latter is often associated with individual h ii regions ( e.g. , * ? ? ? as a result , emission - line gas is usually extinguished more than the stars . by observing 8 starburst galaxies in the local universe , @xcite concluded that @xmath74 and that the slope of the rest - frame uv continuum is related to the total stellar extinction ( in magnitudes ) at 1600 and the total nebular h@xmath0 extinction by @xmath75 with @xmath76 and @xmath77 . without direct measurements of the balmer decrement for large samples of @xmath20 galaxies , it is impossible to confirm this relation for the objects in our sample . nevertheless , using the same g141 grism data studied here , @xcite showed that , indeed , a power - law fit is good representation of the stellar continuum , and the product @xmath78 is at least consistent with the value of 0.162 expected from the @xcite obscuration relation . this law is also supported by several recent surveys ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , though others have found @xmath79 @xcite . nevertheless , it is reasonable to conclude that a @xcite relation , at least in the statistical sense , is applicable to the galaxies in our sample . the top panel of figure [ fig : ratio ] repeats our ly@xmath2/h@xmath0 comparison , but with h@xmath0 corrected for extinction via the @xcite obscuration relation and the assumption that @xmath80 . these values can easily be translated into escape fractions . under case b recombination , every ionization results in the creation of a balmer - line photon , with @xmath81 of these photons coming via an @xmath82 to @xmath10 transision ( i.e. , h@xmath0 ; * ? ? ? * ; * ? ? ? three quarters of these balmer transitions land in a @xmath83 orbital and immediately decay to the ground state via the emission of ly@xmath2 ; the other 25% of the electrons become temporarily trapped in the @xmath84 state before decaying to @xmath85 via two - photon emission . thus , under normal conditions , the ly@xmath2/h@xmath0 ratio should be @xmath86 where @xmath87 is the case b recombination coefficient ( @xmath88 cm@xmath6 s@xmath31 at 10,000 k ) and @xmath90 is the effective case b recombination coefficient for h@xmath0 ( @xmath91 cm@xmath6 s@xmath31 at @xmath92 k ; * ? ? ? * ; * ? ? ? the escape fractions shown in figure [ fig : ratio ] used @xmath93 as the intrinsic ratio of ly@xmath2 to h@xmath0 . in practice , however , these escape fractions are upper limits . if the case b condition is relaxed so that the lyman continuum is optically thin , then the ly@xmath2/h@xmath0 ratio may be boosted to values as large as @xmath94 , thereby lowering . most evidence suggests that at the redshifts considered here , the escape fraction of lyman continuum photons is at most a couple of percent ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but this possibility can not be excluded . similarly , if the ism density approaches @xmath95 cm@xmath26 collisions will redistribute @xmath84 electrons into the @xmath83 state , again enhancing ly@xmath2 relative to h@xmath0 . in the galaxies of the local universe , most h ii regions have electron densities between 1 and 100 cm@xmath26 ( e.g. , * ? ? ? * ) , though this number may be slightly larger in dwarf systems ( e.g. , * ? ? ? shocks or very high densities ( such as in the broad - line regions of agn ) may also increase ly@xmath2 relative to h@xmath0 by creating an environment where the @xmath97 state of neutral material is collisionally populated . finally , our estimates of nebular reddening assume that the observed @xmath20 galaxies have been undergoing vigorous star formation for at least @xmath70 gyr , so that @xmath80 . since all our grism - selected objects have very high h@xmath0 equivalent widths , the intrinsic slopes of their rest - frame uv continua are unlikely to be flatter than this value @xcite . however , @xmath72 could be steeper , up to @xmath98 in the extreme , if the systems have only recently begun their star forming activity @xcite . in this case , our reddening estimates would be underestimated , h@xmath0 would be enhanced , and , once again , our inferred ratios of ly@xmath2 to h@xmath0 would need to be reduced . in summary , our adopted values of @xmath93 and @xmath80 both produce upper limits for the ly@xmath2 escape fraction . any deviations from simple case b recombination or steady - state star formation will only serve to reduce . while factor of @xmath99 errors are theoretically possible for systems younger than @xmath99 myr , in most cases , any systematic error associated with our escape fraction measurements should be less than @xmath100 . as the upper panel of figure [ fig : ratio ] illustrates , the median h@xmath0 emitter detected by the _ hst _ grism has an escape fraction below 7% , with no statistical difference between the results of the two hps fields . note that our constraints on become progressively stronger as the slope of the uv continuum becomes redder . this is simply the result of our extinction law : as the stellar reddening increases , so does the assumed extinction correction for h@xmath0 . the implied increase in h@xmath0 then translates into a decreased limit for the ly@xmath2/h@xmath0 ratio . another way to view the data of figure [ fig : ratio ] is through the mathematics of survival analysis @xcite . several authors have shown that the distribution of ly@xmath2 equivalent widths is exponential in form above a rest - frame equivalent width of @xmath101 @xcite . if the same is true for ly@xmath2 escape fractions , then the computation of the median escape fraction is straightforward . the result is a most likely median value of @xmath102% , where the errors are computed via a markov chain monte carlo sampler @xcite . of course , since this calculation depends on the underlying shape of the distribution , the formal errors on underestimate the true uncertainty in the measurement . perhaps the most instructive way to interpret figure [ fig : ratio ] is through the geometric parameter @xmath103 , which has been defined by @xcite as the ratio of the optical depth of ly@xmath2 to that of the stellar continuum at 1216 . in objects where resonant scattering is important and the escape path for ly@xmath2 photons is long , the likelihood of such a photon encountering a dust grain is large and @xmath104 . conversely , if @xmath105 , ly@xmath2 must have a relatively direct escape route from the galaxy , with a path - length similar to that of the star light . values of @xmath106 point to either anisotropic emission , a spatial offset between the points of origin for ly@xmath2 and the continuum , a clumpy ism @xcite , or significant deviations from case b recombination . the dashed lines figure [ fig : ratio ] display the @xmath107 and 2 relations under the assumption that a @xcite obscuration relation holds , with @xmath108 . as expected , most of the sources have upper limits that are larger than @xmath109 , demonstrating that resonant scattering of ly@xmath2 photons within these galaxies is significant . perhaps as importantly , none of the objects show values of @xmath103 significantly less than one . several studies have found that luminous ly@xmath2 emitting galaxies tend to cluster about the @xmath109 line ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and @xcite found that small values of @xmath103 are common in compact , low - mass laes . but these analyses were for systems selected via their strong ly@xmath2 emission . the sources analyzed here were primarily chosen via their bright [ o iii ] ( or [ o ii ] ) lines . in these more normal star - forming galaxies , @xmath103 is generally greater than one , demonstrating that ly@xmath2 has a difficult time escaping its immediate environment . an alternative approach to exploring the escape fraction of ly@xmath2 photons from the @xmath20 universe is to do so globally , via a comparison of the h@xmath0 and ly@xmath2 luminosity functions . the @xmath5 mpc@xmath6 volume of space surveyed for both h@xmath0 and ly@xmath2 contains 67 galaxies detected in h@xmath0 and 23 identifiable via ly@xmath2 . these numbers are more than sufficient for defining the emission - line luminosity functions of the two galaxy populations and integrating for their total luminosity density . in order to calculate the @xmath20 h@xmath0 luminosity function , we must first estimate the effective area of overlap between the hps and the _ hst _ grism observations . this is not just a simple geometry problem , since , as with all slitless spectroscopy , overlapping sources render a fraction of the survey area unusable . @xcite estimated this fraction via a series of monte carlo experiments , in which realistic magnitude and positional distributions were used to create simulated frames , which were then `` observed '' in the same manner as the original data . at each point on the simulated frames , the amount of contamination was compared to the local sky noise , and all regions where the systematics of spectral subtraction were greater than this noise were excluded from consideration . since sky noise is the dominant source of uncertainty for all @xmath20 observations , this procedure generated a reliable statistical measure of the grism survey s effective area . based on the results of these simulations , we reduced the geometric hps / hst overlap region by 15% to 65 arcmin@xmath34 , and used this new area in our calculation of survey volume . to measure the total h@xmath0 luminosity density at @xmath20 , we began by following the procedures described by @xcite and applied the @xmath110 technique @xcite to the h@xmath0 sources found in cosmos , goods - n , and goods - s . as described in @xcite , the completeness fraction of our _ hst _ grism sample , as a function of h@xmath0 flux , has been calculated via a series of monte carlo simulations ; in summary , the 50% completeness limit for the two goods fields ( in ergs @xmath38 s@xmath31 ) occurs at @xmath111 , while for cosmos , this limit is @xmath112 . in the regions of overlap between the h@xmath0 and ly@xmath2 surveys , 42 galaxies are present above these completeness limits . using these data , we computed @xmath113 , the co - moving volume over which an object with h@xmath0 luminosity @xmath114 would be detected more than 50% of the time . the number density of galaxies in any absolute luminosity bin of width @xmath115 is then @xmath116 where @xmath117 is the inverse of the completeness function , and the summation is performed over all galaxies with luminosities falling within the bin . the top - left panel of figure [ fig : lf ] displays this function , where the uncertainties on the points are from poissonian statistics only . we next fit this function using the maximum - likelihood procedure detailed in @xcite . we began with the assumption that over the redshift range @xmath25 , the h@xmath0 luminosity function can be modeled via a @xcite law @xmath118 with @xmath24 being the characteristic monochromatic luminosity of the epoch . the observed function , of course , does not follow this relation , as incompleteness takes an ever increasing toll at fainter fluxes . hence we define @xmath119 as the schechter function modified by the flux - dependent completeness fraction at each redshift as given by @xcite . from poissonian statistics , the probability of observing @xmath120 galaxies in any given luminosity and volume interval @xmath121 is then @xmath122 where the expectation value @xmath123 . if we let these intervals become differentials , then the likelihood of drawing an observed set of @xmath124 h@xmath0 luminosities from a given schechter function with parameters @xmath24 , @xmath125 , and @xmath2 becomes @xmath126 where , for purposes of our analysis , we define the lower limits of the luminosity integral by where the completeness fraction drops to 50% . the top right panel of figure [ fig : lf ] displays the likelihood contours in @xmath24 and @xmath125 , with the faint - end slope fixed at @xmath127 for consistency with other studies @xcite . to avoid the well - known degeneracy between @xmath24 and @xmath125 , the ordinate of the plot gives the integral of the schechter function ( down to @xmath128 ergs s@xmath31 ) , rather than the traditional coefficient , @xmath125 . the most likely solution , with @xmath129 and @xmath130 mpc@xmath26 , is displayed in the top left of figure [ fig : lf ] . for comparison , the most likely h@xmath0 luminosity function for the full sample of @xmath1 _ hst-_grism selected galaxies in cosmos , goods - n , and goods - s has @xmath131 and @xmath132 . table [ tab : lf ] lists the most - likely @xcite function parameters , their uncertainties , and the total integrated h@xmath0 luminosity density @xmath133 for the @xmath25 regions of overlap between the _ hst _ surveys , hps , and the cdf - s narrow - band survey field . although this latter quantity does require the extrapolation of the @xcite function to zero luminosity , this extension is not an important issue , as galaxies below our detection threshold are predicted to contribute only @xmath134 to the universe s total h@xmath0 luminosity density . for reference , table [ tab : lf ] also lists the h@xmath0 luminosity function parameters for the entire @xmath1 grism survey region of cosmos and goods . although this full volume is roughly four times larger than that for the just the regions of survey overlap , the luminosity function defined by its galaxies is virtually identical to that of the smaller region . the likelihoods given in table [ tab : lf ] and displayed in figure [ fig : lf ] represent only the formal statistical error of our fit . not included are systematic uncertainties associated with the data themselves , most notably , with the completeness corrections . our estimates for the survey area and completeness fraction are based on the simulations performed by @xcite , who also confirmed that metallicity , equivalent width , and redshift are not important factors for determining the detectability of h@xmath0 . this analysis did not , however , take galaxy size into account , and this can be an important factor for slitless spectroscopic surveys . fortunately , for the dataset considered here , the effect is minor . as @xcite have shown , @xmath135 of massive ( @xmath136 ) late - type galaxies at @xmath137 have effective radii smaller than @xmath138 , while all the h@xmath0 galaxies analyzed here have @xmath139 @xcite . in this regime , the completeness of an _ hst _ grism survey is a very weak function of size ( see the analysis of * ? ? ? * ) , and the simulations of @xcite should be valid . based on this result , our calculation of survey area and incompleteness fraction likely carries an additional error of @xmath140 . similarly , because the _ hst _ survey fields are relatively small , the effects of cosmic variance on our luminosity functions are non - negligible . we can estimate this number using the cosmic variance calculator developed by @xcite , which combines press - schechter theory with n - body cosmological simulations to predict field - to - field fluctuations in arbitrary slices ( or pencil beams ) of the universe . according to this estimator , our number counts in the three _ hst _ fields carry an additional uncertainty of @xmath141 , while that for just the 65 arcmin@xmath34 region of overlap between the _ hst-_grism surveys and our two sets of ground - based ly@xmath2 observations is @xmath134 . of course , for estimates of the ly@xmath2 escape fraction , it is the field - to - field variations of galactic properties such as star formation rate and dust content which are the important parameters for the calculation , not simply the number of galaxies present in the region . while we are less able to quantify this number , it seems likely that the effect is significantly less than @xmath134 . the last piece of information needed to compute the h@xmath0 luminosity density of the @xmath20 universe is an estimate of h@xmath0 attenuation . once again , we are limited by the lack of direct knowledge of the galaxies nebular extinction . measurements of the balmer decrement in the @xmath20 universe are rare , especially for samples of emission - line selected galaxies . ( the closest comparison sample to our own that created by @xcite from 128 wfc3 grism - selected galaxies in the @xmath142 universe has very large uncertainties , with a mean value of @xmath143 dex . ) consequently , we must again compute the loss of h@xmath0 statistically from measurements of the stellar continuum . we did this by translating the uv slopes of each h@xmath0 source in our sample into a nebular extinction using the obscuration relations of @xcite , and then examining the distribution of these extinctions . if we just consider the sample of 42 galaxies in the region of space surveyed for h@xmath0 and ly@xmath2 , then the median logarithmic extinction at h@xmath0 is @xmath144 , while the effective extinction , defined via the total amount of h@xmath0 luminosity lost due to dust in all the galaxies , is @xmath145 . for comparison , the logarithmic extinctions found for all @xmath146 galaxies brighter than the 50% h@xmath0 completeness limit in the _ hst _ fields of cosmos , goods - n , and goods - s is @xmath147 ( median ) and @xmath148 ( effective ) . this difference between the median and effective attenuation is not unexpected : as noted by many authors , reddening is strongly correlated with stellar mass and therefore star - formation rate ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? since the brightest h@xmath0 emitters are also those most heavily extinguished by dust , the effective extinction of the sample should be larger than the median extinction . for the remainder of this paper , we will adopt @xmath4 as the total logarithmic extinction at h@xmath0 , while noting that the uncertainty on this number is likely to be @xmath149 dex . the resulting intrinsic h@xmath0 luminosity density and total error for the @xmath20 universe is then @xmath150 ( ergs s@xmath31 mpc@xmath26 ) . our h@xmath0 luminosity function is the first such measurement at @xmath13 . however , there have been previous estimates of the epoch s h@xmath2 luminosity function from deep , narrow - band surveys in the infrared . while @xcite found @xmath151 ( ergs s@xmath31 ) and a total h@xmath2 luminosity density of @xmath152 ( ergs s@xmath31 mpc@xmath26 ) in a 56 arcmin@xmath34 region of goods - s , a much larger ( @xmath99 deg@xmath34 ) study by @xcite inferred @xmath153 and @xmath154 . our best - fit h@xmath0 function , coupled with an effective logarithmic extinction of @xmath4 , a @xcite extinction law ( with @xmath155 ) and an assumed intrinsic h@xmath2/h@xmath0 ratio of 2.86 @xcite implies observed values for @xmath24 and the h@xmath2 luminosity density of @xmath156 ( ergs s@xmath31 ) and @xmath157 ( ergs s@xmath31 mpc@xmath26 ) . in other words , our measurement of @xmath24 is in good agreement with that of @xcite , but our estimated luminosity density is more in line with that found by @xcite . it is somewhat surprising that these three surveys differ by more than 0.3 dex in their determination of luminosity density , but given the small volumes involved ( 5440 , 77,000 , and 108,000 mpc@xmath6 for the @xcite , @xcite , and this survey , respectively ) , and the expected cosmic variance in the number counts ( @xmath158 , and 20% ; * ? ? ? * ) , the result is still reasonable . we can also test our measurement of the h@xmath0 luminosity function by converting it into a star formation rate . the observed h@xmath0 luminosity density in our _ hst _ fields is @xmath159 ( ergs s@xmath31 mpc@xmath26 ) . if we again de - redden this number by @xmath4 and assume an intrinsic h@xmath2/h@xmath0 ratio of 2.86 , then @xmath160 ( ergs s@xmath31 mpc@xmath26 ) . the application of the local h@xmath2 star - formation rate calibration @xcite then yields @xmath161 ( @xmath162 yr@xmath31 mpc@xmath26 ) where the error is dominated by the uncertainty in the reddening correction . this value is generally consistent with most other determinations of the epoch s star - formation rate density ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , although , as pointed out above , it is slightly lower than that found from the h@xmath2 observations of @xcite . to compute the volumetric ly@xmath2 escape fraction of the @xmath20 universe , we repeated the above analysis using the ly@xmath2 emitting galaxies found in our overlapping survey area . for the hps objects , this involved using the estimates of completeness versus line flux computed by @xcite for the 186 separate virus - p pointings covering the _ hst-_grism fields in cosmos and goods - n ; for the narrow - band data , the completeness fraction versus monochromatic flux relation was based on the artificial star experiments performed by @xcite on the region s deep mosaic images . these functions were then folded into the @xmath110 calculation of equation ( [ eq : vmax ] ) and the luminosity function was fit using the maximum - likelihood procedures described by equation ( [ eq : likelihood ] ) . the results are shown in the bottom panels of figure [ fig : lf ] and summarized in table [ tab : lf ] . because our survey fields contain far fewer laes than h@xmath0 galaxies , the ly@xmath2 luminosity function is not as well defined as its h@xmath0 counterpart . indeed , only 11 hps galaxies and 6 narrow - band selected objects have ly@xmath2 detections that are brighter than their frame s 50% completeness limit . as a result , the `` knee '' of the @xcite function is poorly defined and detected only with @xmath163 significance . nevertheless , the best fit function , with @xmath164 ( ergs s@xmath31 ) and @xmath165 ( mpc@xmath26 ) , is similar to that found by @xcite for the entire sample of 49 non - x - ray - emitting hps laes with @xmath166 . our @xmath20 value of @xmath167 is significantly larger that those found by narrow - band surveys @xcite , but , given the vastly larger volume studied by the hps and the increased likelihood of finding rare , exceptionally bright objects , this difference is not a serious concern . despite the rather large uncertainty associated with @xmath167 , the total ly@xmath2 luminosity density of the @xmath20 universe , as defined through equation ( [ eq : sch_int ] ) , is very well defined , with @xmath168 ( ergs s@xmath31 mpc@xmath26 ) . this number is a slight underestimate , since , unlike the _ hst _ grism , the hps and the narrow - band surveys do not unambiguously identify every ly@xmath2 source above a flux limit . to be classified as an lae , a galaxy must also have a rest - frame equivalent width greater than 20 . ( lower equivalent width objects can be confused with foreground sources and are usually ignored . ) to correct for these missing objects , one must estimate the total amount of emission associated with low - equivalent width galaxies . if the equivalent width distribution of such objects follows the exponential function defined by the higher ew laes , then our census may be missing between 20% and 30% of escaping ly@xmath2 photons @xcite . alternatively , if , as suggested by @xcite , the true distribution of ly@xmath2 equivalent widths has a lognormal form similar to that associated with local [ o ii ] emitting galaxies ( e.g. , * ? ? ? * ) , then the missing ly@xmath2 flux would be far less . since observations of lyman - break galaxies appear to support the former possibility ( e.g. , * ? ? ? * ; * ? ? ? * ) , we use this more conservative approach in our calculations and assume that we are missing @xmath134 of the epoch s total ly@xmath2 luminosity density . figure [ fig : escape_prb ] combines our two likelihood functions to show the total ly@xmath2 escape fraction within our @xmath5 mpc@xmath6 survey volume . in the figure , the most probable escape fraction for the @xmath20 universe is @xmath7 , with the bulk of our statistical error arising from the uncertainty in the ly@xmath2 luminosity function . we must note , however , that the systematic errors associated with our measurement may be greater than this . while our assumptions about case b recombination , the history of @xmath20 star formation , and the population of low equivalent width ly@xmath2 emitters suggest that our measurement of is an upper limit , the greatest uncertainty is that associated with internal extinction . as mentioned above , none of the objects studied in this program have measured balmer decrements ; instead , we have estimates of the stellar reddenings derived from the observed slopes of the rest - frame uv continua . these are , at best , indirectly related to the nebular extinctions applicable to our problem @xcite . to produce figure [ fig : escape_prb ] , we have assumed @xmath4 , but this number likely has an uncertainty of @xmath149 dex . this factor adds an additional @xmath8 error onto our estimate , making our systematic error term comparable to our statistical uncertainty . our estimate of @xmath7 for the ly@xmath2 escape fraction is consistent with the study by @xcite , who normalized their @xmath166 lae observations via rest - frame uv measurements of the epoch s star formation rate density . it also agrees with most models for the evolution of with redshift @xcite . but these analyses rely predominantly on indirect measurements : in fact , the only previous study to directly compare ly@xmath2 emission with ly@xmath2 production is that of @xcite . by performing dual narrow - band surveys in ly@xmath2 and h@xmath2 , and estimating nebular extinction via the reddening of the stellar rest - frame uv continuum , @xcite obtained @xmath22% as the ly@xmath2 escape fraction for the @xmath67 universe . our @xmath25 measurements , which also rely on stellar reddenings , encompass a volume @xmath101 times larger than that of the @xcite survey , and are therefore far less sensitive to issues associated with cosmic variance . more importantly , by surveying this larger volume , we have been able to better define the bright end of the ly@xmath2 and h@xmath0 luminosity functions . while our data are not as good as those of @xcite for defining the slope of the faint - end of the luminosity function ( we adopt @xmath127 , rather than fitting for the variable ) , we have many more @xmath169 galaxies , and , unless the @xcite function has a very steep , faint - end slope , it is this latter parameter which is most important for defining the total emission - line flux of course , our observations also have the drawback of being based on h@xmath0 , rather than h@xmath2 , and are thus more sensitive to the effects of dust and internal extinction than the previous study . at present , our measurements of are limited by two factors : the depth of our survey for ly@xmath2 , and our ability to estimate @xmath170 for individual galaxies . both of these issues should improve rather rapidly . at the wavelengths considered here , the hps typically reached a monochromatic flux limit of @xmath171 ergs @xmath38 s@xmath31 , or @xmath172 ( ergs s@xmath31 ) at @xmath21 . unfortunately , as shown in figure [ fig : ratio ] , this is not deep enough to detect many of the emission - line galaxies found by the _ hst _ grism , as our median upper limit on the ly@xmath2 escape fraction is @xmath3 . in other words , our detection threshold lies just above that expected from our analysis of the epoch s volumetric escape fraction . the main hetdex survey , which begins in 2015 , is designed to reach ly@xmath2 flux limits that are a factor of @xmath173 fainter than that for the hps , i.e. , @xmath174 ergs @xmath38 s@xmath31 at @xmath175 . this limit should allow us to directly measure for the typical star - forming galaxy of the epoch , and enable a search for trends with object size , stellar mass , and star formation rate . similarly , our ability to measure the intrinsic balmer line luminosities of the h@xmath0-selected galaxies is rapidly advancing . like most previous studies of ly@xmath2 emission at @xmath176 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , our estimates of nebular extinction are inferred from measurements of stellar reddening , and , while there may be a relation between the two parameters , this unknown does introduce an uncertainty into the calculation . fortunately , instruments are now available which allow simultaneous infrared spectroscopy for large numbers of @xmath20 sources ( e.g. , * ? ? ? * ; * ? ? ? * ) , thereby enabling direct measurements of the galaxies balmer decrements . this capability will remove the most important source of systematic error from the analysis . this work was supported via nsf through grant ast 09 - 26641 . the institute for gravitation and the cosmos is supported by the eberly college of science and the office of the senior vice president for research at the pennsylvania state university . stsci declined to support this use of the 3d - hst treasury program data . cccccc + 10:00:35.54 & @xmath17702:13:03.0 & 1.934 & @xmath178 & @xmath179 & @xmath180 + 10:00:36.67 & @xmath17702:13:07.7 & 2.092 & @xmath181 & @xmath182 & @xmath183 + 10:00:33.96 & @xmath184 + 02:13:16.0 & 2.230 & @xmath185 & & @xmath186 + 10:00:17.30 & @xmath17702:19:26.5 & 2.084 & @xmath187 & @xmath188 & @xmath189 + 10:00:15.73 & @xmath17702:20:28.1 & 1.975 & @xmath190 & @xmath191 & @xmath192 + 10:00:47.16 & @xmath17702:17:44.5 & 2.025 & @xmath193 & @xmath194 & @xmath195 + 10:00:41.65 & @xmath17702:18:00.2 & 2.095 & @xmath196 & @xmath197 & @xmath198 + 10:00:40.82 & @xmath17702:18:22.9 & 2.070 & @xmath199 & & @xmath200 + 10:00:42.91 & @xmath17702:18:25.6 & 2.096 & @xmath201 & @xmath202 & @xmath203 + 10:00:38.64 & @xmath17702:18:36.3 & 1.927 & @xmath204 & @xmath205 & @xmath206 + 10:00:42.21 & @xmath17702:18:48.5 & 2.290 & @xmath207 & @xmath208 & @xmath209 + 10:00:18.68 & @xmath17702:14:59.9 & 2.310 & @xmath210 & & @xmath211 + 10:00:21.92 & @xmath17702:15:40.0 & 2.093 & @xmath212 & @xmath213 & @xmath214 + 10:00:20.69 & @xmath17702:12:53.4 & 2.163 & @xmath215 & @xmath216 & @xmath217 + 10:00:23.79 & @xmath17702:13:10.4 & 2.107 & @xmath218 & @xmath219 & @xmath220 + 10:00:17.22 & @xmath17702:13:38.0 & 2.105 & @xmath221 & @xmath222 & @xmath209 + 10:00:19.19 & @xmath17702:14:06.6 & 2.103 & @xmath223 & @xmath224 & @xmath225 + 10:00:23.61 & @xmath17702:15:57.4 & 2.088 & @xmath226 & @xmath227 & @xmath228 + 10:00:29.58 & @xmath17702:17:02.8 & 1.921 & @xmath229 & @xmath230 & @xmath231 + 10:00:28.64 & @xmath17702:17:48.7 & 2.093 & @xmath232 & & @xmath233 + 10:00:27.24 & @xmath17702:17:31.6 & 2.283 & @xmath234 & @xmath235 & @xmath236 + 10:00:22.88 & @xmath17702:17:14.0 & 2.220 & @xmath237 & @xmath238 & @xmath217 + 10:00:26.61 & @xmath17702:17:14.5 & 2.224 & @xmath239 & @xmath240 & @xmath241 + 10:00:24.22 & @xmath17702:14:11.7 & 2.104 & @xmath242 & @xmath243 & @xmath244 + 10:00:25.45 & @xmath17702:14:27.8 & 2.166 & @xmath245 & @xmath246 & @xmath241 + 10:00:23.41 & @xmath17702:14:32.3 & 2.097 & @xmath247 & @xmath248 & @xmath249 + 10:00:23.23 & @xmath17702:12:30.7 & 2.225 & @xmath250 & @xmath251 & @xmath252 + 10:00:39.12 & @xmath17702:14:51.0 & 2.125 & @xmath253 & @xmath254 & @xmath255 + 10:00:32.33 & @xmath17702:14:53.0 & 1.975 & @xmath256 & @xmath257 & @xmath258 + 10:00:32.60 & @xmath17702:15:35.9 & 2.162 & @xmath259 & @xmath260 & @xmath261 + 10:00:34.08 & @xmath17702:15:54.4 & 2.199 & @xmath262 & @xmath263 & @xmath206 + 10:00:44.93 & @xmath17702:15:53.2 & 2.093 & @xmath264 & @xmath265 & @xmath266 + 10:00:37.02 & @xmath17702:17:47.5 & 1.940 & @xmath267 & @xmath268 & @xmath269 + 10:00:29.81 & @xmath17702:18:49.2 & 2.200 & @xmath270 & & @xmath271 + 10:00:19.55 & @xmath17702:17:56.1 & 2.052 & @xmath272 & @xmath273 & @xmath274 + + 12:36:15.00 & @xmath17762:13:29.9 & 1.998 & @xmath275 & @xmath276 & @xmath269 + 12:36:20.46 & @xmath17762:14:52.2 & 1.999 & @xmath277 & @xmath278 & @xmath279 + 12:36:15.86 & @xmath17762:13:25.7 & 2.086 & @xmath280 & @xmath281 & @xmath282 + 12:36:18.78 & @xmath17762:10:37.3 & 2.264 & @xmath283 & @xmath284 & @xmath285 + 12:36:20.07 & @xmath17762:11:12.4 & 2.004 & @xmath286 & @xmath287 & @xmath288 + 12:36:13.33 & @xmath17762:11:45.2 & 2.258 & @xmath289 & @xmath290 & @xmath291 + 12:36:24.96 & @xmath17762:12:23.6 & 2.216 & @xmath292 & @xmath293 & @xmath249 + 12:36:40.61 & @xmath17762:13:10.9 & 2.051 & @xmath294 & @xmath295 & @xmath296 + 12:36:35.42 & @xmath17762:14:37.6 & 2.006 & @xmath297 & @xmath298 & @xmath299 + 12:36:42.09 & @xmath17762:13:31.4 & 2.018 & @xmath300 & @xmath301 & @xmath302 + 12:36:44.90 & @xmath17762:13:35.6 & 2.234 & @xmath303 & @xmath304 & @xmath305 + 12:36:50.10 & @xmath17762:14:01.2 & 2.235 & @xmath306 & & @xmath307 + 12:36:44.12 & @xmath17762:14:01.9 & 2.273 & @xmath308 & & @xmath309 + 12:36:47.46 & @xmath17762:15:03.5 & 2.083 & @xmath310 & @xmath311 & @xmath312 + 12:36:53.79 & @xmath17762:15:21.7 & 2.022 & @xmath313 & @xmath314 & @xmath261 + 12:36:55.60 & @xmath17762:14:50.7 & 1.975 & @xmath315 & @xmath316 & @xmath217 + 12:36:41.26 & @xmath17762:11:15.6 & 2.062 & @xmath317 & @xmath318 & @xmath319 + 12:36:55.06 & @xmath17762:13:47.1 & 2.233 & @xmath320 & @xmath321 & @xmath322 + 12:36:54.19 & @xmath17762:13:35.9 & 2.263 & @xmath323 & @xmath324 & @xmath325 + 12:37:04.33 & @xmath17762:14:46.2 & 2.220 & @xmath326 & @xmath327 & @xmath328 + 12:37:11.78 & @xmath17762:13:38.9 & 1.914 & @xmath329 & @xmath330 & @xmath206 + 12:37:11.20 & @xmath17762:14:31.5 & 2.191 & @xmath331 & @xmath332 & @xmath333 + 12:37:02.01 & @xmath17762:14:19.1 & 2.291 & @xmath334 & @xmath335 & @xmath209 + 12:36:53.32 & @xmath17762:10:35.9 & 1.975 & @xmath336 & @xmath337 & @xmath338 + 12:37:02.72 & @xmath17762:10:10.8 & 1.989 & @xmath339 & @xmath340 & @xmath341 + 12:37:11.00 & @xmath17762:11:40.1 & 2.270 & @xmath342 & & @xmath343 + 12:37:07.10 & @xmath17762:11:52.6 & 2.275 & @xmath344 & @xmath345 & @xmath249 + 12:37:10.42 & @xmath17762:10:35.5 & 2.161 & @xmath346 & @xmath347 & @xmath274 + + 3:33:00.61 & @xmath34827:40:27.0 & 2.030 & @xmath349 & @xmath350 & @xmath351 + 3:32:40.52 & @xmath34827:49:32.5 & 2.042 & @xmath352 & @xmath353 & @xmath354 + 3:32:34.44 & @xmath34827:47:42.8 & 2.030 & @xmath355 & @xmath356 & + 3:32:41.55 & @xmath34827:48:24.4 & 2.080 & @xmath357 & @xmath358 & @xmath359 + 3:32:42.19 & @xmath34827:48:59.6 & 2.079 & @xmath360 & @xmath361 & @xmath362 + 3:32:28.85 & @xmath34827:52:20.1 & 2.039 & @xmath363 & @xmath364 & + 3:32:13.76 & @xmath34827:43:00.5 & 2.070 & @xmath365 & @xmath366 & @xmath367 + 3:32:45.54 & @xmath34827:53:43.3 & 2.059 & @xmath368 & @xmath369 & @xmath370 + 3:32:26.83 & @xmath34827:46:01.8 & 2.079 & @xmath371 & @xmath372 & + 3:32:25.82 & @xmath373:46:09.3 & 2.076 & @xmath374 & @xmath375 & + 3:32:11.45 & @xmath34827:50:26.7 & 2.068 & @xmath376 & @xmath377 & @xmath378 + 3:32:18.12 & @xmath34827:49:41.9 & 2.034 & @xmath379 & @xmath380 & + 3:32:59.11 & @xmath34827:53:20.7 & 2.030 & @xmath381 & @xmath382 & @xmath383 + 3:32:04.00 & @xmath34827:43:51.6 & 2.044 & @xmath346 & & @xmath384 + 3:32:08.66 & @xmath34827:42:50.2 & 2.075 & @xmath385 & @xmath386 & + 3:32:07.84 & @xmath34827:42:27.2 & 2.049 & @xmath387 & @xmath388 & + 3:32:09.60 & @xmath34827:47:39.9 & 2.040 & @xmath389 & @xmath390 & + 3:32:21.40 & @xmath34827:51:26.2 & 2.037 & @xmath391 & @xmath392 & + 3:32:46.15 & @xmath34827:49:22.6 & 2.032 & @xmath393 & @xmath394 & + 3:32:43.46 & @xmath34827:43:36.5 & 2.078 & @xmath395 & @xmath396 & + 3:32:23.24 & @xmath34827:42:31.6 & 2.078 & @xmath339 & @xmath397 & + 3:32:35.73 & @xmath34827:46:39.0 & 2.073 & @xmath398 & @xmath399 & + 3:32:35.60 & @xmath34827:47:45.4 & 2.030 & @xmath400 & @xmath401 & + 3:32:40.19 & @xmath34827:46:54.6 & 2.069 & @xmath402 & @xmath403 & @xmath404 [ tab : galaxies ] lccc area ( arcmin@xmath34 ) & 275 & 157 & 157 + maximum volume ( mpc@xmath6 ) & 408,000 & 103,000 & 103,000 + number of galaxies & 97 & 43 & 17 + @xmath170 ( median ) & 0.37 & 0.37 & 0.30 + @xmath170 ( effective ) & 0.55 & 0.44 & 0.89 + fixed @xmath2 & @xmath405 & @xmath405 & @xmath405 + + @xmath167 ( ergs s@xmath31 ) & @xmath406 & @xmath407 & @xmath408 + @xmath409 ( @xmath410 ) ( mpc@xmath26 ) & @xmath411 & @xmath412 & @xmath413 + @xmath414 ( ergs s@xmath31 mpc@xmath26 ) & @xmath415 & @xmath416 & @xmath417 + best @xmath125 ( mpc@xmath418 & @xmath419 & @xmath420 & @xmath421 + [ tab : lf ]

there has lately been intense interest in the degree of alignment between the orbits of transiting exoplanets and the spin of their hosts stars . indeed , of 26 transiting exoplanet systems with measurements of the rossiter - mclaughlin ( rm ) effect , eight have now been found to be significantly spin - orbit misaligned in the plane of the sky . unfortunately , the rm effect only measures the angle between the orbit of a transiting exoplanet and the spin of its host star projected in the plane of sky , leaving unconstrained the compliment misalignment angle between the orbit of the planet and the spin of its host star along the line of sight . i use a simple model of stellar rotation benchmarked with observational data to statistically identify ten exoplanet systems from a sample of 75 for which there is likely a significant degree of misalignment along the line of sight between the orbit of the planet and the spin of its host star . i find that hat - p-7 , hat - p-14 , hat - p-16 , hd 17156 , kepler-5 , kepler-7 , tres-4 , wasp-1 , wasp-12 , and wasp-14 are likely spin - orbit misaligned along the line of sight . all ten systems have host stellar masses in the range , and the probability of this occurrence by chance is less than one in ten thousand . in addition , the planets in the candidate misaligned systems are preferentially massive and eccentric . i combine the distribution of misalignment in the plane of the sky from rm measurements with the distribution of misalignment along the line of sight from this analysis , and the result suggests that transiting exoplanets are more likely to be aligned along the line of sight than expected given predictions for the spin - orbit misalignment distribution of a population produced entirely by planet - planet scattering or kozai cycles and tidal friction . for that reason , there are likely two populations of close - in exoplanet systems : a population of aligned systems and a population of apparently misaligned systems in which the processes that lead to misalignment or to the survival of misaligned systems operate more efficiently in systems with massive stars and planets .

1006.2851

the degree of alignment @xmath2 between the orbit of a transiting exoplanet and the spin of its host star can provide insight into the system s formation . in the solar system , the angle @xmath2 between the planets orbits and the spin of the sun is small , about 7@xmath3 ( e.g. * ? ? ? that fact provided a key piece of evidence supporting the earliest models of planet formation in the same disk in which the sun formed . simple models of close - in exoplanet formation involving assembly near the ice line and quiescent migration through a gaseous protoplanetary disk predict that orbits should have only negligible misalignments with the spin of the host star ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? alternatively , simple models without migration but involving dynamical scattering between newly formed planets and planetesimals at several au predict that non - negligible misalignments @xmath2 can occur ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in reality , there will be some degree of migration in all massive planetary systems @xcite , and in certain situations migration in multiple systems can still lead to non - negligible misalignments @xcite . moreover , any multiple systems that migrate will likely be dynamically stable before the dissipation of their gaseous parent protoplanetary disk , but dynamically unstable after dissipation @xcite . the kozai effect @xcite and tidal processes might also contribute ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? despite these complications , the frequency of misaligned systems will provide another constraint on models of planet formation . i describe the geometry of transiting exoplanet systems according to the conventions of @xcite in figure [ fig01 ] . an exoplanet s orbit is defined by the vector @xmath4 and is inclined relative to the line of sight at an angle @xmath5 that can be measured through transit photometry . note that the fact that an exoplanet is observed to transit requires that @xmath6 . the host star s spin is defined by the vector @xmath7 and is inclined at an angle @xmath8 relative to the line of sight . the angle @xmath9 is the angular distance between the projections of @xmath4 and @xmath7 in the plane of the sky , and is observable through the rossiter - mclaughlin ( rm ) effect @xcite . the rm effect the detection of an anomalous blueshift and redshift superimposed on the stellar reflex in radial velocity measurements during transit has lately been used to measure @xmath9 for several transiting exoplanet systems ( e.g. * ? ? ? * and references therein ) . indeed , the latest rm measurements described in triaud et al . ( 2010 , submitted ) show that eight of the 26 systems with rm measurements are significantly spin - orbit misaligned in the plane of the sky . though @xmath8 is not constrained by the rm effect , if all of @xmath5 , @xmath8 , and @xmath9 are known , the calculation of the true deprojected angle @xmath10 between @xmath4 and @xmath7 is trivial . the inclination @xmath8 to the line of sight of stellar spin is difficult to measure for individual stars . if the rotation period of a star is known through photometric observations of starspots moving across its surface , and if the projected rotational velocity @xmath11 has been measured , then @xmath8 is easy to compute ( e.g. * ? ? ? astrosiesmology measurements of @xmath8 are also possible in special circumstances @xcite . unfortunately , period measurements are resource - intensive and require well - sampled and precise photometric time series . they may also be biased toward stars with many starspots and therefore biased toward more active and younger stars . on the other hand , the projection of stellar rotation velocity along the line of sight @xmath11 can be measured cheaply by examining the rotational broadening of absorption lines in a single high - resolution stellar spectrum . if the radius of the star is known from other means , it is easy to forward - model the observable quantity @xmath11 from the stellar rotation period @xmath12 predicted from theoretical or empirical models given a distribution for @xmath13 . monte carlo simulations can then derive the distribution of possible @xmath11 measurements . that distribution can then be compared to observational data for transiting exoplanet systems to determine if the observations are consistent with the alignment of a transiting planet s orbit with the spin of its host star . calculating the rotation period of sun - like stars as a function of mass and time is a tractable problem for stars older than about 650 myr . first , the rotation period of isolated sun - like stars monotonically slows down as they age @xcite . second , even though a population of sun - like stars is formed with a broad distribution of periods @xmath12 @xcite , by the time that population is the age of the hyades and praesepe ( about 650 myr ) the broad distribution has converged to a well - defined function of mass @xcite . two classes of physical models have been proposed to explain this early - time behavior : magnetic breaking by protostellar disks ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and accretion - driven stellar winds ( e.g. * ? ? ? though the physical process that produces this effect in the first @xmath14 myr is not yet known , the data @xcite and theoretical models @xcite are in agreement on the rotational evolution of sun - like stars after this time . indeed , sun - like stars likely lose angular momentum through magnetized stellar winds @xcite , and their rotation slows as @xmath15 , as described observationally as @xmath16 by @xcite . in this paper , i use a simple empirical model to calculate the rotation period of sun - like stars as a function of mass and age . i combine that model with a monte carlo simulation to marginalize over the uncertain masses , radii , and ages of a control sample of 866 stars from the spectroscopic properties of cool stars ( spocs ) catalog of @xcite to determine the expected range of projected rotation @xmath11 for each star in the control sample under the assumption that the inclination distribution of the spocs sample is isotropic . likewise , i perform a similar calculation for the host stars of transiting exoplanets , this time assuming alignment between the host star s spin and the orbit of its exoplanet . i identify those transiting exoplanet systems in which the disagreement between the @xmath11 predicted from the simple empirical model under the assumption of spin - orbit alignment and the observed @xmath11 is larger than the equivalent disagreement for any of the 866 stars in the spocs control sample . in particular , those systems that have anomalously small @xmath11 measurements relative to the measurement errors and the width of simulated @xmath11 distribution are possibly spin - orbit misaligned along the line of sight . i describe the details of my calculation in 2 . i discuss my results in the context of close - in planet formation in 3 . i summarize my findings in 4 . i combine the well defined mass - rotation period relation in the hyades and praesepe ( e.g. * ? ? ? * ) with the relation @xmath15 ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) to compute the expected rotation periods of sun - like stars as a function of mass and age . i use a monte carlo simulation to transform that function into a @xmath17 distribution by marginalizing over the uncertain distributions of stellar mass @xmath0 , radius @xmath18 , age @xmath19 , and the applicable inclination distribution for the sample . as a result , i can compare the observed @xmath20 measurements with the mean @xmath21 estimate from the monte carlo simulation to determine the degree of agreement between the two . in order to determine the degree of disagreement that can be attributed to random effects , i compare the measured values of @xmath20 of sun - like stars in the solar neighborhood from the spocs catalog with the mean @xmath21 from the monte carlo simulation given the stellar parameters of those stars and the standard isotropic inclination distribution . in this way , i use the control sample of spocs stars to determine a threshold level of disagreement that can be expected between the simple empirical model and observation given the observational uncertainties and the imperfections of the model . i then do a similar calculation for the host stars of transiting exoplanets , this time assuming alignment between the spin of the host star and the orbit of its exoplanet . i identify those exoplanet host stars in which the predicted @xmath21 disagrees with the observed value @xmath20 relative to the measurement errors by an amount greater then an equivalent disagreement for any of the 866 stars in the control sample . systems with @xmath20 slower than @xmath21 by an amount large relative to the quoted errors in @xmath20 and the width of the simulated @xmath17 distribution given the uncertainty in stellar mass , radius , and age are either systems with rotation properties unlike any star in the control sample or in which the inclination of the transiting planet s orbit @xmath5 and the inclination of the host star s spin @xmath8 are misaligned . since the spocs sample is representative of the sample of transiting exoplanet host stars , the former is very unlikely . for that reason , the best explanation of the anomalously slow projected rotation is spin - orbit misalignment along the line of sight . i examine a control sample of stars from the spocs catalog of @xcite to quantify the scatter about the simple model . the spocs sample selection is described in detail in @xcite . in summary , the bulk of the sample was initially selected from the hipparcos catalog such that each star has @xmath22 , is no more than three magnitudes above the main sequence , is not a known spectroscopic binary , and has no stellar companion within two arcseconds . subsequent spectroscopic measurements of caii h and k were then used to downselect to those stars with ages @xmath23 gyr , though an additional 100 stars with ages between 50 and 500 myr were later added in . as a result , the spocs sample has both young and evolved stars as indicated in figure 14 of @xcite . in addition , previously unknown spectroscopic binaries were removed from the spocs candidate list , so there are no spectroscopic binaries in the final catalog . short - period low - mass binary companions or exoplanets would have been readily apparent in the original california - carnegie planet search and would have been announced as such long ago . for those reasons , the rotation properties of stars in spocs sample should be unaffected by close companions . moreover , any overlap between the spocs sample and the hosts stars of transiting exoplanets is negligible . collectively , all of these selections and observations suggest that the spocs sample is a fair control sample for this analysis . i model the initial mass - period relation of a 650 myr population of sun - like stars by binning the available hyades and praesepe data @xcite as summarized in @xcite in 0.1 @xmath24 bins . the @xcite data only extends up to about 1.2 @xmath24 , so i supplement it with average rotation periods for more massive field stars as presented in @xcite . i use natural cubic splines to interpolate the binned and supplemented data and find that it minimizes the sum of square residuals relative to high - order polynomial interpolation . i determine the expected rotation period of a sun - like star as a function of mass and age by evolving the initial condition set by the stellar mass according to the relation ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) @xmath25 where @xmath26 can be approximated as a fifth - order polynomial in @xmath0 with coefficients in increasing order ( 5.558509,-77.843724,357.387761,-502.288156,282.908686,-56.329952 ) . i summarize my technique in figure [ fig02 ] . the model described above is likely too simplistic , the relevant input parameters ( stellar mass , radius , and age ) are all uncertain , and the output ( stellar period ) is difficult to observe . for those reasons , i use a monte carlo simulation to determine the range of observable @xmath17 expected for a sample of stars given mass , radius , and age as well as the uncertainties in each of those quantities plus the appropriate inclination distribution for that sample . in order to determine the degree of disagreement between the @xmath21 predicted from the simple empirical model and the observed @xmath20 that can be expected given the observational uncertainties and the imperfections of the simple empirical model , i use the control sample of stars from the spocs catalog . for each star in the spocs catalog , i use the estimated stellar mass and uncertainty @xmath0 and @xmath27 ( column 9 of table 9 ) , estimated stellar radius and uncertainty @xmath18 and @xmath28 ( column 8 of table 9 ) , and age range @xmath29 ( column 14 of table 9 ) . in the monte carlo , i sample each star s mass from a uniform distribution in mass between @xmath30 and @xmath31 , its radius from a uniform distribution in radius between @xmath32 and @xmath33 , and its age from a uniform distribution in age between the given lower and upper age limits . i sample the inclination to the line of sight of each star s rotation axis from the standard random distribution @xmath34 , where @xmath35 is drawn uniformly from the interval [ 0,1 ] . using these parameters , i evolve the initial period given by the randomly selected mass and the initial period - mass relation described in figure [ fig02 ] to the randomly selected age of the system according to equation [ eq1 ] . i then compute @xmath17 using the randomly generated radius of the star and its randomly selected inclination . i repeat this process 1000 times and thereby derive for each star the distribution of possible @xmath17 values , its mean @xmath21 , as well as the width of the distribution @xmath36 . i compare the predicted @xmath21 with the observed @xmath20 relative to the width of the monte carlo distribution @xmath36 and the error in the observed @xmath20 measurement @xmath37 through the rotation statistic @xmath38 : @xmath39 note that large positive ( negative ) values of @xmath38 indicate slower ( faster ) than expected @xmath11 values . stars that are well fit by the model will have small absolute values of @xmath38 . i repeat this calculation on each of the 866 stars from the spocs catalog with @xmath40 , assuming the standard isotropic distribution of inclination . this distribution of @xmath38 is the null hypothesis that i will use to compare to the distribution of @xmath38 measured in the transiting exoplanet systems under the assumption of spin - orbit alignment . i collected the properties of the 80 transiting exoplanets systems known as of may 2010 , of which the 75 listed in table [ tbl-1 ] have measured mass , radius , @xmath20 , and orbital inclination . in this case , the stellar masses and radii are more precisely measured than those available in the spocs catalog , and the inclination of each exoplanet s orbit is known from transit photometry . i again use a monte carlo simulation to determine the range of observable @xmath17 . i use the measured values for each star s mass and radius , and i sample each star s age uniformly between lower and upper bounds for its age from the literature . if there are no age estimates available in the literature , i sample the age of the star uniformly between 650 myr ( the age of the hyades and praesepe ) and its main sequence lifetime . if there are only upper limits on @xmath20 in the literature , i assume that the true value corresponds to the published upper - limit , as this is the most conservative way to approach a search for stars with anomalously slow projected rotation . in those situations , i assume that the error in @xmath20 is 1 km s@xmath41 . i assume that the inclination of the host star s spin relative to the line of sight @xmath8 is the same the inclination of the exoplanet s orbit @xmath5 . i repeat this process 1000 times and compute the rotation statistic @xmath38 for each stellar host ; i report the results of this calculation in table [ tbl-1 ] . i plot the distribution of @xmath38 for both samples in figure [ fig03 ] . the @xmath38 distribution of the control sample peaks near @xmath42 , indicating that the simple empirical model is accurate in that the expected residual between the model and the observations is close to zero . the distribution of @xmath38 falls off quickly as @xmath38 increases , and there are no control stars from the spocs catalog with @xmath43 . however , there does exist a tail of stars with much larger than expected @xmath11 values , and therefore large negative @xmath38 values . in other words , some stars in the control sample of spocs stars are fast rotators ; this is to be expected in any sample of sun - like field stars . on the other hand , there is no tail of spocs stars rotating arbitrarily more slowly than the simple empirical model predicts . for that reason , host stars of exoplanets that are observed to have @xmath43 are apparently unlike any of the 866 stars in the control sample either they rotate more slowly relative to the model than any star in the control sample or their anomalously slow projected rotation is due to a pole - on view of the star and therefore spin - orbit misalignment . i define fast rotators as those stars with @xmath44 , corresponding to the fastest 5% of the @xmath38 distribution for the spocs control sample . note that the fraction of fast rotators in each sample is statistically indistinguishable about @xmath45 in the spocs sample and about @xmath46 in the transiting exoplanet host star sample . the consistency of the two samples in the fraction of stars that are fast rotators is in sharp contrast to the disparity at the other end of the distribution . also , the peak of the @xmath38 distribution of transiting exoplanet systems is offset to more negative values of @xmath38 ( and therefore faster rotation ) than the field star sample . nevertheless , the distribution of the sample mean of @xmath38 for the 90% of the spocs control sample with @xmath47 is @xmath48 ; for the host stars of transiting exoplanets an equivalent calculation yields @xmath49 . i plot @xmath38 as a function of stellar mass for both samples in figure [ fig04 ] . there is a very sharp cutoff in the control sample at @xmath50 at all masses , and this sharp cutoff is due to lack of a constraint on the inclination of individual stars in the control sample . as a result , @xmath51 in equation [ eq2 ] can be large , resulting in small values of @xmath38 . i find that hat - p-7 , hat - p-14 , hat - p-16 , hd 17156 , kepler-5 , kepler-7 , tres-4 , wasp-1 , wasp-12 , and wasp-14 have @xmath52 , indicating @xmath20 too small to be explained by the uncertainties in their masses , radii , or ages if each star s spin and its exoplanet s orbit have the same inclination to the line of sight . unless the host stars of these transiting exoplanets are slower rotators relative to the model prediction than any of the 866 stars in the control sample from the spocs catalog , the most natural explanation for these discrepancies is that the inclination of the spin of these host stars @xmath8 is significantly different that the inclination of their exoplanet s orbit @xmath5 . if these anomalously slow projected rotation values are the result of spin - orbit misalignment , then the difference between @xmath20 and @xmath21 can be used to compute the four possible degenerate values of @xmath8 ( because @xmath53 is not one - to - one for @xmath54 ) . i report these values for the potentially spin - orbit misaligned systems in table [ tbl-2 ] . i also identify corot-2 , hat - p-2 , hd 189733 , kepler-8 , and xo-3 as fast rotators . note that corot-2 @xcite and hd 189733 @xcite are known to be more active than the sun , and therefore potential fast rotators . hat - p-2 has been observed to have a large radial velocity jitter , which may indicate an active and possibly quickly rotating star @xcite . all ten candidate misaligned systems have host stellar mass @xmath55 . again , either these stars have rotation properties unlike any of the 276 control stars in the mass range @xmath56 , or these stars are observed pole - on and therefore spin - orbit misaligned . in order to verify the robustness of this result , i have repeated my calculations with several different sets of input parameters . indeed , there are two potentially uncertain inputs to my simple empirical model : the power - law exponent @xmath57 in the relation @xmath58 and the ages of the stars in the spocs control sample . it is certain that the rotation of isolated sun - like stars slows down with time , so @xmath59 always . to determine if my result is sensitive to @xmath57 , i have repeated my calculation for @xmath60 , @xmath61 , and @xmath62 . if @xmath63 , then all of the systems identified with anomalously slow projected rotation in 2.3 are again identified as anomalously slow projected rotators . if @xmath61 , eight of the original ten systems are identified as anomalously slow projected rotators ( hat - p-7 , hat - p-14 , kepler-5 , kepler-7 , hd 17156 , tres-4 , wasp-12 , and wasp-14 ) . if @xmath62 , only three of the original ten systems are identified as anomalously slow projected rotators ( hat - p-7 , hat - p-14 , and wasp-14 ) . i have also redone my calculation using a completely uninformative prior on the ages of the stars in spocs sample . that is , i assumed that the age of each star was uniformly distributed between 650 myr and its main sequence lifetime . again , all ten systems identified in 2.3 as anomalously slow projected rotators and therefore possibly spin - orbit misaligned are again identified in this case . angular momentum exchange between a host star and its close - in planetary companion is also unlikely to significantly slow the rotation of the host star and produce an anomalously slow projected rotation velocity . for that to occur , the transfer of angular momentum from host star to exoplanet via tidal forces requires the rotation period of the star to be shorter than the orbital period of the planet . indeed , the median orbital period of the transiting planets orbiting the stars identified as anomalously slow projected rotators is @xmath64 days , while the median expected rotation period of their host stars is @xmath65 days . the presence of slightly evolved stars among the host stars of transiting exoplanets will also not produce false identifications as anomalously slow projected rotators . to see why , note that there are also slightly evolved stars in the control sample of spocs stars and recall that the threshold @xmath38 required to identify a star as an anomalously slow projected rotator was set by the maximum disagreement between model and observation in the control sample . in other words , the presence of slightly evolved stars has already been accounted for in setting the threshold . i plot in figure [ fig05 ] the distribution of the spocs sample in a theoretical hr diagram along with those transiting exoplanet hosts with trigonometric parallaxes . i also indicate the locations of those stellar hosts that i identified as anomalously slow projected rotators and therefore potentially spin - orbit misaligned systems . figure [ fig05 ] shows that many of the systems i identified as anomalously slow projected rotators appear to be evolving off of the zero - age main sequence ( zams ) . nevertheless , the fact that there are many stars in the control sample of spocs stars in similar evolutionary states suggests that the effects of stellar evolution on the rotation properties of evolved stars have already been accounted for in my analysis . moreover , there are several hosts of transiting exoplanets that have evolved off of the zams that i do not identify as anomalously slow projected rotators . collectively , these facts suggest that effects of stellar evolution can not explain the rotation properties of the ten systems i identified as anomalously slow projected rotators . additional selection effects in either the spocs control sample or the transiting exoplanet sample are unlikely to affect this result . indeed , the fact that spocs stars were preferentially selected to be those stars with small enough @xmath11 to enable high precision doppler radial velocity measurements biases my control sample to smaller @xmath11 at fixed stellar mass and age . however , since the candidate spin - orbit misaligned stars appear to have anomalously slow projected rotation , the bias acts against the false identification of a star as an anomalously slow projected rotator . though magnitude - limited transit surveys are biased toward stars slightly more massive than the sun , there are 276 stars in the control sample with @xmath66 to compare with . in addition , though targeted transit surveys are biased toward less active stars , the control sample of spocs stars in similarly biased . transit surveys should also not be biased toward evolved stars , as though both transit probability and transit duration increase linearly with host stellar radius ( favoring evolved systems ) , transit depth decreases quadratically with host stellar radius ( favoring unevolved systems ) . for that reason , the chance of detecting a transiting planet is roughly insensitive to host stellar radius . for the anomalously slow projected rotators identified in 2.3 , i list in table [ tbl-2 ] the inclination to the line of sight of the stellar rotation axis if the anomalous projected rotation is due to spin - orbit misalignment along the line of sight . i also note published rm measurements for four of the systems i identify as apparent slow projected rotators ; hat - p-7 and wasp-14 has been identified as a spin - orbit misaligned system @xcite , while hd 17156 and tres-4 are consistent with spin - orbit alignment in the plane of the sky @xcite . though hd 17156 is consistent with spin - orbit alignment in the plane of the sky from rm measurements , my technique measures misalignment along the line of sight . there is other circumstantial evidence that hd 17156 is misaligned along the line of sight . @xcite measured a @xmath67 km s@xmath41 , a stellar radius @xmath68 and used ca ii h and k to infer a rotation period @xmath69 days . if the system has @xmath70 , then a period @xmath69 days and the inferred stellar radius @xmath68 implies @xmath71 km s@xmath41 ( as compared to @xmath72 km s@xmath41 expected based on my simple model ) . clearly , it s possible that in this case the activity indicator gives a spurious period estimate ; still , if the discrepancy is due to misalignment then @xmath73 ( as compared to @xmath74 degrees from my measurement ) . tres-4 is consistent with alignment in the plane of the sky from measurements of the rm effect . however , that is a different angle than that measured through my analysis . it could be that the geometry of the tres-4 system is apparently aligned in the plane of the sky but misaligned along the line of sight ; alternatively , tres-4 could simply be an anomalously slow rotator . i find that ten of the 75 systems in table [ tbl-1 ] have anomalously slow projected rotation and are likely spin - orbit misaligned along the line of sight . in contrast , the latest rossiter - mclaughlin measurements reported and summarized in triaud et al . ( 2010 , submitted ) suggest that eight of the 26 systems with rm measurements have significant spin - orbit misalignment in the plane of the sky . the higher fraction of systems observed to be misaligned by the rm effect is expected , as the rm effect can detect much smaller degrees of misalignment than the coarser technique described here . assume for the moment that the smallest degree of misalignment this technique can identify is @xmath75 ( as suggested by the minimum difference from table [ tbl-2 ] ) . of the 26 systems with measurements of the rm effect , six have @xmath76 . at the same time , the slightly higher incidence of misaligned systems in the sample of rm measurements might be related to the fact that rm measurements are resource intensive . that is , systems that seem to have a higher a priori probability of spin - orbit misalignment based on the current understanding of spin - orbit misalignment ( e.g. eccentric systems ) are more likely to be targeted for rm measurement . the fact that this analysis includes all known transiting exoplanet systems without the targeting biases inherent in the current sample of rm measurements may even produce a less biased spin - orbit misalignment distribution . i identify ten host stars of transiting exoplanet systems in which either the host star is more slowly rotating relative to the simple model than any of the stars in the control sample of spocs stars or in which the star is viewed pole - on and the system is spin - orbit misaligned . this degeneracy can be broken for at least kepler-5 and kepler-7 , as high precision kepler photometry should reveal their rotation periods . if the periods are indeed short , then the small observed @xmath11 indicates spin - orbit misalignment along the line of sight . alternatively , if the observed rotation periods are long , then kepler-5 and kepler-7 are systems in which the host star has somehow lost angular momentum much more quickly than expected by models of sun - like stellar spin - down . high precision photometry from other sources might enable similar measurements for the other systems identified as anomalously slow projected rotators . all ten of the candidate misaligned systems have host stars more massive than the sun . to determine the significance of this observation , i performed a monte carlo simulation in which i randomly selected with replacement ten stars from the 75 listed in table [ tbl-1 ] . i repeated this process 10@xmath77 times . less than 0.01% of the monte carlo trials produced a sample in which all ten stars had @xmath78 . therefore , the probability that all ten candidate misaligned systems would be identified around massive stars by chance is less than one in ten thousand . similarly , the median mass of the planets in the candidate misaligned systems is @xmath79 jupiter - masses , while the median mass of the whole sample of 75 systems is @xmath80 jupiter - masses . interestingly , there may be a hint of the same apparent overrepresentation of massive planets in spin - orbit misaligned systems from rossiter - mclaughlin measurements @xcite . eccentricity also seems to be related to misalignment probability . indeed , of the nine transiting systems with fgk stellar hosts and eccentricity @xmath81 ( hd 80606 , hd 17156 , hat - p-2 , wasp-8 , xo-3 , hat - p-11 , wasp-17 , corot-9 , and hat - p-14 ) , there is now evidence from either measurements of the rm effect or this analysis that six of those systems are significantly misaligned ( hd 80606 , hd 17156 , wasp-8 , xo-3 , wasp-17 , and hat - p-14 ) . the other three systems are unlikely to be observed to be spin - orbit misaligned for a number of reasons : hat - p-2 is likely an intrinsically fast - rotator and therefore unlikely to be identified as misaligned along the line of sight by the method described in this paper even if it does have significant spin - orbit misalignment along the line of sight , hat - p-11 may have too small a radius for measurements of the rm effect , and corot-9 has the largest pericenter distance by far of the known transiting planets and almost certainly has had a different evolutionary history than the other known transiting planets . to determine the significance of the apparent correlation between eccentricity and spin - orbit misalignment , i performed a monte carlo simulation to determine the expected number of systems with eccentricity @xmath81 from a sample of 16 transiting systems the total number of significantly misaligned systems from rm measurements and this analysis expected if spin - orbit misalignment is unrelated to eccentricity . i randomly selected with replacement the eccentricities of 16 planets from the 75 systems listed in table [ tbl-1 ] , and i repeated this process 10@xmath77 times . i find that the mean number of systems with @xmath81 in a sample of 16 systems under the assumption that eccentricity is unrelated to spin - orbit misalignment is @xmath82 . the observation of six systems with @xmath81 from a random sample of 16 systems occurs in less than one in 150 trials , indicating that eccentricity and spin - orbit misalignment are related . in an attempt to derive some constraint on the underlying spin - orbit misalignment distribution in the 75 systems in table [ tbl-1 ] , i performed two completeness calculations for two different @xmath8 distributions resulting from two different @xmath2 distributions : ( 1 ) corresponding to planet - planet scattering ( e.g. * ? ? ? * ) , a case in which the inclination of the host star is completely independent of the alignment of the orbit of its planet @xmath83 , where @xmath84 ; and ( 2 ) corresponding to kozai cycles and tidal friction ( e.g. * ? ? ? * ) , a case in which the distribution of @xmath8 is derived from the @xmath2 distribution as given in figure 10 of @xcite and the @xmath9 distribution as given in the rm compilation in table 5 of triaud et al . ( 2010 , submitted ) . assuming a threshold detectable misalignment that corresponds to the smallest degree of misalignment identified ( @xmath85 ) in 75 systems , i find that the expected number of detections in case ( 1 ) is @xmath86 and in case ( 2 ) it is @xmath87 . even though each individual system in this analysis is securely identified as an anomalously slow projected rotator and therefore possibly spin - orbit misaligned , both completeness calculations are strongly dependent on the smallest degree of misalignment this technique can identify . that quantity is uncertain , and as such more data is necessary before any conclusive statements about distribution of spin - orbit alignments is made . the full sample of kepler transiting planet detections and host star properties available at the end of its four year mission might resolve this issue . collectively , this analysis indicates that predictions of spin - orbit misalignment for a population of close - in planets entirely produced by planet - planet scattering or kozai cycles with tidal friction overpredict the number of misaligned systems . as a result , there seems to be two populations of close - in planets : a population that is spin - orbit aligned and a population that is apparently spin - orbit misaligned . the processes that lead to misalignment or to the survival of misaligned systems seem to operates most efficiently in systems with massive host stars and planets . indeed , the striking appearance of candidate misaligned systems at @xmath88 in figure [ fig04 ] indicates that the processes that lead to spin - orbit misalignment and survival are threshold processes , similar to the rapid increase in exoplanet incidence with host stellar metallicity ( e.g. * ? ? ? * ; * ? ? ? . in this case , the transition to frequent apparent spin - orbit misalignment occurs at the same stellar mass at which sun - like stars with near solar metallicity develop radiative envelopes . this dramatic change in stellar structure will likely strongly effect the poorly - understood tidal processes that play a major role in the formation , evolution , and long - term survival of spin - orbit misaligned systems . there are ten transiting exoplanet systems from a larger sample of 75 systems in which the projected rotation velocity of the host star assuming spin - orbit alignment is apparently smaller relative to a simple empirical model of stellar spin - down than any of 866 stars in a representative control sample of stars in the same mass range from the spectroscopic properties of cool stars ( spocs ) catalog : hat - p-7 , hat - p-14 , hat - p-16 , hd 17156 , kepler-5 , kepler-7 , tres-4 , wasp-1 , wasp-12 , and wasp-14 . unless the host stars of these transiting exoplanets are slower rotators relative to the model prediction than any of the 866 stars in the representative control sample , the anomalously low @xmath11 values are best explained by a pole - on view of the star and therefore spin - orbit misalignment . i only identify significant possible spin - orbit misalignment in systems with host stars more massive than the sun , as in all ten systems the host stars have stellar masses @xmath0 in the range @xmath89 . this observation is significant , as there is less than a one in ten thousand chance of its occurrence by chance . there is evidence that systems with massive and eccentric planets are also those most likely to be significantly spin - orbit misaligned . given current measurements of the rossiter - mclaughlin effect and frequency of misalignment along the line of sight determined in this analysis , models of close - in planet formation invoking planet - planet scattering or kozai cycles and tidal friction alone overpredict the number of misaligned systems . as a result , there seems to be two populations of close - in planets : those that are aligned and those that are apparently misaligned . the process that leads to misalignment or survival of misaligned systems likely operates more efficiently in systems with both massive stars and planets . interestingly , apparently misaligned systems appear to occur preferentially in systems in which the envelope of the host star is radiative , and this transition in stellar structure is likely related through tidal processes to the formation , evolution , and long - term survival of spin - orbit misaligned systems . i thank greg laughlin and connie rockosi for useful comments and conversation . this research has made use of nasa s astrophysics data system bibliographic services . this material is based upon work supported under a national science foundation graduate research fellowship . aigrain , s. , et al . 2008 , , 488 , l43 alonso , r. , et al . 2008 , , 482 , l21 anderson , d. r. , et al . 2010 , , 709 , 159 attridge , j. m. , & herbst , w.1992 , , 398 , l61 bakos , g. . , et al . 2007 , , 670 , 826 bakos , g. . , et al . 2007 , , 671 , l173 bakos , g. . , et al.2009 , , 696 , 1950 bakos , g. . , et al . 2009 , , 707 , 446 bakos , g. . , et al . 2010 , , 710 , 1724 barge , p. , et al . 2008 , , 482 , l17 bean , j. l. , et al . 2008 , , 486 , 1039 beck , j. g. , & giles , p. 2005 , , 621 , l153 borucki , w. j. , et al . 2010 , , 713 , l126 bouchy , f. , et al . 2005 , , 444 , l15 bouchy , f. , et al . 2008 , , 482 , l25 buchhave , l. a. , et al . 2010 , arxiv:1005.2009 burke , c. j. , et al . 2007 , , 671 , 2115 burke , c. j. , et al . 2008 , , 686 , 1331 cameron , a. c. , et al . 2007 , , 375 , 951 charbonneau , d. , et al . 2009 , , 462 , 891 chatterjee , s. , ford , e. b. , matsumura , s. , & rasio , f. a. 2008 , , 686 , 580 choi , p. i. , & herbst , w. 1996 , , 111 , 283 christian , d. j. , et al . 2009 , , 392 , 1585 collier cameron , a. , campbell , c. g. , & quaintrell , h. 1995 , , 298 , 133 daemgen , s. , hormuth , f. , brandner , w. , bergfors , c. , janson , m. , hippler , s. , & henning , t. 2009 , , 498 , 567 deeg , h. j. , et al . 2010 , , 464 , 384 deleuil , m. , et al . 2008 , , 491 , 889 dunham , e. w. , et al . 2010 , , 713 , l136 fabrycky , d. , & tremaine , s.2007 , , 669 , 1298 fabrycky , d. c. , & winn , j. n. 2009 , , 696 , 1230 fernandez , j. m. , holman , m. j. , winn , j. n. , torres , g. , shporer , a. , mazeh , t. , esquerdo , g. a. , & everett , m. e. 2009 , , 137 , 4911 fischer , d. a. , & valenti , j. 2005 , , 622 , 1102 fischer , d. a. , et al . 2007 , , 669 , 1336 ford , e. b. , havlickova , m. , & rasio , f. a. 2001 , icarus , 150 , 303 fossey , s. j. , waldmann , i. p. , & kipping , d. m. 2009 , , 396 , l16 fridlund , m. , et al . 2010 , , 512 , a14 gibson , n. p. , et al . 2008 , , 492 , 603 gillon , m. , pont , f. , moutou , c. , bouchy , f. , courbin , f. , sohy , s. , & magain , p. 2006 , , 459 , 249 gillon , m. , et al . 2009 , , 496 , 259 gillon , m. , et al . 2009 , , 501 , 785 gizon , l. , & solanki , s. k. 2003 , , 589 , 1009 hartman , j. d. , et al . 2009 , , 706 , 785 hebb , l. , et al . 2009 , , 693 , 1920 hebb , l. , et al . 2010 , , 708 , 224 hellier , c. , et al . 2009 , , 690 , l89 hellier , c. , et al . 2009 , , 460 , 1098 ida , s. , & lin , d. n. c. 2008 , , 673 , 487 ida , s. , & lin , d. n. c. 2008 , , 685 , 584 irwin , j. , & bouvier , j. 2009 , iau symposium , 258 , 363 jenkins , j. m. , et al . 2010 , arxiv:1001.0416 johns - krull , c. m. , et al . 2008 , , 677 , 657 johnson , j. a. , et al . 2008 , , 686 , 649 johnson , j. a. , winn , j. n. , albrecht , s. , howard , a. w. , marcy , g. w. , & gazak , j. z. 2009 , , 121 , 1104 joshi , y. c. , et al . 2009 , , 392 , 1532 juri , m. , & tremaine , s.2008 , , 686 , 603 kawaler , s. d. 1988 , , 333 , 236 keppens , r. , macgregor , k. b. , & charbonneau , p. 1995 , , 294 , 469 koch , d. g. , et al . 2010 , , 713 , l131 koenigl , a. 1991 , , 370 , l39 kovcs , g. , et al . 2007 , , 670 , l41 kozai , y. 1962 , , 67 , 591 kraft , r. p. 1967 , , 150 , 551 latham , d. w. , et al . 2009 , , 704 , 1107 latham , d. w. , et al . 2010 , , 713 , l140 laughlin , g. , wolf , a. , vanmunster , t. , bodenheimer , p. , fischer , d. , marcy , g. , butler , p. , & vogt , s. 2005 , , 621 , 1072 lger , a. , et al . 2009 , , 506 , 287 lin , d. n. c. , bodenheimer , p. , & richardson , d. c. 1996 , , 380 , 606 lister , t. a. , et al . 2009 , , 703 , 752 marcy , g. w. , butler , r. p. , fischer , d. a. , & vogt , s. s. 2004 , extrasolar planets : today and tomorrow , 321 , 3 marigo , p. , girardi , l. , bressan , a. , groenewegen , m. a. t. , silva , l. , & granato , g. l. 2008 , , 482 , 883 marzari , f. , & weidenschilling , s. j. 2002 , icarus , 156 , 570 matt , s. , & pudritz , r. e. 2005 , , 632 , l135 mccullough , p. r. , et al . 2006 , , 648 , 1228 mccullough , p. r. , et al . 2008 , arxiv:0805.2921 mclaughlin , d. b. 1924 , , 60 , 22 mcnally , d. 1965 , the observatory , 85 , 166 melo , c. , santos , n. c. , pont , f. , guillot , t. , israelian , g. , mayor , m. , queloz , d. , & udry , s. 2006 , , 460 , 251 mestel , l. 1968 , , 138 , 359 moutou , c. , et al . 2008 , , 488 , l47 nagasawa , m. , ida , s. , & bessho , t.2008 , , 678 , 498 narita , n. , et al . 2007 , , 59 , 763 narita , n. , sato , b. , hirano , t. , & tamura , m. 2009 , , 61 , l35 narita , n. , sato , b. , hirano , t. , winn , j. n. , aoki , w. , & tamura , m. 2010 , arxiv:1003.2268 noyes , r. w. , et al . 2008 , , 673 , l79 nutzman , p. , charbonneau , d. , winn , j. n. , knutson , h. a. , fortney , j. j. , holman , m. j. , & agol , e. 2009 , , 692 , 229 ogihara , m. , & ida , s. 2009 , , 699 , 824 ohta , y. , taruya , a. , & suto , y. 2005 , , 622 , 1118 pl , a. , et al . 2008 , , 680 , 1450 pl , a. , et al . 2009 , , 700 , 783 pl , a. , et al . 2010 , , 401 , 2665 papaloizou , j. c. b. , & terquem , c. 2001 , , 325 , 221 pollacco , d. , et al . 2008 , , 385 , 1576 pont , f. 2009 , , 396 , 1789 pont , f. , et al . 2009 , , 502 , 695 prosser , c. f. , et al . 1995 , , 107 , 211 radick , r. r. , thompson , d. t. , lockwood , g. w. , duncan , d. k. , & baggett , w. e. 1987 , , 321 , 459 rauer , h. , et al . 2009 , , 506 , 281 rossiter , r. a. 1924 , , 60 , 15 santos , n. c. , et al . 2004 , , 426 , l19 sato , b. , et al . 2005 , , 633 , 465 scholz , a. , & eislffel , j. 2007 , , 381 , 1638 skillen , i. , et al . 2009 , , 502 , 391 skumanich , a. 1972 , , 171 , 565 shporer , a. , et al . 2009 , , 690 , 1393 shu , f. , najita , j. , ostriker , e. , wilkin , f. , ruden , s. , & lizano , s. 1994 , , 429 , 781 snellen , i. a. g. , et al . 2009 , , 497 , 545 southworth , j. 2008 , , 386 , 1644 southworth , j. , et al . 2009 , , 396 , 1023 southworth , j. , et al . 2009 , , 707 , 167 sozzetti , a. , torres , g. , charbonneau , d. , latham , d. w. , holman , m. j. , winn , j. n. , laird , j. b. , & odonovan , f. t. 2007 , , 664 , 1190 sozzetti , a. , et al . 2009 , , 691 , 1145 tassoul , j .- l . 2000 , stellar rotation / jean - louis tassoul cambridge ; new york : cambridge university press , 2000 . ( cambridge astrophysics series ; 36 ) , terquem , c. , & papaloizou , j. c. b. 2002 , , 332 , l39 terquem , c. , & papaloizou , j. c. b. 2007 , , 654 , 1110 thommes , e. w. , & lissauer , j. j.2003 , , 597 , 566 torres , g. , et al . 2007 , , 666 , torres , g. , winn , j. n. , & holman , m. j. 2008 , , 677 , 1324 torres , g. , et al . 2010 , , 715 , 458 triaud , a. h. m. j. , et al . 2009 , , 506 , 377 valenti , j. a. , & fischer , d. a.2005 , , 159 , 14 ward , w. r. 1997 , icarus , 126 , 261 weber , e. j. , & davis , l. , jr . 1967 , , 148 , 217 west , r. g. , et al . 2009 , , 137 , 4834 winn , j. n. , et al . 2005 , , 631 , 1215 winn , j. n. , et al . 2007 , , 133 , 1828 winn , j. n. , et al . 2008 , , 683 , 1076 winn , j. n. , et al . 2009 , , 693 , 794 winn , j. n. , johnson , j. a. , albrecht , s. , howard , a. w. , marcy , g. w. , crossfield , i. j. , & holman , m. j. 2009 , , 703 , l99 yu , q. , & tremaine , s. 2001 , , 121 , 1736 lccccccccccc corot-1 & 0.95 & 1.11 & @xmath90 & @xmath90 & 5.20 & 1.00 & 85.10 & 2.17 & 0.87 & -2.28 & 1 + corot-2 & 0.97 & 0.90 & 0.20 & 4.00 & 11.85 & 0.50 & 87.84 & 3.10 & 1.27 & -6.43 & 2,3 + corot-3 & 1.37 & 1.56 & 1.60 & 2.80 & 17.00 & 1.00 & 85.90 & 14.92 & 1.21 & -1.32 & 4 + corot-4 & 1.16 & 1.17 & 0.70 & 2.00 & 6.40 & 1.00 & 90.00 & 6.23 & 0.93 & -0.12 & 5,6 + corot-5 & 1.00 & 1.19 & 5.50 & 8.30 & 1.00 & 1.00 & 85.83 & 2.05 & 0.12 & 1.04 & 7 + corot-6 & 1.05 & 1.03 & 1.00 & 3.30 & 7.50 & 1.00 & 89.07 & 3.62 & 0.62 & -3.29 & 8 + corot-7 & 0.93 & 0.87 & 1.20 & 2.30 & 3.50 & @xmath90 & 80.10 & 2.57 & 0.24 & -0.91 & 9 + corot-9 & 0.99 & 0.94 & 0.15 & 8.00 & 3.50 & @xmath90 & 89.99 & 2.66 & 1.54 & -0.46 & 10 + gj 436 & 0.45 & 0.46 & 1.00 & 10.00 & 2.40 & @xmath90 & 85.80 & 1.03 & 0.35 & -1.30 & 11,12,13 + gj 1214 & 0.16 & 0.21 & 3.00 & 10.00 & 2.00 & @xmath90 & 88.62 & 5.65 & 0.96 & 2.63 & 14 + hat - p-1 & 1.13 & 1.12 & 0.70 & 5.20 & 3.75 & 0.58 & 86.28 & 4.06 & 1.16 & 0.24 & 12,15 + hat - p-2 & 1.36 & 1.64 & 2.10 & 3.10 & 20.80 & 0.30 & 86.72 & 13.60 & 0.77 & -8.72 & 16 + hat - p-3 & 0.94 & 0.82 & 0.10 & 6.90 & 0.50 & 0.50 & 87.24 & 2.19 & 1.24 & 1.26 & 17 + hat - p-4 & 1.26 & 1.59 & 3.60 & 6.80 & 5.50 & 0.50 & 89.90 & 5.81 & 0.53 & 0.42 & 18 + hat - p-5 & 1.16 & 1.17 & 1.20 & 4.70 & 2.60 & 1.50 & 86.75 & 4.31 & 0.82 & 1.00 & 12,19 + hat - p-6 & 1.29 & 1.46 & 1.70 & 2.80 & 8.70 & 1.00 & 85.51 & 9.09 & 0.66 & 0.33 & 12,20 + hat - p-7 & 1.47 & 1.84 & 1.20 & 3.20 & 3.80 & 0.50 & 85.70 & 32.66 & 4.61 & 6.23 & 21 + hat - p-8 & 1.28 & 1.58 & 2.40 & 4.40 & 11.50 & 0.50 & 87.50 & 7.77 & 0.67 & -4.45 & 22 + hat - p-9 & 1.28 & 1.32 & 0.20 & 3.40 & 11.90 & 1.00 & 86.50 & 9.95 & 3.87 & -0.49 & 23 + hat - p-11 & 0.81 & 0.75 & 2.40 & 12.40 & 1.50 & 1.50 & 88.50 & 1.00 & 0.24 & -0.33 & 24 + hat - p-12 & 0.73 & 0.70 & 0.50 & 4.50 & 0.50 & 0.40 & 89.00 & 1.62 & 0.52 & 1.72 & 25 + hat - p-13 & 1.22 & 1.56 & 4.30 & 7.50 & 2.90 & 1.00 & 83.40 & 4.54 & 0.36 & 1.54 & 26 + hat - p-14 & 1.39 & 1.47 & 0.90 & 1.70 & 8.40 & 0.50 & 83.50 & 20.57 & 1.91 & 6.16 & 27 + hat - p-16 & 1.22 & 1.24 & 1.20 & 2.80 & 3.50 & 0.50 & 86.60 & 6.32 & 0.78 & 3.03 & 28 + hd 17156 & 1.26 & 1.45 & 2.30 & 3.70 & 2.60 & 0.50 & 86.20 & 6.86 & 0.48 & 6.15 & 29,30 + hd 80606 & 0.97 & 0.98 & @xmath90 & @xmath90 & 1.80 & 1.00 & 89.29 & 2.08 & 0.85 & 0.21 & 31,32,33 + hd 149026 & 1.29 & 1.50 & 1.00 & 2.80 & 6.00 & 0.50 & 85.40 & 10.32 & 1.51 & 2.71 & 12,34,35 + hd 189733 & 0.81 & 0.76 & 2.40 & 12.00 & 3.32 & 0.05 & 85.76 & 1.03 & 0.23 & -9.75 & 12,36,37 + hd 209458 & 1.12 & 1.16 & 2.40 & 3.90 & 4.70 & 0.16 & 86.55 & 3.66 & 0.26 & -3.45 & 12,38 + kepler-4 & 1.22 & 1.49 & 3.00 & 6.00 & 2.20 & 1.00 & 89.76 & 5.04 & 0.49 & 2.55 & 39 + kepler-5 & 1.37 & 1.79 & 2.40 & 3.60 & 4.80 & 1.00 & 86.30 & 14.63 & 0.87 & 7.41 & 40 + kepler-6 & 1.21 & 1.39 & 2.80 & 4.80 & 3.00 & 1.00 & 86.80 & 4.95 & 0.38 & 1.82 & 41 + kepler-7 & 1.35 & 1.84 & 2.50 & 4.50 & 4.20 & 0.50 & 86.50 & 12.51 & 1.05 & 7.13 & 42 + kepler-8 & 1.21 & 1.49 & 2.34 & 5.34 & 10.50 & 0.70 & 84.07 & 5.30 & 0.64 & -5.48 & 43 + ogle - tr-10 & 1.14 & 1.17 & 0.10 & 7.20 & 7.00 & 1.00 & 83.87 & 4.52 & 2.70 & -0.86 & 12,44,45 + ogle - tr-56 & 1.23 & 1.36 & 1.90 & 4.20 & 3.20 & 1.00 & 79.80 & 5.72 & 0.65 & 2.12 & 12,44,45 + ogle - tr-111 & 0.85 & 0.83 & 2.20 & 14.00 & 5.00 & @xmath90 & 88.10 & 1.11 & 0.29 & -3.73 & 12,44,45 + ogle - tr-113 & 0.78 & 0.77 & 10.80 & 14.00 & 5.00 & @xmath90 & 89.40 & 0.73 & 0.03 & -4.27 & 12,44,46 + ogle - tr-132 & 1.31 & 1.32 & 0.10 & 2.70 & 5.00 & @xmath90 & 83.30 & 14.01 & 6.69 & 1.33 & 12,44,45 + ogle2-tr - l9 & 1.52 & 1.53 & 0.00 & 0.66 & 39.33 & 0.38 & 79.80 & 128.20 & 205.72 & 0.43 & 47 + tres-1 & 0.88 & 0.81 & 0.90 & 7.10 & 1.08 & 0.30 & 88.40 & 1.64 & 0.47 & 0.99 & 12,48,49 + tres-2 & 0.98 & 1.00 & 2.90 & 7.70 & 2.00 & 1.00 & 83.62 & 1.92 & 0.27 & -0.08 & 12,50,51 + tres-3 & 0.93 & 0.83 & 0.10 & 3.70 & 1.50 & 1.00 & 81.85 & 2.88 & 1.53 & 0.76 & 52 + tres-4 & 1.40 & 1.81 & 2.50 & 4.40 & 8.50 & 0.50 & 82.86 & 16.49 & 1.34 & 5.60 & 51,52 + wasp-1 & 1.30 & 1.52 & 2.40 & 3.60 & 5.00 & @xmath90 & 88.25 & 8.58 & 0.50 & 3.20 & 12,45,53 + wasp-2 & 0.89 & 0.84 & 0.00 & 14.00 & 5.00 & @xmath90 & 84.80 & 1.61 & 1.72 & -1.71 & 12,51,53 + wasp-3 & 1.24 & 1.31 & 0.70 & 3.50 & 13.40 & 1.50 & 85.06 & 7.37 & 1.70 & -2.66 & 54,55 + wasp-4 & 0.85 & 0.87 & 2.00 & 9.00 & 2.00 & 1.00 & 89.35 & 1.37 & 0.29 & -0.61 & 56 + wasp-5 & 0.96 & 1.03 & 1.10 & 9.80 & 3.50 & 1.00 & 85.80 & 2.02 & 0.63 & -1.25 & 56,57 + wasp-6 & 0.88 & 0.87 & 4.00 & 18.00 & 1.40 & 1.00 & 88.47 & 1.02 & 0.22 & -0.37 & 58 + wasp-7 & 1.28 & 1.24 & @xmath90 & @xmath90 & 17.00 & 2.00 & 89.60 & 6.98 & 2.13 & -3.43 & 59 + wasp-8 & 1.03 & 0.95 & 3.00 & 5.00 & 2.00 & 0.60 & 88.52 & 2.29 & 0.17 & 0.46 & 60 + wasp-10 & 0.70 & 0.78 & 0.60 & 1.00 & 6.00 & @xmath90 & 86.90 & 2.82 & 0.20 & -3.12 & 61 + wasp-11/hat - p-10 & 0.83 & 0.79 & 4.10 & 11.70 & 0.50 & 0.20 & 88.60 & 0.99 & 0.15 & 1.96 & 62 + wasp-12 & 1.35 & 1.57 & 1.00 & 3.00 & 2.20 & 1.50 & 83.10 & 14.26 & 2.19 & 4.55 & 63 + wasp-13 & 1.03 & 1.34 & 3.60 & 14.00 & 4.90 & @xmath90 & 86.90 & 2.25 & 0.44 & -2.43 & 64 + wasp-14 & 1.21 & 1.31 & 0.50 & 1.00 & 2.80 & 0.57 & 84.32 & 10.45 & 1.05 & 6.39 & 65,66 + wasp-15 & 1.18 & 1.48 & 2.60 & 6.70 & 4.00 & 2.00 & 85.50 & 4.40 & 0.57 & 0.19 & 67 + wasp-16 & 1.02 & 0.95 & 0.10 & 8.10 & 3.00 & 1.00 & 85.22 & 2.71 & 1.63 & -0.15 & 68 + wasp-17 & 1.20 & 1.38 & 0.40 & 3.90 & 9.00 & 1.50 & 87.80 & 6.95 & 2.25 & -0.76 & 69 + wasp-18 & 1.28 & 1.23 & 0.00 & 2.00 & 11.00 & 1.50 & 86.00 & 15.25 & 14.39 & 0.29 & 70,71 + wasp-19 & 0.96 & 0.94 & 1.00 & 14.50 & 4.00 & 2.00 & 80.50 & 1.58 & 0.59 & -1.16 & 72 + wasp-21 & 1.01 & 1.06 & 7.00 & 17.00 & 1.50 & 0.60 & 88.75 & 1.43 & 0.18 & -0.11 & 73 + wasp-22 & 1.10 & 1.13 & 1.00 & @xmath90 & 3.50 & 0.60 & 89.20 & 3.21 & 0.95 & -0.26 & 74 + wasp-24 & 1.13 & 1.15 & 0.00 & 3.70 & 6.96 & @xmath90 & 85.71 & 7.46 & 11.68 & 0.04 & 75 + wasp-25 & 1.00 & 0.95 & 0.40 & 4.60 & 3.00 & 1.00 & 87.70 & 3.06 & 1.06 & 0.04 & 76 + wasp-26 & 1.12 & 1.34 & 4.00 & 8.00 & 2.40 & 1.30 & 82.50 & 3.06 & 0.31 & 0.49 & 77 + wasp-28 & 1.08 & 1.05 & 3.00 & 8.00 & 2.20 & 1.60 & 89.10 & 2.39 & 0.33 & 0.11 & 78 + wasp-29 & 0.82 & 0.85 & 10.00 & @xmath90 & 1.50 & 0.60 & 87.96 & 0.80 & 0.06 & -1.15 & 79 + wasp-33 & 1.50 & 1.44 & 0.00 & 0.50 & 90.00 & 10.00 & 87.67 & 131.62 & 200.45 & 0.21 & 80 + xo-1 & 1.03 & 0.93 & 0.10 & 4.10 & 1.11 & 0.67 & 89.06 & 3.76 & 1.96 & 1.28 & 12,45,81 + xo-2 & 0.97 & 0.98 & 3.90 & 8.70 & 1.30 & 0.30 & 88.90 & 1.69 & 0.20 & 1.08 & 82,83 + xo-3 & 1.21 & 1.38 & 2.00 & 3.40 & 18.54 & 0.17 & 84.20 & 5.79 & 0.44 & -27.04 & 84,85 + xo-4 & 1.32 & 1.56 & 1.50 & 2.70 & 8.80 & 0.50 & 88.70 & 11.76 & 1.02 & 2.61 & 86 + xo-5 & 1.00 & 1.11 & 7.70 & 9.30 & 0.70 & 0.50 & 86.80 & 1.71 & 0.05 & 2.01 & 87,88 + lcccccccc hat - p-7 b & 85.70 & 6.7 & 170 & 190 & 350 & -177.5 & @xcite + hat - p-14 b & 83.50 & 24 & 160 & 200 & 340 & @xmath90 & + hat - p-16 b & 86.60 & 34 & 150 & 210 & 330 & @xmath90 & + hd 17156 b & 86.20 & 22 & 160 & 200 & 340 & 10 & @xcite + kepler-5 b & 86.30 & 19 & 160 & 200 & 340 & @xmath90 & + kepler-7 b & 86.50 & 20 & 160 & 200 & 340 & @xmath90 & + tres-4 & 82.86 & 31 & 150 & 210 & 330 & -6.3 & @xcite + wasp-1 b & 88.25 & 36 & 140 & 220 & 320 & @xmath90 & + wasp-12 b & 83.10 & 8.8 & 170 & 190 & 350 & @xmath90 & + wasp-14 b & 84.32 & 15 & 160 & 200 & 340 & -33.1 & @xcite +

in this work , we explore a new direction by complementing the game - theoretic applications of nonlocal correlations through appropriately formulated games using temporal quantum correlations . in the context of bayesian games , we show the way temporal correlations can be utilized , thereby leading to quantum strategies that are impossible to simulate in a classical world . furthermore , by educing some intriguing characteristics of projective measurements performed between two temporally separated measurements on a given system , we also construct some specific types of biased bayesian games . finally , we impart a new dimension to the cooperative bayesian nonlocal game introduced in [ http://www.nature.com/ncomms/2013/130703/ncomms3057/full/ncomms3057.html[nat . comm . 4 , 2057 ( 2013 ) ] ] through expedient inclusion of temporal correlations in the epr - bohm setting .

1601.03006

in recent years , several categories of quantum games ( @xcite-@xcite,@xcite ) have been constructed , all of which indubitably reveal the superior potentiality of certain strategies accessible only in the quantum domain . harnessing nonlocal resources for game - theoretic implementations ( @xcite-@xcite ) has been a primary cynosure of research in this direction . for instance , a number of interesting nonlocal games ( @xcite-@xcite,@xcite-@xcite,@xcite ) have been studied by setting up close associations with bell s inequalities . nonlocality has been one of the most captivating hallmarks of quantum mechanics that has been employed extensively for information processing tasks , as for example , in teleportation ( @xcite-@xcite ) and cryptographic protocols ( @xcite,@xcite ) . tersely stated , nonlocality concerns the phenomenon whereby two spatially separated agents sharing a pair of entangled qubits can generate correlations that can not be reproduced by any local realistic theory . hence the correlations are said to be nonlocal . this feature was first explicitly demonstrated by john bell in the epr - bohm setting through the formulation of testable inequalities ( @xcite,@xcite,@xcite ) characterizing a specific type of constraint which local correlations are bound to satisfy but which is violated by some quantum correlations . + a particular category of nonlocal games known as bayesian chsh games was originally discussed by brunner and linden ( @xcite ) , and very recently , by anna pappa et al ( @xcite ) , albeit in a different setting , through which the connection between bayesian games ( @xcite,@xcite ) and bell nonlocality was made much more categorical . in our work , we transcend the relevance of spatial correlations and highlight the role of temporal correlations with a similar rationale of probing its game - theoretic applications in the context of bayesian games . just like a violation of the bell - chsh inequality ( @xcite,@xcite ) paves the way for characterizing nonlocal correlations between two spatially separated measurements , violations of the leggett - garg inequality ( lgi ) ( @xcite-@xcite,@xcite-@xcite,@xcite,@xcite,@xcite,@xcite ) and the temporal chsh inequality ( @xcite,@xcite,@xcite ) decidedly exhibit the existence of certain types of temporal correlations in the quantum domain arising due to the inherent invasiveness of quantum mechanical measurements . central to the game - theoretic application of temporal quantum correlations is the appreciation of the crucial distinction between ideal non - invasive measurements realizable in the classical domain and inherently invasive measurements characteristic of the quantum world . nonlocal cooperative bayesian games , as discussed in @xcite , exploit violations of the chsh inequality ( @xcite ) to demonstrate the existence of _ super - classical _ payoffs . by _ super - classical _ payoffs , we imply expected payoffs of the players that lie beyond the range of payoffs feasible in the classical domain . such payoffs can be secured by operating with nonlocal correlations ( characterizable through the chsh inequality ) which , in turn , necessitates a cooperation between two active participants . + in a similar vein , we formulate a cooperative _ temporal bayesian game _ by deploying violations of the leggett - garg inequality to illustrate the existence of _ super - classical _ payoffs . it is a derivative of the xor - type game ( @xcite ) so devised as to entail a bijective connection between the expected payoff and the three - term expression pertaining to the third - order leggett - garg inequality ( @xcite ) . the inequality requires measuring a dichotomic physical quantity @xmath0 ( having a discrete spectrum of values @xmath1 ) at three different times @xmath2 , @xmath3 , and @xmath4 respectively ( @xmath5 ) and computing the complete set of two - time correlation functions @xmath6 to forge an expression of the form @xmath7 . the lgi mandates that for a specific class of measurements that do not cause any disturbance to the fundamental ontic state ( @xcite-@xcite ) of the system @xmath8 must be bounded above by unity : @xmath9 although initially proposed with the intention of testing our classical intuition of the macroscopic world , the lgi is interesting enough for microscopic objects , mostly due to an intimate connection between violations of the same and the behaviour of a physical system under measurement ( @xcite ) . we formally introduce the connection between bayesian games and the leggett - garg inequality in section [ s3 ] , but only after we recapitulate , in a nutshell , the essence of the two - party bayesian framework in section [ s2 ] , as this model would serve as the quintessential bedrock in setting up the temporal version . + subsequently , with the objective of making temporal bayesian games a little bit more interesting , we expand the sphere of participants in the customary bayesian setting and provide a suitable connotation to the objectives of those actively participating in it . this modification involves insetting a conflict of interest between two asymmetric groups of players . it is easy to appreciate , as has already been stated earlier , that the primary incentive behind constructing quantum bayesian games ( @xcite,@xcite ) is to underscore how quantum correlations can help to outmanoeuvre classical strategies . however , one might also be tempted into considering games in which the effective " quantumness of the correlations may be viewed as the ultimate deciding factor in settling a conflict of interests . by effective " quantumness , we make reference to the perceived nature of the correlations as ascertained through the machinery of suitable correlation inequalities ( like the lgi ) . consequently , in sections [ s4 ] and [ s5 ] , we construct a few biased " games of this sort , in the context of which we demonstrate some interesting attributes of intervening " projective measurements that make it possible , at least superficially , to quell the quantumness of temporal correlations , thereby precluding the possibility of achieving super - classical payoffs , even when the games are played quantum mechanically and the correlations are manifestly quantum . thus , the purpose of this novel formulation is not to simply throw light on the quantum - classical divide as regards the quantitative nature of correlations . instead , we are motivated by the urge to probe quantum correlations in its own right without obligating a comparison , qualitative or quantitative , with classical correlations . taking a cue from the observations pertaining to a third - party intervention in the temporal bayesian game , we examine , in section [ s6 ] , the effect of intervening measurements in a spatio - temporal setting and delineate its game - theoretic applicability . this feature is unravelled in the familiar ( nonlocal ) bayesian chsh game ( @xcite ) fittingly reworked by introducing a time - gap between the measurements effected by alice and bob on their respective qubits , in order that the relevance of a third - party intervention at an intermediate instant may be investigated . to get started , we lay out , succinctly , the general setting in which quantum bayesian games ( @xcite,@xcite ) have hitherto been formulated . the game consists of two players alice and bob along with an outside party usually called the referee . the latter is , however , not an active participant in the game . the modus operandi of the game is fairly simple , with the referee designated the task of asking questions to each of alice and bob who are supposed to provide answers in response . the respective sets of questions from which the referee chooses his questions for alice and bob are denoted @xmath10 and @xmath11 , while the sets of possible answers that alice and bob can come up with are denoted @xmath12 and @xmath13 respectively . contingent on the pair of answers @xmath14 provided to a given pair of questions @xmath15 , the referee allocates a payoff to the two players . let us denote the corresponding payoff function as @xmath16 , where @xmath17 . that is , mathematically , @xmath18 as is always the case for cooperative games , higher the payoff , better for the two players engaged in a cooperation . the question - answer rounds are repeated for an ideally infinite number of times with the questions following a well - defined long - time probability distribution represented by @xmath19 @xmath20 and the corresponding responses the conditional distribution @xmath21 @xmath20 . the probabilities are obviously assumed to be normalized , i.e. @xmath22 and @xmath23 @xmath24 . the average payoff @xmath25 can then be easily computed : @xmath26 where @xmath27 and @xmath28 are @xmath29 matrices with elements given by @xmath30_{rs}=\alpha(r , s|k , l)$ ] and @xmath31_{rs}=\mu(r , s|k , l)$ ] . @xmath31 $ ] is referred to as the payoff matrix . here , by @xmath32 we denote the cardinality of the set @xmath33 . the arguments @xmath34 and @xmath35 of the expected payoff function show its dependence on the probability distributions of the choices of both the questions as well as the responses . + and @xmath36 represent the possible measurement settings of alice and bob respectively . what setting would be chosen by either party is determined by the question posed to the latter . @xmath37 and @xmath38 are the respective questions posed by the referee to alice and bob . the outcomes of measurements performed by the two parties are two - valued , @xmath1 . @xmath39 and @xmath40 are the answers provided by alice and bob depending on the measurement outcomes respectively observed.,width=302,height=226 ] such games are called games with incomplete information because , while the players have prior knowledge of the distributions @xmath35 , @xmath34 and the payoff function @xmath41 , they can not communicate between themselves in order to know each other s questions and answers . however , they can connive together before the start of the game and decide upon a strategy that could help them in their quest for a high payoff . such a strategy is usually referred to as shared randomness in the classical context , while in the quantum setting a predetermined strategy involves a suitable choice of shared quantum state and measurement schemes . + if the players deploy independent strategies based purely on the questions asked to them , then @xmath21 is of the product form : @xmath42 , where @xmath43 and @xmath44 are the marginals of the joint distribution . however , it is also possible to adopt correlated strategies based on some classical advice ( the source of shared randomness ) , parametrized by @xmath45 and delivered to both the players by an external advisor . the advice @xmath45 can be distributed according to some distribution @xmath46 . each player can then choose a strategy contingent on the question posed to him and on the common advice . so , for classical correlated strategies , we have @xmath47 in the quantum context , it is possible to generate correlations that do not admit this decomposition . for instance , the epr - bohm nonlocal correlations are inconsistent with such a restrictive decomposition rule . this inconsistency is , however , not endemic to spatial correlations only - there exist temporal quantum correlations as well that do not adhere to this restriction . the incompatibility is manifested through violations of suitable correlation inequalities characterizing some algebraic constraints on correlations of the form [ el ] . the relevance of any correlation inequality ( such as the chsh inequality ) that places a constraint on a certain category of correlations can be made explicit in the bayesian framework by appropriately mapping the questions to measurement settings and the answers to the observed outcomes ( @xcite ) . as an immediate consequence of the correlation inequality , the expected payoff arising from the corresponding category of correlations is forced to satisfy an inequality ( equivalent to the original correlation inequality ) of the following form : @xmath48 in the following section , we establish a connection between a variant of the xor - type game discussed by brunner and linden ( @xcite ) , and the leggett - garg inequality ( @xcite ) which encapsulates a restriction on temporal correlations resulting from ideal noninvasive measurements characteristic of classical systems . to construct this game , rather than considering just two individual players we envisage two groups of players calling them @xmath49 and @xmath50 respectively , with each group comprising a total of @xmath51 members . we label these players @xmath52 and @xmath53 . also , we now have a group of @xmath51 referees , @xmath54 , with the @xmath55 referee @xmath56 assigned the task of interrogating the @xmath55 pair of players @xmath57 - @xmath58 is questioned earlier and @xmath59 is questioned at a later time . the purpose behind tweaking the usual scenario of the nonlocal chsh game will soon become clear . + pairs of players @xmath60 are interrogated by @xmath51 referees . @xmath61 asks questions to @xmath58 and @xmath59 . groups @xmath49 and @xmath50 cooperate using a predetermined strategy with an aim to maximize their payoff.,width=340,height=264 ] for the ease of establishing correspondences between the questions and the measurement choices , and between the answers and the measurement outcomes , we choose @xmath62 , and @xmath63 . in the limit @xmath64 , we assume that we have well - defined normalized probability distributions @xmath19 and @xmath21 . we define the payoff function as follows : @xmath65 then , the average payoff obtained by the groups a and b can be computed as @xmath66 where the averaging is carried out by considering the relevant payoffs secured by all of the @xmath51 pairs @xmath67 , @xmath68 . + let us now presume that temporal correlations of measurements performed on the same system are manipulated by the two groups of players in the following manner : for the pair of questions @xmath14 , @xmath58 and @xmath59 perform projective spin measurements on a shared quantum bit and notify their answers depending on the outcomes observed in the respective cases . temporal correlations are defined by correlations between the observed outcomes of measurements effected on the same system at different instants in time . here , we shall concern ourselves with two - point correlations arising from measurements performed at two distinct times . as we have already made it clear , there is a bijective mapping between the outcomes and the responses , and therefore , we shall use @xmath69 to signify not only the pair of responses but also the pair of observed outcomes . so we invoke the following correspondence : @xmath70 ( the set of observed outcomes ) . that is , the first element of either set corresponds to an observed outcome of @xmath71 and the second to @xmath72 . similarly , we use @xmath73 and @xmath74 to denote the answers as well as the observed outcomes . + for the questions @xmath75 and @xmath76 posed to @xmath58 , he performs measurements of the same spin projection , say along @xmath77 , at times @xmath2 and @xmath3 respectively , and for the questions @xmath76 and @xmath78 made to @xmath59 , the latter performs measurements along @xmath77 , at times @xmath3 and @xmath4 respectively , on a shared two - level quantum system . the measurement rules in this game are delineated in fig . + can be asked one of the questions @xmath79 and @xmath80 ; @xmath59 can be asked one of the questions @xmath80 and @xmath81 . if @xmath79 is asked , @xmath58 performs measurement at @xmath2 ; if @xmath80 is asked , @xmath82 performs a measurement at @xmath3 ; if @xmath81 is asked , @xmath59 performs a measurement at @xmath4 . @xmath83,width=340,height=188 ] it is easy to see that the temporal correlator @xmath84 can now be expressed as : @xmath85 defining the variable @xmath86 and invoking the normalization condition @xmath87 , we have the following compact relation : @xmath88 the probability distribution @xmath19 is defined as follows : @xmath89 the consequence of such a definition is that if one of the two players @xmath58 and @xmath59 is asked the question @xmath76 corresponding to his / her set of questions , the other will definitely not be asked the question @xmath76 from his / her set . the rest of the question pairs , @xmath90 , @xmath91 and @xmath92 are all equi - probable . note that this gives either member of the pair @xmath93 the chance to unambiguously make out the question posed to his / her partner , if and when he / she is asked the question @xmath76 from his / her set . however , the players wo nt be knowing each other s individual responses and therefore , will not be able to decide upon a pair of answers that would fetch them the higher payoff of + 1 . they would thus be compelled to play the game probabilistically . further , also note that since @xmath94 , there is no possibility of @xmath58 and @xmath59 making the same measurement at the same time . + having stated all the relevant parameters , we define @xmath95 so that upon substitution of @xmath96 and @xmath97 in @xmath98 we obtain : @xmath99 where @xmath100 resembles the well known expression corresponding to the leggett - garg inequality ( lgi ) @xmath101 ( @xcite ) . the lgi is based on the assumptions of realism and noninvasive measurability , but holds true whenever the joint probabilities of outcomes obtained on measuring the same physical quantity at two different times are factorizable . however , if the temporal correlations are quantum ( @xcite,@xcite,@xcite ) in nature , where measurement invasiveness does not usually allow the probabilities of outcomes at two different times to be independent of one another , @xmath8 can attain a highest value of @xmath102 . + it thus becomes evident in this backdrop that while ideal noninvasive ( classical ) measurements can yield an average payoff of at most @xmath103 ( approx . ) , manipulation of invasive quantum measurements can lead to a maximum average payoff of @xmath104 ( a staggering @xmath105 hike in the expected payoff ) . alternatively speaking , if the game is played using resources in which there exists no correlation between temporally separated measurements performed on a system , the average payoff is bounded above by a numerical value of @xmath103 . however , if we make use of quantum resources so that measurements performed are inherently invasive in nature , we can go beyond this classical upper constraint by about @xmath105 . also @xmath106 represents a quantum correlated nash equilibrium as this is the highest possible payoff attainable by @xmath49 and @xmath50 using quantum resources and the players wo nt have any incentive to switch to an alternative strategy . we shall now continue to consider a scenario identical to the bayesian lgi game with a minor alteration of its rules . in place of the group of @xmath51 referees , we consider a group @xmath107 of @xmath51 players @xmath108 not only performing the same task as the referees in the previous game but also actively participating in the game . however , we no more wish to treat @xmath16 as the payoff function assigned to @xmath58 and @xmath59 for the pair of responses @xmath14 on being asked the pair of questions @xmath15 . instead , we define it as points earned by the pair @xmath57 " under the same circumstances . + in this context , we construct the quantity @xmath109 , where @xmath110 and @xmath111 corresponds to the upper bound for the leggett - garg inequality . but in this new game , we insist on the active participation of not only the members of @xmath49 and @xmath50 , but also of those belonging to the group @xmath107 . if @xmath49 and @xmath50 win the game , the group @xmath107 loses , and vice - versa . thus , on the one side , we have a cooperation between @xmath49 and @xmath50 , both of whom have conflict of interests with the third group @xmath107 . the following winning criterion is imposed for @xmath49 and @xmath50 : @xmath112 that is , if the average number of points earned by @xmath49 and @xmath50 exceeds the classical upper limit of @xmath103 , they win and @xmath107 loses the game . + as we have observed in the original version of the game , ( in the limit @xmath64 ) @xmath49 and @xmath50 can make use of suitable temporal quantum correlations which lead to a violation of the leggett - garg inequality and can definitely win this game . we , therefore , consider the following question : is this game perfectly biased in favour of @xmath49 and @xmath50 , so that the group @xmath107 is doomed to lose ? the answer is an emphatic no " . there , in fact , exists a nice strategy for @xmath107 that would shift the balance completely in its favour . + to this end , we elucidate an advantageous feature of _ intervening _ measurements performed at some time(s ) _ between _ two temporally separated measurements , all of which are carried out on the same system . in the ordinary lgi game corresponding to a given run , @xmath58 is assumed to perform a measurement at time @xmath113 ( say ) , following which , @xmath59 makes a measurement at a time @xmath114 ( say ) on the same system . but here we wish to investigate a situation where , over and above these measurements , one or more number of measurements are carried out on that system at some _ intermediate _ times . we shall deal only with projective measurements for simplicity . a further simplification is attained by setting the hamiltonian of the system , @xmath115 , to zero , so that the collapse of the state vector upon a measurement is the only dynamics present in the qubit . however , in that case , there is no such restriction placed on the directions of measurements effected by @xmath58 and @xmath59 ( recall that in the original game we had imposed the condition that the same spin projection is measured at different times ) . in the heisenberg picture , the observable @xmath116 being measured by @xmath58 and @xmath59 evolves unitarily from @xmath113 to @xmath117 in conformity with the rule @xmath118 but setting @xmath119 causes the state to be frozen in time , so , in the heisenberg picture , the observables do nt evolve either . as a result , we have @xmath120 . that is why , we relax the restriction of measuring the same observable in this context and allow for the measurement of three observables corresponding to three arbitrarily chosen directions of spin . + assume now that a total of @xmath121 intermediate spin measurements are executed on the given system at times @xmath122 along the directions @xmath123 respectively , where @xmath124 . further suppose that @xmath58 and @xmath59 perform their spin measurements along @xmath125 and @xmath126 respectively and denote the intermediate measurement outcomes , in chronological order , by @xmath127 . under such a condition , it can be shown ( see appendix [ ii ] ) that we are led to the following expression for the temporal correlator between the outcomes obtained by @xmath49 and @xmath50 : @xmath128 where @xmath129 . for only one intermediate measurement performed in the direction @xmath130 , the following form can be inferred : @xmath131 herein we wish to manifest a novel feature of correlators of the above form . a straightforward substitution into the third - order leggett - garg expression reveals that the correlators defined in @xmath132 are incapable of violating the lgi , i.e. @xmath133 . in other words , an intervening projective measurement corresponding to each of the sub - ensembles defining @xmath134 and @xmath135 ensures that the leggett - garg inequality is satisfied and the invasiveness of the measurement performed by alice apparently " disappears . @xmath136 note that we have explicitly assumed here that the same projective measurement ( along @xmath130 ) is carried out at some intermediate time as regards to each of the three sub - ensembles . equation [ e15 ] is easy to verify , however for a simple proof , one is referred to appendix [ ii ] . + performs a spin measurement along @xmath125 at @xmath113 ; @xmath59 measures along @xmath126 at @xmath117 . @xmath137 intervenes at an intermediate time @xmath138 performing a measurement along @xmath130 . this intervention modifies the correlations between the measurement outcomes of @xmath58 and @xmath59 according to equation [ e14 ] . @xmath139,width=302,height=151 ] the apparent reversion to classicality as a consequence of an intervening measurement is precisely what could be exploited in this game if the group @xmath107 aspires to shift the bias of the game in its favour . as is clearly discernible at this juncture , the members of @xmath107 could conspire , in unison , to induce the _ same _ projective measurement on the system shared by each of the pairs @xmath57 so as to guarantee that any advantage gained by the groups @xmath49 and @xmath50 over the classical upper bound for the average number of points earned vanishes . in this case , since @xmath101 , or equivalently @xmath140 , @xmath49 and @xmath50 would never be able to win the game , no matter what spin measurements each of the pairs @xmath57 carries out on the shared system . we have thus illustrated the existence of a simple but useful strategy available to @xmath107 that could effectively rob @xmath141 of the counterfactual super - classical advantage that would have been achieved had there been no intervention . in fact , it does nt matter either as to what intermediate measurement is performed by each of the referees ; equation @xmath142 holds for any unit vector @xmath130 . therefore , as long as @xmath137 performs a projective spin measurement in between @xmath113 and @xmath117 on the system shared by @xmath58 and @xmath59 , it is guaranteed that @xmath107 will outsmart @xmath49 and @xmath50 . this game too has a similar setting as the lgi one ( [ s3 ] ) , albeit with a few essential modifications . in this case , we choose all of the sets @xmath143 and @xmath13 to be @xmath144 and continue to denote , by @xmath16 , the number of points earned by @xmath58 and @xmath59 when they provide the pair of responses @xmath69 to the pair of questions @xmath145 asked by @xmath137 . however , the function @xmath41 is also modified ( @xcite ) , as delineated below : @xmath146 the distribution function corresponding to any pair of questions is taken to be uniform : @xmath147 the rules remain the same as the biased lgi game , apart from the fact that @xmath148 is now equal to @xmath149 , where @xmath150 is the upper bound of the temporal chsh inequality ( @xcite,@xcite,@xcite ) : @xmath151 in the temporal chsh scenario , there are two parties with the respective measurement choices @xmath152 and @xmath153 . the first party measures one variable from the set @xmath152 at an earlier time @xmath113 and the second party measures another from the set @xmath153 at a later time @xmath117 , and correlators corresponding to each of the @xmath154 pairs of measurements , @xmath155 , @xmath156 , are estimated . all variables are assumed to be dichotomic , i.e. @xmath157 and measurements pertain to a single qubit . + in the context of this game , @xmath58 is assumed to perform a projective measurement at @xmath113 and @xmath59 another projective measurement at @xmath114 on the same system . the measurement rules are outlined in the following table : + [ cols="^,^,^ " , ] + * s. m. * @xmath158 spin measurement + * t. m. * @xmath158 time of measurement if the temporal correlations between the measurement outcomes obtained by @xmath58 and @xmath59 are utilized , then in view of the fact that the relation @xmath159 continues to hold as we persist with the mappings @xmath160 , we uncover the following relation for @xmath161 : @xmath162 and accordingly , an upper bound for ideal classical measurements is obtained : @xmath163 and @xmath59 perform spin measurements on the qubit at @xmath113 and @xmath117 respectively . @xmath137 performs multiple interventions at intermediate times @xmath164 on the same qubit that is shared by @xmath58 and @xmath59 . quite remarkably , for arbitrary number of interventions , the temporal chsh inequality pertaining to the correlations between the measurements of @xmath58 and @xmath59 is always satisfied.,width=340,height=151 ] but suitable quantum temporal correlations will result in a maximum value of about 0.707 ( @xmath165 increase ) for @xmath166 , as the tsirelson s bound ( @xcite,@xcite ) for the temporal chsh inequality happens to be @xmath167 ( identical to the spatial scenario ( @xcite ) ) . therefore , as per the rules of the game , whenever the members of @xmath49 and @xmath50 operate with correlations that precipitate in a violation of the temporal chsh inequality , they win the game and @xmath107 ends up on the losing side . however , recall that in the biased lgi game ( [ s4 ] ) , we proposed a novel strategy for the group @xmath107 enabling it to wrest out a win with certainty . could that strategy be employed in this context as well ? indeed , it can . we observe , once again , that by performing an intermediate measurement it is possible to quell the super - classical advantage of the correlations exploited by @xmath49 and @xmath50 . not only that , the strategy in this case is found to extremely robust , as we shall soon be able to acknowledge . if we compute the chsh expression corresponding to the temporal correlators ( [ e13 ] ) pertaining to the situation where arbitrary number of intermediate projective measurements are performed on the shared qubit , we find that it is also bounded above by @xmath76 . @xmath168\leqslant 2\end{aligned}\ ] ] thus , quite remarkably , even when multiple intervening measurements are executed at different intermediate times on the system , the temporal chsh inequality is still satisfied . this follows directly from the fact that @xmath169 is bounded above by 1 , for all positive integral values of @xmath121 . consequently , it can be made out quite clearly that following any intermediate measurement executed by @xmath137 , @xmath59 wo nt be able to retrieve any super - classical advantage by performing any additional ( intermediate ) measurement(s ) . thus , the balance tilts irrevocably in the favour of @xmath107 precluding any possibility of @xmath49 and @xmath50 winning the game . we have already mentioned before that the connection between bayesian games and bell nonlocality was made explicit in the seminal work of brunner and linden ( @xcite ) , in which the chsh inequality ( @xcite ) was viewed through the lens of bayesian games . here we propose a slender variation of the nonlocal chsh game by casting it in a similar setting as the temporal chsh one . plus , we extend the number of participants to three : the active participants are alice , bob and charlie with the first two cooperating with each other against the third . charlie asks questions to alice and bob following a distribution @xmath19 . the functions @xmath16 and @xmath19 are identical to those discussed in the biased temporal chsh game ( [ s5 ] ) : @xmath146 @xmath147 the winning condition for alice and bob continues to be dictated by @xmath170 , where the average is calculated from a sample of sufficiently large number of questions and the corresponding answers given in response . alice and bob share two suitably entangled qubits for which the bell - chsh inequality is violated . alice has access to one of the qubits and bob to the other . each player performs measurement on the qubit that he / she can access . the measurement rules too remain the same as those stated in the biased temporal chsh game with one significant difference : alice and bob are hereby supposed to carry out measurements on different qubits and not on the same qubit . but like in the previous case , alice always performs her measurement at an earlier time @xmath113 while bob at @xmath171 . an interesting feature of this game is that the correlations so generated between the measurement outcomes obtained by alice and bob are not merely nonlocal " , but also temporal " in nature since there happens to be a temporal separation " between the two measurements . + as one can easily verify for oneself , because these measurements involve commuting observables ( alice and bob perform local measurements on their qubits ) , the introduction of a temporal separation does not alter the inherently nonlocal correlations between the two qubits - in fact , the correlations behave as if both measurements are simultaneous ( see appendix [ iv ] ) . that is , the correlator between @xmath172 measured at @xmath173 and @xmath174 measured at @xmath175 is independent of the magnitude of @xmath176 . then @xmath177 , as before , regardless of the magnitude of the temporal separation . + however , because of the temporal separation , a novel prospect arises in the biased version of the game , where the winning criterion is identified with the inequality @xmath170 . once again , it is revealed that by incorporating intermediate measurements , the nonlocal correlations can be made to abide by the chsh inequality , provided the intervention is made on bob s qubit . in that case , the correlators take the following form ( refer to appendix [ iv ] for a derivation ) : @xmath178 and bob measures on qu2 at @xmath117 . charlie intervenes at an intermediate time @xmath179 on one of the two qubits . if he intervenes on alice s qubit , the correlations between alice and bob s measurements are unaltered . however , if the intervention is made on bob s qubit , the correlations are changed , albeit in a suitable manner so as to respect the bell - chsh inequality.,width=340,height=226 ] where @xmath180 refers to the correlation between the two outcomes when @xmath121 number of intervening measurements are carried out on bob s system , @xmath181 to the correlation between the outcomes obtained by alice and the outcomes due to the first intervening measurement performed on bob s system , and @xmath182 . note that @xmath183 is related to @xmath169 defined earlier , as @xmath184 and again , like in the temporal chsh scenario , @xmath185 is found to be bounded above by @xmath76 ( see appendix [ iv])for an arbitrary number of intervening measurements - the scope of gaining super - classical advantage on part of alice and bob is perennially lost making it a completely biased game . @xmath186\leqslant 2\end{aligned}\ ] ] one should understand that this feature would have been impossible had there been no temporal separation between the measurements performed by alice and bob . it is only through this fact that we can realize the true essence of integrating temporal correlations in our game . + as a closing remark , we would like to comment on the case in which interventions are made on alice s qubit and emphasize on the fact that in this scenario , the correlators between alice and bob s measurement outcomes remain invariant in form ( see appendix [ iv ] for a brief discussion on this ) . thus , it is absolutely necessary for charlie to effect an intervening measurement only on bob s qubit in between the measurements performed by alice and bob . he gains nothing by performing a measurement on alice s qubit . to put things into perspective , the essence of our endeavour has been to look beyond the much - discussed relevance and utility of nonlocal correlations for information processing and game - theoretic implementations . quantum theory is rife with spooky , counter - intuitive features , and nonlocality , or more precisely , the conflict with a local realistic world view , is just one of them . though admittedly , from the point of view of quantum foundations and its applications , the existence of nonlocality has usually captured the imagination of quantum physicists and philosophers more than anything else , there has been a plethora of literature centering this phenomenon . in particular , an overwhelming majority of information processing tasks and quantum games bank on the potential of nonlocal resources . given this state of affairs , our exploration was driven by the motivation of presenting the applicability of temporal correlations realizable in the quantum realm , albeit in a fairly elementary form , to the restricted context of bayesian games . + game theory ( @xcite,@xcite ) as a mathematical device appertains to the analytical treatment of the decision - making strategies available , in principle , to the active participants of a game . looking at correlation inequalities in quantum theory through the lens of games enables us to uncover innovative correlated strategies leading to interesting outcomes when the games are played using properly synchronized quantum devices . in the cooperative nonlocal bayesian game ( @xcite ) , the players share nonlocal resources , manipulation of which leads to correlated strategies that can not be replicated using local resources and can successfully outperform all available classical strategies that rely on local resources . consequently , these strategies make it possible for the participating parties to earn payoffs higher than the classical upper bound . in the bayesian lgi game ( [ s3 ] ) , appropriately correlated strategies are realized through temporal correlations that engender in a violation of the leggett - garg inequality ( [ e1 ] ) . these correlations stem from the property of measurement - invasiveness observed in quantum systems . rigorously stated , invasiveness of measurements is a categorically nonclassical feature ; while measurement on a classical system can be carried out by effecting an arbitrarily small perturbation on the ontological state of the system , in the quantum regime , a measurement will almost always lead to a disturbance in the ontic state , assuming , of course , that there exists an underlying reality to any quantum system . + in sections [ s4 ] and [ s5 ] we have highlighted an interesting disentangling " effect of intermediate projective measurements carried out on a two - level quantum system between two projective measurements separated in time . in the leggett - garg scenario ( [ s4 ] ) , where we consider the expression @xmath187 , we observe that under the insertion of an intervening projective measurement , any possibility of a violation of the lgi is done away with . this provides the group @xmath107 in our biased lgi game with a quantum " strategy to successfully thwart the cooperation of @xmath49 and @xmath50 . that this contrivance on part of @xmath107 is being branded quantum " is consistent with the fact that resources used in our game are unmistakably quantum and so are all the measurement schemes . consequently , this semblance of a reversion to classicality is a phenomenon noticed exclusively within the quantum paradigm without taking recourse to any classical devices . the groups @xmath49 and @xmath50 are obviously not interested in utilizing classical correlations ; it is the group @xmath107 that has a conflict of interest with the former and is therefore , intent upon curbing any super - classical advantage that @xmath49 and @xmath50 might get out of the quantum resources in their possession . + it is further demonstrated in section [ s5 ] that the temporal correlators are modified , as they should be , in such a fashion as to suppress any ( counterfactual ) violation of the temporal chsh inequality ( @xmath188 ) . what s more , in this scenario , the correlation inequality is satisfied for any number of intermediate measurements . this provides @xmath107 the possibility of winning the biased temporal chsh game unconditionally , as once an intervention is made , the shared system can not be manipulated any further to generate correlations capable of violating the relevant inequality . + finally , in section [ s6 ] , we introduce a temporal separation in the customary epr - bohm type setting in which , once again , intermediate intervention provides the novel possibility of reverting to correlations that can be simulated by local correlations . this is because any correlation that respects the spatial chsh inequality ( @xcite ) can be mimicked using appropriately chosen local correlations . this possibility is made available only in a hybrid spatio - temporal setting where there is a time gap between the measurements of alice and bob ( and therefore , between the times when the two cooperators are asked their respective questions ) . + we conclude with a few remarks in relation to our attempt to exemplify some of the applicative facets of temporal quantum correlations . we have chosen to deal explicitly with the restricted class of sharp projective measurements effected on two - level quantum systems , obviously for the sake of mathematical convenience . consequently , we leave open the possibility of exploring the corresponding correlations when the measurement schemes are different - as for example , when unsharp measurements , povm , etc . are employed , and/or when higher dimensional systems are utilized . one could also consider the potential applications of temporal correlations in information processing tasks , namely in the context of cryptography , quantum communication and communication complexity . in such scenarios , it could be worth investigating the relevance of an intervening measurement . an intervention could possibly be construed as an attack on effective quantumness or an act of impeding quantum advantage , as has been demonstrated in the games we formulated . however , if such interventions could enhance quantum advantage in certain game - theoretic settings or information processing tasks may also be probed . the author is grateful to dipankar home and som kanjilal for fruitful discussions and for their insightful suggestions in course of this work . 9 e. w. piotrowski and j. sladkowski , http://www.sciencedirect.com/science/article/pii/s0378437102008427[physica a 312 ( 2002 ) 208 - 216 ] . e. w. piotrowski and j. sladkowski , http://www.actaphys.uj.edu.pl/_old/vol32/pdf/v32p3873.pdf[acta phys . b32 ( 2001 ) 3873 - 3879 ] . chiu fan lee and n. johnson , http://arxiv.org/abs/quant-ph/0203043[quant-ph/0203043 ] . j. eisert , m. wilkens and m. lewenstein , http://journals.aps.org/prl/abstract/10.1103/physrevlett.83.3077[phys . 83 ( 1999 ) 3077 - 3088 ] . d. a. meyer , http://journals.aps.org/prl/abstract/10.1103/physrevlett.82.1052[phys . 82 ( 1999 ) 1052 - 1055 ] . j. eisert and m. wilkens , http://www.tandfonline.com/doi/abs/10.1080/09500340008232180#.vop_avl9600[j . 47 ( 2000 ) 2543 - 2556 ] . a. a. iqbal and d. abbott , http://www.sciencedirect.com/science/article/pii/s0375960110006572[phys . lett . a 374 , 3155 ( 2010 ) ] t. cheon and a. iqbal , http://journals.jps.jp/doi/abs/10.1143/jpsj.77.024801[j . japan 77 , 024801 ( 2008 ) ] . a. iqbal , t. cheon , and d. abbott , http://www.sciencedirect.com/science/article/pii/s0375960108014096[phys . lett . a 372 , 6564 ( 2008 ) ] . a. p. flitney et al . , http://www.sciencedirect.com/science/article/pii/s0375960108017246[phys . a 373 , 521 ( 2009 ) ] . j. briet , and t. vidick , 2013 , http://arxiv.org/abs/1108.5647[comm . 321 , 181 ] . c. d. hill , a. p. flitney , and n. c. menicucci , http://www.sciencedirect.com/science/article/pii/s0375960110008315[phys . lett . a 374 , 3619 ( 2010 ) ] . c. schmid et al . , http://iopscience.iop.org/article/10.1088/1367-2630/12/6/063031/meta;jsessionid=068f106b4bfb365435e5d24394ffa249.c4.iopscience.cld.iop.org[new j. phys . 12 , 063031 ( 2010 ) ] . j. m. chappell , a. iqbal and d. abbott , http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0036404[plos one 7 , e36404 ( 2012 ) ] . a. iqbal , s. weigert , http://iopscience.iop.org/article/10.1088/0305-4470/37/22/012/pdf[j . phys . a 37 ( 2004 ) a. iqbal , t. cheon , http://journals.aps.org/pre/abstract/10.1103/physreve.76.061122[phys . e 76 ( 2007 ) 061122 ] . a. iqbal , d. abbott , http://journals.jps.jp/doi/abs/10.1143/jpsj.78.014803[j . 78 ( 1 ) ( 2009 ) 014803 ] . n. brunner and n. linden , http://www.nature.com/ncomms/2013/130703/ncomms3057/full/ncomms3057.html[nat . commun . 4 , 2057 ( 2013 ) ] . n. d. mermin , http://homepages.physik.uni-muenchen.de/~vondelft/lehre/09qm/lec21-22-bellinequalities/mermin1990a.pdf[am . 58 , 731 ( 1990 ) ] . z. bar - jossef , t. s. jayram , and i. kerenidis , _ proceedings of the 36th annual acm symposium on theory of computing _ ( 2004 ) , pp . 128 - 137 . h. buhrman , o. regev , g. scarpa , and r. de wolf , _ proceedings of the 26th ieee annual conference on computational complexity _ ( 2011 ) , pp . 157 - 166 . a. pappa , n. kumar , t. lawson , m. santha , s. zhang , e. diamanti , and i. kerenidis , http://journals.aps.org/prl/abstract/10.1103/physrevlett.114.020401[phys . 114 , 020401 ( 2015 ) ] . s zhang , proceedings of the 3rd innovations in theoretical computer science conference , ( itcs ) : 39 - 59 ( 2012 ) . j. von neumann and o. morgenstern , _ the theory of games and economic behavior _ , princeton university press , princeton ( 1944 ) . m. j. osborne , _ an introduction to game theory _ , oxford university press , usa ( 2003 ) . j. c. harsanyi , management science 14 ( 3 ) : 159 - 183 ( part i ) : 14 ( 5 ) 320 - 334 ( part ii ) , 14 ( 7 ) : 486 - 502 ( part iii ) , ( 1967 ) . t. cheon and i. tsutsui , http://arxiv.org/abs/quant-ph/0503233[phys . lett . a348 ( 2005 ) 147 - 152 ] . bennett , c. h. , g. brassard , c. cr?peau , r. jozsa , a. peres , and w. k. wootters , http://journals.aps.org/prl/abstract/10.1103/physrevlett.70.1895[phys . rev . lett . 70 , 1895 ( 1993 ) ] . popescu , s. , http://journals.aps.org/prl/abstract/10.1103/physrevlett.72.797[phys . 72 , 797 ( 1994 ) ] . cavalcanti , d. , a. acin , n. brunner , and t. vertesi , 2013a , http://journals.aps.org/pra/abstract/10.1103/physreva.87.042104[phys . a 87 , 042104 ] . horodecki , r. , m. horodecki , and p. horodecki , http://www.sciencedirect.com/science/article/pii/0375960196006391[phys . a 222 , 21 , ( 1996 ) ] . ekert , a. , http://journals.aps.org/prl/abstract/10.1103/physrevlett.67.661[phys . 67 , 661 ( 1991 ) ] . acin , a. , s. massar , and s. pironio , http://arxiv.org/abs/quant-ph/0605246[2006c , new j. phys 8 , 126 ] . bell , physics 1 , 195 ( 1964 ) . clauser , j. f. , m. a. horne , a. shimony , and r. a. holt , http://journals.aps.org/prl/abstract/10.1103/physrevlett.23.880[phys . 23 , 880 , ( 1969 ) ] . bell , j. s. , _ speakable and unspeakable in quantum mechanics _ cambridge university press ( 2004 ) . freedman , s. , and j. clauser , phys . 28 , 938 ( 1972 ) . aspect , a. , p. grangier , and g. roger , http://journals.aps.org/prl/abstract/10.1103/physrevlett.49.91[1982b , phys . 49 , 91 ] . a. j. leggett and a. garg , http://journals.aps.org/prl/abstract/10.1103/physrevlett.54.857[phys . 54 , 857 ( 1985 ) ] . a. j. leggett , http://iopscience.iop.org/article/10.1088/0953-8984/14/15/201/meta{}j . matter 14 , r415 ( 2002)[j . ; condens . matter 14 , r415 ( 2002 ) ] . paz and g. mahler , http://journals.aps.org/prl/abstract/10.1103/physrevlett.71.3235[phys . 71 , 3235 ( 1993 ) ] . c. d. tesche , http://journals.aps.org/prl/abstract/10.1103/physrevlett.64.2358[phys . 64 , 2358 ( 1990 ) ] . a. peres , http://journals.aps.org/prl/abstract/10.1103/physrevlett.61.2019[phys . 61 , 2019 ( 1988 ) ] . a. j. leggett and a. garg , link : phys . 63 , 2159 ( 1989)[phys 63 , 2159 ( 1989 ) ] . a. palacios - laloy , f. mallet , f. nguyen , p. bertet , d. vion , d. esteve , and a. n. korotkov , http://www.nature.com/nphys/journal/v6/n6/fig_tab/nphys1641_f1.html[nat . phys . 6 , 442 ( 2010 ) ] . n. d. mermin , http://scitation.aip.org/content/aip/magazine/physicstoday/article/38/4/10.1063/1.880968[physics today 38 , 38 ( 1985 ) ] . j. e. mooij , http://www.nature.com/nphys/journal/v6/n6/full/nphys1698.html[nat . phys . 6 , 401 ( 2010 ) ] . j. kofler and c. brukner , http://journals.aps.org/prl/abstract/10.1103/physrevlett.99.180403[phys . 99 , 180403 ( 2007 ) ] . j. kofler and c. brukner , http://journals.aps.org/prl/abstract/10.1103/physrevlett.101.090403[phys . lett . 101 , 090403 ( 2008 ) ] . c. brukner , s. taylor , s. cheung , and v. vedral , http://arxiv.org/abs/quant-ph/0402127[quantph/042127 ( 2004 ) ] . s. marcovitch and b. reznik , http://arxiv.org/abs/1107.2186[arxiv:1107.2186 ( 2011 ) ] . g. c. knee , s. simmons , e. m. gauger , j. j. morton , h. riemann , n. v. abrosimov , p. becker , h .- j . pohl , k. m. itoh , m. l. thewalt , g. a. d. briggs , and s. c. benjamin , http://www.nature.com/ncomms/journal/v3/n1/full/ncomms1614.html[nat . commun . 3 , 606 ( 2012 ) ] . j. dressel , c. j. broadbent , j. c. howell , and a. n. jordan , http://journals.aps.org/prl/abstract/10.1103/physrevlett.106.040402[phys . 106 , 040402 ( 2011 ) ] . r. w. spekkens , http://journals.aps.org/pra/abstract/10.1103/physreva.71.052108[phys . rev . a 71 , 052108 ( 2005 ) ] . t. rudolph , http://arxiv.org/abs/quant-ph/0608120[quant-ph/0608120 ( 2006 ) ] . n. harrigan and t. rudolph , http://arxiv.org/abs/0709.4266[arxiv:0709.4266 ( 2007 ) ] . n. harrigan and r. spekkens , http://arxiv.org/abs/0706.2661[found phys . 40 , 125 ( 2010 ) ] . o. j. e. maroney , http://arxiv.org/abs/1207.3114[arxiv:1207.3114 ( 2012 ) ] . v. athalye , s. s. roy , and t. s. mahesh , http://journals.aps.org/prl/abstract/10.1103/physrevlett.107.130402[phys . 107 , 130402 ( 2011 ) ] . t. fritz , http://iopscience.iop.org/article/10.1088/1367-2630/12/8/083055#metrics[new j. phys . 12 , 083055 ( 2010 ) ] . clive emary , neill lambert , franco nori , http://arxiv.org/abs/1304.5133[arxiv:1304.5133 ] . costantino budroni and clive emary , http://journals.aps.org/prl/abstract/10.1103/physrevlett.113.050401[phys 113 , 050401 ( 2014 ) ] . f. de zela , http://journals.aps.org/pra/abstract/10.1103/physreva.76.042119[phys . rev . a 76 , 042119 ( 2007 ) ] . b. tsirelson , lett . phys . 4 , 93 ( 1980 ) . [ [ i ] ] i ~ a crucial result was used in the context of the biased bayesian lgi game ( [ s4 ] ) that when a single intervening projective measurement is carried out on the system shared by @xmath58 and @xmath59 , the resulting correlations between the outcomes obtained by @xmath49 and @xmath50 are rendered incapable of violating the leggett - garg inequality ( [ e1 ] ) . in the context of the biased temporal chsh game ( [ s5 ] ) , this feature was claimed to be holding true for any number of interventions . before we can prove this result , we need to work out an expression for the temporal correlator between the outcomes obtained by two parties . we begin by reiterating the following assumptions : + ( i ) a total of @xmath121 intermediate spin measurements are executed on the given system at times @xmath122 along the directions @xmath123 respectively , where @xmath124 . + ( ii ) @xmath58 and @xmath59 perform their spin measurements along @xmath189 and @xmath190 respectively getting outcomes @xmath191 and @xmath192 respectively . ( we have tweaked the notations a bit for our convenience . ) we denote the intermediate measurement outcomes , in chronological order , by @xmath127 . + the temporal correlator between the outcomes obtained by @xmath58 and @xmath59 are calculated as : here , @xmath194 , @xmath195 ( the initial state of the system ) and @xmath196 is the two - dimensional identity operator . to begin simplifying this problem , we first take into account the last two trace - terms in @xmath197 is simply the projector - decomposition of the spin observable along @xmath190 , i.e. @xmath199 , we have : @xmath200 here we have made use of the fact that @xmath201 . next we carry out the summation over @xmath202 so that another partially simplified expression is obtained : @xmath203 continuing in this fashion , we find , at a certain stage : @xmath204 this is identical to the expression in [ e14 ] with @xmath205 , @xmath206 . recall that we had defined the quantity @xmath169 as being equal to @xmath207 and so , @xmath208 . if only one intervention is made , it is easily seen that the correlator - expression is given by : @xmath209 where the direction of intermediate spin measurement is along @xmath210 . this particular expression ( for the case when only one intermediate measurement is performed on the system ) was stated by brukner et al ( @xcite ) . here we have derived a generalized simplified expression for the temporal correlator when an arbitrary number of projective measurements are carried out between the the ones effected by @xmath58 and @xmath59 . + the following section justifies why the correlators can never violate the leggett - garg inequality , when only one intervention is made on the qubit . as we have just derived , the temporal correlator under consideration is suitably modified when an intermediate projective spin measurement is performed on the system , so that we have the following form : let us now investigate the resulting lgi expression @xmath211 . taking @xmath212 , @xmath213 , we have : @xmath214\\ & = \cos\phi_2(\cos\phi_1+\cos\phi_3)-\cos\phi_1\cos\phi_3\end{aligned}\ ] ] then , clearly @xmath215 , where @xmath216 , and @xmath217 . consequently , if we can show that @xmath218 is never greater than @xmath75 ( which is the classical upper bound for the leggett - garg inequality ) , then basically we are done . + * case 1 : @xmath219 . * so we probe the upper bound of @xmath220 . note that the expression @xmath221 is monotonically increasing in @xmath222 , since @xmath223 . @xmath224 in this section we show that the temporal correlators between the observed outcomes of @xmath49 and @xmath50 as a consequence of an arbitrary number of intermediate measurements on the shared systems are found to abide by the temporal chsh inequality . but before that , we briefly recapitulate the temporal version ( @xcite,@xcite,@xcite ) of the chsh inequality . + the temporal chsh inequality is usually derived from two premises : ( a ) _ realism ( r ) _ @xmath158 the results of measurement performed on a given physical system are determined by well - defined pre - existing properties pertaining to the system that exist independent of observation , and ( b ) _ non - invasive measureability ( nim ) _ @xmath158 an ideal measurement does not perturb the ontic state of the system under consideration . stated somewhat differently , it is possible , in principle , to extract all relevant information about the state of the system with arbitrarily small disturbance on its subsequent dynamics . strictly speaking , a third assumption also goes into the derivation : _ induction_. essentially this means that the measurement outcome at any point in time does not depend on what is measured on the system at some other instant of time . + consider a quantum system described by an underlying hilbert space @xmath230 and dynamics governed by the hamiltonian @xmath230 . we have two sets of dichotomic observables @xmath231 and @xmath232 , each with eigenvalues @xmath1 . alice has the choice of measuring one out of @xmath233 at time @xmath113 and bob can measure any one out of @xmath234 at time @xmath114 . then the temporal chsh expression @xmath235 is given by : @xmath236 where @xmath237 ( @xcite ) is the temporal correlator of the outcomes @xmath238 and @xmath239 obtained on measuring the observables @xmath240 and @xmath241 at times @xmath113 and @xmath117 respectively . now , starting from the assumptions of _ realism _ and _ noninvasive measureability _ , we shall show that @xmath242 . to that end , we define ontic states @xmath45 ( @xcite-@xcite,@xcite ) with @xmath243 being the probability ( density ) that the system is prepared in the ontic state @xmath45 . so @xmath244 . let @xmath245 be the outcome function that signifies the probability of obtaining @xmath246 on measuring the system in the ontic state @xmath45 at time @xmath247 . similarly , let @xmath248 be the outcome function that gives the probability of obtaining @xmath249 on measuring the system in @xmath45 at @xmath247 . also let @xmath250 denote the probability that on measuring @xmath0 at time @xmath247 the ontic state @xmath45 is changed to @xmath251 . consequently , the joint probability @xmath252 is obtained as : @xmath253 under the assumption of _ nim _ , @xmath254 . so the joint probability becomes : this completes the proof of the temporal chsh inequality in an ontic model framework . that is , any theory with an underlying ontological framework in which measurements do not disturb the ontic state of the system is bound to satisfy the inequality @xmath260 . + we now shift our focus to the case when intervening projective measurements are executed on the system under consideration , in which case the correlators are altered : with @xmath263 , @xmath264 , @xmath265 , @xmath266 , where we have taken the measurements made by alice and bob to be along @xmath267 and @xmath268 respectively . as can be easily seen from the above expression , @xmath269 and since , @xmath208 , it follows that @xmath270 . in other words , intervening projective measurements ensure that the chsh inequality is satisfied and invasiveness of the measurement performed by alice apparently " disappears . this property remains valid for arbitrarily large @xmath121 , which rules out the achievability of super - classical correlations once an intervening measurement is performed - further intermediate measurements can not be implemented that could possibly retrieve the lost correlations . finally , we focus on the game ( [ s6 ] ) that we formulated in a spatio - temporal setting . an apposite modification was invoked by conveniently incorporating a temporal separation between the measurements performed by alice and bob . in the conventional spatial chsh setting , alice chooses to perform a measurement on the qubit in her possession along one of the two directions @xmath271 and @xmath272 , while bob measures along one of the two directions @xmath273 and @xmath274 on his qubit . both measurements are performed ( usually ) at the same time . in the new version that we have considered , we have assumed alice to be performing her measurement at an earlier time @xmath113 and bob at a later time @xmath117 . we show here that this temporal separation does not alter the correlation between the outcomes obtained by alice and bob . denoting @xmath275 as the corresponding correlator and labelling the respective outcomes as @xmath276 and @xmath277 , we observe that : @xmath278 here @xmath279 $ ] , @xmath280 $ ] , and @xmath281 is the state of the biqubit system to which the initial state @xmath282 collapses when alice obtains the outcome @xmath276 upon measuring along @xmath125 . @xmath283 substituting in the expression for the correlator , we obtain : @xmath284 what happens if we insert some intermediate measurements on alice s subsystem ? we show , for simplicity , the case when only a single intermediate measurement is performed along @xmath130 . the corresponding outcomes are labelled as @xmath173 and the projectors as @xmath285 $ ] . a straightforward calculation then yields : where @xmath287 is the biqubit state formed when @xmath173 is obtained as an outcome for the intermediate measurement . + that is , in this case too , the correlators do not suffer any change . it is also easy to convince oneself that the form remains invariant even when an arbitrary number of intermediate measurements are performed on alice s subsystem . however , the correlators are substantially modified when the intermediate measurements are performed on bob s subsystem . once again , we demonstrate the case for a single intermediate measurement . then the chsh expression looks like : @xmath289 from which @xmath270 follows . + when multiple intervening measurements are performed , we have : @xmath290 where @xmath291 and bell s inequality is still satisfied ( this follows , again , from the consideration that @xmath292 ) . + as has already been emphasized earlier , it is necessary to perform the intermediate measurement(s ) on the second qubit ( in the possession of bob ) , if one wishes to suppress any kind of super - classical correlation . this feature can be nicely construed in the following manner . once alice has performed a measurement on her qubit , the resulting bipartite state becomes separable ( whether or not it was so to begin with ) , and any further measurement performed on alice s qubit does not have any effect whatsoever on the state of bob s qubit . consequently the original correlations ( local or nonlocal ) between the outcomes of alice ( already obtained ) and the counterfactual ( yet to be obtained ) outcomes of bob s measurement remain unaffected by any intervention on alice s qubit following the completion of her own measurement . on the contrary , if any intermediate measurement is performed on bob s qubit , it does tamper with the state of his qubit following the collapse effected by alice s measurement on her qubit at an earlier time . as a result , the correlations between alice s and bob s outcomes are subjected to a modification , but interestingly , the ensuing correlations are found to comply with the chsh inequality .

at redshifts , quasar absorption - line constraints on space - time variations in the fine - structure constant , , rely on the comparison of mgii and feii transition wavelengths . one potentially important uncertainty is the relative abundance of mg isotopes in the absorbers which , if different from solar , can cause spurious shifts in the measured wavelengths and , therefore , . here we explore chemical evolution models with enhanced populations of intermediate - mass ( i m ) stars which , in their asymptotic giant branch ( agb ) phase , are thought to be the dominant factories for heavy mg isotopes at the low metallicities typical of quasar absorption systems . by design , these models partially explain recent keck / hires evidence for a smaller in absorption clouds than on earth . however , such models also over - produce n , violating observed abundance trends in high- damped lyman- systems ( dlas ) . our results do not support the recent claim of ashenfelter , mathews & olive ( 2004b ) that similar models of im - enhanced initial mass functions ( imfs ) may simultaneously explain the hires varying- data and dla n abundances . we explore the effect of the im - enhanced model on si , al and p abundances , finding it to be much - less pronounced than for n. we also show that thec/c ratio , as measured in absorption systems , could constitute a future diagnostic of non - standard models of the high - redshift imf . [ firstpage ] quasars : absorption lines stars : agb nucleosynthesis

astro-ph0501168

in the past few years , evidence has emerged that the fine - structure constant , @xmath5 , may have been smaller in high - redshift quasar ( qso ) absorption systems than the value measured today on earth ( e.g. * ? ? ? * ; * ? ? ? * hereafter ) . a possible explanation for the lower - redshift half of this result is that the abundances of the heavy mg isotopes ( @xmath6 mg and @xmath7 mg ) in the absorbers are much higher , relative to that of @xmath8 mg , than solar values . recently , @xcite proposed and expanded upon ( * ? ? ? * hereafter ) a chemical evolution model with an initial mass function ( imf ) strongly enhanced at intermediate masses ( ims ) , whereby highly super - solar abundances of heavy mg isotopes are produced via asymptotic giant branch ( agb ) stars . in this paper we explore the side - effects of this model with a view to identifying possible observational signatures other than increased heavy mg isotope abundances . the paper is organised as follows . the remainder of this section summarizes the qso absorption - line evidence for and against variations in @xmath1 and describes the sensitivity of that evidence to variations in isotopic abundances . section [ nucleosynthesis ] briefly describes the nucleosynthesis of mg isotopes in stars of different masses . in section [ model ] we detail our chemical evolution models and compare them with those used by . section [ results ] presents our main results for the predicted evolution of the total and isotopic abundances of various elements typically observed in qso spectra . in each case we compare these with available data from qso absorption - line and local stellar and interstellar medium ( ism ) studies and discuss their dominant uncertainties . section [ discussion ] assesses the overall validity of the im - enhanced models in light of our new results and presents other arguments for and against such models . section [ conclusions ] gives our main conclusions . the universality and constancy of the laws of nature rely on the space - time invariance of fundamental constants , such as @xmath1 . therefore , since @xcite and @xcite first suggested the time - variation of the newton gravitational constant , a great diversity of theoretical and experimental exploration of possible space - time variations in fundamental constants has been pursued ( e.g. see review in * ? ? ? high resolution spectroscopy of absorption systems lying along the lines - of - sight to background qsos has provided particularly interesting constraints on variations in @xmath1 over large spatial and temporal baselines . early work focused on the alkali - doublet ( ad ) method : since the relative wavelength separation between the two transitions of an ad is proportional to @xmath10 ( e.g. * ? ? ? * ) , comparison between ad separations seen in absorption systems with those measured in the laboratory provides a simple probe of @xmath1 variation . several authors ( e.g. @xcite ; @xcite ) applied the ad method to doublets of several different ionic species ( e.g. civ , siii , siiv , mgii and aliii ) . the strongest current ad constraints come from analysis of many siiv absorption systems in @xmath11 spectra : @xmath12 ( * ? ? ? * systems ; @xmath13 ) and @xmath14 ( * ? ? ? * systems ; @xmath15 ) , for @xmath16 and @xmath17 the values of @xmath1 in the absorption system(s ) and in the laboratory respectively . ] considerable recent interest has focused on the many - multiplet ( mm ) method introduced by @xcite and @xcite . the mm method is a generalization of the ad method , constraining changes in @xmath1 by utilizing many observed transitions from different multiplets and different ions associated with each qso absorption system . it holds many important advantages over the ad method , including an effective order - of - magnitude precision gain stemming from the large differences in sensitivity of light ( e.g. mg , si , al ) and heavy ( e.g. fe , zn , cr ) ions to varying @xmath1 . at low redshift ( @xmath18 ) , mg lines , whose red transition wavelengths ( @xmath19 ) are relatively insensitive to changes in @xmath1 , act as anchors against which the larger expected shifts in the bluer ( @xmath20 ) fe transition wavelengths can be measured . at higher , transitions from si and al provide anchor lines distributed in wavelength space amongst a variety of cr , fe , ni and zn transitions which shift by large amounts in both positive and negative directions as @xmath1 varies . this diversity at high- ensures greater reliability in the face of simple systematic effects compared with the low- mg / fe systems . the mm method , applied to keck / hires qso absorption spectra , has yielded very surprising results , with the first tentative evidence for a varying @xmath1 by @xcite becoming stronger with successively larger samples ( @xcite ; @xcite ) . the most recent keck / hires constraint comes from 143 absorption systems over the range @xmath21 : @xmath22 . this result is quite internally robust : it comprises three different observational samples and approximately equal low- and high- subsamples , all of which give consistent results . stubbornly , it has also proven resistant to a range of potential instrumental and astrophysical systematic effects @xcite . intriguingly , @xcite ( see also * ? ? ? * ) have analysed 23 mg / fe absorption systems in higher signal - to - noise ratio ( s / n ) spectra _ from a different telescope and spectrograph _ , the vlt / uves , claiming a precise , null result over the range @xmath23 : @xmath24 . @xcite and @xcite also find null results in individual uves absorbers . the discrepancy between the vlt / uves and keck / hires results is yet to be resolved . however , it is important to note that low - order distortions of the keck / hires wavelength scale , as might be expected from simple instrumental systematic errors , produce opposite effects on the low- and high- samples and so can not fully explain the hires uves difference @xcite . if the hires result is incorrect , the nature of the contributory systematic errors must be subtle and somewhat conspiratorial . @xcite first identified the potential systematic error introduced into the mm method if the relative isotopic abundances of crucial anchor elements like mg and si underwent strong cosmological evolution . this is because the absorption lines of the different isotopes are spaced widely enough ( @xmath25 ) in these light ions to affect the measured line centroids ( fig . [ alpha_iso_fig ] , left panel ) . calculated from their hires absorption line data as a function of the assumed relative isotopic abundances of mg and si . the results are presented in the right panel of fig . [ alpha_iso_fig ] . note the relative insensitivity of in the high- systems to the si heavy isotope ratio , . as noted above , this is expected because of the greater diversity of transitions and line - shifts available in the high- regime . it is therefore important to note that evolution in the si isotopic abundance can not explain the high- hires results . until recently , no information about the isotopic shifts in transitions from heavier ions used at high- ( e.g. crii , feii , niii , znii ) were available . however , due to their higher masses , the shifts are expected to be smaller with respect to the sensitivity of the lines to varying @xmath1 . indeed , the recent isotopic shift calculations of @xcite confirm this . however , fig . [ alpha_iso_fig ] emphasises the strong dependence of in the low- systems on the mg heavy isotope ratio , . the importance of this potential systematic effect therefore depends entirely on the evolution of mg isotopic ratios in qso absorption systems . unfortunately , no direct measurement of in qso absorbers is currently feasible due to the small separation of the isotopic absorption lines . fortunately however , mg is one of the few elements for which stellar isotopic abundances can be measured through molecular absorption lines , in this case transitions of mgh . @xcite and @xcite have shown that observed stellar values of generally decrease with decreasing [ fe / h ] , as predicted in the galactic chemical evolution models of @xcite . the ( normal imf ) model we present in section [ model ] also predicts such a decrease ( fig . [ iso_grid_fig ] ) . this has been used to argue that in the low - metallicity environments of mg / fe absorbers ( typically , @xmath26}\!\sim\!-1.0 $ ] ) , one should expect sub - solar values of and so , if anything , one expects the low - values of to be _ too positive _ @xcite . contrary to this trend , @xcite have found very high values of for some giant stars in the globular cluster ngc 6752 , which has metallicity @xmath27}\!\sim\!-1.6 $ ] . one star had @xmath28 ( cf . @xmath29_\odot\!=\!0.27 $ ] ) . since i m agb stars are thought to produce significant quantities of @xmath6 mg and @xmath7 mg ( see section [ nucleosynthesis ] ) , @xcite proposed that low - metallicity , i m agb stars may have polluted this globular cluster . this prompted @xcite to propose a chemical evolution model with a strongly enhanced population of i m stars as a possible explanation for the low- hires varying-@xmath1 results . in the following sections we describe in detail the nucleosynthesis of mg isotopes and construct a chemical evolution model , similar to that of , to investigate the various effects of an im - enhanced imf . for the elements c , mg , si , ti , cr , fe , ni , zn , which are used in varying @xmath30 analyses . isotopic abundances are shown on a logarithmic scale relative to the dominant isotope for that element . different curves illustrate the predicted behaviour of the various isotopes from a standard solar radius chemical evolution model . symbols show the corresponding solar values . in general , the minor isotopes accrue more slowly in the ism with respect to the dominant isotope for each element . a notable exception is ni , for which @xmath31ni/@xmath32ni declines over time . the departure of ni from the general rule of thumb may reflect uncertainties in its nucleosynthesis , including an overproduction from sne ia , that have been pointed out in the literature ( e.g. timmes et al . 1995 ; iwamoto et al . massive stars culminating in type ii sne are responsible for most of the mg isotopes in the present - day galaxy . however , there is evidence that intermediate - mass stars may dominate the production of the neutron - rich mg isotopes in the metal - poor regime . according to standard models of stellar nucleosynthesis , the yield of the heavy mg isotopes , , scales with the initial stellar metallicity . conversely , the generation of from sne ii operates fairly independently of initial metallicity . since massive stars alone are insufficient to account for the higher than expected values of / detected in metal - poor stars ( gay & lambert 2000 ; yong et al . 2003b ) , it has been suggested that there is a supplemental source of the neutron - rich mg isotopes . karakas & lattanzio ( 2003 ) have shown that and production is substantial in metal - poor intermediate - mass stars ( imss ) . at low metallicities , asymptotic giant branch stars are believed to generate and from @xmath1-capture onto @xmath33ne triggered by he - shell thermal pulsing . temperatures at the base of the convective envelope in 46m@xmath34 stars can be high enough to burn via hot bottom burning ( hbb ) as well as synthesise large amounts of and . based on the recent nucleosynthetic calculations from karakas & lattanzio ( 2003 ) , agb stars have been shown to produce sufficient quantities of to resolve the discrepancy between local stellar observations and previous model predictions ( fenner et al . for the present study , we have incorporated the same grids of low- and intermediate - mass stellar yields that successfully reproduced the solar neighbourhood mg isotopic evolution . the distribution of elements and isotopes as a function of time and radius was simulated for a milky way - like disk galaxy by numerically solving the classic set of equations ( as described by tinsley 1980 ) governing gas infall , star formation , stellar evolution and nucleosynthesis . in particular , we present predictions for ( i ) the solar annulus ( @xmath35 8.5 kpc ) because it is the solar neighbourhood for which we have the most comprehensive set of empirical constraints , and ( ii ) the outer disk ( @xmath35 16.5 kpc ) because it is possible that many qso absorption systems correspond to the outer regions of spiral disks ( e.g. dessauges - zavadsky et al . 2004 ) . to summarise the details of the model : we define @xmath36 as the mass surface density of species @xmath37 at time @xmath38 and radius @xmath39 , and assume that its rate of change of is given by : @xmath40 where the three terms on the right - hand side correspond to the stellar ejecta , gas infall , and star formation , respectively . the star formation rate , @xmath41 , varies with the square of the gas surface density in our models , consistent with the empirical schmidt ( 1959 ) law . @xmath42 denotes the stellar yield of @xmath37 ( in mass units ) from a star of mass @xmath43 and metallicity @xmath44 , @xmath45 is the initial mass function , and @xmath46 is the mass fraction of element @xmath37 . by definition , the sum of @xmath46 over all @xmath37 is unity . the total surface mass density is identical to the integral over the infall rate . the lower and upper stellar mass limits , @xmath47 and @xmath48 , are 0.08 m@xmath34 and 60 m@xmath34 , respectively , while @xmath49 is the main - sequence lifetime of a star of mass @xmath43 . we split the first term into three equations that deal separately with low- and intermediate - mass stars , type ia supernova ( sn ) progenitors , and massive stars . further details of the numerical code employed in this study can be found in fenner & gibson ( 2003 ) , fenner et al . ( 2003 ) and fenner , prochaska & gibson ( 2004 ) . we assumed that the milky way - like disk galaxy formed during two main gas accretion episodes . the first occurs on a rapid timescale ( @xmath50 0.5 gyr ) and is associated with the formation of the halo and thick disk , while the second episode occurs on a longer timescale and fuels the formation of stars in the disk . for simplicity , we have assumed no prior metal - enrichment of the gas infalling onto the disk , although there is some evidence from observations of high - velocity clouds that gas falling into the galaxy may contain traces of heavy elements ( e.g. wakker et al . 1999 ; gibson et al . 2001 ; sembach et al . exponentially decaying infall rates have been adopted , such that the evolution of total surface mass @xmath51 density is given by @xmath52 where the infall rate coefficients @xmath53 and @xmath54 are chosen in order to reproduce the present - day surface mass density of the halo / thick disk and thin disk components , which we take to be 10 and 45m@xmath34pc@xmath55 , respectively . the adopted timescales for the infall phases are @xmath56=0.1gyr and @xmath57=9.0gyr at the solar radius @xmath58=8.5kpc . disk formation starts after an initial delay @xmath59gyr . the ` inside - out ' functional form for @xmath60 ( romano et al . 2000 ) is adopted , whereby the timescale for disk gas accretion increases linearly with radius . the milky way age is taken to be 13gyr . the shape of the stellar initial mass function ( imf ) influences the quantity of galactic material locked up in stars of different masses , which in turn determines the rate at which different elements are released into the ism . the models presented in section [ results ] compare the kroupa , tout & gilmore ( 1993 ) three - component imf with one enhanced in agb stars at low metallicities . the shape of the im - enhanced imf at time @xmath61 is illustrated in fig . [ imf_fig ] . the height and width of the agb bump was chosen to approximate that in . it is centred on 5 m@xmath34 with a narrow mass range , although not as narrow as the model 1 , for which the extra agb population consists almost entirely of 5@xmath621 m@xmath34 stars ( compare our fig . [ imf_fig ] with their figure 3 ) . their preferred models ( i.e. models 2 and 3 ) adopt a wider im - bump and we also favour a broader peak , on the grounds that it is more physically realistic . in our im - enhanced models , the amplitude of the im - bump decays exponentially with increasing gas metallicity . this differs from , who adopted time - dependent decay . imposing a metallicity- rather than time - dependent im - enhancement was designed to reflect the different physical properties of metal - poor gas clouds . the decreased cooling rate and magnetic field strength in metal - poor material is expected to influence the mass distribution of newly formed stars . for consistency with model 1 , our scale - factor for the exponentially decaying im - bump was chosen to correspond to the predicted metallicity at the solar radius at time = 0.2 gyr . this corresponds to @xmath63}\!\sim\!-1.3 $ ] , or roughly 1/20th solar metallicity . the transformation from a time- to a metallicity - dependent ims burst makes it reasonable to apply the same imf formalism across all galactic radii . because the outer disk has a more protracted star formation rate and builds up metals more slowly than in the solar radius , the effects of an ims burst would be suppressed by adopting a fixed timescale . instead , conversion to metallicity - dependence prolongs the temporal duration of the ims burst at outer radii and amplifies the chemical signature of an im - enhanced imf . the imf is composed of an agb - biased bump superimposed on a standard kroupa , tout & gilmore ( 1993 ) 3-component power - law . the height and width of the agb bump was chosen to emulate that in . the amplitude of the bump decays exponentially with increasing gas metallicity.,width=302 ] _ low and intermediate mass stars ( lims ) : _ for stars less massive than 8 m@xmath34 , we incorporated the yields from the stellar evolution and nucleosynthesis code described in karakas & lattanzio ( 2003 ) , supplemented with yields for additional isotopes from hydrogen to sulfur and for the iron - peak , as well as unpublished yields for the metallicity @xmath64=0.0001 . the wide range of heavy elements and isotopes incorporated in these agb nucleosynthesis models allows us to self - consistently predict the contribution from low- and intermediate - mass stars ( lims ) to the elemental and isotopic abundance of many species of interest in dla and varying @xmath1 studies . these include c , n , mg , al , si , ni and fe . we also ran identical chemical evolution models using two alternative sets of lims yields : those of ( i ) van den hoek & groenewegen ( 1997 ) and ( ii ) marigo ( 2001 ) ( using mixing length parameter @xmath65 for consistency with ) . these two alternative inputs do not predict yields for elements beyond o , but they provide an indication of the uncertainty in the cno yields , as will be discussed in section [ uncertainties ] . _ type ia supernovae ( sne ia ) : _ we adopted a recalculation of the thielemann , nomoto , & yokoi ( 1986 ) w7 model by iwamoto et al . ( 1999 ) to estimate the yields from sne ia . it was assumed that 4% of binary systems involving intermediate and low mass stars result in sne ia , since this fraction provides a good fit to the solar neighbourhood ( e.g. alib ' es , labay , & canal 2001 ; fenner & gibson 2003 ) . _ massive stars : _ for stars more massive than 810 m@xmath34 that end their lives in violent supernova explosions , we implemented the yields from woosley & weaver ( 1995 ) . their yields span metallicities from zero to solar and cover the mass range 1140m@xmath34 . since the upper mass limit of our stellar imf extends beyond this mass range , we extrapolated the 35 and 40 m@xmath34 yields for @xmath66 40 m@xmath34 . we took the lower energy ` a ' models for stars @xmath67m@xmath34 and the higher energy ` b ' models for heavier stars . taking note of the suggestion by timmes , woosley , & weaver ( 1995 ) that the ww95 mass - cuts may have penetrated too deeply within the iron core , we have uniformly halved the iron yields from these models . in the uncertain mass range 812 m@xmath34 , we make the conservative assumption that these stars do not synthesise new heavy elements , but expel material with the same heavy element abundance pattern as the gas from which they were born . _ stellar lifetimes : _ we adopt metallicity - dependent main - sequence lifetimes calculated by schaller et al . although stars lose material over the course of their evolution via stellar winds and planetary nebulae , this model assumes that all the mass loss takes place at the end of the main - sequence phase . our predictions are not significantly affected by this simplification . we now list some key differences between our model and that of ashenfelter et al . ( 2004b ) and describe the impact on the results : * * agb yields : * we have adopted the low- and intermediate - mass stellar yields from karakas & lattanzio ( 2003 ) , with additional yields for isotopes up to s as well as the fe - peak , along with unpublished yields for the metallicity @xmath64=0.0001 . thus , the agb yields of species such as n and the mg isotopes are internally self - consistent . added the mg isotopic yields from karakas & lattanzio ( 2003 ) onto the marigo ( 2001 ) c , n and o yields . although this means that their n yields were not drawn from the same agb models as the mg yields , the predicted evolution of n is very similar in either case , as discussed in section [ cn_uncertainties ] . * * type ia sne yields : * employed a metallicity - dependent type ia supernovae rate from kobayashi , tsujimoto & nomoto ( 2000 ) that prohibits the formation of sne ia below a minimum [ fe / h ] threshold of @xmath681.1 . this metallicity - dependence was postulated mostly on theoretical grounds . observationally however , @xmath1/fe ratios in dwarf galaxies ( e.g. shetrone et al . 2001 ) and s / zn ratios in dlas ( pettini , ellison , steidel & bowen 1999 ) provide evidence for sn ia activity below [ fe / h ] @xmath69 @xmath681 . furthermore , the trend of [ o / fe ] with [ fe / h ] seen in local stars can be satisfactorily explained without imposing a sn ia metallicity threshold ( e.g. alib ' es et al . thus , the present study does not impose a metallicity threshold and the sn ia rate is calculated following the method of greggio & renzini ( 1983 ) and matteucci & greggio ( 1986 ) . while a [ fe / h ] @xmath69 @xmath681 sn ia threshold would have only a minor effect on the solar radius results , the outer disk remains metal - poor for at least several gyr and would be more sensitive to the precise sn ia prescription , as will be discussed in section [ results ] . * * standard imf : * in recalculating the model of timmes et al . ( 1995 ) , adopted the salpeter ( 1955 ) single power - law imf . we adopt a three - component kroupa et al . ( 1993 ) law , which is flatter at lower masses and steeper at the high end of the imf . this empirically derived function leads to a lower overall effective yield that is more consistent with local data ( pagel 2001 ) . although the choice of imf does influence the evolution of the chemical species of interest in this investigation , the impact is not large enough to affect our final conclusions . model 1 from also differs from our im - enhanced model through the inclusion of an exponential 0.5 gyr time scale before the onset of the `` normal '' stellar imf component . thus , at the very earliest times , most of the star - forming gas ends up inside stars with @xmath70 5 m@xmath34 . in contrast , we retain our normal imf component at all times ( akin to s model 2 ) . * * im - enhanced imf : * while the im - bump decays exponentially on a fixed timescale ( ranging from 0.2 to 0.4 gyr depending on the model ) , we transformed this from a time to a metallicity dependence . as discussed in section [ imf ] , the assumption that the shape of the imf is governed primarily by the chemical composition of the star forming gas cloud is presumably more physically justified than imposing a uniform time - dependence . converting from the time to the metallicity domain allows us to apply the same imf prescription to systems with various star formation histories , such as the slowly evolving outer galactic disk , which we investigate in this paper . * * dual - phase infall : * assumed that the galactic disk forms during a _ single _ phase of gas infall , in order to directly compare with the results of timmes et al . we model the milky way formation using a _ dual_-phase infall scheme , since this has been shown to provide a better match than single - infall models to the number distribution of g- and k - dwarfs in the solar neighbourhood ( e.g. chiappini et al . we stress that our final conclusions are largely insensitive to our choice of infall scheme . we now present the predicted chemical evolution for the solar radius and the outer galactic disk in the case of an enhanced population of intermediate - mass stars at low metallicity . these results are compared with those obtained using a normal imf and plotted against observations of local stars and dlas , where applicable . in all the following figures , solid and dashed lines denote the evolution at the solar ( 8.5 kpc ) and outer radii ( 16.5 kpc ) , respectively . models adopting a normal imf are shown with thin lines , while thick lines correspond to models with the im - enhanced imf , as illustrated in fig . [ imf_fig ] . the evolution of the type ia supernova rate predicted by the model described in section [ model ] is shown in the upper panel of fig . [ snia_fig ] . the incidence of type ia sne in the solar region ( solid lines ) is predicted to have risen to a peak about 7 gyr ago and steadily declined thereafter . in contrast , the predicted sn ia rate in the outer disk ( dashed lines ) has continued to increase up to the present - day . this difference reflects the more rapid and efficient conversion of gas into stars in regions of higher surface density . a consequence of an imf that increases the number of intermediate - mass stars is the birth of more type ia sne progenitors . type ia sne are understood to be associated with binary systems of low- and intermediate - mass stars , in which the mass lost by the more evolved star is accreted by its white dwarf ( wd ) companion until the wd can no longer be sustained by electron degenerate pressure and a violent explosion ensues . comparing the thick and thin solid lines in the upper panel of fig . [ snia_fig ] , it can be seen that the im - enhanced imf has no significant affect on the local solar neighbourhood sn ia rate . in the outer disk , however , the agb - enhancement elevates the sn ia rate by up to a factor of six between 0.51 gyr and about a factor two at 3 gyr before converging with the normal imf model after @xmath696 gyr ( compare the thick and thin dashed lines ) . the different impact of the im - enhanced imf on the solar and outer radii is due to a faster build - up of metals in the solar radius relative to the outer disk . this causes the im - bump to decay more quickly and leads to only a minor increase in the sn ia rate . in contrast , the slower rate of enrichment in the outer radius leads to a longer - lasting im - bump , which has a significant impact on the sn ia rate at early times . im - enhanced imfs at low metallicity could leave an observable trace on the cosmic sn ia rate , which could be used to discriminate between these models ( fields et al . 2001 ) . type ia sne are major producers of iron and are responsible for 1/32/3 of the solar fe content ( e.g. timmes et al . thus , any increased incidence of sn ia leads inevitably to the production of more fe , as seen in the lower panel of fig . [ snia_fig ] where the evolution of [ fe / h ] is plotted . the inclusion of an enhanced i m agb population increases the abundance of iron in the ism of the outer disk by a factor of 23 between 13 gyr . once again , the solar radius is less sensitive to the im - enhanced imf because the amplitude of the im - bump decays on a shorter timescale . as well as being sn ia progenitors , agbs leave behind white dwarf remnants that can be indirectly detected through microlensing experiments . we found that the im - enhanced imf increases the present - day number of wd remnants by only about 5% and 10% for the solar and outer radii , respectively . thus , the number of wds is not expected to be a sensitive discriminant of the different imf scenarios . this echos the galactic halo results of gibson & mould ( 1997 ) and fields , freese , & graff ( 2000 ) , who found that stellar c and n abundances provide far stronger constraints than wd counts on any enhancement in the number of lims at early times . it should be noted that although only predict the evolution at the solar radius , if their im - enhanced model were applied to the outer disk we expect that they would _ not _ predict an increase in either the sn ia rate or [ fe / h ] because their sn ia prescription prohibits the formation of type ia sne in metal - poor environments . the square shows the position of the sun , while lines have the same meaning as in the panel above . due to a faster build - up of metals in the solar radius relative to the outer disk , the im - bump decays more quickly and leads to only a minor increase in the sn ia rate and , consequently , the [ fe / h ] evolution . in contrast , the slower rate of enrichment in the outer radius leads to a longer - lasting im - bump , which has a significant impact on the sn ia rate and fe content at early times.,title="fig:",width=302 ] vs time . the square shows the position of the sun , while lines have the same meaning as in the panel above . due to a faster build - up of metals in the solar radius relative to the outer disk , the im - bump decays more quickly and leads to only a minor increase in the sn ia rate and , consequently , the [ fe / h ] evolution . in contrast , the slower rate of enrichment in the outer radius leads to a longer - lasting im - bump , which has a significant impact on the sn ia rate and fe content at early times.,title="fig:",width=302 ] figure [ mgiso_fig ] shows the sensitivity of ( @xmath6mg+@xmath7mg)/@xmath8 mg versus [ fe / h ] to the choice of imf for both the solar ( 8.5 kpc ) and outer ( 16.5 kpc ) radius models . predictions are plotted against abundances observed in nearby dwarf stars by gay & lambert ( 2000 ; circles ) and yong et al . ( 2003b ; diamonds ) . it has been suggested that the production of heavy mg isotopes by imss is needed to explain the observations in metal - poor stars ( e.g. timmes et al . 1995 ; goswami & prantzos 2000 ; alib ' es et al . 2001 ) . indeed , fenner et al . ( 2003 ) found that imss were responsible for most of the heavy mg isotopes in the solar neighbourhood for @xmath27}\!<\!-1 $ ] . the inclusion of agb nucleosynthesis within the framework of the standard imf model provides a good match to both datasets at low [ fe / h ] , as illustrated by the thin solid line in fig . [ mgiso_fig ] . at higher [ fe / h ] , the normal imf solar model matches the gay & lambert ( 2000 ) data but not those of yong et al . ( 2003b ) . we caution that our predictions are best compared against the gay & lambert ( 2000 ) sample because yong et al . ( 2003b ) used kinematics to preferentially select halo and thick disk stars . consequently , their sample contains relatively few thin disk members and perhaps some stars belonging to an accreted component ( although inspection of figure 13 from yong et al . 2003 indicates that the fraction of accreted stars is small if one employs the gratton et al . 2003 criteria to specify the accreted component ) . an uncertainty in the @xmath6 mg and @xmath7 mg yields of @xmath620.2 dex is illustrated for the im - enhanced solar radius model by the shaded region in fig . [ mgiso_fig ] , and will be discussed further in section [ mg_uncertainties ] . although the mean ( @xmath6mg+@xmath7mg)/@xmath8 mg for all stars with @xmath27}\!<\!-1 $ ] is about 0.15 , fig . [ alpha_iso_fig ] demonstrates that a value 59 times larger than this is required to explain the measured by mfw04 , under the assumption of a solar @xmath6mg:@xmath7 mg ratio ) . estimated that a ratio of ( @xmath6mg+@xmath7mg)/@xmath8 mg = 0.62 would remove the need for any time - variation in @xmath1 . this is lower than our calculated value of 1.1 @xmath62 0.3 ( see fig . [ alpha_iso_fig ] ) because theirs is only a rough approximation and ours is based on the qso absorption - line spectra . however , to aid comparison , we followed in imposing an im - enhanced imf capable of elevating ( @xmath6mg+@xmath7mg)/@xmath8 mg to @xmath690.62 for the solar radius model ( thick solid line ) . the behaviour of the mg isotopic ratios in our im - enhanced solar radius model is very similar to model 1 , reaching a maximum of ( @xmath6mg+@xmath7mg)/@xmath8 mg = 0.63 at @xmath71}\!=\!-1.65 $ ] . the prolonged impact of the im - enhanced imf in the more metal - poor outer disk leads to a maximum ( @xmath6mg+@xmath7mg)/@xmath8 mg ratio of 0.92 at @xmath27}\!=\!-1.7 $ ] . this is seven times higher than the corresponding ratio in the case of the normal imf . if we assume that ( i ) the level of agb stellar enhancement is a function of gas metallicity and declines with the build - up of metals , and ( ii ) that many qso absorption systems consist of slowly evolving objects such as dwarfs or outer disks of spirals , then these findings are qualitatively consistent with a scenario in which much higher neutron - rich mg isotopic abundances are found in qso absorption systems than in local stars of the same metallicity . mg+@xmath7mg)/@xmath8 mg as a function of [ fe / h ] . as for fig [ snia_fig ] , the solid and dashed lines show predictions for the solar radius and outer disk , respectively . thin lines denote a normal imf model and thick lines represent the model with an im - enhanced imf . circles and diamonds show stellar data from gay & lambert ( 2000 ) and yong et al . ( 2003b ) , respectively . the square shows the solar value . the shaded area indicates a @xmath620.2 dex uncertainty range for the thick solid line.,width=302 ] the types of agb stars responsible for expelling enough heavy mg isotopes at early times to explain an apparent variation in @xmath1 are also thought to be the main factories of primary nitrogen . thus , an enhanced agb population should leave a strong imprint on the n abundances in local stars and in dlas . the predicted variation of [ n/@xmath1 ] with [ @xmath1/h ] is plotted against dla measurements in fig . [ nonsi_fig ] . the model results are plotted using si as the reference element , while for the centuri ' on et al . ( 2003 ) data , the @xmath1-element is o , si or s. it is reassuring that the normal imf solar model ( thin solid line ) passes through the dla data before rising to roughly solar value . however , no single homogeneous chemical evolution model will account for the broad spread in the dla data and , worryingly , the outer disk normal imf model ( thin dashed line ) overproduces n with respect to the data . since the normal imf models ( thin lines ) already tend to produce more n than is observed in dlas , it is not surprising that the im - enhanced imf models ( thick lines ) overproduce n by more than an order of magnitude . an indication of the sensitivity of these results to the stellar nucleosynthesis models is given by comparing the thick solid line with the thin dotted line , which corresponds to the im - enhanced solar radius model in the case of van den hoek & groenewegen ( 1997 ) ims yields . section [ cn_uncertainties ] describes the uncertainties in these predictions in greater detail . ] as a function of [ @xmath1/h ] . dashed and solid lines have the same meaning as in fig [ snia_fig ] . model results employ si as the reference @xmath1-element . data points with limits and error bars show dla measurements by centuri ' on et al . ( 2003 ) , where the @xmath1-element is either o , si , or s. the thin dotted line shows the im - enhanced solar radius model in the case of van den hoek & groenewegen ( 1997 ) ims yields rather than karakas & lattanzio ( 2003 ) , to give an indication of the uncertainties associated with n production in the stellar models . , width=302 ] inspection of figs . [ mgiso_fig ] & [ nonsi_fig ] reveals that the empirical constraints imposed by dla nitrogen abundances are strongly violated by the models capable of producing sufficient @xmath73 mg relative to @xmath8 mg to mimic the variation in the @xmath1 obtained by mfw04 . encountered the same problem with their model 1 , which they sought to rectify in model 2 by increasing star formation efficiency ( sfe ) by a factor of @xmath692.5 . figure [ nsfe_fig ] shows how our [ n/@xmath1 ] versus [ @xmath1/h ] varies with a factor of 2.5 increase in sfe . the agreement with the data is improved , but [ n/@xmath1 ] is still too high , except at the lowest metallicities . the curve corresponding to our increased sfe model in fig . [ nsfe_fig ] is extremely similar to the dot dash line in figure 15 from . we note that their models 1 and 2 differ not just in sfe , but in the location , width , amplitude , and decay timescale of the imf im - bump . nevertheless , we wish to point out a troublesome consequence of such an increase in sfe that is robust to those differences . figure [ gasdens_fig ] shows the present - day gas surface density profile expected in the case of normal or enhanced sfe . the standard model agrees with the data from dame ( 1993 ) to within a factor of two . however , a factor of 2.5 increase in the sfe leads to severe gas depletion , in conflict with the observations . while the chemical evolution model that applies to qso absorption systems need not satisfy the empirical constraints from the milky way , we caution that increasing star formation efficiency in order to alleviate overproduction of n might not be appropriate given the gas - rich nature of many absorption systems . ] vs [ @xmath1/h ] to the star formation efficiency . the solid line represents the im - enhanced solar radius model , as seen in fig [ nonsi_fig ] . the dotted line shows the effect of increasing star formation ( sf ) efficiency by a factor of 2.5 . increased sf efficiency improves the agreement with the dla data ( symbols ) , but still overproduces n at [ @xmath1/h ] @xmath69 -1.,width=302 ] along with any significant contribution to the abundance of n and the heavy mg isotopes in metal - poor environments , low- and intermediate - mass stars should leave an additional observable chemical signature in the form of high @xmath3c abundance relative to @xmath4c . metal - poor agb stars with mass @xmath69 4 m@xmath34 are a major source of @xmath3c , produced during hot bottom burning when the cn cycle converts @xmath4c into @xmath3c . since the important factories of @xmath4c are less massive 23m@xmath34 stars , the @xmath3c/@xmath4c ratio in the ejecta of imss is strongly mass - dependent , peaking sharply between 4 to 5 m@xmath34 . figure [ ciso_fig ] shows the predicted evolution of @xmath3c/@xmath4c as a function of [ fe / h ] . for both the solar and outer radii models , the introduction of an im - enhanced imf increases the @xmath3c/@xmath4c ratio 46-fold in the metallicity range corresponding to the greatest enhancement of the heavy mg isotopes . to estimate the uncertainty in these results due to the stellar yields , we also ran identical models using the lims yields from van den hoek & groenewegen ( 1997 ) and marigo ( 2001 ) , depicted with thin dotted and dot - dashed lines , respectively . high values of @xmath3c/@xmath4c may conflict with the symmetry of civ line profiles observed in some qso absorbers and could impose a further constraint on the shape of the imf at early times . this will be discussed in more detail in section [ conclusions ] . c/@xmath4c as a function of [ fe / h ] . symbols and lines have the same meaning as in fig [ snia_fig ] . the thin dotted and dot - dashed lines show the im - enhanced solar radius model in the case of van den hoek & groenewegen ( 1997 ) and marigo ( 2001 ) ims yields , respectively . , width=302 ] since transitions from si are important anchor lines in the many - multiplet method , we show the predicted behaviour of ( @xmath74si+@xmath75si)/@xmath76si as a function [ fe / h ] in fig . [ siiso_fig ] . the relative abundance of si isotopes is not affected by an enhanced early population of agb stars to the extent of n and the heavy mg and c isotopes . indeed , ( @xmath74si+@xmath75si)/@xmath76si remains well below the solar value ( open square ) for both the local and outer disk models ( thick lines ) . nevertheless , the measured in high- qso absorption systems is largely independent of the si isotopic composition , as highlighted in section [ ss : agb ] . thus , even a dramatic elevation in ( @xmath74si+@xmath75si)/@xmath76si due to agb stars would not account for the high - redshift results from . si+@xmath75si)/@xmath76si as a function of [ fe / h ] . symbols and lines have the same meaning as in fig [ snia_fig].,width=302 ] aluminium is another element whose production by agb stars is expected to be important ( e.g. karakas & lattanzio 2003 ) . the synthesis of al via the mg - al chain is particularly efficient in metal - poor 46m@xmath34 stars . the evolution of [ al / fe ] with [ fe / h ] in the ism is depicted in fig . [ al_fig ] . observations of al in both dlas ( solid squares ) and nearby stars ( circles and open squares ) covering a wide range of metallicities make this element a potentially useful discriminant of the different imf models . unfortunately , most of the dla measurements are lower limits , owing to line saturation ( prochaska & wolfe 2002 ) . the normal imf models ( thin lines ) pass through the lowermost dla _ detections _ , but only the im - enhanced imf models ( thick lines ) can satisfy the highest dla upper limits . the power of al to constrain the shape of the early imf is diminished by the inability of either solar radius model to match the stellar observations at high and low metallicity . the overproduction of al with respect to metal - poor halo stars and the underproduction relative to thin - disk stars is a problem of standard galactic chemical evolution models that has been noted by other authors ( e.g. timmes et al . 1995 ; alib ' es et al . 2001 ) . however , we note that local thermodynamic equilibrium ( lte ) calculations underestimate the al abundance in metal - poor stars . the open squares in fig . [ al_fig ] would be about 0.6 dex higher if non - lte effects were included ( e.g. gehren et al . 2004 ) . as a function of [ fe / h ] . dashed and solid lines have the same meaning as in fig [ snia_fig ] . dla measurements from prochaska et al . are plotted with solid square symbols , while circles and open squares show stellar observations from edvardsson et al . ( 1993 ) and cayrel et al . ( 2004 ) , respectively . , width=302 ] the production of phosphorus from imss leads to a @xmath690.5 dex increase in [ p / fe ] when the im - bump is added to the standard imf , as shown in fig . [ p_fig ] . stars in the 46m@xmath34 mass range are the chief culprits for the additional @xmath77p , which is understood to be generated through neutron - capture onto si followed by @xmath78-decay to @xmath77p ( j. lattanzio , private communication ) . as is the case for al , the difference between the two imf models is significant , but the dla measurements do not allow the elimination of either scenario . in the case of p , the dla data is very scant and consists mostly of upper limits since the pii line is often blended in the ly@xmath1 forest ( dessauges - zavadsky et al . nevertheless , we present our predictions for al and p in the event that the growing collection of dla measurements may eventually provide a more stringent test of the models presented in this paper . as a function of [ fe / h ] . dashed and solid lines have the same meaning as in fig [ snia_fig ] . dla measurements from prochaska et al . are plotted with square symbols , while the circle is from outram , chaffee , & carswell ( 1999).,width=302 ] denissenkov & herwig ( 2003 ) and denissenkov & weiss ( 2004 ) showed that the production of @xmath6 mg and @xmath7 mg in a typical 5 m@xmath34agb star is fairly robust to the number of thermal - pulses , the hbb temperature , and the third dredge - up efficiency . the final envelope abundance of @xmath6 mg and @xmath7 mg in their series of models agreed within @xmath69 0.1 dex and @xmath69 0.2 dex , respectively . fenner et al . ( 2004 ) estimated the sensitivity of the agb yields of @xmath6 mg and @xmath7 mg to the mass - loss prescription , finding them reduced by @xmath69 0.2 dex when the vassiliadis & wood ( 1993 ) mass - loss law was replaced with a steadier rate from reimers ( 1975 ) ( with @xmath79 on the agb ) . the treatment of convection in agb models has recently been investigated by ventura ( 2004 ) , who found that the yields were very sensitive to the adoption of standard mixing length theory ( mlt ) versus the full spectrum of turbulence ( fst ) model . the latter case results in greater mass - loss rates , shorter agb lifetimes , less third dredge - up , and reduced yields of heavy elements including cno and the mg isotopes . the fst model faces serious hurdles of its own , since it predicts a deficit of na in agb ejecta and a _ positive _ o - na correlation : at odds with observations of globular cluster stars believed to be polluted by agbs ( ventura , dantona , & mazzitelli 2004 ) . a further uncertainty relates to the presence of a @xmath3c pocket , which is neglected in the agb models implemented in this study . one might expect that the inclusion of this additional source of neutrons would increase the production of @xmath6 mg and @xmath7 mg . however the impact of a @xmath3c pocket on nucleosynthesis in metal - poor ims should be marginal in comparison to the effects of hbb ( m. lugaro , _ private communication _ ) . despite these unknowns , the ability of our standard galactic chemical evolution model to reproduce the mg isotopic ratios in metal - poor stars by adding in the karakas & lattanzio ( 2003 ) yields , suggests that their mg predictions are reasonably accurate . the total production of @xmath73 mg with respect to @xmath8 mg from type ii sne is uncertain to a similar degree due to internal uncertainties in the stellar models as well as sensitivity to the shape and upper mass limit of the imf . fenner et al . ( 2003 ) found that a salpeter ( 1955 ) imf leads to 50per cent higher present - day @xmath7mg/@xmath8 mg ratio than in the case of a kroupa et al . ( 1993 ) function , because the salpeter law gives rise to a greater fraction of massive stars ( see their figs . 5 and 6 ) . the shaded region in fig . [ mgiso_fig ] indicates the @xmath62 0.2 dex uncertainty in the results due to errors in both the agb and sn ii contribution . to estimate the uncertainty in the predicted evolution of n and the @xmath3c/@xmath4c ratio , we ran identical models using two alternative sets of ims yields : those of van den hoek & groenewegen ( 1997 ) and marigo ( 2001 ) . the marigo yields for n were in close agreement with those of karakas & lattanzio ( 2003 ) while the implementation of the van den hoek & groenewegen ( 1997 ) yields produced even more n. the dotted line in fig . [ nonsi_fig ] displays the results from an im - enhanced solar radius model using the van den hoek & groenewegen ( 1997 ) yields in place of karakas et al . ( i.e. compare dotted with thick solid curve ) . it is clear that the problem of excess nitrogen is further exacerbated in this case . the dotted and dot - dashed lines in fig . [ ciso_fig ] show the behaviour of @xmath3c/@xmath4c assuming van den hoek & groenewegen ( 1997 ) and marigo ( 2001 ) ims yields , respectively . the marigo yields generate the highest @xmath3c/@xmath4c peak [ with ( @xmath3c/@xmath4c)@xmath80 = 0.073 ] and those of van den hoek & groenewegen ( 1997 ) the lowest [ with ( @xmath3c/@xmath4c)@xmath80 = 0.042 ] . the region encompassed by these two extreme models is indicative of the level of uncertainty afflicting the models . @xcite pointed out that measurements are sensitive to the isotopic composition of mg in the gas - phase of the low-@xmath81 absorbers used in varying-@xmath1 studies . standard models of chemical evolution predict that the abundance of the neutron - rich isotopes of mg relative to @xmath8 mg decreases with lookback time , making the result even more significant . however , demonstrated that a sufficiently enhanced early population of intermediate - mass stars can raise ( @xmath6mg+@xmath7mg)/@xmath8 mg to the supersolar levels needed to render the result null . a major problem with this scenario is the almost inevitable overproduction of n when compared against the dla data ( fig . [ nonsi_fig ] ) . this is because the agb stars responsible for significant @xmath73 mg production are also important sources of n. there are various ways to reconcile an im - enhanced imf with the low n abundances in dlas , but none are entirely satisfactory . for instance , since nucleosynthesis within very metal - poor imss is not well understood , the theoretical n yields may be overestimated . however , in this case , it would be difficult to explain the high n abundance found in some of the most iron - depleted stars in our galaxy ( e.g. norris , beers & ryan 2000 ; christlieb et al . 2004 ) . another way to mitigate the problem of excess [ n/@xmath1 ] is through increased star formation efficiency , however this leads to severe gas depletion that may be inconsistent with the gas - rich nature of dlas ( fig . [ gasdens_fig ] ) . finally , the history of many dlas may be marked by periods of supernova - driven outflows , but it is not clear how such galactic winds could preferentially remove n but not the heavy mg isotopes . thus , the observed n abundance in qso absorbers still stands as a robust test of a putative im - enhanced early imf . the enhanced @xmath3c/@xmath4c ratio predicted by these types of chemical evolution models ( fig . [ ciso_fig ] ) holds promise as a future probe of the agb contribution to qso absorption line abundances . @xcite has calculated the isotopic shifts in the ubiquitous civ @xmath821548/1550 alkali doublet , finding the @xmath83c line to lie @xmath84 to the blue of the @xmath4c line . recent calculations by berengut et al . ( in preparation ) confirm these large shifts to within 1per cent precision and show the @xmath3c@xmath4c separation to be @xmath85 . enhancements as large as @xmath86 may already be ruled out by the symmetric civ line profiles observed in @xmath87 ( @xmath88 ) spectra of highly ionized qso absorbers with simple velocity structure ( e.g. * ? ? ? however , placing limits at lower @xmath3c abundances may be unconvincing unless the laboratory wavelengths of the civ doublet transitions can be measured with better precision than 0.4@xcite . if the calculated isotopic separations for transitions of other c species ( e.g. ci and cii ) prove to be as large as those for civ , they may provide a more reliable probe of the @xmath3c/@xmath4c ratio in gas more closely related to dlas and , therefore , the chemical evolution models presented here . given that many qso absorbers are thought to probe slowly evolving systems such as dwarf galaxies and the outer regions of galactic disks , we simulated chemical evolution at large galactocentric radii , where metal - enrichment is more gradual . we found that the chemical signature from an enhanced ims population at low metallicities was more pronounced in the outer disk , with respect to the solar neighbourhood . this follows directly from the assumed metallicity - dependence of the ims enhancement . thus , our im - enhanced models predict that systems with the slowest build - up of metals will bear the strongest signature of agb - pollution . the chemical abundance constraints imposed by qso absorption systems do not support excess numbers of agb stars being formed in the types of systems from which derived a varying @xmath1 . moreover , a significantly enhanced agb population in the local galaxy is ruled out by the sub - solar ratios of the neutron - rich mg isotopes to @xmath8 mg in nearby metal - poor stars . we now summarise other arguments for and against the imf having a non - standard shape in the early universe . motivated by the large observed scatter and possible trends with metallicity of local and high- d abundance determinations , fields et al . ( 2001 ) and prantzos & ishimaru ( 2001 ) speculated that d could be destroyed within stars without an accompanying increase in metals , provided that : ( i ) the earliest populations of stars were strongly enhanced in imss , and ( ii ) zero - metallicity imss do not release their synthesized c and n. prantzos & ishimaru ( 2001 ) emphasize that if the second provision is not met , c and n abundances should be highly enhanced in high- systems if their d has been depleted through astration alone . they note that this is inconsistent with the dla data . indeed , based on the n/@xmath1 trend in dlas , prochaska et al . ( 2002 ) presented a case for a top - heavy imf , with the birth of imss actually _ suppressed _ in many qso absorption systems . the abundance of c and n in galactic stars and dlas provides one of the strongest constraints on any enhancement in the population of lims in the early galaxy ( gibson & mould 1997 ; fields , freese , & graff 2000 ) . however , this is contingent upon how much c and n is released by extremely metal - poor imss . fujimoto et al . ( 2000 ) provide evidence favouring the suppression of c and n yields from zero - metallicity imss , whereas chieffi et al . ( 2001 ) reach the opposite conclusion . abia et al . ( 2001 ) propose that an early imf peaked between 38 m@xmath34 helps explain the presence of metal - poor galactic stars with enhanced c and n abundances , presuming that lims release c and n in large quantities . thus , the stellar models do not offer a cohesive picture of the level of c and n enrichment from metal - poor lims , nor do observations of dlas and metal - poor stars clarify the situation . however , if agbs are responsible for @xmath73 mg enhancements large enough to mimic a varying @xmath1 in the redshift range @xmath18 , they will tend to have had a much higher initial metallicity than primordial . thus , their yields are not subject to the very large uncertainties afflicting zero - metallicity models . evidence from microlensing surveys initially led to speculation that white dwarf remnants contribute up to half of the mass of the milky way halo ( e.g. bennett et al . 1996 ; alcock et al . 1995 ) . populating the halo with such a large fraction of white dwarfs requires the early imf to be heavily biased toward stars with initial mass in the range 18m@xmath34 . adams & laughlin ( 1996 ) constructed imfs capable of producing a white - dwarf dominated halo , and adopted imf parameters that were consistent with their deduced constraints . however , the white dwarf contribution to the dynamical mass of the galactic halo has since been drastically revised downwards to less than a few percent ( gibson & mould 1997 ; flynn , holopainen & holmberg 2003 ; garcia - berro et al . 2004 ; lee et al . 2004 ) , obviating the need to invoke non - standard imfs . theoretical models of star formation suggest that the imf of primordial gas would be biased toward higher mass stars ( e.g. abel , bryan & norman 2000 ; kroupa 2001 ; hernandez & ferrara 2001 ; mackey , bromm & hernquist 2003 ; clarke & bromm 2003 ) . this is due to diminished cooling efficiency in metal - poor gas . however , recent hydrodynamical simulations of the collapse and fragmentation of primordial gas have predicted bimodal primordial imfs with peaks at about 100 and 1 m@xmath34 ( nakamura & umemura 2001 ) and 30 and 0.3 m@xmath34 ( omukai & yoshii 2003 ) . empirical support exists for an early imf skewed toward high stellar masses in extragalactic environments . for instance , to explain the high metal content of the intra - cluster medium various top - heavy imfs have been proposed ( e.g. gibson & matteucci 1997 ; elbaz , arnaud & vangioni - flam 1995 ) . the photometric properties of ellipticals have also been well - explained assuming a top - heavy imf ( arimoto & yoshii 1987 ) . note that their model is not inconsistent with the notion of a very early population of very massive stars that enrich the ism to @xmath89 . in light of the uncertainties besetting theoretical simulations of star formation and the interpretation of extragalactic photometric data , careful analysis of the nucleosynthetic enrichment patterns of qso absorption systems may be a vital step in deciphering the shape of the imf in different astrophysical environments . an early imf heavily biased towards intermediate - mass stars is a potential explanation of the @xmath1-variation apparent in keck / hires qso spectra . during their agb phase these stars produce large quantities of @xmath73 mg which later pollute qso absorption systems at @xmath90 , causing an apparent shift in the mg absorption lines and a spurious @xmath91 . however , such a strong contribution from agbs should be evident in the detailed abundance patterns obtained in a growing number of dlas . using chemical evolution models of a milky way - like spiral galaxy at different radii , we have tested the idea that agb stars were more prevalent in the metal - poor universe . the predicted chemical ramifications of an enhanced agb population are not supported by the available data , with the strongest argument against severe agb pollution coming from the low [ n/@xmath1 ] values observed in dlas . contrary to the conclusions of , we do not find consistent model parameters which simultaneously explain the keck / hires varying-@xmath1 and abundance data . we also contend that a variety of other arguments for invoking an im - enhanced imf are , at best , unclear . future measurements of ( or limits on ) the @xmath3c/@xmath4c ratio from line - profile asymmetries and further dla observations of other elements synthesised by agb stars , such as al and p , may provide additional constraints on the imf s shape at early epochs . the sensitivity of low-@xmath81 mg / fe measurements to isotopic abundance variations in mg ( fig . [ alpha_iso_fig ] ) and the clear link to the nucleosynthetic history of qso absorbers demonstrates the importance of careful measurements , even in the absence of real variations in @xmath1 . if is measured to be zero in @xmath90 absorbers via independent means ( e.g. by combining hi 21-cm and molecular absorption lines ; * ? ? ? * ) then future samples of mg / fe absorbers might reveal mg isotopic abundance variations in two ways : ( i ) through the bulk relative line shifts between mg and fe transitions studied previously and , ( ii ) if we assume that agb pollution of the absorbers is not uniform , through an increased scatter in the line shifts . note that no increased scatter is observed in the current hires or uves samples (; the authors are extremely grateful to maria lugaro and amanda karakas for many valuable discussions . we thank the anonymous referee for comments that helped improve this paper . financial support from the australian research council ( arc ) is gratefully acknowledged . we also acknowledge the monash cluster computing laboratory , the victorian partnership for advanced computing and the australian partnership for advanced computing for use of supercomputing facilities . mtm thanks pparc for support at the ioa under the observational rolling grant . yf thanks the afuw@xmath68sa for their support through the daphne elliot bursary . d. a. , potekhin a. y. , ivanchik a. v. , 2000 , in dunford r. w. , gemmel d. s. , kanter e. p. , kraessig b. , southworth s. h. , young l. , eds , aip conf . 506 , x - ray and inner - shell processes . argonne national laboratory , argonne , il , usa , p. 503

we present new _ chandra _ x - ray observations of seven low - mass black holes ( ) accreting at low - bolometric eddington ratios between . we compare the x - ray properties of these seven low - mass active galactic nuclei ( agn ) to a total of 73 other low - mass agn in the literature with published _ chandra _ observations ( with eddington ratios extending from ) . we do not find any statistical differences between low- and high - eddington ratio low - mass agn in the distributions of their x - ray to ultraviolet luminosity ratios ( ) , or in their x - ray spectral shapes . furthermore , the distribution of low- agn displays an x - ray weak tail that is also observed within high- objects . our results indicate that between , there is no systematic change in the structure of the accretion flow for active galaxies hosting black holes . we examine the accuracy of current bolometric luminosity estimates for our low- objects with new _ chandra _ observations , and it is plausible that their eddington ratios could be underestimated by up to an order of magnitude . if so , then in analogy with weak emission line quasars , we suggest that accretion from a geometrically thick , radiatively inefficient ` slim disk ' could explain their diverse properties in . alternatively , if current eddington ratios are in fact correct ( or overestimated ) , then the x - ray weak tail would imply that there is diversity in disk / corona couplings among individual low - mass objects . finally , we conclude by noting that the distribution for low - mass black holes may have favorable consequences for the epoch of cosmic reionization being driven by agn .

1605.00742

virtually every large galaxy harbors a supermassive black hole ( smbh ; @xmath7 ) in its nucleus ( e.g. , * ? ? ? however , understanding in detail the mechanism(s ) that formed the first primordial ` seed ' black holes , and determining the evolutionary path that those seeds took to grow to smbh sizes , remains a fundamental problem ( e.g. , * ? ? ? * ; * ? ? ? we currently know of @xmath840 quasars at @xmath9 with @xmath1010@xmath11 smbhs ( see , e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? black hole seeds produced by the collapse of population iii stars would have @xmath121000 @xmath13 ( e.g. , * ? ? ? * ; * ? ? ? * ) , and one generally needs to invoke super - eddington accretion onto such low - mass seed black holes in order to explain the presence of high - redshift quasars @xcite . on the other hand , more massive seeds ( @xmath14 ) , as could be produced from the direct collapse of gas clouds at the centers of galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ) , provide another feasible way to cultivate rapid growth . we refer to @xmath14 black holes here as `` massive black holes '' ( mbhs ) . l c c c c c c c c [email protected] & 7 & 0.0414 & 20.4 & 26.3 & 5.8 & @xmath16 & 15015 & 15.7 + [email protected] & 19 & 0.0310 & 20.5 & 26.7 & 6.1 & @xmath18 & 15017 & 10.0 + [email protected] & 40 & 0.0537 & 20.6 & 26.7 & 6.0 & @xmath19 & 15016 & 25.5 + [email protected] & 53 & 0.0295 & 20.6 & 26.6 & 6.2 & @xmath20 & 15018 & 9.0 + [email protected] & 56 & 0.0290 & 20.3 & 26.4 & 6.2 & @xmath21 & 15019 & 9.0 + [email protected] & 82 & 0.0491 & 20.1 & 26.6 & 6.0 & @xmath18 & 15020 & 20.5 + [email protected] & 111 & 0.0473 & 20.1 & 26.8 & 6.1 & @xmath19 & 15021 & 21.8 [ tab : obslog ] over the past 10 - 15 years , samples of mbh - powered active galactic nuclei ( agn ) have rapidly evolved from discoveries of a handful of individual objects such as ngc 4395 @xcite , pox 52 @xcite , and henize 2 - 10 @xcite to the production of systematically assembled catalogs drawn from large sky surveys ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the largest mbh catalogs so far have utilized the sloan digital sky survey ( sdss ; * ? ? ? * ) , particularly ( * ? ? ? * hereafter ) and @xcite who select mbh agn based on the presence of broad h@xmath22 emission in optical spectra . they use the widths and luminosities of broad h@xmath22 to estimate black hole masses and bolometric luminosities via virial scaling techniques @xcite . both studies require @xmath23 , and their catalogs include 229 and 309 @xcite objects . @xcite also recover a sizable number of mbh candidates ( 151 objects ) with a slightly different approach . they specifically target dwarf galaxies by restricting host galaxy stellar masses to @xmath24 . the vast majority ( 136 ) of their mbh agn candidates are then selected via photoionization signatures of black hole activity via narrow emission line ratios ( only a small fraction of their 151 agn candidates display broad h@xmath22 ) . the smallest mbh discovered so - far through optical searches weighs in at only @xmath25 @xcite . it is natural to turn to the x - ray waveband for insight into the properties of the accretion flows that feed mbhs , since x - ray emission is a nearly universal feature of accretion . so far , the most extensive x - ray follow - up is based on 67 objects that were observed with the _ chandra _ x - ray telescope ( @xcite , hereafter ; also see @xcite for an _ xmm - newton _ view of about a dozen mbh - powered agn ) . these x - ray observations show that mbhs tend to have higher x - ray to ultraviolet ( uv ) luminosity ratios on average compared to ( @xmath26 ) type 1 quasars . the higher x - ray to uv ratios are generally consistent with expectations from accretion disk / corona models ( e.g. , * ? ? ? * ; * ? ? ? * ) , where lower black hole masses yield higher accretion disk temperatures , which results in less uv emission near 2500 ( combined with less efficiently compton - cooled coronae , this yields higher x - ray to uv flux ratios ; see ) . however , as a population , mbh agn show a large dispersion in their x - ray to uv luminosity ratios , and they display a puzzling `` x - ray weak '' tail . the above x - ray studies only targeted mbhs at high eddington ratios is the bolometric luminosity , which calculate based on the luminosity of broad h@xmath22 emission ( see their section 3 ) . the eddington luminosity , @xmath27 for ionized hydrogen , is the maximum luminosity before radiation pressures halts accretion , assuming a spherical geometry and isotropic radiation . ] ( @xmath28 ) . here , we present new x - ray observations of seven mbhs at eddington ratios up to an order of magnitude lower ( @xmath29 ) . by increasing the dynamic range in eddington ratio , we can search for trends between x - ray properties and @xmath4 . we describe our sample selection and x - ray observations / analysis in section [ sec : sample ] . x - ray results are presented in section [ sec : res ] , where we also discuss the optical emission line properties of our targets ( section [ sec : optical ] ) , as well as bolometric corrections for mbhs ( section [ sec : lbol ] ) . our results are discussed in section [ sec : disc ] . for consistency with previous _ follow - up , all optical spectroscopic measurements are taken from , and we generally follow the recipes of when deriving x - ray properties . we adopt the same cosmology as and : @xmath30 km s@xmath31 mpc@xmath31 , @xmath32 , and @xmath33 @xcite . chandra targets were selected from the catalog of mbhs , from which we consider low - eddington ratio agn with @xmath34 ( for estimating bolometric luminosities , assume that @xmath35 , where @xmath36 is the agn luminosity at 5100 ; please refer to sections [ sec : aox ] and [ sec : lbol ] for details ) . there are 17 mbhs in with such low eddington ratios . to improve x - ray efficiency , we further restrict our target list to only include nearby galaxies ( @xmath37 , which is comparable to the average redshift of objects targeted by ) , providing seven _ chandra _ targets . these seven galaxies were observed by _ chandra _ during cycle-14 ( proposal i d 14700673 ; pi gallo ) . the galaxy properties and observations are summarized in table [ tab : obslog ] . throughout the text , we refer to each galaxy by the catalog number assigned by ( see column 2 of table [ tab : obslog ] ) . l c c c c c c c c 7 & & & @xmath38@xmath39 & @xmath38@xmath40 & @xmath38@xmath41 & @xmath38@xmath42 & @xmath38@xmath43 & @xmath38@xmath44 + 19 & & & @xmath38@xmath40 & @xmath38@xmath45 & @xmath38@xmath46 & @xmath38@xmath47 & @xmath38@xmath48 & @xmath38@xmath49 + 40 & @xmath50 & @xmath51 & @xmath52 & @xmath53 & @xmath54 & @xmath55 & 40.8 & 41.1 + 53 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & @xmath60 & @xmath61 & 40.9 & 40.8 + 56 & @xmath62 & @xmath63 & @xmath64 & @xmath65 & @xmath66 & @xmath60 & 40.6 & 40.9 + 82 & & & @xmath38@xmath39 & @xmath38@xmath39 & @xmath38@xmath67 & @xmath38@xmath68 & @xmath38@xmath43 & @xmath38@xmath69 + 111 & @xmath70 & @xmath71 & @xmath72 & @xmath73 & @xmath74 & @xmath75 & 42.4 & 42.8 [ tab : xraycnt ] each galaxy was placed at the aimpoint of the s3 chip on the advanced ccd imaging spectrometer ( acis ; * ? ? ? the data were telemetered in faint mode and reduced with the _ chandra _ interactive analysis of observations software v4.7 ( ciao ; @xcite ) . we reprocessed each _ chandra _ observation , applying the latest calibration files ( caldb v4.6.7 ) . the event files were then filtered to only retain events with grades 0,2,3,4,6 and energies between 0.38 kev . we examined light curves of the acis background , and we removed time periods where the background deviated by more than 3@xmath76 from the mean , using the lc_sigma_clip routine in ciao . only 200 s were removed from obsids 15015 and 15016 ( galaxies 7 and 40 ) , and 400 s from obsid 15020 ( galaxy 82 ) . the effective exposure times for each observation are listed in table [ tab : obslog ] . we ran the automated point - source detection tool wavdetect on 0.3 - 8 kev images of the s3 chip , adopting wavelet scales of 1.0 , 1.4 , 2.0 , 2.8 , and 4.0 pixels , a point spread function ( psf ) map calculated at 2.3 kev , and a significance threshold of 10@xmath77 ( which corresponds to one expected false detection across the s3 chip ) . within the positional accuracy of a source located at the aimpoint of the acis detector ( @xmath80@xmath784 ) , four observations contained an x - ray point source at a location consistent with the position of the sdss optical galaxy ( galaxies 40 , 53 , 56 , and 111 ) . images of each _ chandra _ target ( over 0.5 - 8 kev ) are displayed in figure [ fig : xrayims ] . all x - ray sources are point - like , with no signs of nearby diffuse emission or nearby ( off - nuclear ) neighbors . we associate all four x - ray detections with the agn ( see sections [ sec : res ] ) . next , we created soft- ( 0.5 - 2 kev ) and hard - band ( 2 - 8 kev ) images of the s3 chip ( these bands were chosen to match the analysis of ) . the numbers of soft and hard x - ray counts were extracted over circular regions with 5@xmath79 radii , centered on the nuclear x - ray source ( or the sdss optical position for the three observations where wavdetect did not find an x - ray source ) . the local background was estimated over an annulus concentric with the source extraction region , with inner and outer radii of 10 and 20@xmath79 , respectively . the x - ray source in galaxy 111 is significantly brighter than the others , so we adopted a 10@xmath79 radius source extraction circle , and a background annulus with inner and outer radii 15 and 25@xmath79 for that observation . in table [ tab : xraycnt ] , we list the total numbers of counts extracted within each circular region , and the net count rates in both the soft and hard bands . we typically obtained @xmath801050 net counts in each band , except for galaxy 111 where we obtained @xmath80300 and 800 net soft- and hard - counts , respectively . for the four galaxies with associated x - ray emission , we assign uncertainties on the total numbers of counts using poisson statistics @xcite . for the other three galaxies , we use the estimated number of background counts in each source aperture ( scaled from the measured number of background counts in each sky annulus ) and the bayesian method of @xcite to calculate the number of counts required to detect an x - ray source at the 95% confidence level ( typically 3 - 6 counts ) . we confirm that no x - ray point source is present for these three galaxies , and we include 95% upper limits on the net count rates in table [ tab : xraycnt ] . finally , we estimate unabsorbed x - ray fluxes ( and luminosities ) using the portable , interactive multi - mission simulator ( pimms ) and a power - law model , where we adopt the best - fit photon index ( @xmath81 ) and column density ( @xmath82 ) described in section [ sec : xrayspec ] . for the three galaxies lacking an x - ray source , we provide flux and luminosity upper limits by assuming @xmath83 ( the average value from ) and the galactic @xmath82 from the @xcite maps ( see table [ tab : obslog ] ) . for the four galaxies with nuclear x - ray detections , we extract x - ray spectra over 0.3 - 8 kev using the ciao tool specextract . we use the interactive spectral interpretation system ( isis ; * ? ? ? * ) to analyze the spectra . given the typically low number of counts , and in order to compare to the spectral analysis of , we only attempt to fit an absorbed power - law model to each spectrum ( phabs*powerlaw in isis ) . we utilize cash statistics @xcite , with the background included in each spectral fit . the spectral fits are shown in figure [ fig : xspec ] , and the best - fit spectral parameters are reported in table [ tab : xrayspec ] , including the photon index ( @xmath81 ) and column density ( @xmath82 ) . error bars are quoted at the 90% level ( corresponding to a change in the cash statistic of @xmath84 for one parameter of interest ) . two sources ( galaxies 53 and 111 ) display moderate absorption , with intrinsic column densities @xmath8010@xmath85 @xmath86 . for the other two galaxies with x - ray detections ( 40 and 56 ) , the column density converges toward zero during the spectral fits . if we instead fix @xmath82 to the galactic value from the @xcite maps for those two sources , the best - fit @xmath81 values do not change significantly for either source ( @xmath87 and @xmath88 for galaxies 40 and 56 , respectively ) . for consistency , in table [ tab : xrayspec ] we always quote @xmath81 values for the fits where @xmath82 is allowed to vary . the x - ray source in galaxy 111 has a high count rate , and it appears to suffer mildly from the effects of photon pileup . for sources with high count rates , two or more photons may hit a ccd detector region before the frame is read out ( every 3.2 sec for our observations ) , and the multiple photons are registered as only a single event . one effect of pileup is that the observed x - ray spectrum may appear harder than the intrinsic one , because of energy migration , where the registered event has an energy equal to the sum of the multiple `` piled '' photons . to correct for this effect , the best - fit model parameters for galaxy 111 in table [ tab : xrayspec ] are reported using the @xcite pileup model.0.3 kev ( i.e. , we do not apply an 8 kev high - energy filter ) ; see http://cxc.harvard.edu/ciao/why/filter_energy.html . ] the pileup fraction is @xmath89 ( as determined from the best - fit pileup model ) . the numbers of net counts and count rates reported for galaxy 111 in table [ tab : xraycnt ] are the observed numbers , without any pileup corrections ; the reported fluxes and luminosities are pileup corrected ( calculated within isis using the best - fit pileup model ) . , where @xmath90 is the cash statistic ) . the best - fit parameters are listed in table [ tab : xrayspec ] . ] l c c c c 7 & & & @xmath38@xmath91 & @xmath38@xmath92 + 19 & & & @xmath38@xmath93 & @xmath38@xmath94 + 40 & & @xmath95 & @xmath96 & @xmath97 + 53 & @xmath98 & @xmath99 & @xmath96 & @xmath100 + 56 & & @xmath101 & @xmath96 & @xmath102 + 82 & & & @xmath38@xmath103 & @xmath38@xmath104 + 111 . the probability of retaining @xmath105 events that are ` piled ' together is @xmath106 . ] & @xmath107 & @xmath108 & @xmath109 & @xmath110 [ tab : xrayspec ] throughout the remainder of the text , we compare our observations to a total of 73 other mbhs with _ chandra _ coverage . the bulk of this comparison sample includes 67 higher eddington ratio objects , composed of 49 _ chandra _ observations originally published by , 10 from @xcite , and 8 from @xcite . ( we refer to these as agn , even though 18/67 of these objects were originally identified as agn by @xcite ) . we collectively refer to these 67 objects as the _ sample_. x - ray information for these 67 objects are taken directly from or @xcite , unless stated otherwise . the remaining 6 archival objects come from a search of the literature for lower - eddington ratio mbhs with _ chandra _ coverage . observations of four objects are presented by @xcite , whose sample includes two sources selected from the @xcite catalog ( sdss [email protected] and sdss j112637.74 + 513423.0 ; both of these sources also appear in ) , and two sources that were selected by @xcite from the sdss data release 5 ( sdss j074345.47 + 480813.5 and sdss j130456.95 + 395529.7 ) . these four sources span @xmath111 , as calculated by ( * ? ? ? * see their table 1 ) . we also consider objects with new _ observations presented by @xcite . to avoid duplicating a similar parameter space as , we exclude objects from @xcite with @xmath112 . of their six sources with new _ observations , two remain sdss j121629.13 + 601823.5 ( @xmath113 ) and sdss j132428.24 + 044629 ( @xmath114 ) . although both of these sources were initially identified as mbh agn by , neither would have been included in our cycle-14 _ chandra _ program because their eddington ratios are above our @xmath34 criterion ( plus , sdss j121629.13 + 601823.5 is at too high of a redshift ; @xmath115 ) . we include them here nevertheless to improve statistics , as they span a similar range in @xmath116 as the four objects from @xcite . x - ray information for these combined six archival observations are taken directly from @xcite and @xcite , unless stated otherwise . three of our _ chandra _ targets do not have x - ray detections , and a total of 15 other objects compiled in our comparison sample ( section [ sec : samp : parent ] ) are reported as non - detections in the literature ( 12 from the high-@xmath4 sample , two from @xcite , and one from @xcite ) . since different detection thresholds are applied in each paper , we re - analyze all 15 archival observations here , so that we can compare upper limits on x - ray fluxes consistently between our analysis and the archival ones . we repeat a similar data reduction on these 15 archival objects as described in section [ sec : obs ] , with the primary difference being that we use a @xmath117 aperture for extracting source counts , in order to minimize the background in each aperture . results of the photometry in soft ( 0.5 - 2 kev ) and hard ( 2 - 8 kev ) images are presented in table [ tab : xlimphot ] , where we also repeat the analysis for our three cycle 15 _ chandra _ targets adopting @xmath117 apertures . in the final column of table [ tab : xlimphot ] , we tabulate whether each source is considered an x - ray detection at the 95% and 99% level over the full 0.5 - 8 kev band ( according to the confidence interval tables in @xcite ) . we find 6/18 objects are detected at the 95% level , and 4/18 at the 99% level in each 0.5 - 8 kev image . for the six objects detected at the 95% confidence level , we adopt the fluxes quoted in table [ tab : xlimphot ] throughout the remainder of the text . the x - ray fluxes are calculated within pimms , assuming an absorbed power - law with @xmath83 and @xmath82 set to the galactic value from @xcite . the fluxes are calculated over the full 0.5 - 8 kev band , after applying a 92% aperture correction , in order to account for the fraction of x - ray photons excluded by our choice of a @xmath117 source aperture ( these aperture corrections are based on the enclosed energy fraction at 2.3 kev at the _ chandra _ acis - s3 aimpoint ; including these aperture corrections produces fluxes consistent with the larger apertures that we adopted in section [ sec : obs ] ) . for the 12 sources that remain undetected at the 95% confidence level , we include 95% upper limits on the full - band x - ray fluxes in table [ tab : xlimphot ] . l c c c c c c c c c c 003552.26 + 011249.4 & 15015 & 15.9 & 0 & 0.06 & 0 & 0.09 & @xmath118 & n & n & this work + [email protected] & 9235 & 4.7 & 1 & 0.02 & 1 & 0.1 & @xmath119 & n & n & y14 + [email protected] & 15017 & 10.0 & 0 & 0.06 & 0 & 0.06 & @xmath120 & n & n & this work + 094310.12 + 604559.1 & 5661 & 5.0 & 2 & 0.07 & 0 & 0.1 & @xmath121 & n & n & gh07c + 095306.81 + 365028.0 & 15020 & 20.9 & 0 & 0.1 & 0 & 0.2 & @xmath122 & n & n & this work + 095330.53 + 562653.4 & 11452 & 2.0 & 3 & 0.01 & 0 & 0.01 & @xmath123 & y & n & d12 + 105755.66 + 482502.0 & 11455 & 2.0 & 1 & 0.01 & 0 & 0.02 & @xmath124 & n & n & d12 + 112637.74 + 513423.0 & 9234 & 4.7 & 1 & 0.05 & 0 & 0.02 & @xmath125 & n & n & y14 + 114343.76 + 550019.3 & 11460 & 1.8 & 0 & 0.0 & 0 & 0.01 & @xmath126 & n & n & d12 + 115138.24 + 004946.4 & 7735 & 4.7 & 0 & 0.03 & 1 & 0.09 & @xmath125 & n & n & d09 + 121629.13 + 601823.5 & 13860 & 24.0 & 21 & 0.09 & 5 & 0.2 & @xmath127 & y & y & g14 + 131926.52 + 105610.9 & 11470 & 1.8 & 5 & 0.03 & 2 & 0.0 & @xmath128 & y & y & d12 + [email protected] & 11474 & 1.8 & 0 & 0.03 & 1 & 0.07 & @xmath129 & n & n & d12 + 144507.30 + 593649.9 & 7738 & 4.7 & 2 & 0.02 & 6 & 0.1 & @xmath130 & y & y & d09 + 153656.44 + 312248.1 & 11476 & 1.8 & 0 & 0.02 & 1 & 0.05 & @xmath129 & n & n & d12 + 163159.59 + 243740.2 & 11483 & 1.9 & 24 & 0.05 & 8 & 0.02 & @xmath131 & y & y & d12 + 170246.09 + 602818.9 & 7739 & 4.7 & 0 & 0.03 & 2 & 0.1 & @xmath121 & n & n & d09 + [email protected] & 5667 & 4.7 & 2 & 0.1 & 2 & 0.1 & @xmath132 & y & n & gh07c [ tab : xlimphot ] l c c c c c c c c c c 95% conf . & 12 & 77.9 & 5 & 0.5 & @xmath133 & 6 & 1.0 & @xmath134 & @xmath135 & 22.1(@xmath3822.3 ) + 99% conf . & 14 & 84.6 & 10 & 0.6 & @xmath136 & 8 & 1.1 & @xmath137 & 1.0@xmath138 & 21.8(@xmath3822.1 ) + [ tab : xstack ] next , we perform a stacking analysis on the 12 sources that remain undetected at the 95% level , and on the 14 sources not detected at the 99% level . the stacked signals are presented in table [ tab : xstack ] . for both subsets , we obtain x - ray detections in both the soft and hard energy bands at @xmath1099% confidence . to confirm that the stacked detection is not an artifact of improper background estimation ( see , e.g. , * ? ? ? * ; * ? ? ? * ) we perform the following test . for the stack of 99% non - detections ( 14 objects ) , we blindly displace the center of each image s source extraction region by @xmath1395 , in a randomly chosen direction for each image ( the magnitude of this offset is chosen to avoid overlap with each background extraction region and the original source region ) . we then repeat the stacking analysis . we find only one stacked count in the soft band , and one stacked count in the hard band , which is consistent with the expected background levels of 0.6 soft and 1.1 hard counts . finally , we use the observed hardness ratio of the stacked signals ( defined by @xmath140/@xmath141 , where @xmath140 and @xmath141 are the net count rates in the hard and soft bands , after incorporating 88 and 95% aperture corrections , respectively , appropriate at 1 and 4.5 kev for @xmath117 extraction regions ) to explore the average spectral properties of the x - ray non - detected objects . we first assume an absorbed power - law in pimms , with the column density fixed to @xmath142 , which is the ( exposure - time weighted ) average of the galactic @xmath82 values for each source ; the range of @xmath81 that can explain the observed hardness ratios of the stacked samples are reported in column ( 10 ) of table [ tab : xstack ] . we also perform a similar test fixing @xmath143 to identify the range of @xmath82 that can replicate the observed hardness ratios ( see column 11 of table [ tab : xstack ] ) . the uncertainties on @xmath81 and @xmath82 estimated in this manner are large , and we can not break the degeneracy between column density and photon index with so few stacked counts . still , we can confidently assert from this analysis that the population of x - ray non - detected mbh agn have , on average , a relatively hard observed x - ray spectrum ( either caused by a photon index flatter than most of the x - ray detected objects , by a modest amount of intrinsic absorption , or a combination of the two ) . we associate all four x - ray detections with nuclear mbhs , as described below . all four sources have hard x - ray luminosities @xmath144 . these x - ray luminosities are as expected given the eddington ratios derived by for our _ chandra _ targets : for @xmath145 and @xmath146 ( see table [ tab : obslog ] ) , we expect @xmath147 if 10% of the bolometric luminosity is emitted in the hard x - ray band . excluding galaxy 111 , both the soft and hard x - ray luminosities of our _ chandra _ targets generally populate the faint end of the high-@xmath4 mbh sample ( figure [ fig : xray ] ) . for completeness , we show the best - fit photon indices for the four x - ray detections in figure [ fig : xray]c , which cover a similar range as the high-@xmath4 sample . sample are shown for reference ( open histograms ) . for clarity , only high-@xmath4 objects with x - ray detections are included . ( a ) hard x - ray luminosity from 2 - 8 kev . ( b ) soft x - ray luminosity from 0.5 - 2 kev . ( c ) best - fit photon spectral index @xmath81 ( only four _ chandra _ targets with x - ray detections are shown ) . ( d ) x - ray to uv luminosity ratio parameterized by @xmath3 . more negative numbers are `` x - ray weaker . '' ] it is very unlikely that a stellar mass black hole ( or even multiple stellar mass black holes within the _ chandra _ point spread function ) could produce the observed amount of x - ray emission from each galaxy . the high - luminosity tail of a galaxy s x - ray binary ( xrb ) population contains the vast majority of known ultraluminous x - ray sources ( ulxs ; i.e. , most ulxs are stellar mass black holes accreting at super - eddington rates ; e.g. , @xcite ) . accounting for a combination of super - eddington accretion and mild beaming , it is difficult for a stellar mass black hole to attain an x - ray luminosity @xmath1010@xmath148 @xcite . the largest ulx catalogs observationally confirm that objects with @xmath149 are extremely rare and very unlikely attributed to luminous xrbs ( * ? ? ? * ; * ? ? ? * also see @xcite for a review ) . furthermore , the brightest ulxs are typically found in galaxies with high specific star formation rates , which contain ( short - lived ) high - mass x - ray binary populations ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? ulxs in galaxies with little star formation arise from low - mass xrbs ( likely applicable to 3/4 of our galaxies with x - ray detections , as judged from their sdss optical colors ; see section [ sec : optical ] ) , and they appear to always have @xmath150 @xcite . ulxs are also @xmath810 times less common in such galaxies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , since the number of low - mass xrbs scales with the total stellar mass instead of the specific star formation rate @xcite . thus , the presence of hard x - rays at @xmath1010@xmath148 strongly advocates that our four x - ray detected galaxies host agn . one of the most common ways to quantify the relative amount of accretion power outputted in the x - ray waveband is through the @xmath3 parameter , which measures the ratio of x - ray to uv luminosity @xcite . we adopt @xmath151 , where @xmath152 and @xmath153 are ( unabsorbed ) x - ray and uv luminosity densities at rest - frame 2 kev and 2500 , respectively . we use pimms and the adopted spectral parameters for each source ( see section [ sec : xrayspec ] ) to calculate @xmath152 . for consistency with , @xmath153 calculations are based on broad h@xmath22 line luminosities ( @xmath154 ) as follows . first , we take the @xmath154 measurements and use the relation @xmath155 @xcite to determine the luminosity of the agn continuum at 5100 ( @xmath156 ) , and we then assume that the continuum follows a power - law of the form @xmath157 @xcite to estimate @xmath153 . we list @xmath3 values in table [ tab : xrayspec ] , and the @xmath3 distribution is shown in figure [ fig : xray]d . @xcite and show that the high-@xmath4 sample has on average harder ( i.e. , less negative ) @xmath3 values compared to quasars with larger black hole masses ( i.e. , mbhs are systematically `` x - ray brighter '' ) . explain this as being driven primarily by black hole mass , caused by lower - mass black holes being fed by higher temperature accretion disks . find evidence for a weak anti - correlation between @xmath3 and @xmath158 among the high-@xmath4 sample , which supports their interpretation ( an anti - correlation between @xmath3 and @xmath159 has similarly been observed among quasars ; e.g. , @xcite ) . however , also show that @xmath3 is not as hard as expected , if one were to extrapolate from the well - known anti - correlation between @xmath3 and @xmath153 that is defined by more luminous ( and massive ) quasars ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? a potential flattening of the @xmath160 relation at low luminosities was also hinted at by @xcite , and seen by @xcite as well for a sample of 13 low ionization nuclear emission line region ( liner ) galaxies . while most of our _ chandra _ targets are also x - ray brighter than luminous quasars , they tend to be x - ray weaker ( on average ) than expected for their uv luminosities excluding galaxy 111 , the six other targets have @xmath161 ( see table [ tab : xrayspec ] ) , where @xmath162 is the average @xmath3 expected from the @xcite @xmath160 relation . our targets @xmath163 values correspond to being x - ray weaker than @xmath162 at the @xmath164@xmath76 level ( see table [ tab : xrayspec ] ; the rms @xmath76 deviation for each object is tabulated from @xcite ) . finally , we note that galaxy 111 is the x - ray brightest mbh from observed so far , but it does not have an unusual x - ray spectrum or column density that would indicate that it is fundamentally different than the other objects . all seven _ chandra _ targets are formally below the broad line detection thresholds of , who primarily require @xmath165 and ew@xmath166>15 $ ] ( @xmath167 is the h@xmath22 line flux , @xmath168 is the rms deviation of the continuum subtracted spectrum , and ew@xmath166 $ ] is the h@xmath22 equivalent width ; see @xcite for details ) . our targets were included in the sample because visual inspection of their spectra ( by ) revealed broad and interesting h@xmath22 , and they were subsequently flagged by as less secure `` candidate '' black holes ( which they term as their _ c _ sample ) . we assess our targets agn identifications here by re - examining their optical properties in conjunction with our new _ chandra _ x - ray constraints . as an initial cross - check , we consult the results of an independent optical analysis by @xcite , who analyzed the sdss spectra of @xmath867000 emission line galaxies . @xcite detected broad h@xmath22 in all of our targets except for galaxy 7 . and seyfert - like narrow emission lines . ] we therefore operate under the assumption that only for galaxy 7 could the broad h@xmath22 seen by during visual inspection be a statistical `` false positive . '' that is , we consider the claim for the presence of broad h@xmath22 in these objects to be robust . we stress , however , that broad h@xmath22 on its own does not prove that an agn is present , especially in star forming galaxies where there could be sources of contamination from young stars and/or supernovae ( see e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * baldassare et al . subm . ) . in figure [ fig : bpt ] we show the location of each of our targets in the baldwin - phillips - terlevich ( bpt ) diagram ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , which uses narrow emission line diagnostics to separate purely star forming galaxies from galaxies with harder ( agn - like ) ionizing continua . for reference , we include the high-@xmath4 sample in figure [ fig : bpt ] . three of our _ chandra _ targets have narrow optical line ratios typical of seyfert galaxies ( galaxies 19 , 40 , and 111 ) . these three galaxies also appear to have red sdss optical colors ( see bottom panels of figure [ fig : bpt ] ) , which suggests negligible star formation ( although , see @xcite for a high spatial resolution morphological study that indicates that all three galaxies contain a disk component ) . the combination of the narrow line diagnostics and the presence of broad h@xmath22 emission highly suggests that all three of these galaxies host an agn . galaxies 40 and 111 also display x - ray emission that confirms their agn nature ( see section [ sec : nuclear ] ) . we note that the absence of x - ray emission is not sufficient to exclude the presence of an agn . although we do not detect x - rays from galaxy 19 , we have a sensitive limit ( @xmath169 ) , which places it in the x - ray weak tail defined by the high-@xmath4 sample . ( see color bar ) . open star symbols denote x - ray non - detections . the solid curve shows the `` maximum starburst line '' from @xcite , derived from pure stellar photoionization models . galaxies above the solid curve are `` seyfert - like . '' the dashed curve shows the empirical dividing line between star forming and active galaxies @xcite . agn from the high-@xmath4 sample are overplotted for comparison ( circles ) , with open symbols denoting x - ray non - detections . the bottom panels show sdss @xmath170 color composite images of each of our _ chandra _ targets . the green scale bar at the top left of each image represents 5@xmath79 . ] none of the other four galaxies show obvious optical narrow - line agn signatures in figure [ fig : bpt ] . however , two of these galaxies are still very likely agn based on their x - ray properties ( galaxies 53 and 56 ; we note that @xcite independently classify galaxy 53 as an mbh agn from broad h@xmath22 in its optical sdss spectrum ) . galaxies 7 and 82 do not show x - ray emission or agn - like narrow line ratios , they have blue sdss optical colors , and it is unclear if galaxy 7 displays broad h@xmath22 . we are therefore uncertain on the agn nature for galaxies 7 and 82 , and we consider them to be low - confidence agn _ candidates _ here . we note that galaxy 7 is not included in the low-@xmath158 sample of @xcite , but galaxy 82 is included in their sample . among the six mbhs with low-@xmath4 for which we take x - ray information from the literature , the two from @xcite were also flagged by as part of their _ c _ sample . following similar arguments as above , we consider both sources to be agn here : both have agn - like @xmath171 [ ] /h@xmath172 measured by , and nuclear x - ray point sources . @xcite discuss classification of their four targets in detail , and we consider all four of their objects to be bona fide mbh agn . in summary , we consider galaxies 40 and 111 to be agn based on their optical and x - ray properties , and galaxy 19 to be an x - ray weak agn . galaxies 53 and 56 appear to be agn that lack agn - like narrow - emission lines , and we consider galaxies 7 and 82 to be ( lower - confidence ) agn candidates . throughout the text , we generally adopt the eddington ratios calculated by for our _ chandra _ targets . for these calculations , a bolometric correction of @xmath173 @xcite is adopted , where @xmath36 is the luminosity of the agn continuum at 5100 ( which is estimated from the broad h@xmath22 line luminosity ; see section [ sec : aox ] ) . we adopt the @xmath4 estimates here to ease comparison with the literature , as @xmath4 estimates for the high-@xmath4 sample are also drawn from . it is possible that this specific choice of bolometric correction could systematically over- or under - estimate @xmath4 . however , such a bias would affect the the entire sample in the same direction , on average . therefore , by calculating @xmath4 in a consistent manner between all subsets , we can reliably search for trends as a function of ( relative ) eddington ratio , as long as we bear in mind the possibility for a systematic offset for the entire sample when interpreting the results . at the moment , bolometric corrections for mbh agn are still poorly constrained , and empirical constraints will require multiwavelength , high - spatial resolution imaging of the nuclear emission for a sizable sample of objects . in the meantime , we explore three other methods of applying bolometric corrections to estimate @xmath174 for agn , which are described below and summarized in table [ tab : lledd ] : 1 . @xmath175 : suggest that the bolometric correction could include a black hole mass term , such that @xmath176 ( see their equation 3 ) . for a 10@xmath177 mbh , this leads to @xmath178 , which increases the estimates from by @xmath801.2 dex ( see column 3 of table [ tab : lledd ] ) . @xmath179 : we also estimate @xmath4 from the strength of [ ] line emission in the sdss optical spectra . following @xcite , we assume @xmath180 @xmath181 [ ] . when calculating @xmath181 [ ] , we include a correction for extinction by estimating @xmath182 from the balmer - decrement in the sdss spectra ( assuming a @xcite extinction curve and @xmath154/@xmath183 = 3.1 , which is typically adopted for agn to include photoionization + collisional enhancement of hydrogen lines ; e.g. , @xcite ) . since this @xmath174 estimate assumes photoionization by an agn continuum , we only perform the calculation for the three targets with agn - like narrow emission line ratios in figure [ fig : bpt ] ( galaxies 19 , 40 , 111 ; see column 4 of table [ tab : lledd ] ) . @xmath184 : we also estimate @xmath4 from the observed x - ray luminosities . following ( * ? ? ? * see their section 5 ) , we apply an x - ray bolometric correction for mbhs of @xmath185 ( see column 5 of table [ tab : lledd ] ) . both the mass - dependent and @xmath181[]-based bolometric corrections suggest more rapidly accreting agn than estimated by . the x - ray based bolometric corrections generally predict lower eddington ratios ( except for galaxy 111 , the lone object with a positive @xmath186 value ) . the discrepancy between different methods of calculating @xmath4 serves as a quantitative guide for the degree to which systematics may affect our adopted @xmath116 estimates . we stress , however , that the discrepancies between different @xmath174 estimates do not imply that the values must be incorrect or biased . rather , the discrepancies motivate the need for high - resolution broadband imaging of the nuclei of a large sample of mbh agn , to provide observational constraints to calibrate bolometric corrections against . l c c c c 7 & @xmath171.7 & @xmath170.4 & & @[email protected] + 19 & @xmath171.6 & @xmath170.5 & @xmath170.8 @xmath170.6 & @[email protected] + 40 & @xmath171.5 & @xmath170.3 & @xmath171.1 @xmath170.8 & @xmath172.1 @xmath171.6 + 53 & @xmath171.8 & @xmath170.7 & & @xmath172.6 @xmath172.2 + 56 & @xmath172.0 & @xmath170.9 & & @xmath172.5 @xmath172.1 + 82 & @xmath171.6 & @xmath170.4 & & @xmath38 - 2.4 + 111 & @xmath171.5 & @xmath170.4 & @xmath170.5 @xmath170.3 & @xmath170.5 @xmath170.0 + [ tab : lledd ] as a function of eddington ratio for our _ chandra _ targets ( red filled star symbols ) , the four mbh agn with @xmath187 from ( * ? ? ? * blue triangles ) , the two mbh agn with @xmath187 from ( * ? ? ? * cyan upside down triangles ) , and the high-@xmath4 sample ( circles ) . arrows denote upper limits on @xmath3 . the open star symbols represent our two less confident agn candidates ( galaxies 7 and 82 ; see section [ sec : optical ] ) . ] here , we compare the x - ray properties of the lower - eddington ratio agn to the high-@xmath4 sample ( differences between the x - ray properties of accreting mbhs vs. smbhs have already been explored in depth by ; also see @xcite ) . we generally exclude the two lower - confidence agn candidates ( galaxies 7 and 82 ) from the following discussion , unless stated otherwise . our low-@xmath4 sample therefore contains 11 objects , and the high-@xmath4 comparison samples contains 67 objects . as noted in section [ sec : lbol ] , we adopt the @xmath4 estimates from , as that provides the most straightfoward method for uniformly comparing @xmath4 across the entire 78 object sample . while these estimates may be systematically offset from the `` true '' @xmath4 values , the estimates provide a reliable tracer for the _ relative _ eddington ratios between objects in the full sample . we can not calculate @xmath4 from @xmath181 [ ] across the entire sample because not all objects show narrow - line agn signatures . the full 78 object sample shows a large dispersion in @xmath3 ( see section [ sec : wlq ] ) , which implies that applying an x - ray based bolometric correction for determining @xmath4 is not straightforward , and would likely require an undetermined correction that is dependent on @xmath3 . report on the lack of a correlation between @xmath3 and @xmath116 within just the high-@xmath4 sample ( also see @xcite ) . after extending the dynamic range in @xmath116 by almost an order of a magnitude , we do not see any correlation either ( figure [ fig : aox ] ) . we perform a linear regression on @xmath3 vs. @xmath116 ( including upper limits on @xmath3 ; @xcite ) , and we find a slope consistent with zero ( @xmath188 ) . furthermore , generalized kendall s @xmath189 and spearman s @xmath190 correlation tests on the combined 78 object sample both indicate that no correlation is present at the @xmath191 level . we incorporate upper limits on the x - ray non - detections when running these correlation tests by using the astronomy survival analysis ( asurv ) package rev 1.2 @xcite , which implements the statistical methods presented in @xcite . the lack of a correlation within just the 67 object high-@xmath4 sample reported by is therefore not solely due to their limited dynamic range . in addition , we do not see any evidence for statistically different x - ray properties between the low- and high-@xmath4 samples . a peto - prentice test ( run through asurv to incorporate upper limits ) indicates that the low- and high - eddington ratio samples do not follow statistically different distributions in @xmath3 ( @xmath192 ; also see figures [ fig : xray]d and [ fig : aox ] ) . we also estimate the average @xmath3 values for each distribution using the kaplan - meier estimator in asurv . the 11 lower-@xmath4 agn have a mean @xmath193 , which is comparable to @xmath194 for the 67 object high-@xmath4 sample . furthermore , for our four _ chandra _ objects to which we could fit a spectrum , we similarly do not see any meaningful differences in the spectral properties between the two samples ( figure [ fig : xray]c ) . we might have expected to see a correlation between ( hard x - ray ) @xmath81 and @xmath116 , as is observed for luminous quasars with @xmath195 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but proper investigation will require a sample with tighter x - ray spectral constraints , and thus deeper x - ray observations . as a function of @xmath153 for our low-@xmath4 _ chandra _ targets ( red star symbols ; excluding the less - confident agn in galaxies 7 and 82 ) , the four @xcite low-@xmath4 objects ( blue triangles ) , the two @xcite low-@xmath4 objects ( cyan upside down triangles ) , and the 67 objects from the high-@xmath4 mbh sample ( circles ) . for comparison , we plot samples of more massive black holes , including wlqs ( at @xmath196 ; orange diamonds ) , `` normal '' type 1 sdss quasars from ( * * squares ) , and the best - fit @xmath197 relation from ( * ? ? ? * solid line ) . all open symbols denote upper limits on @xmath3 . ] the lower-@xmath4 mbhs appear to show as wide of a spread in @xmath3 as the high-@xmath4 sample ( see figure [ fig : aox ] ) . given that very few mbhs display signs of x - ray absorption ( and those that do typically have moderate column densities , @xmath198 @xmath86 ) , some mbhs might be intrinsically x - ray weak . they could be similar to the nearby ( z=0.192 ) narrow - line seyfert 1 galaxy phl 1811 , which may have a smaller or quenched x - ray corona ( e.g. , * ? ? ? * ; * ? ? ? intriguingly , the range in @xmath3 displayed by mbhs is also reminiscent of the x - ray properties of weak emission line quasars ( wlqs ; see , e.g. , @xcite ) . wlqs are higher - redshift ( mostly known at @xmath199 ) , unobscured , radio - quiet quasars that have unusually weak high - ionization broad emission lines ( especially ly@xmath22 and @xmath2001549 ; see , e.g. , @xcite ) . approximately 50% of known wlqs are significantly x - ray weaker than expected for their uv luminosities ( with @xmath201 ; see @xcite ) . in figure [ fig : aoxwlq ] , we compare the @xmath3 distribution of mbhs to wlqs as a function of @xmath153 , with typical sdss type 1 quasars from @xcite plotted for reference . we include 38 wlqs that were optically - selected from the sdss , that have @xmath196 ( to ensure sdss spectroscopic coverage of the high - ionization emission line ) , and that have _ chandra _ x - ray observations presented by @xcite and @xcite . were initially selected as high - redshift analogs to phl 1811 , which are nearly always x - ray weak and appear to share many similarities to wlqs in their optical and uv spectra . for convenience , we refer to all objects studied by @xcite and @xcite as wlqs . ] the lack of wlqs at @xmath202 hz@xmath31 in figure [ fig : aoxwlq ] is largely ( but unlikely solely ) due to the restriction in redshift . it is clear from figure [ fig : aoxwlq ] that both the mbh and wlq populations display a larger dispersion in @xmath3 compared to ` normal ' type 1 sdss quasars . to quantify the dispersion , we use the kaplan - meier estimator . since we do not see any statistical difference in the x - ray properties between the low- and high-@xmath4 mbh samples , we consider the entire 78 object mbh sample in the following , in order to improve statistics . we find that the @xmath203 - 75@xmath204 percentiles of the @xmath3 distributions for the 78 mbhs and 38 wlqs span a range of @xmath205 and @xmath206 , respectively ( errors are the standard deviations on the @xmath203 and 75@xmath204 percentile @xmath3 values added in quadrature ) ; when limiting the comparison @xcite sdss quasars to similar luminosities as the wlq sample ( 120 type 1 quasars with @xmath207 hz@xmath31 ) , luminous quasars have 25@xmath204 - 75@xmath204 percentiles spanning only @xmath208 in @xmath3 . the larger dispersion hints at a potential difference in accretion properties between mbh agn and ` normal ' type 1 quasars . in the next subsection , we consider an analogy with wlqs to explore one potential mechanism for the larger dispersion in @xmath3 , namely accretion via a slim disk . for @xmath209 , accretion is expected to take place in the `` slim disk '' regime ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , where advective cooling losses are comparable to radiative losses , and the accretion flow becomes geometrically thick and radiatively inefficient ( see , e.g. , section 6 of * ? ? ? * for a brief review ) . @xcite propose that the unusual x - ray properties of wlqs ( i.e. , the large fraction of x - ray weak objects ) may be related to accretion via slim disks , as described below . @xcite initially suggested that wlqs contain a column of x - ray shielding gas that is very local to the black hole ( interior to the broad emission line region ) ; this x - ray shielding gas can produce highly x - ray weak objects at certain orientations , and x - ray normal objects at other orientations . @xcite went on to physically associate this shielding gas with the inner edge of a ( geometrically thick ) slim disk . in this picture , all wlqs are fed by an inner slim disk , but only the ones oriented such that we are viewing the central engine through the `` puffed up '' disk material appear to be x - ray weak . for these x - ray weak objects , the direct x - ray continuum will be highly absorbed , and any detected x - ray emission should be dominated by reflected / scattered light @xcite . this scenario results in the population of wlqs as a whole displaying a large observed dispersion in @xmath3 . it is tempting to appeal to a similar scenario to explain the large dispersion in @xmath3 observed for mbhs . however , if such a scenario were to apply to mbh agn , then we expect to see a systematic change in the x - ray properties of the 78-object mbh sample around @xmath210 , as the accretion flow transitions between a geometrically thick slim disk and a geometrically thin ( radiatively efficient ) @xcite disk . in particular , we should see less dispersion in @xmath3 at @xmath211 , in the context of a scenario where x - rays at these lower eddington ratios originate from inverse compton scattering of disk uv photons off a hot corona that is energetically coupled to a thin disk ( e.g. , * ? ? ? * ) . for @xmath211 , the higher uv flux from the thin disk would increase the number of inverse compton scatterings , thereby cooling the corona and also producing mildly steeper x - ray spectra on average ( e.g. , * ? ? ? a systematic change in the x - ray properties , as described above , is not observed among the mbh sample . however , when comparing @xmath116 estimates from different methods in table [ tab : lledd ] , we can not exclude the possibility that the eddington ratios adopted for both the low- and high-@xmath4 mbh samples are systematically underestimated , by perhaps up to an order of magnitude . if so , then nearly all of the high-@xmath4 sample would be in the super - eddington regime , and several of the lower-@xmath4 objects would fall close to the proposed `` slim disk '' transition . furthermore , we could also be systematically underestimating @xmath4 for our _ chandra _ targets if their virial - based @xmath158 estimates happen to be too large . thus , we can not exclude the possibility that the entire mbh agn sample could feasibly be more rapidly accreting than they appear to be in figure [ fig : aox ] , in which case wlqs may provide useful insight into understanding accretion onto the mbh sample considered here . we note that limited information on eddington ratios for wlqs seems to point toward @xmath212 ; @xcite ) . our x - ray stacking analysis of the x - ray non - detected mbhs yields a relatively hard x - ray spectrum ( @xmath213 ) . for the objects with small @xmath3 , the inner edge of the slim disk modifies the `` intrinsic '' x - rays associated with the agn corona , and any observed x - rays are likely dominated by reflection / scattering , such that a hard x - ray spectrum is expected . among wlqs , an x - ray stacking analysis of the subpopulation of x - ray weak objects reveals a hard x - ray spectrum as well ( @xmath214 ; * ? ? ? * also see @xcite ) . the similarly hard ( average ) spectra of the subsets of x - ray weak mbh agn and x - ray weak wlqs may futher support a wlq analogy . of course , the error bars on @xmath81 for our stacked mbhs are large ( see table [ tab : xstack ] ) . still , we find this to be an intriguing result , motivating a need for tighter x - ray spectral constraints for a sample of mbh agn spanning both high- and low- @xmath3 , in order to rigorously compare @xmath81 as a function of @xmath3 . if the wlq analogy holds , then we require samples of mbhs that are accreting even more weakly than our low-@xmath4 sample , in order to search for a slim - to - thin disk transition by searching for systematic changes in the x - ray properties described earlier ( i.e. , less dispersion in @xmath3 and steeper @xmath81 at lower eddington ratios in the thin disk regime ) . the prototype mbh ngc 4395 @xcite has a well - determined bolometric luminosity @xmath215 from a highly sampled broadband spectrum @xcite , providing @xmath216 for @xmath217 @xcite . intriguingly , ngc 4395 is not only accreting below the expected slim - to - thin disk transition at 0.10.3 @xmath4 , but it is also near / below another critical accretion regime at @xmath80.01@xmath218 where the disk is expected to switch from a thin disk to a radiatively inefficient accretion flow ( riaf ) , ) and low - accretion rate riafs ( @xmath219 ) are both radiatively inefficient , the physical reasons for their radiative inefficiencies are quite different . slim disks are radiatively inefficient largely because of photon trapping effects at near - eddington luminosities ; in most low - accretion rate riaf models , the low radiative efficiency is mainly due to weak coulomb coupling in the accretion flow ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? for other variants of riafs at low - accretion rates , also see , e.g .. @xcite . ] as described below . for agn fed by a thin disk , the x - ray photon index @xmath81 is correlated with eddington ratio when @xmath220 ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? below @xmath80.01 @xmath218 , @xmath81 and @xmath4 are anti - correlated , so that agn show the hardest x - ray spectra around @xmath221 , which may indicate a transition from a thin disk to a riaf around 1% @xmath218 ( e.g. , @xcite ; note that @xmath81 eventually plateaus to @xmath222 at the lowest eddington ratios , e.g. , @xcite ) . this x - ray spectral behavior is observed for stellar mass black holes in x - ray binary systems as well ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , suggesting that it is a universal feature of black hole accretion , regardless of black hole mass . ngc 4395 supports this trend for mbhs , as it displays a hard photon index @xmath223 at @xmath224 @xcite . we have no reason to suspect that the bolometric luminosity of ngc 4395 could be biased in the same manner as for the other mbhs ( since its @xmath174 is calculated from an observed broadband spectrum of the nucleus ) . the physical mechanism for the small @xmath81 for ngc 4395 is therefore different than the small @xmath81 discussed earlier in the context of a slim disk . this intriguing trend is of course far from robust being based on a single source , and it further motivates a need for high signal - to - noise x - ray spectra for mbhs across a wide range of @xmath4 . a sizeable population of lower-@xmath4 mbh agn is unlikely accessible from optical - selection techniques , however , and recovering such objects will require complementary multiwavelength searches . high - spatial resolution x - ray surveys ( especially when combined with the radio ) are a promising avenue for revealing weakly accreting black holes ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * also note galaxies 53 and 56 in the current work that show x - ray signatures of an agn , but lack optical photoionization signatures of activity . ) . success has also been achieved through infrared surveys ( e.g. , * ? ? ? * ) , as well as fast variability @xcite . we stress that it is unclear if the optically - based @xmath4 measurements are indeed systematically underestimated , and the above wlq analogy is only meant to represent one possibility . if the adopted @xmath4 estimates are accurate , then there does not appear to be a distinct eddington ratio that marks a transition in radiative efficiency . in that case , a range of accretion disk / corona properties may contribute to the dispersion in @xmath3 , which may include a substantial number of intrinsically x - ray weak agn . to confirm or refute current @xmath4 estimates , broadband spectral energy distributions for a substantially larger number of mbhs are required to properly constrain the bolometric corrections for this population . such a project will require high - resolution imaging across the entire electromagnetic spectrum to separate the agn from the host galaxy ( see for further discussion , as well as , e.g. , @xcite , for examples of well - sampled broadband spectra ) . we find no evidence for a difference in the x - ray properties of mbhs from @xmath225 . we argue that either ( 1 ) optically - based @xmath4 estimates are systematically underestimated and nearly all mbhs accrete from a geometrically thick , radiatively inefficient slim disk ; or ( 2 ) there is variety in the accretion details among individual objects , but there is no evidence for a systematic change in accretion properties at a specific eddington ratio . if mbhs indeed accrete from a slim disk , then super - eddington accretion could provide a mechanism for growing smbhs in the early universe ( e.g. , * ? ? ? finally , @xcite recently showed that a faint high - redshift agn population could produce enough flux to account for the epoch of reionization , provided that agn x - ray - to - uv luminosity ratios in the early universe are not too hard ( @xcite explicitly adopt @xmath226 ) . from an empirical perspective ( and regardless of the accretion mode ) , the observed flattening of the @xmath227 relation at low luminosities implies that accretion onto mbhs produces a broadband continuum in line with requirements for agn driven reioniziation . we thank the referee for suggestions that improved the presentation of this paper . we thank ohad shemmer for helpful discussions , and anna conlon for advice on our survival analysis . support for this work was provided by the national aeronautics and space administration through chandra award number go3 - 14116x issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . support for aer was provided by nasa through hubble fellowship grant hst- hf2 - 51347.001-a awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas 5 - 26555 . this research has made use of software provided by the chandra x - ray center ( cxc ) in the application package ciao . this research has made use of data obtained from the chandra data archive , and software provided by the chandra x - ray center ( cxc ) in the application packages ciao , chips , and sherpa . , a. , mcdowell , j. c. , allen , g. e. , et al . 2006 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol . 6270 , society of photo - optical instrumentation engineers ( spie ) conference series , 1 , g. p. , bautz , m. w. , ford , p. g. , nousek , j. a. , & ricker , jr . , g. r. 2003 , in society of photo - optical instrumentation engineers ( spie ) conference series , vol 4851 , x - ray and gamma - ray telescopes and instruments for astronomy . , j. e. truemper & h. d. tananbaum , 2844 , m. , isobe , t. , & feigelson , e. 1992 , in astronomical society of the pacific conference series , vol . 25 , astronomical data analysis software and systems i , ed . d. m. worrall , c. biemesderfer , & j. barnes , 245

the quantum scattering of magnon bound states in the anisotropic heisenberg spin chain is shown to display features similar to the scattering of solitons in classical exactly solvable models . localized colliding gaussian wave packets of bound magnons are constructed from string solutions of the bethe equations and subsequently evolved in time , relying on an algebraic bethe ansatz based framework for the computation of local expectation values in real space - time . the local magnetization profile shows the trajectories of colliding wave packets of bound magnons , which obtain a spatial displacement upon scattering . analytic predictions on the displacements for various values of anisotropy and string lengths are derived from scattering theory and bethe ansatz phase shifts , matching time evolution fits on the displacements . the time evolved block decimation ( tebd ) algorithm allows for the study of scattering displacements from spin - block states , showing similar scattering displacement features .

1507.08624

the study of classical dynamics in nonlinear media has proven to be a source of astonishing surprises over the last century . two observations , both based on numerical simulations , have challenged prejudices and fundamentally altered traditonal ways of thinking . first , the famous observation of fermi , pasta , ulam , and tsingou@xcite of a simple nonlinearly - coupled set of oscillators showing nontrivial recurrences , has shattered the long - held assumption that all nonlinear dynamical systems ergodically explore their full phase space . second , the pioneering numerical analysis of zabusky and kruskal @xcite on the korteweg - de vries equation @xcite demonstrated that this equation supports excitations , which they coined ` solitons ' , displaying a number of surprising fundamental features . the solitons are localized in space , with a form remaining stable under time evolution which sees them moving uniformly at a speed linearly proportional to their amplitude . additionally the astounding characteristic was observed of solitons emerging intact from mutual scattering processes , during which they simply `` ` pass through '' one other without losing their identity',@xcite the only effect of the collision being a relative spatial displacement as compared to their free propagation . the proper understanding of solitons in nonlinear media ultimately led to the development of the classical inverse scattering method , @xcite which is the overarching framework for classical integrable models . nonlinear classical systems find their quantum mechanical analogue in the shape of interacting many - body systems . in parallel to the classical case , some quantum models have been shown to be special , in the sense of being exactly solvable using the quantum version of the inverse scattering method @xcite ( one might also say integrable , though the quantum notion of integrability is not as well - defined as its classical counterpart @xcite ) . fundamental representatives of this family are the heisenberg spin chain , solved by bethe using what is now known as the bethe ansatz,@xcite along with the lieb - liniger model of @xmath0-interacting bosons on a line.@xcite for the latter , particle - like excitations called lieb type i modes @xcite exist due to the interparticle repulsion , along with type ii hole - like excitations visualized as holes in an effective fermi sea . it is possible to distinguish the presence of type i and ii modes in correlated bosonic gases in optical lattices using bragg spectroscopy.@xcite quantum magnets such as the heisenberg spin chain similarly carry particle - like magnon modes when the magnetization is close to saturation . in the limit of small magnetization , hole - like modes again appear , which in zero field are known as spinons.@xcite their dynamics can be experimentally observed using inelastic neutron scattering.@xcite the heisenberg chain supports distinct bound states of magnons , whose dynamics has been investigated theoretically@xcite and has recently been observed experimentally.@xcite one could view such excitations as the quantum equivalents of classical solitons . this equivalence is however only partial : on the one hand , these quantum mechanical modes represent exact eigenstates and are stable under time evolution ; on the other hand , being exact eigenstates of translationally - invariant systems , they are not spatially localized . that said , as is usually the case in quantum mechanics , it is possible to adopt a ` complementary ' picture and create spatially localized excitations by forming wave packets of fundamental excitations by linearly combining states over a range of differing momenta . locality however comes at a price : the wave packet , mixing together states at different energies , will disperse and is thus not stable over long timescales , unlike its classical counterpart . we use the term quasisoliton for such a wave packet construction , an example of which was recently studied in the context of the lieb - liniger model,@xcite while their mutual scattering has been studied in quantum spin chains.@xcite the spectroscopic methods traditionally employed to experimentally study condensed - matter systems typically provide momentum- and energy - resolved measurements . however , current experimental developments provide motivation to obtain a better theoretical understanding of spatially localized dynamics . time - resolved experiments are now able to track quantum many - body systems at timescales smaller than relaxation timescales , particularly in experiments with ultracold atoms,@xcite but also in nmr setups,@xcite and potentially in pump - probe spectroscopy experiments.@xcite as a result , non - dissipative dynamics of many - body systems out of equilibrium is now a rapidly growing field of experimental as well as theoretical investigation.@xcite novel cold - atom experimental techniques for spatially resolved manipulation and observation at the single - site level @xcite have opened the door to explicit high - resolution tracking of spatial propagation phenomena . moreover , an experiment with interacting bosonic atoms has highlighted the interaction - induced longevity of repulsive pairs . @xcite this has motivated increased theoretical attention to the ( anti-)binding of localized excitations and interactions of these bound clusters , both in itinerant systems@xcite and in spin chains.@xcite very recently , the spatial dynamics of itinerant clusters has been studied in experiment . @xcite propagation of quantum solitons in bose - hubbard chains has also been studied numerically . @xcite in ref . , the scattering of a magnon wave packet on approximate bound eigenstates of @xmath1 particles was studied numerically in the heisenberg chain . in the present work , we present an algebraic framework and exact calculations based on bethe ansatz . we therefore consider quantum scattering of localized excitations over a ferromagnetic background in the bethe ansatz solvable anisotropic spin-1/2 heisenberg chain ( xxz model ) @xcite @xmath2 . \label{eq : xxzhamiltonian}\ ] ] the xxz model is experimentally realizable , for example in the setups of refs . and , which use two hyperfine states of bosonic atoms in the mott phase to experimentally realize the spin - up and spin - down states . the effective model is a nearly isotropic ( @xmath3 ) heisenberg chain , while an experimental setup with variable anisotropy @xmath4 is under development.@xcite in addition , the xxz model has been shown to describe josephson junction arrays of the flux qubit type,@xcite and may also be realizable in optical lattices@xcite or with polaritons in coupled arrays of cavities.@xcite it is conceivable that these or similar experimental setups may provide time - resolved observations of propagating and interacting localized excitations . the parameter @xmath5 in hamiltonian is given by the exchange interaction of two neighboring electrons or , in the aforementioned experimental setup , by the exchange interaction between two neighboring bosons in an optical lattice at unit filling . its sign does not matter for our purposes here;@xcite for definiteness , we set it to @xmath6 . we distinguish two regions for the anisotropy parameter @xmath4 , namely the planar @xmath7 ( @xmath8 ) or axial @xmath9 ( @xmath10 ) cases . due to the interaction term ( @xmath11 ) the xxz chain displays a whole zoology of fundamental excitations : isolated down spins can form spatially bound states ( as was understood by bethe already in his original publication @xcite ) , whose bond size decreases as @xmath4 increases . these excitations are often referred to as ` string states ' , as the rapidities describing bethe states containing such bound multi - magnons appear as approximately equally - spaced vertical strings in the complex plane ( the precise set of available bound states depends on the value of @xmath4 ; at the isotropic point @xmath12 and in the axial regime , all string lengths are allowed ) . we construct spatially localized wave packets of @xmath1 bound magnons using linear combinations of these string states of length @xmath1 with gaussian - distributed momenta . we call them ` @xmath1-string wave packets ' . we further demonstrate that exact methods based on the algebraic bethe ansatz @xcite provide a framework to evaluate the time - dependent expectation value of the local magnetization @xmath13 algebraically , which can be used to track those localized magnon - like wave packets . we investigate their stability and mutual scattering using a combination of scattering theory , bethe ansatz , and numerically exact calculations . at large anisotropy @xmath14 , magnon bound states resemble having downturned spins on neighboring sites . an @xmath1-string wave packet is thus closely approximated by a consecutive block of @xmath1 downturned spins . while this correspondence breaks down at smaller @xmath4 , this provides motivation to study the evolution of states with downturned spins on a consecutive block of sites . in addition , this is exactly the type of initial state prepared in experiments.@xcite scattering of such blocks has been explored in ref . , where a spatial displacement of two sites was observed for the block at several @xmath4 , and explained at large @xmath4 in terms of energy conservation . in the present work , we connect scattering phase shifts with trajectory displacements in order to provide a bethe ansatz derivation of the observed displacements . the paper is organized as follows . in sec . [ sec : xxz ] we will introduce the concepts of string solutions and the scattering phase shifts associated with @xmath1-strings , along with details on the algebraic bethe ansatz@xcite based evaluation of the time - dependent expectation value of the local magnetization @xmath13 , which can be used to track localized magnon - like excitations of the spin chain . in sec . [ sec : wave packets ] , we elaborate on the construction , stability and time evolution of scattering of @xmath1-string wave packets . we also compare with the stability of consecutive - site spin blocks . in sec . [ sec : displacement ] we consider the scattering trajectory displacements , derive analytical results for them and compare with numerical measurements . the @xmath15 limit is treated analytically , comparing with numerical data for spin block initial states . the appendices provide details involving scattering theory and details on obtaining the phase shift directly from the phase of the time - evolving wave function . in this section we present the basic formulas of the bethe ansatz for the spin-1/2 xxz model and explain the concept of string solutions and the scattering phase shift associated with @xmath1-strings . in subsec . [ sec : timeevoaba ] , expressions are given for the time evolved expectation value of the local magnetization by using results from algebraic bethe ansatz . the method used here relies on the availability of determinant expressions for matrix elements between bethe states.@xcite this last subsection is relatively technical and could be skipped on first reading . the eigenstates of the xxz spin chain can be constructed via the bethe ansatz,@xcite and have the form @xmath16 where @xmath17 denotes the number of downturned spins and therefore fixes the magnetization . the sum over @xmath18 is a sum over all permutations of @xmath17 objects and the amplitudes @xmath19 are related to the scattering phases . the set of @xmath17 complex rapidities @xmath20 completely determines the bethe state and is simply related to the physical energy and momentum . by imposing periodic boundary conditions , i.e. @xmath21 for @xmath22 in the hamiltonian , each set of rapidities @xmath23 corresponding to an @xmath17-magnon eigenstate must obey bethe equations , @xmath24 the different definitions of @xmath25 , @xmath26 and @xmath27 for various regions in anisotropy @xmath4 are given in tab . [ table : kerneldef ] . within bethe ansatz , the momenta of single downturned spins can be parametrized in terms of rapidities , @xmath28 = \pi - \theta_1(\lambda)\,.\ ] ] .definitions of functions appearing in bethe ansatz for different regions in anisotropy @xmath4 . [ cols="<,^,^,^",options="header " , ] by invoking schrdinger s equation @xmath29 for a bethe state consisting of a single downturned spin , the magnon dispersion relation @xmath30 is easily derived . in the case of just two single magnons , their scattering phase shift @xmath31 can be obtained from the permutation of two magnons in the schrdinger equation , @xmath32 where @xmath18 is the identity and @xmath33 interchanges the two indices @xmath34 and @xmath35 . furthermore , the magnon momenta @xmath36 and @xmath37 as well as the scattering phase shift @xmath38 are parametrized by the two rapidities @xmath39 and @xmath40 . the sets of rapidities solving bethe eqs are self - conjugate and arrange themselves in patterns of string solutions , @xmath41 where the string center @xmath42 and @xmath43 is the internal label of a rapidity within a string of length @xmath1 and parity @xmath44 . in the planar regime @xmath8 , periodicity of the trigonometric functions also allows for string centers to be located on the line @xmath45 , resulting in negative parity strings ( @xmath46 ) . this type of strings will be left out of consideration for the analysis of scattering magnons , restricting to @xmath47 . at finite size , solutions are not exactly given by strings , but rather contain string deformations @xmath48 , under the constraint that the full set of rapidities @xmath23 remains self - conjugate . in the cases considered here the deviations are exponentially small in system size and it is therefore sufficient to take the limit of vanishing deviations . in this limit , the product of the bethe equations of all rapidities within a string reduces to the bethe - gaudin - takahashi equations,@xcite which are similar to the bethe equations but given in terms of the @xmath1-string centers @xmath49 . in logarithmic form they read @xmath50 where @xmath51 denotes the number of @xmath1-strings present , satisfying @xmath52 . the logarithmic scattering kernels @xmath26 are defined in tab . [ table : kerneldef ] and the scattering phase shift between two individual strings of arbitrary length is given by ( for @xmath53 we consider only strings with positive parity ) @xmath54 the logarithmic form of the bethe - gaudin - takahashi equations allows for the introduction of string quantum numbers @xmath55 , obeying an exclusion principle for all bethe states , meaning that every bethe state is characterized by a unique set of string quantum numbers . by considering the limit of sending a string center to infinity , the maximum allowed string quantum number can be derived . these limiting quantum numbers define the dimensions of sub - sectors of the hilbert space containing a specific string content . moreover , in the planar case @xmath53 , the existence of strings with a specific length @xmath1 in the spectrum is determined by the anisotropy.@xcite therefore restrictions on the availability of @xmath1-string wave packets as well as on their momenta@xcite are present in the planar case @xmath53 , depending on the value of @xmath4 . we get all bethe states with the desired string content by solving the bethe - gaudin - takahashi equations using an iterative algorithm for all combinations of allowed string quantum numbers . after obtaining the rapidities , the energy of a bethe state containing strings is easily computed as @xmath56 where @xmath57 is the derivative of @xmath58 . the energy contribution @xmath59 of a string of length @xmath1 to the energy @xmath60 is @xmath61 where @xmath62 for @xmath53 , @xmath63 , and @xmath64 , respectively . the momentum @xmath65 of an @xmath1-string with string center @xmath66 is given by @xmath67 in the following we use the convention @xmath68 . note that for a single @xmath1-string the minimum of the energy dispersion is always at @xmath69 , i.e. at momentum @xmath70 . the total momentum of a bethe state can be extracted from its string quantum numbers , @xmath71 in the thermodynamic limit and for a finite number @xmath17 of rapidities , each separate bound magnon represented by a string quantum number @xmath55 can be associated to a single particle momentum @xmath72 @xmath73 to summarise , the rapiditites belonging to each eigenstate are obtained by iteratively solving the bethe - gaudin - takahashi equations . they can be used to evaluate the determinant expressions for the normalised matrix elements of the following section . time evolution of the expectation value of the local magnetization @xmath13 is performed by making use of the algebraic bethe ansatz.@xcite the time dependent wave function is computed using unitary time evolution in a basis of bethe states @xmath74 , @xmath75 where the coefficients @xmath76 are determined by the initial state @xmath77 , which is given in sec . [ sec : wave packets ] , eq . , for the construction of @xmath1-string wave packets . the expectation value of the local magnetization at site @xmath78 is given by @xmath79 as the states @xmath74 , @xmath80 are bethe states determined respectively by sets of rapidities @xmath81 , @xmath82 that obey bethe eqs ( here we have @xmath17 rapidities in both sets since the operator @xmath83 does not change the magnetization ) , the matrix elements are given by the normalised expressions obtained from algebraic bethe ansatz @xmath84 here we make use of the determinant expressions obtained in ref . , @xmath85}{\prod\limits_{\substack{k , l=1\\ k < l}}^m \phi_0(\mu_k-\mu_l)\phi_0(\lambda_l-\lambda_k)}\ , , \label{eq : matrixelements}\end{aligned}\ ] ] with @xmath86 and @xmath25 defined in tab . [ table : kerneldef ] . the entries of the matrices @xmath87 and @xmath88 are given by @xmath89^n\prod_{\substack{k=1\\ k\neq a}}^m \phi_{-2}(\mu_k-\lambda_b)\big]\ , , \\ p_{ab}(\{\mu\},\{\lambda\ } ) & = \frac{\phi_2(0)}{\phi_{-1}(\mu_a)\phi_1(\mu_a)}\prod_{k=1}^m \phi_2(\lambda_k-\lambda_b)\,.\end{aligned}\ ] ] the normalization @xmath90 is computed from the gaudin determinant,@xcite @xmath91^m \prod\limits_{\substack{k , l=1\\ k \neq l}}^m \frac{\phi_2(\lambda_k-\lambda_l)}{\phi_0(\lambda_k-\lambda_l ) } \det \phi(\{\lambda\})\ , , \label{eq : gaudinnorm}\ ] ] where the gaudin matrix is given by the jacobian of the bethe equations , @xmath92\nonumber \\ & \quad + \theta_2^{\prime}(\lambda_a-\lambda_b)\ , . \end{aligned}\ ] ] the time - dependent expectation value of the local magnetization is then obtained by evaluating the double sum over matrix elements in eq . . by construction , the double sum only includes eigenstates with the same particle content , which is not large for the few - magnon states we will consider . as a result , the double summation is still tractable at lattice sites @xmath93 . in the case of dealing with string solutions for the magnon bound states , reduced determinants for strings described in ref . must be used . the strings described in the previous section do not correspond one to one to localized bound states of downturned spins , but rather are translationally invariant constituents of bethe eigenstates . in order to create states of @xmath1 bound magnons with localized magnetization features , we construct gaussian wave packets by summing over single @xmath1-string states ( labeled by the string center @xmath94 ) with momenta distributed around @xmath95 , @xmath96 where @xmath97 is a normalization constant . unitary time evolution under hamiltonian implies an expression of the velocities of the wave packets by expanding the dispersion relation around @xmath98 to first order ( see appendix [ app : sec_scattering ] ) , @xmath99 the prefactor can be expanded for large anisotropy , @xmath100 implying that wave packets constructed from higher strings have a lower velocity in real space . in the remainder of this section we will analyze the stability of @xmath1-string wave packets , along with the stability of another form of a localized multi - magnon , a consecutive - site spin block . the two constructions are closely related to each other for large @xmath4 . subsequently , scattering processes of @xmath1-string wave packets are visualised by computing the time evolution of the local magnetization . furthermore , we describe a method for direct observation of the phase accumulated by the wave function of a finite spin chain during a scattering process . the center of the wave packet @xmath101 and its width @xmath102 are respectively given by the expectation value and the variance of the position operator @xmath103 in the continuum limit , the sum over all possible @xmath1-string momenta @xmath104 can be approximated by an integral . for the width of the gaussian wave packet in real space we obtain @xmath105 ^ 2 |\psi(0)\rangle } \approx \frac{\alpha}{2 } \,.\ ] ] string solutions , being associated with bound states of magnons with exponentially decaying wave functions , will add exponential terms to the shape of the @xmath1-string wave packets in real space . by inserting the complex momenta of the individual constituents of the string solutions to the bethe wave function , it can be shown that for example a 2-string state with momentum @xmath106 contains exponentials in the bethe wave function reading @xmath107 , with an effective binding length @xmath108 . this average distance between the constituent particles in the string states provides a lower bound to how localized the wave packets constructed out of such string states can be . for @xmath109 , the wave packet will start to lose its gaussian shape in real space in favour of a simple exponential decay around the wave packet center . this issue can be circumvented by choosing large enough gaussian widths , but will require much higher system sizes . the effective binding length becomes smaller at higher anisotropy , making the problem of the extra exponentially decaying shape to the gaussian wave packet construction relevant only at @xmath110 . induced by the nonlinear dispersion relation of the magnons , the width of a wave packet in real space will furthermore increase in the course of time ( see eq . ) , @xmath111 obtained by expanding the dispersion relation ( eq . ) to second order around the average momentum , yielding @xmath112 . the broadening of @xmath1-string wave packets in real space is therefore described by the initial width @xmath113 and @xmath114 to first order , all strings of arbitrary length are stable at momenta @xmath115 , but possess a non - trivial dependence on the anisotropy for other momenta . moreover , the broadening of 1-string wave packets is not influenced by the anisotropy . the anisotropy dependent factor can again be approximated in the large anisotropy limit given by eq . , showing that the stability of the wave packets increases with increasing anisotropy and string length . in fig . [ fig : lifetime ] the magnetization profile of a diffusing 2-string wave packet with zero group velocity computed from algebraic bethe ansatz is shown . we used as average momentum @xmath116 since the energy dispersion relation has its minimum there . furthermore , fitted parameters on the time - dependent wave packet width are compared to the theoretical values of @xmath117 for 2- and 3-strings . the diverging behaviour at low anisotropy of both @xmath118 and the effective binding length of strings constrain the applicability of scattering theory results for the planar regime @xmath8 . around @xmath116 , where the energy dispersion has its minimum ) , in a chain of @xmath119 sites , calculated from algebraic bethe ansatz matrix elements . right : prefactor @xmath120 of the @xmath121 dependence of the wave packet width . theoretical curve , eq . with @xmath116 and @xmath122 , as a function of @xmath4 . the data points are retrieved by fitting @xmath120 from the decaying wave packets from the bethe ansatz time evolutions . ] a similar or even more drastic dispersing behaviour can be observed for blocks of @xmath1 adjacent sites . [ fig : decay ] shows the time evolution of a block of 20 upturned spins in a ferromagnetic chain of downturned spins for different anisotropy parameters @xmath4 . the results shown in figs . [ fig : decay ] and [ fig : scattering ] were obtained using the time evolving block decimation ( tebd ) algorithm . @xcite in the axial regime @xmath123 , the initial state mostly projects onto many - magnon bound states , namely @xmath1-strings with @xmath124 , which for such anisotropy values are tightly bound and thus have a large overlap with the initial spin block . a quantitative analysis of the overlap between the initial spin block and large strings is provided by refs . and for a comparable situation involving a prepared domain wall state containing @xmath17 consecutive down - spins on a polarized background . the overlaps between the @xmath17-spin block state and a few string configurations are considered , where the normalization saturation becomes entirely dominated by the @xmath17-string states with increasing anisotropy . since the spin blocks mostly contain large strings for large anisotropy , they display slow dispersion ( see eq . ) , meaning that the initial spin block remains more or less intact in time over long time scales . however , as the isotropic point @xmath12 is approached , the nature of the overlaps drastically changes . eigenstates with combinations of smaller strings start carrying a larger fraction of the total overlap with the initial state . under time evolution , one thus sees spacetime propagation lines corresponding to shorter strings , which disperse more rapidly . by the time one has entered the planar regime , @xmath125 , the initial spin block decomposes into all available string lengths including the most rapidly - dispersing 1-string states , leading to a rapid dispersion of the magnetization throughout the ` light cone ' defined by the maximal group velocity of the 1-strings . sites for different values of anisotropy , computed using tebd . around the isotropic point @xmath63 , the behaviour of the spin block changes drastically , as the spin block becomes less tightly bound and will start to decay into smaller strings . ] a comparison of figs [ fig : lifetime ] and [ fig : decay ] ( both at @xmath116 ) shows that spin blocks decay faster than @xmath1-string wave packets of the previous subsection , which has a twofold explanation . first , an @xmath1-string wave packet is a superposition of @xmath1-string bethe states which have no decay channel into strings of smaller lengths . in contrast , in the superposition of the initial spin block state all smaller string lengths are allowed and actually present . second , in the momentum distribution of the spin block , momenta belonging to high velocities are not gaussian - like suppressed . therefore , states with high velocities ( @xmath126 ) are more dominant in the spin block state than in the @xmath1-string wave packet both with @xmath116 . however , the spin - block state and the @xmath1-string wave packets become similar at higher @xmath4 , such that the analytic predictions on scattering displacement becomes applicable to both cases . the pre - scattering bound magnon initial states @xmath77 are composed of two gaussian wave packets , where the construction relies on allocation of individual momenta to distinct strings within a single bethe state according to eq . . we localize two gaussian wave packets labeled by @xmath127 with average momenta @xmath128 around two well - separated lattice sites @xmath129 where @xmath130 for simplicity , we label bethe states here only by the two string centers @xmath94 and @xmath131 instead of the whole set @xmath132 . the two centers are respectively uniquely determined by the momenta @xmath133 and @xmath134 via eq . . due to the energy dispersion of the bound magnons , the relative velocity of the wave packets is maximized at @xmath135 . [ fig : heatmaps ] shows time - dependent magnetization profiles for scattering of 1- and 2-string wave packets respectively , computed from the algebraic bethe ansatz matrix elements described in sec . [ sec : timeevoaba ] . a close examination of the profiles shows distinctive features akin to soliton scattering , namely that the wave packets emerge out of a collision intact , but spatially displaced . this displacement will be quantified in the next section . illustrating scattering @xmath1-string wave packets , computed using algebraic bethe ansatz , for @xmath119 , @xmath136 , @xmath135 and @xmath137 . left : scattering of 1-string wave packets ( single magnons ) . right : scattering of 2-string wave packets ( bound magnons ) . ] [ fig : initial ] [ fig : heatmaps ] the real - time scattering trajectories in this work are analyzed using the idea that the scattering between two localized wave packets is well - described by the scattering phase shift corresponding to the average momenta of the two wave packets . this idea can also be verified through direct observation of the phase accumulated by the wave function of a finite chain during a scattering process . we therefore compare the evolution of an interacting chain ( @xmath138 ) with the evolution of a non - interacting chain ( @xmath139 ) , each containing two localized 1-string wave packets . the overlap between the two wave functions , @xmath140}(t ) | \psi^{[\delta]}(t ) \rangle \,,\ ] ] gives the phase acquired due to the interaction between the magnons . the value of @xmath4 is here indicated in the superscript . in appendix [ app : phaseshiftdirect ] we show how the phase of the quantity @xmath141 after a single scattering event , calculated using numerical exact diagonalization , matches the bethe ansatz phase shift of eq . , even for wave packets that are spatially well localized . such overlaps between time - evolved wave functions are currently not directly accessible by bethe ansatz . the initial state constructed from bethe states is prepared as two wave packets of bound states of an arbitrary finite number of magnons with initial average positions and momenta @xmath142 . the @xmath143- and @xmath1-string wave packets are constructed separately at large separation , such that their motion before and after scattering can be considered to be free . in particular , the motion of the center of each wave packet is ( see appendix [ sec : app_general ] ) @xmath144 given in units of lattice distance , where the velocity @xmath145 was defined in eq . . note that , in the case of single magnon scattering for example , a negative average momentum @xmath128 yields a positive velocity @xmath145 and vice versa ( see eq . with @xmath5 positive ) . the displacement @xmath146 can be obtained by expanding the scattering phase shift around the average momenta , see also eq . , and is therefore given as @xmath147 where we introduced the notation with subscript @xmath78 to refer to the momentum derivative @xmath148 of the scattering phase shift @xmath31 . assumes that all scatterings occur without particle production , which is the case for the integrable model we are dealing with . an analytic expression for the displacement as a function of anisotropy and incoming momenta can be extracted from the bethe ansatz scattering phase . the phase shift of two bound magnons of arbitrary length is obtained from the scattering kernel @xmath149 of the bethe - gaudin - takahashi eqs , which consists of a sum over the functions @xmath150 , @xmath151 , defined in tab . [ table : kerneldef ] , see also eq . . for the scattering of an @xmath1-string wave packet labeled by @xmath34 with an @xmath143-string wave packet labeled by @xmath35 , the displacements on the trajectories of the @xmath1- and @xmath143-string wave packets ( @xmath127 respectively ) are given as a function of momenta as @xmath152 the @xmath1- and @xmath143-strings with centers @xmath94 and @xmath131 carry momenta @xmath153 and @xmath154 , respectively . the wave packets are located such that @xmath155 and @xmath156 . we first discuss the planar case @xmath157 . by inverting eq . , we express the string center of an @xmath1-string in terms of its momentum as @xmath158 due to eq . , expression for the displacement consists of a sum of the momentum derivative of the functions @xmath150 of which the individual terms are computed as @xmath159 for the scattering phase shift of two 1-string wave packets with @xmath160 , the displacement is given as a function of anisotropy as @xmath161 the latter result could have been also obtained directly by taking the derivative of eq . . similarly , the scattering displacement of two 2-string wave packets becomes @xmath162 the validity of the latter equation only extends to the region where @xmath163 , as 2-strings with momentum @xmath164 do not exist for lower anisotropy@xcite , which can be shown from the anisotropy dependent maximum string quantum numbers . for the regime @xmath123 , eqs and hold as well , since the @xmath165 for both regimes are just rotated in the complex rapidity plane with respect to each other . starting from @xmath165 for @xmath64 with @xmath166 therefore yields identical results for the scattering displacements . the displacements for the scattering of an 1-string wave packet at a 3-string wave packet is given by @xmath167 the displacement in the trajectories induced by scattering effects is easily deduced from the time evolution data of fig . [ fig : heatmaps ] . for symmetric cases with identical particles , the average position of the magnetization of a single wave packet can be computed on one half of the system as a function of time , @xmath168 the result for the scattering between two @xmath35-string wave packets is plotted in fig . [ fig : scatt ] . note that the plotted average location of the wave packet in fig . [ fig : scatt ] by means of eq . does not resemble the actual trajectories when the wave packets spatially overlap with each other , as the trajectories are not properly defined during the scattering event . a linear fit with the same slope is applied to the propagation of the average position of the wave packet before and after scattering . the horizontal difference between the two straight lines is the displacement of the wave packet due to scattering effects . the procedure of measuring the displacements by means of eq . was performed for multiple values of anisotropy @xmath4 . moreover , these results can be compared with eq . , where the displacements for the situations of fig . [ fig : scatt ] are specifically given by eqs and as function of anisotropy . the results are plotted in fig . [ fig : shift ] as well , showing agreement between both approaches . the explicit time evolution relying on algebraic bethe ansatz matrix elements provides confirmation of the analytical predictions for the displacements . , by computing the average location of the wave packets according to eq . and taking the horizontal difference of the linear fits . right : scattering displacement ( in units of lattice distance ) for single and bound magnons as function of anisotropy . measured data from algebraic bethe ansatz time evolution of magnetization ( see left panel ) compared to the derivative of scattering phase shifts , see eqs - . ] [ fig : scatt ] besides symmetric scattering situations , colliding distinct @xmath1-string and @xmath143-string wave packets can be constructed and traced in the time - evolved magnetization profile . [ fig : multiplot_one_nstring ] shows the scattering between an 1- and a 3-string wave packet . two situations are distinguished , the former with both wave packets at maximal velocity at @xmath135 , where the larger string moves much slower because of its effective mass , see eq . . the latter situation consists of an incoming 1-string wave packet scattering on a stationary wall of a 3-string wave packet at @xmath169 . the corresponding analytic scattering displacements , eqs - , are shown adjacent to the time evolution plots in fig . [ fig : multiplot_one_nstring ] . the scattering displacements are measured from the time evolution by imposing a gaussian fit on the wave packet after scattering and comparing the average location to the time evolution of a single wave packet without scattering . in the lower right panel ( @xmath169 ) , the fitting procedure of the average location of the 3-string wave packet becomes less accurate for decreasing @xmath4 . in the planar regime where @xmath53 , we encounter substantial limitations ( as described in sec . [ sec : stability ] ) on both the construction of the scattering wave packets , as well as on the comparison to results of scattering theory . due to the effective binding length of the individual constituents of the string states , the tails of the magnetization profile of the wave packets start overlapping with each other significantly at low @xmath4 , invalidating important assumptions of scattering theory which include asymptotic separation of the wave packets before and after scattering . comparison of the measured displacements to scattering theory is therefore not meaningful for higher strings in the planar regime . only going to much larger system sizes would resolve the aforementioned issue . for a single magnon wave packet with momentum @xmath170 scattering against a 3-string wave packet with momentum @xmath171 ( top ) and @xmath172 ( bottom ) respectively , with @xmath119 , @xmath136 and @xmath137 . right : corresponding scattering displacements ( in units of lattice distance ) as function of anisotropy , see eqs - , where the points are measured displacements obtained from the time evolution data . ] the scattering displacement turns out to have a particularly simple form at large anisotropy . in particular , it was found in ref . that when a propagating @xmath1-particle cluster is incident on a larger block , the block is displaced by @xmath173 sites . this can be explained from the bethe ansatz results presented in previous subsections , by taking the ising limit @xmath174 of eq . for the displacement of an @xmath1-string wave packet scattering at an @xmath143-string wave packet . first , we obtain for all @xmath175 @xmath176 which is independent of @xmath1 , @xmath143 , @xmath133 , and @xmath134 . using eq . for the phase shift between an @xmath1-string and an @xmath143-string eventually yields @xmath177 thus , the scattering displacement for unequal wave packets ( @xmath178 ) is equal to twice the number of particles in the smaller wave packet . the leading order term of the ising limit in eq . can be given for the case where @xmath179 as @xmath180 yielding an error estimate for eq . . a systematic expansion for all momenta becomes cumbersome due to the summation in eq . and is left out of consideration . if @xmath181 , the leading contribution is given by the third line of . if further @xmath182 , the leading order terms will be formed by higher powers like @xmath183 for @xmath184 or @xmath185 for @xmath186 . [ fig : scattering ] shows tebd results of the scattering of a two - spin block excitation with a block consisting of 10 adjacent spins at @xmath187 . the initial 2-spin block was created by upturning two neighboring spins at lattice sites 2 and 3 . the first site ( with open boundary conditions ) is energetically inaccessible for large @xmath4 ; hence the 2-spin block travels to the right and is incident on the 10-spin block . the displacement is clearly by 4 sites , as predicted by eq . . a similar displacement by 2 sites in the case of a single incident particle was highlighted in ref . . although the 2-spin block and the 10-spin block are not explicity prepared as wave packets in this case , at large @xmath4 , string states are tightly bound , and therefore these blocks may be understood intuitively to be close to 2-string wave packets and 10-string wave packets . the ising limit eq . thus provides a satisfactory explanation to the shift observed in these numerical experiments . even more , it was observed @xcite that the behaviour of the scattering displacement of a large spin - block state close to the isotropic point at @xmath188 still resembles the scattering behaviour of the large @xmath4 limit by shifting the spin - block by two lattice sites upon scattering . in order to explain this from bethe ansatz , we take the limit of eq . for the scattering of a large string ( @xmath189 ) with center @xmath190 and an 1-string with center @xmath191 at all values of @xmath64 , @xmath192 \left(1 + \mathcal{o}(e^{-s \zeta})\right ) \left(1 + \mathcal{o}(e^{-m \zeta})\right ) \,,\end{gathered}\ ] ] where @xmath175 can only take on the values @xmath193 or @xmath194 , due to eq . . the latter equation finally gives for the displacement of the large @xmath143-string at @xmath64 , @xmath195 implying that already at @xmath196 ( @xmath197 ) , the scattering displacement of a large spin block should still be close to two sites , as is in agreement with the aforementioned observation . . the block of 10 upturned spins is shifted by four lattice sites upon impact of the 2-string like bound magnon . this is in correspondence with eq . , yielding a shift of four caused by a 2-string on a larger string . in this work , we have studied the quantum analogue of soliton - like scattering phenomena in the anisotropic spin-1/2 heisenberg chain , by utilising the algebraic bethe ansatz . we considered quantum scattering of localized excitations , created from linear combinations of bethe states with gaussian - distributed momenta , constructing wave packets of @xmath1 bound magnons . this construction allows to study scattering phenomena of wave packets containing an arbitrary number of bound magnons . exact methods based on the algebraic bethe ansatz provide a framework to evaluate the time - dependent expectation value of the local magnetization profile , which allows for a spatial tracking of the localized excitations . this explicit unitary time evolution of the initial state relies on the availability of determinant expressions for matrix elements of local spin operators . the algebraic bethe ansatz time evolution of colliding wave packets of bound magnons displays a spatial displacement in the trajectories of the wave packets under scattering , consistent with scattering theory results . for different values of anisotropy , fits on the displacements of the time evolved trajectories are in agreement with analytical results on the displacement from the derivative of the bethe ansatz scattering phase shifts , for several combinations of string lengths . the scattering phase shift can also be measured directly as well for the scattering between two localized single - magnon wave packets , again matching phase shift expressions provided by bethe ansatz . using tebd , scattering displacements from spin - block states have been studied , showing similar scattering features and validating the analytic predictions of the ising limit for the scattering displacement . the experimental realizability of real time tracking of localized excitations in the heisenberg spin chain@xcite might provide an opportunity to study dynamical scattering phenomena of ( bound ) magnons . a possible manifestation of such phenomena might be provided by the soliton - like scattering effects analysed in this work . the results on the scattering displacements can be extended to other bethe ansatz solvable models . finally , the time evolution method relying on matrix element expressions from algebraic bethe ansatz can be used to construct other initial states in spin chains as well and to study their respective relaxation phenomena . we would like to thank i. bloch , f. essler and g. mussardo for useful discussions . r.v . , d.f . , m.b . , and j .- s.c . acknowledge support from the foundation for fundamental research on matter ( fom ) and from the netherlands organization for scientific research ( nwo ) . m.g . and h.g.e . acknowledge support from the fwf , sfb vicom f4104 . the authors would furthermore like to thank the mpi pks for hospitality during the workshop `` quantum systems out of equilibrium '' in 2013 , and the crm for their hospitality during the beyond integrability " workshop in 2015 in this appendix , we review some general results from quantum scattering theory , emphasizing the direct connection between physically observable quantities and ( derivatives of ) the scattering phase shift.@xcite we consider an initial state with two well - separated quasiparticles ( e.g. magnons and magnon bound states ) with almost well - defined positions @xmath198 and momenta @xmath199 . in the asymptotic region , i.e. when the distance between the two wave packets is much larger than the radius of the interaction , the time evolution is free . hence , the centers of the wave packets @xmath200 translate rigidly . the scattering between the two particles however introduces a displacement proportional to the derivative of the scattering phase @xmath201 . more precisely , in the asymptotic regions , the motion of the center of each wave packet is @xmath202 where @xmath203 is the displacement , while the group velocity @xmath145 is given by the derivative of the dispersion relation , i.e. @xmath204 . similarly , the scattering has an effect on the width of each wave packet @xmath205 ^ 2 \rangle$ ] . for gaussian wave packets we have to first order @xmath206 where @xmath207 , while @xmath208 and @xmath209 . an analogous formula holds for @xmath210 as well . scattering also builds up correlations between the ( initially uncorrelated ) gaussian wave packets , as can be seen from the time evolution of the correlator @xmath211\ , \left [ \hat{x}_2(t)-x_2(t)\right]\rangle$ ] , @xmath212 & \quad \textrm{after scatt . } \label{eq_corr } \end{array}\right.\ ] ] all the aforementioned quantities carry information about the derivatives of the scattering phase and can be in principle measured in a scattering experiment . these results hold for any one dimensional theory with stable particles . for the xxz spin chain the stability of magnon bound states above the binding energy threshold is preserved by the integrability of the theory . in this appendix , we review some general results for the scattering of two particles ( magnons , magnon bound states , etc.),@xcite and derive eqs - . for simplicity , we consider the scattering of two distinguishable particles in a continuum integrable model . the same results can be obtained for identical particles . so , let us consider two particles with asymptotic momenta @xmath133 and @xmath134 and different dispersion relations @xmath213 , @xmath214 in an infinite volume ( zero density ) . the statement of the coordinate bethe ansatz , see eq . , is that in the asymptotic region where the two particles are very far apart the eigenfunctions of the system are plane waves , @xmath215 where @xmath216 is the scattering matrix and @xmath201 the scattering phase shift . at zero density , the energy is simply @xmath217 let us briefly comment on the structure of the wave function . first of all , as we discussed in the main text for the xxz model , bound states are characterized by complex conjugate rapidities , which leads to exponentially decaying terms in the bethe wave function with respect to the relative coordinate . this is a feature of bound states that are characterized by a center of mass coordinate . in what follows , we do not denote these exponentially decaying terms and label the bound states only with the position of the center of mass . for elementary particles , eq . is a consequence of the conservation of energy and momentum in one dimension , and as such it is valid for any model with a sufficiently short - range potential . instead , if one of these particles is not elementary but it is a bound state , then for a general theory the previous simple form of the wave function is not true anymore . the bound state can decay and scattering be diffractive . however , there exist models ( such as the xxz spin chain ) for which the scattering is always non diffractive . hence , the coordinate bethe ansatz describes a complete set of asymptotic eigenfunctions , as thoughtfully discussed in sutherland s book.@xcite for such theories , bound states can not decay , but are protected by integrability . let us consider the scattering problem . at time @xmath218 the two particles are far apart and have ( almost ) well - defined positions @xmath219 and momenta @xmath128 , @xmath127 . without loss of generality , we may assume that @xmath220 and @xmath156 , where @xmath145 is the group velocity , @xmath221 when they are far apart , the two wave packets move with velocities @xmath145 . if @xmath222 , the evolution is always free , while for @xmath156 at some time the two particles become close and the interaction plays a role . the time evolution of a two - body wave function is given by @xmath223 \label{eq_time}\,,\end{aligned}\ ] ] where @xmath224 @xmath225 is a complete set of eigenfunctions , and @xmath226 is the initial wave function . now , we consider an initial state which is factorized , @xmath227 where @xmath228 since we are interested in computing also the spreading in time of the wave packet , we assume that the functions @xmath229 are gaussian wave packets with minimal indetermination , @xmath230 where @xmath231 . this assumption may be dropped if we are interested only in the displacement , eq . . up to now , everything is exact . however , we can make the following approximations : * since the initial state @xmath226 is sharply peaked around @xmath232 and @xmath233 with @xmath220 , the only relevant contributions to come from the regions around these two points . we can then use the asymptotic formula for @xmath234 with @xmath235 . inserting eq . into , and using definition yields @xmath236 * since the functions @xmath237 , @xmath127 , in are peaked around @xmath238 , we can expand the dispersion relations @xmath239 around these momenta in the integrand of eq . . moreover , for @xmath240 we may substitute the asymptotic expression for @xmath225 into and similarly expand the scattering phase shift @xmath201 . up to second order these expansions read @xmath241 with @xmath242 and @xmath243 , and we sum over repeated indices . these expansions up to the second order are physically meaningful only if we can ignore the distortion of the wave packet , and so they are no more valid for times long enough . taking advantage of these approximations , we are now in the position to derive eqs - . before scattering , the two wave packets propagate freely . since they start around @xmath232 and @xmath233 with @xmath220 , they are centered around @xmath244 with @xmath245 for small times . thus , we can use the asymptotic formula of @xmath234 valid for @xmath235 in eq . , and hence perform the gaussian integrations thus obtaining the pre - scattering results , eqs - . similarly , long after the scattering , we have @xmath246 . thus , taking advantage of the proper asymptotic formula for the eigenfunctions and the expansion of the scattering phase shift we also obtain the post - scattering formulas - . left : time evolution of @xmath247 , performed using numerical exact diagonalization for @xmath248 sites . two single - magnon wave packets are prepared with initial width @xmath249 and opposite momenta @xmath250 ; here @xmath251 and @xmath252 . right : the bethe ansatz phase shift ( lines ) are compared with the phase of @xmath253 ( symbols , shifted by @xmath254 ) , obtained from the wave function of the time - evolved @xmath248 chain . the comparison is shown for three @xmath255 values . ] a prominent theme of this work is that real - time scattering between two localized wave packets is well - described by the phase shift corresponding to the average momenta of the two wave packets . in this appendix we connect real - time scattering data for two single - magnon ( 1-string ) wave packets in a finite chain directly to the definition of the scattering phase shift , namely , that the scattering phase shift is the phase picked up by the system wave function during the scattering process . to demonstrate the robustness of this idea , we show results from a stringent situation of rather small wave packets ( width of a few sites ) in a rather small chain ( @xmath256 sites ) , far from the usual idealized limit of infinitely extended excitations . here the hamiltonian of the xxz model is used in the following form : @xmath257\,.\ ] ] the extra terms compared to eq . are convenient for considering the phase of the time - evolving wave function in the two - magnon sector . when evolving with the hamiltonian , there is a constant accumulation of @xmath4-dependent phase in the time evolution even when the magnons are spatially separated ; this is avoided with the above form . with this form , the interaction affects the phase of the chain wave function only when the magnons meet each other . in the initial state , each magnon wave packet is prepared as a gaussian , localized approximately around @xmath258 and @xmath259 respectively , with opposite momenta @xmath260 . the initial state is thus @xmath261 e^{-ikx } \\ \times \sum_y s_y^{- } \ ; \exp\left[-\frac{(y - y_0)^2}{\alpha^2}\right ] e^{+iky } \quad { \lvert\uparrow\uparrow \ldots \uparrow\rangle } \qquad\end{gathered}\ ] ] with @xmath262 and @xmath263 near the center of the left half and right half of the chain . here @xmath264 and @xmath265 are used as discrete site indices , and @xmath266 is a normalization constant . the definition of the width @xmath113 is chosen to be consistent with @xmath113 in the main text . to avoid excessive dispersion , we use @xmath255 near @xmath106 . this preparation ensures that the two wave packets will move toward each other , collide , and continue on after the scattering , as shown in the left panel of fig . [ fig : phaseshift_from_wf_phase ] . the effect of the interaction @xmath267 should be felt only when the particles are crossing each other . this interaction gives a phase shift to the wave function during the scattering . the wave function at time @xmath268 is @xmath269}(t)\rangle } = e^{-ih_{\delta}t } { \lvert\psi^{[\delta]}(0)\rangle } = e^{-ih_{\delta}t } { \lvert\psi(0)\rangle}\ ] ] ( same initial state for every @xmath4 ) . we consider the overlap @xmath270}(t)\vert\psi^{[\delta]}(t)\rangle}\ ] ] at some appropriately chosen ` final ' time @xmath271 . evolution with the @xmath139 ( non - interacting ) hamiltonian gives a ` reference ' state . we are interested in the phase accumulated in the evolution with the @xmath138 hamiltonian in comparison with the reference state , namely the phase of @xmath272 . the final time @xmath273 is chosen such that the particles have completed their scattering , but have not reached the boundaries of the chain . so , there are no edge effects . for example , in the process shown in fig . [ fig : phaseshift_from_wf_phase](left ) , it would be reasonable to compare overlaps at @xmath274 . the phase of @xmath272 should approximate the scattering phase shift , @xmath275 the shift @xmath254 is in accordance with the convention used in this paper , e.g. , in eq . . for @xmath276 , the bethe ansatz phase shift is obtained from eq . to be @xmath277 fig . [ fig : phaseshift_from_wf_phase](right ) compares the phases obtained from the time - evolving wave function of a finite chain containing two relatively narrow wave packets , with the bethe ansatz phase shift expressions which are strictly valid for delocalized excitations . even with wave packets as narrow as @xmath278 , the bethe ansatz expressions match extremely well the phase acquired in real - time evolution . 99 ifxundefined [ 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 e. fermi , j. pasta , and s. ulam , studies of nonlinear problems . alamos scientific laboratory report no . la-1940 , ( 1955 ) . focus issue : the fermi - pasta - ulam " problem - the first 50 years , chaos * 15 , * ( 2005 ) . g. gallavotti , _ the fermi - pasta - ulam problem : a status report _ , springer ( 2008 ) . m. lewenstein , a. sanpera , v. ahufinger , _ ultracold atoms in optical lattices : simulating quantum many - body systems _ , oxford university press ( 2012 ) . i. bloch , j. dalibard , and s. nascimbne , nature physics * 8 , * 267 ( 2012 ) . i. bloch , j. dalibard , w. zwerger , rev . 80 * , 885 ( 2008 ) . t. kinoshita , t. wenger , and d. s. weiss , nature * 440 , * 900 ( 2006 ) . m. greiner , o. mandel , t. w. hnsch , and i. bloch , nature * 419 * , 51 ( 2002 ) . a. l. cavalieri , n. mller , t. uphues , v. s. yakovlev , a. baltuka , b. horvath , b. schmidt , l. blmel , r. holzwarth , s. hendel , m. drescher , u. kleineberg , p. m. echenique , r. kienberger , f. krausz , and u. heinzmann , nature * 449 * , 1029 ( 2007 ) . s. wall , d. brida , s. r. clark , h. p. ehrke , d. jaksch , a. ardavan , s. bonora , h. uemura , y. takahashi , t. hasegawa , h. okamoto , g. cerullo , and a. cavalleri , nature phys . * 7 * , 114 ( 2011 ) . l. perfetti , p. a. loukakos , m. lisowski , u. bovensiepen , h. berger , s. biermann , p. s. cornaglia , a. georges , and m. wolf , phys . lett . * 97 * , 067402 ( 2006 ) . s. iwai , m. ono , a. maeda , h. matsuzaki , h. kishida , h. okamoto , and y. tokura , phys . rev . lett . * 91 * , 057401 ( 2003 ) . f. schmitt , p. s. kirchmann , u. bovensiepen , r. g. moore , l. rettig , m. krenz , j .- h . chu , n. ru , l. perfetti , d. h. lu , m. wolf , i. r. fisher , and z .- x . shen , science * 321 * , 1649 ( 2008 ) . j. eisert , m. friesdorf , and c. gogolin , nature physics * 11 * , 124 ( 2015 ) . a. polkovnikov , k. sengupta , a. silva , m. vengalattore , rev . 83 * , 863 ( 2011 ) . j. dziarmaga , adv . * 59 * , 1063 ( 2010 ) . w. s. bakr , a. peng , m. e. tai , r. ma , j. simon , j. i. gillen , s. flling , l. pollet , and m. greiner , science * 329 * , 547 ( 2010 ) . j. f. sherson , c. weitenberg , m. endres , m. cheneau , i. bloch , and s. kuhr , nature * 467 * , 68 ( 2010 ) . d. petrosyan , b. schmidt , j. r. anglin , and m. fleischhauer , phys . rev . a * 76 * , 033606 ( 2007 ) ; * 77 * , 039908 ( 2008 ) . m. valiente and d. petrosyan , j. phys . b * 41 * , 161002 ( 2008 ) . l. jin , b. chen , and z. song , phys . a * 79 * , 032108 ( 2009 ) . f. heidrich - meisner , s. r. manmana , m. rigol , a. muramatsu , a. e. feiguin , and e. dagotto , phys . a * 80 * , 041603(r ) ( 2009 ) . r. a. pinto , m. haque , and s. flach , phys . a * 79 * , 052118 ( 2009 ) . m. valiente , d. petrosyan , and a. saenz , phys . a * 81 * , 011601(r ) ( 2010 ) . r. khomeriki , d. o. krimer , m. haque , and s. flach , phys . a * 81 * , 065601 ( 2010 ) . a. deuchert , k. sakmann , a. i. streltsov , o. e. alon , and l. s. cederbaum , phys . a * 86 * , 013618 ( 2012 ) . l. f. santos and m. i. dykman , new j. phys . * 14 * , 095019 ( 2012 ) . m. i. ivanchenko , r. khomeriki , and s. flach , eur . * 98 * , 66002 ( 2012 ) . j. p. ronzheimer , m. schreiber , s. braun , s. s. hodgman , s. langer , i. p. mcculloch , f. heidrich - meisner , i. bloch , and u. schneider , phys . lett . * 110 * , 205301 ( 2013 ) . l. vidmar , s. langer , i. p. mcculloch , u. schneider , u. schollwck , and f. heidrich - meisner , phys . b * 88 * , 235117 ( 2013 ) . s. langer , f. heidrich - meisner , j. gemmer , i. p. mcculloch , and u. schollwck , phys . b * 79 * , 214409 ( 2009 ) . m. haque , phys . rev . a * 82 * , 012108 ( 2010 ) . s. jesenko and m. nidari , phys . b * 84 * , 174438 ( 2011 ) . v. alba , k. saha , and m. haque , j. stat . ( 2013 ) p10018 . c. karrasch , j. e. moore , and f. heidrich - meisner , phys . b * 89 * , 075139 ( 2014 ) . a. sharma and m. haque , phys . rev . a * 89 * , 043608 ( 2014 ) . w. liu and n. andrei , phys . * 112 * , 257204 ( 2014 ) . u. schneider , l. hackermller , j. p. ronzheimer , s. will , s. braun , t. best , i. bloch , e. demler , s. mandt , d. rasch , and a. rosch , nature physics * 8 * , 213 ( 2012 ) . a. s. srensen , e. altman , m. gullans , j. v. porto , m. d. lukin , and e. demler , phys . a * 81 * , 061603(r ) ( 2010 ) .

we use the star formation history map of the large magellanic cloud recently published by harris & zaritsky to study the sites of the eight smallest ( and presumably youngest ) supernova remnants in the cloud : sn 1987a , n158a , n49 , and n63a ( core collapse remnants ) , 0509.5 , 0519.0 , n103b , and dem l71 ( type ia remnants ) . the local star formation histories provide unique insights into the nature of the supernova progenitors , which we compare with the properties of the supernova explosions derived from the remnants themselves and from supernova light echoes . we find that all the core collapse supernovae that we have studied are associated with vigorous star formation in the recent past . in the case of sn 1987a , the time of the last peak of star formation ( 12 myr ) matches the lifetime of a star with the known mass of its blue supergiant progenitor ( ) . more recent peaks of star formation can lead to supernovae with more massive progenitors , which opens the possibility of a type ib / c origin for snrs n158a and n63a . stars more massive than are very scarce around snr n49 , implying that the magnetar sgr 0526 in this snr was either formed elsewhere or came from a progenitor with a mass well below the threshold suggested in the literature . three of our four ia snrs are associated with old , metal poor stellar populations . this includes snr 0509.5 , which is known to have been originated by an extremely bright type ia event , and yet is located very far away from any sites of recent star formation , in a population with a mean age of gyr . the type ia snr n103b , on the other hand , is associated with recent star formation , and might have had a relatively younger and more massive progenitor with substantial mass loss before the explosion . we discuss these results in the context of our present understanding of core collapse and type ia supernova progenitors .

0902.2787

the identification of the progenitor stars of supernova ( sn ) explosions is one of the central problems of stellar astrophysics . in the case of core collapse supernovae ( cc sne : types ii , ib , ic , and derived subtypes ) the progenitors are known to be massive ( @xmath5 ) stars whose inner cores collapse to a neutron star or a black hole . in a few cases , it has been possible to constrain the properties of the progenitor star using pre - explosion images or the turn - off masses of compact clusters @xcite , but there are still many open issues regarding which stars lead to specific subtypes of cc sne ( for an extended discussion and a complete set of references , see * ? ? ? * ; * ? ? ? in the case of thermonuclear ( type ia ) sne , the situation is much more complex . although a co white dwarf ( wd ) in some kind of binary is almost certainly the exploding star , the exact nature of the progenitor system has never been firmly established , either theoretically or observationally ( see * ? ? ? * and references therein ) . when direct identifications are not possible , the properties of the progenitors can be constrained using the stellar populations around the exploding stars . a number of studies have done this for sne in nearby galaxies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but this approach has important limitations . first , the available information , be it photometric ( e.g. * ? ? ? * ) or spectral ( e.g. * ? ? ? * ) , is usually integrated over the entire host galaxy , although local measurements at the sn sites have been made for a small number of objects ( e.g. * ? ? ? this effectively ignores the metallicity and stellar age gradients that must be present in the host . second , even in surveys that work with complete host spectra , the stellar populations are not resolved . among other things , this means that the stellar light is weighted by luminosity , which can conceal many important properties of the stellar populations . in practice , the information that can be obtained from this kind of observations is restricted to average metallicities and ages , unless sophisticated fitting techniques are used to extract the star formation history ( sfh ) of the host ( see * ? ? ? ideally , one would want to study _ resolved _ stellar populations associated with sn progenitors . the information that can be obtained in this way is much more detailed and reliable , but it requires focusing on very nearby sne . the present work is the first in a series of papers aimed at constraining the fundamental properties of cc and ia sn progenitors in the magellanic clouds by examining the stellar populations at the locations of the supernova remnants ( snrs ) left behind by the explosions . to do this , we take advantage of the large amount of observational data accumulated on the stellar populations of the clouds , in particular the star formation history ( sfh ) maps published by @xcite for the smc and @xcite ( henceforth , hz09 ) for the lmc . to identify the sites of recent sne , we rely on the extensively observed population of snrs in the mcs @xcite . much information about the sn explosions can be extracted from the observations of snrs of both ia and cc origin @xcite , and in some cases this information can be complemented by light echoes from the sne themselves @xcite . in this first installment , we focus on the eight youngest snrs in the lmc : sn1987a , n158a , n63a , and n49 ( cc snrs ) ; [email protected] , [email protected] , n103b and dem l71 ( ia snrs ) . this paper is organized as follows . in [ sec : targets ] we describe the criteria that have led to the selection of our eight target snrs . in [ sec : sntosnr ] we review their types and the characteristics of the parent sne that can be inferred from their observational properties . in [ sec : map ] we review the fundamental details of the sfh map of the lmc presented in hz09 . in [ sec : relevance ] we discuss the relevance that the local sfh has for the properties of the sn progenitors , given our knowledge about the global sfh and the stellar dynamics of the lmc . in [ sec : sfh ] we examine the local sfhs for the target snrs , with specific comments relating each sfh to the sne that originated the snrs . in [ sec : disc ] we discuss the impact that our findings have in the context of our current understanding of cc and ia sn progenitors . finally , in [ sec : conc ] we present our conclusions and we outline some avenues for future research . we will focus on young snrs because they are usually ejecta - dominated and still contain a great deal of information about their parent sne - in particular , the risk of mistyping young cc and ia snrs is minimal ( see [ sec : sntosnr ] ) . identifying the youngest snrs in a given set , however , is not trivial . among the snrs in the lmc , only one has a known age ( sn 1987a ) , and only three ( [email protected] , [email protected] , and n103b ) have more or less accurate age estimates from light echoes @xcite . in the absence of consistent age estimates for all objects , size is the best criterion to select the youngest ones . much information about the snr population in the lmc can be found in the _ rosat _ atlas by @xcite , but the snr sizes in particular are not reliable and must be revised . sizes of snrs with sharp outer boundaries are overestimated due to the large _ rosat _ psf ( e.g. [email protected] , * ? ? ? * ) , while sizes of diffuse snrs are underestimated due to the low _ rosat _ effective area ( e.g. n23 , * ? ? ? we have searched the literature for more recent _ chandra _ observations to constrain the lmc snr sizes , and we have selected the eight smallest objects ( sizes @xmath6 arcmin , see table [ tab-1 ] ) . the age estimates listed in table [ tab-1 ] merit a few comments . for snrs without sn or light echo information , ages are calculated from the snr size assuming a specific model for the snr dynamics , which can introduce large uncertainties . in particular , the standard dynamical models for young snrs ( e.g. * ? ? ? * ) ignore the effect of cosmic ray acceleration at the forward shock . it is now widely accepted that energy losses due to cosmic ray acceleration can affect the size of young snrs in a noticeable way @xcite , which implies that calculations based on unmodified snr dynamics can overestimate the age by as much as 20@xmath7 ( see 5.2 in * ? ? ? * for a discussion ) . in figure [ fig-1 ] , we illustrate the location of our eight target snrs within the large scale structure of the lmc using the data from field 13 of the southern h - alpha sky survey atlas ( shassa , * ? ? ? two snrs , sn 1987a and n158a , are located in the 30 dor region , the most prominent active star forming region in the lmc . two more , n49 and n63a , are in the northern part of the disk , embedded in the north blue arm discussed in hz09 and @xcite . snrs [email protected] and n103b are in the outer parts of the lmc bar . the last two objects , [email protected] and dem l71 , are in rather inconspicuous parts of the lmc disk , in the area called the northwest void by hz09 . more specific discussions about the location of each snr will be given in [ sec : sfh ] . typing snrs as cc or type ia can be an uncertain and treacherous business . both cc and ia sne deposit a similar amount of kinetic energy ( @xmath8 erg ) in the ambient medium ( am ) , which often makes it impossible to distinguish mature cc from mature ia snrs based on their size or morphology alone . a much more reliable way to type snrs is to examine the evidence left behind by the explosion itself : x - ray spectrum from the sn ejecta and am , snr dynamics , and properties of the compact object or pulsar wind nebula ( pwn ) , if present . in general , this can only be done for relatively young objects ( but see * ? ? ? * ; * ? ? ? * ) . by using methods along these lines , we have been able to determine the type of all the objects in our list with a high degree of confidence , and in some cases even the sn subtypes within the broader cc and ia categories . in this section we will discuss the classification and sn subtypes of our target snrs , but before going into the details of each object , it is important to mention the work of @xcite . these authors attempted to type _ all _ the lmc snrs known at the time by noting the distance from each object to hii regions and ob associations . although this is a very crude method , their conclusions regarding the cc or type ia nature of our target snrs coincide with ours , except in the case of snr n103b , which will be discussed in detail in [ sec : sntosnr : iasnrs ] and [ sec : sfh : ia ] . our theoretical understanding of core collapse sne is still incomplete @xcite . in particular , the mapping between progenitor mass and cc sn subtype is uncertain , because key processes like stellar mass loss and binary interactions are not well understood @xcite . to set the stage for further discussions , the stellar evolution models of @xcite for single stars of lmc metallicity ( @xmath9 ) predict that stars between @xmath10 and @xmath3 will explode as red supergiants , retaining most of their h envelope and becoming type iip sne , stars between @xmath11 and @xmath12 will lose a large part of their envelopes and explode as type iil or type iib sne , and stars above @xmath12 will lose all their envelopes and become naked cc sne of types ib and ic . within naked cc sne , there is some evidence that type ic sne , which are linked to long duration gamma - ray bursts @xcite come from more massive stars than type ib sne @xcite . stars that retain a massive h envelope but explode as blue supergiants instead of red supergiants form a separate class , often referred to as sn 1987-like events . the lifetimes associated with these stellar masses range between 41 myr for an isolated @xmath13 star and 5.4 myr for an isolated @xmath12 star ( always taking the @xmath9 models from * ? ? ? * ) . in principle , mass loss will be facilitated by binary interactions , leading to fewer red supergiants and more type ib / c sne in binary systems , but stellar evolution calculations that include these effects are subject to an entirely different set of uncertainties @xcite . from the point of view of the snrs , the complex and turbulent structure of most young cc snrs makes a quantitative interpretation of the x - ray spectrum in terms of specific explosion models and progenitor scenarios very challenging ( e.g. see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . many times , it is hard to infer the sn subtype from the observational properties of the snr , but the large intrinsic diversity of cc sne as a class can often be used to some advantage in snr studies . @xcite argues that several aspects of snr evolution are expected to be very different depending on the subtype of the parent sn : mixing in the ejecta , fallback onto the central compact object , expansion of the pwn , interaction with the csm , and photoionization of the am by shock breakout radiation . using arguments along these lines , @xcite inferred from the positions of the fluid discontinuities , the presence of high velocity h , and the extent of the clumpy photoionized pre - sn wind in the cas a snr that its progenitor must have been a type iin or type iib event . this ` prediction of the past ' was later confirmed by the spectroscopy of the light echo of the cas a sn @xcite , which is very similar to the spectrum of the type iib sn1993j . although this agreement is certainly encouraging , we must insist that studies based on snrs are still a long way from providing a robust method of subtyping cc sne - as an example , the type iin / iib classification of cas a by @xcite was challenged by @xcite , who argued for a type ib progenitor . in the following paragraphs , we examine each of the four target cc snrs in more detail . for a summary , see table [ tab-2 ] . [ [ snr - n49 ] ] snr n49 + + + + + + + this snr harbors one of only two magnetars known outside the milky way : sgr 0526@xmath066 . in principle , the presence of a compact object should immediately classify this object as a cc snr , but the association between this magnetar and the snr has been controversial @xcite . even disregarding the compact object , the ejecta emission shows significantly enhanced abundances from o and si , but a comparatively small amount of fe @xcite , and the snr is located within the ob association lh53 @xcite . taken together , these arguments lend strong support to a cc origin . the complex filamentary structure of the shocked am suggests that dense material surrounded the sn at the time of the explosion @xcite , which favors a progenitor with a slow wind , maybe a type iip sn . unfortunately , this is just an educated guess , because the complex multi - phase x - ray emission of the snr and the poorly known age make the interpretation of the observations very challenging . [ [ snr - n63a ] ] snr n63a + + + + + + + + this snr has no detected compact object , although the upper limits do not exclude the presence of a low - activity pwn @xcite . it is embedded in the large hii region n63 , and it also appears to be located within an ob association ( ngc 2030 , * ? ? ? * ) , making a cc type very likely . the size , morphology , and x - ray spectrum show evident signs of a complex interaction with a highly structured am , and they seem to indicate that it is expanding into a large cavity @xcite , which suggests a massive progenitor with a fast wind , maybe a type ib / c sn . however , there is no additional evidence to support this conclusion because the x - ray emission , which is dominated by the shocked am , reveals very little about the properties of the sn ejecta @xcite . [ [ sn-1987a ] ] sn 1987a + + + + + + + + the classification and subtype of sn 1987a are obvious from the sn spectroscopy . the vast amount of information available on this object is summarized in @xcite and other publications in the same volume . for our purposes , it suffices to mention that the progenitor of this sn is known to have been a blue supergiant star , sk @xmath14 , whose initial mass has been estimated at @xmath15 @xcite , and might have been part of a close binary system @xcite . [ [ snr - n158a ] ] snr n158a + + + + + + + + + this object harbors a well - observed pwn that types it as a cc snr and constrains its age to be @xmath16800 yr @xcite . the sn subtype classification has been rather controversial . the snr dynamics indicate that the shock wave is moving into dense , clumpy csm similar to what can be found around a massive wolf - rayet star , and the presence of strong o and s lines in the x - ray spectrum of the innermost ejecta reveal that at least some of the heavy elements in the ejecta have not fallen back onto the central neutron star @xcite . this would favor a massive type ib / c progenitor , but both the detection of h in the pwn filaments @xcite and a recent re - analysis of the ejecta emission @xcite seem to indicate that the progenitor might have been in the @xmath17 range , implying a type iip explosion ( for more detailed discussions , see * ? ? ? * ; * ? ? ? our current understanding of type ia sn progenitors is still extremely sketchy @xcite , but several interesting trends have been inferred from the bulk properties the host galaxies . any theoretical model for type ia progenitors must account for the fact that type ia sne explode in elliptical galaxies with very little star formation ( sf ) , but at the same time the rate of ia events in star forming galaxies appears to scale with the specific star formation rate @xcite . moreover , type ia sne exploding in elliptical galaxies are on average dimmer that those exploding in star forming galaxies @xcite . @xcite and @xcite used these observational facts to postulate _ two _ populations of type ia sn progenitors : a ` prompt ' population with short delay times ( of the order of a few hundred myr ) , associated with recent sf and leading to somewhat brighter sn ia , and a ` delayed ' population with longer delay times ( of the order of gyr ) , not associated with recent sf and leading to somewhat dimmer sn ia . in this two - component model , the specific type ia sn rate in a given galaxy is expressed as @xmath18 , with @xmath19 being the total stellar mass in the galaxy and @xmath20 the specific star formation rate . it is important to stress that the observed rates do not _ require _ the existence of two components - in some theoretical scenarios , type ia sne from a single progenitor channel can explode with both very short and very long delay times @xcite . nevertheless , there have been several attempts to associate the prompt and delayed progenitor populations to the two leading theoretical scenarios for type ia sne : single degenerate ( sd ) systems , in which the wd accretes material from a non - degenerate companion and double degenerate ( dd ) systems , in which the wd accretes material from another wd . so far , none of these attempts has succeeded ( see f.i . * ? ? ? * ; * ? ? ? * ) , and the identity of type ia sn progenitors remains a mystery . despite all the uncertainties regarding their progenitors , type ia sn explosions as a class are much more homogeneous and have less intrinsic dispersion than cc sne . in particular , there is a simple relationship between the structure of the ejecta and the peak brightness of the sn that is well reproduced by one - dimensional delayed detonation ( ddt ) explosion models @xcite . this allows to map the vast majority of sn ia onto a sequence of bright to dim events based on the amount of @xmath21ni that they synthesize . generally speaking , type ia snrs are also much less turbulent than cc snrs , and most of them seem to be interacting with a relatively unmodified am @xcite , although there are exceptions like the kepler snr @xcite and snr n103b ( * ? ? ? * , see also the discussion in [ sec : conc - ia ] ) . thanks to this set of circumstances , it is generally easier to interpret the x - ray emission of type ia snrs quantitatively in terms of specific explosion models , provided that the dynamic evolution of the snr and the nonequilibrium processes in the shocked plasma are properly taken into account @xcite . it is also possible to estimate the brightness of the parent event from the mass of @xmath21ni synthesized by the preferred ddt explosion model , as shown by @xcite in the case of the tycho snr . these four objects were classified as type ia snrs by @xcite based on their lack of compact object or pwn and the general properties of their x - ray emission , which is dominated by fe lines and has only weak or absent lines from o. in order to confirm these classifications and derive the sn subtype , it is necessary to perform an in - depth analysis of the ejecta emission , as done by @xcite for snr [email protected] . the x - ray emission of the other three objects will be the subject of a forthcoming publication ( * ? ? ? * henceforth bh09 ) , but the main results of that analysis are presented in the following paragraphs , and summarized in table [ tab-3 ] . [ [ snr - dem - l71 ] ] snr dem l71 + + + + + + + + + + + this is the oldest object in our type ia snr list , and the only one without a light echo age estimate . @xcite determine an age of @xmath22 from the snr dynamics , and yet the x - ray spectrum appears dominated by shocked fe from the sn ejecta , specially in the center @xcite . the old age of this snr makes the analysis of the ejecta emission somewhat challenging , but bh09 find that it can be reproduced by ddt models for normal type ia sne . [ [ snr - n103b ] ] snr n103b + + + + + + + + + this object was initially classified as a cc snr by @xcite based on its location at the edge of the hii region dem 84 and 40 pc away from the ob association ngc 1850 . however , the x - ray spectrum is strongly suggestive of a type ia origin @xcite . the snr is also remarkable in that it shows a strong east - west asymmetry @xcite , which has been interpreted as a sign of some kind of csm interaction , mainly by analogy to the kepler snr . this asymmetry also makes the ejecta analysis challenging , but bh09 find a relatively good match to the spectrum using ddt models for moderately bright type ia sne with a snr age close to the 800 yr estimated from the light echo by @xcite . [ [ snr-0509 - 67.5 ] ] snr [email protected] + + + + + + + + + + + + + + + + + + + + + + + + + + + next to sn1987a , this snr has the most secure subtype classification in the lmc . in 2008 , two teams analyzed independently the optical spectrum of the sn light echo @xcite and the x - ray emission and dynamics of the snr @xcite , and came to the same conclusion : snr [email protected] was originated @xmath23 yr ago by an exceptionally bright type ia sn that synthesized @xmath24 of @xmath21ni . this agreement is a very important validation of the modeling techniques introduced in @xcite that bh09 apply to the other three target type ia snrs , and in particular of the capability of the models to recover the sn subtype . [ [ snr-0519 - 69.0 ] ] snr [email protected] + + + + + + + + + + + + + + + + + + + + + + + + + + + the final object in our type ia snr list has an estimated age of @xmath25 yr from its light echo @xcite . its x - ray emission is well reproduced by a moderately bright type ia sn model that synthesizes @xmath26 of @xmath21ni ( bh09 ) . the sfh map that we use in the present work is described in full detail in hz09 . the map was elaborated using four band ( _ u _ , _ b _ , _ v _ , and _ i _ ) photometry from the magellanic clouds photometric survey @xcite , which has a limiting magnitude between @xmath27 and @xmath28 in _ v _ , depending on the local degree of crowding in the images . data from more than 20 million stars was assembled to produce color - magnitude diagrams in 500 @xmath29 cells encompassing the central @xmath30 of the lmc ( see figure 4 in hz09 ) , and then the starfish code @xcite was applied to derive the local sfh for each cell . cells with enough stars in them , like the eight cells that contain our target snrs , were further subdivided into four @xmath31 subcells . the sfh of each cell is given at thirteen lookback times between @xmath32 ( 6.3 myr ) and @xmath33 ( 17.8 gyr ) , and it is broken into four metallicity bins : @xmath34 and @xmath35 . for reference in further discussions , we reproduce the sfh of the entire lmc from hz09 in figure [ fig-2 ] . the error bars on the total sfh represented with the gray shaded area are dominated by crowding effects ( see 3.3 in hz09 ) . although no metallicities were fitted for ages below 50 myr , the plots display the canonical lmc metallicity ( @xmath9 ) in the most recent age bins , which is reasonable in view of the high degree of homogeneity in the metallicity of the ism and the young stars in the lmc @xcite . a detailed discussion of the sfh of the lmc and its interpretation in the context of the lmc s past history can be found in 5 of hz09 . for our purposes , it suffices to note that , after an initial episode of sf in the distant past , the lmc went into a quiescent period that lasted until 5 gyr ago , and since then it has been forming stars at an average rate of @xmath36 , with episodes of enhanced sf at 2 gyr , 500 myr , 100 myr , and 12 myr . from figure [ fig-2 ] , it is obvious that the vast majority of the stars in the lmc have ages above 1 gyr . most of these old stars have metallicities of one tenth solar or lower . during the lifetime of a galaxy , several processes naturally mix the stellar populations . these include both internal processes like the ` churning ' of the disk by spiral arms @xcite and external processes like tidal interactions and mergers @xcite . in this context , the properties of the stellar population ( and hence the sfh ) in the neighborhood of a young snr will only be representative of the sn progenitor up to a certain lookback time , @xmath37 . in principle , @xmath37 can be calculated for each location within a galactic disk provided there is a viable dynamic model that includes all the relevant processes . unfortunately , no such model exists for the lmc , despite the wealth of observational information available . the lmc disk is warped @xcite and might also be flared @xcite , and it has a rich history of tidal interactions with the smc and ( maybe ) the milky way , which may have important effects on the stellar dynamics ( see * ? ? ? * ; * ? ? ? the vestigial arms seen in hi @xcite are probably originated by these tidal interactions , but the details of this process are not well understood @xcite . even the nature of the most prominent feature in the disk - the lmc bar - and its role in the dynamics of the galaxy are unclear @xcite . without a reliable way to calculate @xmath37 for each of the subcells that contain our target snrs , all we can do is estimate the relevant timescales for a number of different processes . the physical size of the subcells in the hz09 map is @xmath38 pc ( assuming @xmath39 kpc , * ? ? ? * ) , and the velocity dispersions for the young disk and old disk populations determined by @xcite are @xmath10 and @xmath40 , respectively . thus , the length of time that it would take an average star of the young ( old ) disk to drift from one subcell to the next in the absence of restoring forces , @xmath41 , is 43 ( 16 ) myr . these timescales are not relevant for the progenitors of cc sne , which should belong to the young disk , but they will be very important for sn ia progenitors , which could be quite older . in any case , @xmath37 should be much larger than @xmath41 , because ( a ) some regions of the lmc are more homogeneous than others , which means that stars have to drift over larger distances in order to find substantially different stellar populations , and ( b ) there are restoring forces like gravity that maintain the structural integrity of the disk and act to limit stellar drift . we have quantified the spatial homogeneity of the stellar populations around our target snrs in figure [ fig-3 ] . we plot the absolute value of the relative differences in the stellar populations as a function of the distance from the center of each of the eight subcells that contain the cc and ia snrs in our list . to calculate the relative differences , we have integrated the sfh in each neighboring subcell , taking all the time bins were the differences between the neighbor and the snr subcell were statistically significant ( i.e. , the error bars did not overlap ) up to a lookback time of 1.1 gyr , and then divided by the total number of stars formed in the snr subcell . at each distance , the relative difference is the mean of the relative differences between the snr subcell and all the subcells at that distance . figure [ fig-3 ] shows that some of our target snrs are in remarkably homogeneous regions of the lmc disk . these include the cc snrs n49 and n63a in the blue arm and the type ia snr [email protected] in the northwest void , with average relative differences in the stellar populations below @xmath42 within 1400 pc of the central subcell . this distance translates into @xmath41 values of 215 and 80 myr for young and old disk stars in the absence of restoring forces . the effect of restoring forces on the value of @xmath37 is more difficult to estimate . if no chaotic processes intervene , neighboring stars will tend to move together through the disk , which explains why some lmc structures like the bar and the northwest arm show up in the hz09 map with lookback times as large as 1 gyr ( see their figure 8) . this long survival time is not restricted to large structures - in the solar neighborhood , there is evidence that several groups of old stars ( 2 to 8 gyr ) are moving together through the disk of the milky way @xcite . but smaller structures like young stellar clusters seem to disappear on timescales of the order of @xmath43 myr in the lmc @xcite , which is roughly equivalent to the dynamic crossing time of the lmc disk . we will adopt this value as a figure of merit for @xmath37 in a single subcell of the sfh maps . since the result of @xcite applies to young stars , this implies that restoring forces increase the value of @xmath41 by at least a factor @xmath44 , but we stress that this is just a very rough estimate . we conclude that the relevance of the local sfhs for type ia sn progenitors will depend on both the homogeneity of the stellar populations around each subcell ( figure [ fig-3 ] ) and the age of the progenitors . if the lifetime of type ia progenitors in the prompt channel is as short as the 180 myr claimed by @xcite , it should be possible to use the local sfhs of type ia snrs in the lmc to explore their properties . for progenitors with longer lifetimes , the stellar context of each snr should be taken into account . objects like snr [email protected] might allow exploration of timescales up to several hundred myr , but snrs like [email protected] probably will not . the local sfhs in the subcells containing the eight snrs in our list are plotted in figures [ fig-4 ] and [ fig-6 ] . the lifetime of an isolated @xmath13 star with @xmath9 from @xcite has been indicated by a dashed vertical line in all the plots for illustrative purposes . for simplicity , we have collapsed all the sfh bins at ages above 2 gyr into a single bin at 10 gyr . several interesting average quantities can be calculated from the local sfhs . we have listed two such quantities in tables [ tab-2 ] and [ tab-3 ] : the average metallicity of all the stars formed in the subcell , @xmath45 and their average age @xmath46 . these averages are always dominated by the large population of old stars in each subcell , and therefore they are irrelevant for the properties of cc sn progenitors - the values in table [ tab-2 ] are merely provided for comparison with the values in table [ tab-3 ] ( see discussion in [ sec : disc ] ) . the average values for the entire lmc are @xmath47 and @xmath48 gyr . the most salient feature of the sfhs around the four cc snrs ( figure [ fig-4 ] ) is that they are strongly dominated by intense bursts of star formation in the recent past ( @xmath49 myr ) . this is of course expected , and in the cases where the snrs have been typed for their close proximity to young stellar clusters ( notably , snr n63a ) , it does not reveal any new information . however , the timing of these bursts and their intensity will determine the properties of the population of massive stars that can be found at each location , and hence the likelihood of each cc sn subtype . to highlight these aspects , we display the most recent bins of the sfhs associated with the cc snrs in greater detail in figure [ fig-5 ] , alongside the lifetimes of isolated massive stars with @xmath9 from @xcite . we have convolved the three most recent sfh bins with a standard salpeter imf to calculate the fraction of massive stars that are exploding now as cc sne ( @xmath50 ) from progenitors in three mass intervals : @xmath10 to @xmath51 , @xmath52 to @xmath53 and above @xmath2 . we list these fractions in table [ tab-2 ] for each of the cc snrs . the interval cuts are the stellar masses whose lifetimes correspond to the upper edges of the first and second age bins in the sfhs : 9.4 and 18.9 myr . we remind the reader that these values of @xmath50 are calculated using isolated star models that do not take into account the potentially large effects of binarity on stellar evolution . an entirely different but also potentially serious problem comes from the fact that massive stars are notoriously difficult to study using photometry alone @xcite . because starfish uses _ all _ the stars ( not just the massive ones ) in each subcell to calculate the sfr at each age , the values of @xmath50 might not be severely affected by this , but to this date there has been no systematic calibration of the starfish results for young stellar populations . to reflect these and other caveats , we do not list error bars on the @xmath50 values , which should only be regarded as approximate . [ [ snr - n49 - 1 ] ] snr n49 + + + + + + + the integrated sfh for the north blue arm region that contains snr n49 and snr n63a is dominated by a coherent episode of low - metallicity star formation 100 myr ago ( see 5.1.3 and figure 16 in hz09 ) , which is apparent in the corresponding panels of figure [ fig-4 ] . this is the reason why the values of @xmath45 and @xmath46 for snrs n49 and n63a are lower than those of the other snrs . many parts of the north blue arm have also had noticeable star formation activity in the last 40 myr , although not as intense as in the more prominent star forming regions of the lmc , 30 dor and constellation iii . snr n49 is in one such region , which had a moderately intense sf burst 12 myr ago , but very little sf activity in the most recent bin centered at 6.3 myr ( see figure [ fig-5 ] ) . from these properties of the sfh , the expectation is that the majority of the cc sn progenitors in this subcell should be stars between @xmath52 and @xmath53 ( see table [ tab-2 ] ) . even taking the upper limit of the sfr in the most recent bin and the lower limit on the bin at 12 myr , the fraction of cc sn progenitors with masses above @xmath53 remains below @xmath54 , with all the caveats associated to the calculated values of @xmath50 . this is in good agreement with the properties of the snr discussed in [ sec : sntosnr : ccsnrs ] , and it has interesting implications for the association of snr n49 with sgr 0526@xmath066 . magnetars are thought to be originated by very massive ( @xmath55 ) stars @xcite , but such stars appear to be very scarce around snr n49 . one possibility is that the magnetar was formed elsewhere and the association is coincidental . @xcite examined this issue in detail , and came to the conclusion that the link between snr n49 and sgr 0526@xmath066 is considerably less convincing than those of other magnetars in snrs . more recently , @xcite performed a nir survey of the area around snr n49 and identified a young stellar cluster ( sl 463 ) at a projected distance of @xmath56 pc northeast of sgr 0526@xmath066 that could have been the birthplace of the magnetar . this cluster does fall partially on a neighboring subcell of the hz09 map with intense sf at 6.3 myr , consistent with the 5 to 20 myr age estimates for sl 463 by @xcite , and much more promising as a birthplace of massive cc progenitors ( @xmath57 above @xmath53 ) . as pointed out by @xcite , if this hypothesis is true the magnetar must have been ejected from its birthplace with a certain velocity , and should have a measurable proper motion ( see their 3.2 ) . another possibility is that the association between snr n49 and sgr 0526@xmath066 is indeed real , but not all magnetars have stellar progenitors more massive than @xmath3 . [ [ sn-1987a-1 ] ] sn 1987a + + + + + + + + the sfh associated with this snr is of particular interest because , together with the known mass of the progenitor , @xmath1 , it can provide some test of the robustness of our sfh approach to cc sn progenitors . @xcite conducted a detailed study of the immediate neighborhood of sn1987a within the 30 dor region using _ hst _ data , and found a loose young cluster with an age of @xmath58 myr , which they identified as the birthplace of sn1987a s progenitor . the local sfh drawn from the hz09 map is indeed dominated by an intense sf episode 12 myr ago , in good agreement with the results of @xcite . from the sfh , we expect @xmath59 of the cc sn progenitors in this region to be stars between @xmath52 and @xmath53 ( see figure [ fig-5 ] ) . this agreement is encouraging , and indicates that some information about the progenitor mass of cc sne can be recovered from the sfh maps of hz09 . [ [ snr - n63a-1 ] ] snr n63a + + + + + + + + like snr n49 , snr n63a is located in a part of the blue arm with sf activity in the recent past . the sfh in this subcell , however , is different from that around snr n49 in that it is dominated by the most recent bin centered at 6.3 myr . as a result , @xmath60 of the cc sn progenitors in this subcell are expected to be stars more massive than @xmath53 . with a salpeter imf , roughly @xmath61 of these stars will be in turn more massive than @xmath12 , which makes a type ib / c origin for snr n63 plausible , as suggested by the properties of the snr . if this were true , snr 63a would be one of the youngest nearby type ib / c snrs , and a closer examination of this object to locate the elusive compact object and study the ejecta emission in greater detail would be of the highest interest . [ [ snr - n158a-1 ] ] snr n158a + + + + + + + + + this object is also located in 30 dor , but in a region with even more vigorous recent sf than the neighborhood of sn 1987a . according to our estimates , @xmath62 of the cc sn progenitors in this subcell should be stars more massive than @xmath53 . again , this is compatible with the relatively massive progenitor suggested by the snr properties ( see [ sec : sntosnr : cc ] ) , but unfortunately the temporal resolution of the sfh does not allow us to discriminate between a type iip progenitor with @xmath63 @xcite and a type ib / c progenitor with @xmath64 @xcite . from the point of view of the stellar population around snr 158a , both hypotheses are equally plausible . the local sfhs around the four type ia snrs are displayed in figure [ fig-6 ] . since the age of the progenitors is not known _ a priori _ , we also display the integrated sfhs in figure [ fig-7 ] to provide a more intuitive picture of the makeup of the stellar populations in these locations and how they have evolved with time . with one exception ( snr n103b , see below ) , these sfhs are very different from those associated with the cc snrs . the sfhs of snrs dem l71 , [email protected] , and [email protected] show very little ( but nonzero ) activity in the last 200 myr , resulting in old and metal poor stellar populations , typical of the regions of the lmc without ongoing star formation . the sfhs of these three snrs are also punctuated by bursts of sf at 600 myr and 2 gyr whose prominence varies from object to object , but this is probably more related to the global properties of the lmc ( see figure [ fig-2 ] ) than to the type ia sn progenitors . given the coincidence of the lmc disk crossing time and the upper limit for the delay time of short - lived sn ia progenitors ( 180 myr according to * ? ? ? * see [ sec : relevance ] ) , it is possible to use the a+b model introduced in [ sec : sntosnr : ia ] and the local sfhs to predict what fraction of type ia sne will come from prompt and delayed progenitors in each subcell ( @xmath65 ) . for the prompt component , we have calculated an average star formation rate by integrating the sfhs between 64 myr ( the minimum time necessary to produce a co wd at the metallicity of the lmc , * ? ? ? * ) and 180 myr , and we have multiplied it by the value of @xmath66 from @xcite , @xmath67 . for the delayed component , we have simply multiplied the number of stars in each subcell by the value of @xmath68 from @xcite , @xmath69 . the values of @xmath65 for each type ia snr are listed in table [ tab-3 ] , and can be compared with the values obtained for the whole lmc using the integrated sfh from figure [ fig-2 ] : @xmath70 and @xmath71 . even more so than in the case of cc sne , we stress that these numbers should be considered with caution , because they make strong assumptions about the properties of type ia sn progenitors . in particular , increasing the upper limit to the age of the prompt component from 180 to 300 myr results in an increase of the fraction of _ delayed _ type ia progenitors between 8 and 15 percentual points , depending on the snr . at a first glance , it is somewhat surprising that the sn ia rates from prompt and delayed progenitors are always comparable both in the whole lmc and at the locations of the individual type ia snrs . this can be easily understood by examining figure 6 in @xcite : for low values of the sfr , the _ specific _ rate of type ia sne from the prompt and delayed components is very similar , and low - intensity sf has been widespread across the lmc during the last 5 gyr ( see [ sec : map ] and hz09 ) . [ [ snr - dem - l71 - 1 ] ] snr dem l71 + + + + + + + + + + + the exceptionally low rate of sf between 64 and 180 myr in the subcell containing this snr makes it the most likely object ( @xmath72 probability ) to be associated with a delayed type ia sn progenitor . unfortunately , dem l71 is also the oldest type ia snr , and detailed information about its parent sn is hard to extract from the observations . in particular , the ejecta emission is not remarkable in any way , and seems consistent with normal type ia sn explosion models ( bh09 ) . the forward shock is running into material with a factor @xmath73 density range @xcite , but this can be easily explained by inhomogeneities in the ism . the snr dynamics do not indicate any substantial modification of the am by the progenitor @xcite . [ [ snr - n103b-1 ] ] snr n103b + + + + + + + + + of all the sfhs associated with type ia snrs that we discuss here , that of snr n103b is without doubt the most remarkable . this snr is in a region of the lmc bar that has seen vigorous sf activity in the recent past , with a prominent extended peak between 100 and 50 myr ( probably associated to the nearby cluster ngc 1850 , * ? ? ? * ) and a more recent one 12 myr ago . the intensity of this last peak is even larger than those associated with sn 1987a and snr 158a in the 30 dor region . it is not surprising that @xcite mistyped snr n103b as a cc snr before a good quality x - ray spectrum was available , based on this evident association with recent sf . this does not necessarily imply that the progenitor of snr 103b was a young star , although the predicted fraction of prompt ia progenitors ( @xmath74 ) is higher for this snr than for any of the others . however , if the properties of the snr itself are taken into account , the association with recent sf becomes more intriguing . @xcite noted a number of similarities between the strong lateral asymmetry of snr n103b and that of kepler s snr . in the case of snr n103b , it is unclear whether this asymmetry is due to an interaction with an ism structure ( e.g. , the nearby hii region dem 84 ) or to some kind of csm modified by the sn progenitor . in the kepler snr , however , it has been shown that the asymmetry is indeed due to interaction with a csm modified by the type ia sn progenitor ( see * ? ? ? * and references therein ) , suggesting that either the progenitor or its companion must have been relatively massive . from the shape of the sfh , a relatively young ( less than 150 myr ) and metal - rich ( @xmath9 ) progenitor for snr n103b seems a likely possibility . [ [ snr-0509 - 67.5 - 1 ] ] snr [email protected] + + + + + + + + + + + + + + + + + + + + + + + + + + + this snr is in a region that hz09 called the northwest void due to its conspicuous lack of recent star formation . in fact , hz09 argue that the old and metal - poor stars in this part of the lmc are representative of the primordial stellar population of the lmc . moreover , the snr is in a very homogeneous region ( see figure [ fig-3 ] ) , with neighboring subcells having very similar stellar populations . the closest subcells with noticeable sf activity ( above @xmath75 ) at @xmath76 myr are roughly 1 kpc away from the snr . this would not be noteworthy for a generic type ia snr , but thanks to the work of @xcite and @xcite we know that snr [email protected] was originated by an exceptionally bright type ia sn which synthesized @xmath24 of @xmath21ni . this seems to be at odds with the conventional wisdom regarding prompt and delayed type ia sn progenitors , which holds that exceptionally bright type ia sne are usually associated with younger stellar populations @xcite , but in fact the local sfh predicts equal contributions from prompt and delayed progenitors even in this remarkably quiet part of the lmc . if this snr did have a relatively young and massive progenitor , however , it appears to have left the am around it relatively undisturbed @xcite . [ [ snr-0519 - 69.0 - 1 ] ] snr [email protected] + + + + + + + + + + + + + + + + + + + + + + + + + + + this snr falls in line with [email protected] and dem l71 in having a low sfh at all times , although the local stellar population appears to be more metal rich than in the other type ia snrs ( see figure [ fig-7 ] ) . other than that , snr [email protected] is unremarkable both in its overall structure and dynamics and in the properties of its ejecta emission . from the local sfh , delayed type ia progenitors are slightly favored over prompt ones , but not significantly . the combination of snr studies and local sfhs that we have introduced in this paper is a promising method to further our understanding of cc sn progenitors , but it needs to be refined before it can have a significant impact on this field of research . two major issues need to be addressed . first , the techniques used to determine the subtype of cc sne from their snrs are still too crude to provide consistent results , even for well observed objects like the cc snrs in our sample . and second , the ability of tools like starfish to recover the sfh from mixed stellar populations at the ages that are relevant for the evolution of cc sn progenitors ( below 40 myr ) using photometric data needs to be firmly established . it would be interesting to investigate the possibility of an increased temporal resolution at very early times ( less than 10 myr ) in order to distinguish between type iil / b and type ib / c sn progenitors , but this might require support from spectroscopic surveys like the vlt - flames @xcite . we hope to address these issues in future publications . despite the limitations of the method , we have been able to obtian some interesting results . in general , we can say that the properties of the snrs and the local sfhs are compatible with each other , allowing for the large uncerainties discussed above . among our target objects , the sfh around snr n49 seems to indicate that stars more massive than @xmath53 are scarce in this part of the lmc . this opens two possibilities : either very massive ( @xmath55 ) stars are not necessary to produce magnetars , or the source sgr 0526@xmath066 is not in fact associated with snr n49 , as suggested by @xcite . the sfh around snr n63a seems compatible with a very massive sn progenitor , maybe a type ib / c sn , but other possibilities can not be discarded . for snr 158a , the temporal resolution of the hz09 maps is too coarse to resolve the issue of the progenitor . in any case , our findings stress the importance of revisiting and reanalyzing the x - ray emission from these young cc snrs in detail to learn more about the sn explosions that originated them . we have found that the combination of snr studies and local sfhs is a powerful tool to explore the properties of type ia sn progenitors . the x - ray emission from type ia snrs is well understood in terms of explosion physics , and the ability of snr studies to recover the type ia sn subtype ( i.e. , bright vs. dim ) has been demonstrated by @xcite . moreover , the combination of starfish and a data set like the mcps is ideally suited to characterize the stellar population in different parts of the lmc . among the type ia snrs that we have studied , one ( snr n103b ) is found in a region that has experienced vigorous sf in the recent past , but the other three are associated to old and metal - poor stellar populations . on a first inspection , it would be tempting to establish connections between these two kinds of environments and the prompt and delayed components to type ia progenitors proposed by @xcite , but we have found that the situation is more complex than that . even in regions with a remarkably small amount of recent sf , it is very hard to isolate objects that arise unambiguously from delayed type ia progenitors . this is due to the high efficiency of the still unknown mechanisms that turn co wds from relatively young , intermediate - mass stars into type ia sn progenitors of the prompt component @xcite . in order to identify with some confidence a type ia snr as a product of the prompt or delayed component , it would have to be either in a region with very vigorous sf or in a pristine region with no measurable sf activity in the appropriate range of ages . since even elliptical galaxies appear to have some residual sf during their entire lifetimes @xcite , isolating the delayed type ia progenitors in the local universe to study their properties in detail might be a very challenging task , unless a type ia sn is found in a nearby globular cluster @xcite . our results allow us to test specific theoretical predictions about type ia progenitors , like the claim by @xcite that the type ia sn rate should be very low for metallicities lower than a tenth of the solar value . this is based on the so - called ` accretion wind ' scenario for single - degenerate type ia progenitors , which requires a minimum opacity in the material transferred from the companion to the wd @xcite . the average _ stellar _ metallicity is close to this value or even much lower around three of the four type ia snrs that we have examined , which seems hard to reconcile with the results of @xcite , although we stress that all the regions that we examined do contain a small number of stars with higher metallicities . similar concerns about this prediction and its implications have been raised by @xcite , who found several type ia sne in low - metallicity dwarf galaxies . the accretion wind scenario also makes strong predictions about the shape of the csm around type ia progenitors that are not substantiated by the dynamics of known type ia snrs @xcite . in this context , an interesting possibility is opened by the recent work of @xcite , which allows one to make _ direct _ measurements of the metallicity of type ia sn progenitors using mn and cr lines in the x - ray spectra of young snrs . if this technique could be applied to the lmc snrs , we would be able to contrast the results with the properties of the stellar populations , and test theoretical ideas about the role of metallicity in different kinds of type ia sne ( e.g. * ? ? ? two of the snrs we have examined have remarkable properties with important implications for type ia sn progenitors . the unusual morphology of snr n103b ( see * ? ? ? * and references therein ) , which is strongly suggestive of some kind of csm interaction , has become even more noteworthy in light of the vigorous recent sf revealed by the local sfh . it appears that snr n103b might be a member of an emerging class of type ia snrs with csm interaction that could be associated with relatively young and massive progenitors that lose an appreciable amount of mass before exploding as type ia sne . this class would include the kepler snr @xcite in our galaxy and other lmc snrs @xcite , but the local sfhs around these objects should be examined to confirm this possibility . evident signs of csm interaction , however , can not be found in other well studied type ia objects like tycho and sn 1006 in our own galaxy or the other three lmc snrs that we have analyzed here @xcite , indicating that a majority of type ia progenitors do not modify their surroundings in a noticeable way . since it is unlikely that _ all _ these other objects have had progenitors from the delayed component , we are left with two possibilities : either the amount of mass loss during the pre - sn evolution of prompt type ia progenitors has a large dynamic range or there is more than one way to produce type ia sne with short delay times . the properties of snr [email protected] are also remarkable , for entirely different reasons . @xcite found @xmath77 for this sn , which translates into to a @xmath78 magnitude at maximum light close to @xmath79 @xcite . yet , the sn exploded in a large region of the lmc with very little sf in the recent past ( see figure [ fig-3 ] ) . the stars in the subcell that contains this snr are on average very old ( @xmath80 gyr ) and metal - poor ( @xmath81 ) . @xcite do find some relatively bright type ia sne associated with old stellar populations , but all their objects with peak @xmath78 magnitude above @xmath82 and ages above 5 gyr have large error bars on the age axis ( see their figure 5 ) . thus , snr [email protected] probably the first _ bona fide _ example of an exceptionally bright type ia sn associated with an old stellar population . we note that our measurement of @xmath46 should be very reliable , because it has not been derived from a luminosity - weighted spectrum . it is important to stress that these bulk properties of the stellar population around snr [email protected] do not preclude a relatively young progenitor for this object . during the age range that we have adopted for prompt type ia progenitors , @xmath83 of stars were formed in the subcell that contains snr [email protected] . with a salpeter imf , roughly @xmath84 of this mass is in the 4 to 6 @xmath85 range ( the zams masses that give co wds on timescales shorter than 180 myr according to * ? ? ? * ) , which results in a few hundred co wds from young stars . this number may seem small , but observational constraints on the percentage of co wds that eventually explode as sn ia are high ( 2 to 40 % according to * ? ? ? * ) , making a prompt progenitor for snr [email protected] a perfectly reasonable possibility . the fact that an object like snr [email protected] appears in a sample of only four snrs implies that bright type ia sne in old stellar populations may not be an exceptional occurrence , which should be taken into account when examining the contribution of bright and dim type ia sne in cosmological studies . our results underline the dangers of trying to understand the behavior of type ia sn progenitors by studying only the bulk properties of unresolved stellar populations in distant galaxies . if the lmc had been a distant type ia sn host , two objects with such radically different sfhs as snrs n103b and [email protected] would have been assigned the same age and metallicity . even average quantities obtained from resolved stellar populations like @xmath46 and @xmath45 can be misleading if they are used by themselves to characterize the properties of sn progenitors - compare the values for cc and type ia snrs from tables [ tab-2 ] and [ tab-3 ] . in this paper , we have presented the first systematic study of the stellar populations around cc and type ia snrs in the lmc . our ultimate goal is to use all the available information on the x - ray emitting snrs and the resolved stellar populations of the magellanic clouds to improve our understanding of cc and type ia sn progenitors , their final evolutionary stages , the sn explosions that mark their demise , and the aftermath of these explosions . in that broad context , this work only represents a first exploration of the many possibilities that are opened by recent theoretical and observational advances in both snr research and stellar population studies . we plan to pursue this line of research in the future , increasing the sample of objects and refining the techniques that we have presented here . we have found that the local sfhs around the cc snrs in our sample ( n49 , sn 1987a , n63a , and n158a ) are always dominated by significant episodes of sf in the recent past ( @xmath49 myr ) , as expected from previous observational and theoretical work . the timing and intensity of these sf episodes can provide interesting constraints on the masses of cc sn progenitors , but more work is needed to explore the full potential of this method . the local sfhs have also allowed us to study the ages and metallicities of the stellar populations around our target type ia snrs ( dem l71 , n103b , [email protected] , and [email protected] ) in great detail . we have found that type ia sne explode in a variety of environments , ranging from old and metal - poor populations to sites with vigorous sf in the recent past . using the two - component model proposed by @xcite , we have explored the relationship between specific properties of type ia sne and their parent stellar populations . we have seen that extremely bright type ia sne can explode very far away from any significant star formation activity ( snr [email protected] ) , and that type ia sne associated with young stellar populations might sometimes experience significant mass - loss before they explode ( snr n103b ) . recent studies of extragalactic type ia sn rates and our own findings suggest that reality is probably too complex to be explained with the popular two - component progenitor model . if this is so , high - quality sfhs for type ia snrs obtained from resolved stellar populations like the ones we present here should provide an excellent observational constraint on new ideas about type ia progenitors . this work has benefited from many conversations with a large number of colleagues , including steve bickerton , jeremy goodman , jim gunn , jack hughes , raul jimnez , vicky kaspi , dan maoz , maryam modjaz , evan scannapieco , jerry sellwood , and kris stanek . we are also grateful to the referee , stephen smartt , for many helpful suggestions that improved the quality of our manuscript . support for this work was provided by nasa through _ chandra _ postdoctoral fellowship award pf6 - 70046 issued by the _ chandra _ x - ray observatory center , which is operated by the smitsonian astrophysical observatory for and on behalf of nasa under contract nas8 - 0360 . dz acknowledges support from nasa ltsa grant nng05ge82 g . jlp is supported by nsf grant ast-0707982 . [email protected] & dem l71 & 05h 05 m 42s & -67d 52 m 39s & 1.4x1.0 ( h03 ) & @xmath22 ( g03 ) & snr dynamics + [email protected] & n103b & 05h 08 m 59s & -68d 43 m 35s & 0.46 ( b07 ) & @xmath16860 ( r05 ) & light echo + [email protected] & & 05h 09 m 31s & -67d 31 m 17s & 0.48 ( b07 ) & @xmath86 ( b08,r05 ) & light echo + [email protected] & & 05h 19 m 35s & -69d 02 m 09s & 0.52 ( b07 ) & @xmath25 ( r05 ) & light echo + [email protected] & n49 & 05h 26 m 00s & -66d 04 m 57s & 1.4 ( p03 ) & @xmath87 ( v06 ) & snr dynamics + sn 1987a & sn 1987a & 05h 35 m 28s & -69d 16 m 11s & 0.03 ( n08 ) & 22 & sn + [email protected] & n63a & 05h 35 m 44s & -66d 02 m 14s & 1.1 ( w03 ) & @xmath88 ( h98 ) & snr dynamics + [email protected] & n158a & 05h 40 m 11s & -69d 19 m 55s & 1.3x0.7 ( h01 ) & @xmath16800 ( c05 ) & pulsar n49 & ob ( c88 ) , co ? ( g01 ) & unknown ( iip ? , b07 ) & 0.0018 & 6.0 & @xmath89 & @xmath90 & @xmath89 + sn 1987a & sn & 1987-like / iipec & 0.0043 & 8.7 & @xmath27 & @xmath91 & @xmath92 + n63a & ob ( c88 ) & unknown ( ib / c ? , h98 ) & 0.0013 & 6.4 & @xmath89 & @xmath11 & @xmath93 + n158a & co ( k89 ) & iip ( w08 ) or ib / c ( c05 ) & 0.0036 & 7.8 & @xmath11 & @xmath94 & @xmath95 dem l71 & x - ray ( h95 ) & normal ? ( bh09 ) & 0.0022 & 8.3 & 28 & 72 + n103b & x - ray ( h95 ) & bright ? ( bh09 ) & 0.0023 & 8.1 & 73 & 27 + [email protected] & x - ray ( h95 ) & very bright ( b08,r08 ) & 0.0014 & 7.9 & 49 & 51 + [email protected] & x - ray ( h95 ) & bright ( bh09 ) & 0.0032 & 7.7 & 41 & 59

a commutative but not cocommutative graded hopf algebra , based on ordered rooted trees , is studied . this hopf algebra generalizes the hopf algebraic structure of unordered rooted trees , developed by butcher in his study of runge kutta methods and later rediscovered by connes and moscovici in the context of non - commutative geometry and by kreimer where it is used to describe renormalization in quantum field theory . it is shown that is naturally obtained from a universal object in a category of non - commutative derivations , and in particular , it forms a foundation for the study of numerical integrators based on non - commutative lie group actions on a manifold . recursive and non - recursive definitions of the coproduct and the antipode are derived . it is also shown that the dual of is a hopf algebra of grossman and larson . contains two well - known hopf algebras as special cases : the hopf algebra of butcher connes kreimer is identified as a proper subalgebra of using the image of a tree symmetrization operator . the hopf algebra of the free associative algebra is obtained from by a quotient construction .

math0603023

since cayley @xcite in 1857 and 100 years later merson @xcite and followed shortly after by butcher @xcite , it has been known that rooted trees are extremely useful for structuring algebras of differential operators and elementary differentials . in 1972 butcher @xcite produced the far - reaching result that runge kutta methods form a group . this was later named the butcher group in the paper by hairer and wanner @xcite , who made significant contributions to this theory . the butcher group is defined on the dual of the tree space and it was pointed out by dr @xcite that there exists a one - to - one correspondence between the butcher group and the commutative graded hopf algebra of unordered rooted trees . the hopf algebra of unordered rooted trees has had far - reaching applications in various areas of mathematics and physics . in 1998 a hopf subalgebra was discovered by connes and moscovici @xcite during work on an index theorem in non - commutative geometry and by kreimer @xcite in the renormalization method of quantum field theory . further collaborations between connes and kreimer @xcite have lead to other surprising results ; notably a connection with the riemann hilbert problem was established . brouder @xcite realized that the mathematical structure of connes kreimer was the same as that of butcher . grossman and larson @xcite also developed a cocommutative graded hopf algebra on a general class of rooted trees . it was shown by foissy @xcite and hoffman @xcite that the commutative hopf algebra of butcher and connes kreimer was the dual of the cocommutative hopf algebra of grossman larson , which corrected the original result of panaite @xcite . murua @xcite has developed series expansions of elementary differential operators and shown among other results that the logarithm of such a series is equivalent to the series expansions obtained from backward error analysis . recently a great deal of interest has been focused on developing numerical methods which preserve geometric properties of the exact flow . in particular , lie group integrators , which describe integrators that use lie group actions on manifolds , were originally proposed by crouch and grossman @xcite followed shortly after by lewis and simo @xcite . integrators of this type are now known as lie group integrators ; a survey of these methods is given in @xcite . series expansions for various classes of lie group methods have been developed ; these expansions are generally used to analyze order . munthe - kaas @xcite constructed the order conditions for a special subclass of lie group methods , where the computations are performed in a lie algebra , which is a linear space . later , @xcite it was shown that the classical order conditions could be used along with a certain transformation . owren and marthinsen @xcite developed the general order conditions for the crouch grossman methods with their analysis being based on ordered rooted trees . recently , owren @xcite derived the order conditions for the commutator free lie group methods @xcite , which were derived to overcome some of the problems associated with computing commutators . in this paper we aim to construct a commutative graded hopf algebraic structure , which can be used to analyze the order of all lie group methods . the outline of this paper is as follows : in section [ sect : derivation ] we will introduce ordered trees and forests and describe some useful operations on them . we will motivate the present hopf algebra as a universal object in a category of non - commutative derivations , and also briefly discuss lie butcher theory which will be treated in more detail in @xcite . in section [ sect : hopf ] we develop the hopf algebra of ordered trees , giving both recursive and non - recursive definitions of the coproduct and antipode , using certain cutting operations on ordered rooted trees . we show that the hopf algebra described in this paper is the dual of the grossman larson hopf algebra , thus generalizing the result of hoffman @xcite . finally , in section [ sect : homo ] we use a symmeterization operator to provide an injective hopf algebra homomorphism from the unordered trees into the ordered trees , establishing the former as a sub - algebra of the latter . in this section we will define an algebra @xmath3 spanned by forests of ordered ( and possibly colored ) rooted trees . this algebra is a universal ( ` free ' ) object in a general category of non - commutative derivation algebras , and plays a role in symbolic computing with lie butcher series similar to the role of _ free lie algebras _ @xcite in symbolic computing with lie algebras . let @xmath4 denote the set of _ ordered colored rooted trees _ , and @xmath5 denote the ( empty and nonempty ) words over the alphabet @xmath4 , henceforth called the set of empty and non - empty _ forests_. it should be noted that , unlike the classical butcher theory , the ordering of the branches in the trees in @xmath4 is important , and likewise the ordering of the trees within the forest @xmath5 . the basic operations involved in building @xmath4 and @xmath5 are : * create the empty forest @xmath6 . * create a longer forest from shorter forests by concatenation , @xmath7 . * create a tree from a forest by adding a root node , @xmath8 . in the instance where we wish to color the nodes using a set of colors @xmath9 , we introduce an indexed family of root adding operations @xmath10 for all @xmath11 . the inverse operation whereby we create a forest from a tree by removing the root node is written @xmath12 . this operation extends to @xmath5 by @xmath13 and @xmath14 . the total number of forests , with @xmath15 nodes colored in @xmath16 different ways , is defined by modifying the definition of the well known catalan numbers @xmath17 see a000108 in @xcite for various combinatorial representations of the catalan numbers . for a forest @xmath18 we define the _ degree _ , @xmath19 , as the number of trees in @xmath20 as : @xmath21 and the _ order _ , @xmath22 , as the total number of nodes in all the trees of @xmath20 as : @xmath23 we let @xmath24 denote the linear space of all finite @xmath25-linear combinations of elements in @xmath5 . this vector space is naturally equipped with an inner product such that all forests are orthogonal , @xmath26 for @xmath27 and @xmath18 , we let @xmath28 denote the coefficient of the forest @xmath20 , thus @xmath29 can be written as a sum @xmath30 where all but a finite number of terms are zero . the space of all infinite sums of this kind is denoted @xmath31 and is the dual space of @xmath3 , that is @xmath32 we again let @xmath33 denote the value of @xmath34 on @xmath35 . the operations @xmath36 , @xmath37 and concatenation extend to @xmath3 by linearity and the distributive law of concatenation , that is @xmath38 the vector space @xmath3 with the concatenation product and the grading @xmath39 forms a graded associative algebra @xmath40 , where @xmath41 denotes the linear combination of forests with @xmath42 trees . alternatively , it is possible to grade this algebra using @xmath43 ; in this case , @xmath41 denotes the linear combination of forests with the same number of nodes . now we introduce a _ left grafting product _ which has the algebraic structure of a derivation . [ defn : grafting ] for @xmath44 , define the left grafting @xmath45\in n$ ] by the following recursion formulae , where @xmath46 and @xmath47 : [ eq : grafting ] @xmath48 & = 0 , \\ \tau[\omega\widetilde{\omega } ] & = ( \tau[\omega])\widetilde{\omega}+ \omega(\tau[\widetilde{\omega}]),\\ \tau[b^+_i(\omega ) ] & = b^+_i(\tau[\omega])+b^+_i(\tau\omega ) , \end{split}\ ] ] @xmath49 & = a,\\ ( \tau\omega)[a ] & = \tau[\omega[a ] ] - ( \tau[\omega])[a ] . \end{split}\ ] ] the definition of left grafting is extended to the general case @xmath45 $ ] by bilinearity . it is useful to understand left grafting directly rather than via the recursive definition . from we verify that if @xmath46 and @xmath18 then @xmath50 $ ] is a sum of @xmath22 words , each word obtained by attaching the root of @xmath51 with an edge to the left side of a node of @xmath20 . @xmath52=\aaabbb\aababb\aabb+ \ab\aaabbababb\aabb+\ab\aabaaabbbb\aabb+\ab\aaaabbbabb\aabb+ \ab\aababb\aaabbabb+\ab\aababb\aaaabbbb\ ] ] from we see that if @xmath53 and @xmath18 then @xmath54 $ ] is obtained by first left - grafting @xmath55 to all nodes of @xmath20 and then left - grafting @xmath56 to all the nodes of the resulting expression , except to the nodes coming originally from @xmath55 . @xmath57=\aababb\aabb+ \aabb\aababb+\aabb\aaabbb+\aabb\aababb+\ab\aabababb+ \ab\aabaabbb+\aabb\aaabbb+\ab\aabaabbb+\ab\aaababbb\ ] ] [ lem:1 ] if @xmath58 and @xmath18 then @xmath59 $ ] is a sum of @xmath60 words obtained by , in the order @xmath61 , attaching the root of the tree @xmath62 with an edge to the left side of any node in @xmath20 . in particular we have @xmath63 = b^+_i(\omega ) , \qquad\mbox{for all $ \omega \in\of $ .}\ ] ] equations and imply that for any @xmath64 and @xmath65 , we have the leibniz rule and a composition rule of the form @xmath66 & = d[a]{b } + a d[{b } ] \\ \label{eq : composit } d[a[{b } ] ] & = da[{b } ] + d[a][{b}].\end{aligned}\ ] ] thus @xmath67 acts as a first degree derivation on @xmath3 . [ defn : gl ] the grossman larson ( gl ) product @xmath68 is defined as @xmath69 , \qquad\mbox{for all $ \omega,\widetilde{\omega}\in\of$,}\ ] ] and is extended to the general @xmath70 for @xmath71 by linearity . since @xmath72\right)$ ] , the gl product can be understood by adding an invisible root to @xmath73 ( turning it into a tree ) , and left - grafting @xmath20 onto all nodes of @xmath74 , including the invisible root . the root is then removed from each of the resulting trees , with the gl product resulting in a total of @xmath75 forests . some examples of the grafting product from definition [ defn : grafting ] and the gl product from definition [ defn : gl ] are given in table [ tab : glproduct ] . in fact , the theory of grossman and larson @xcite is formulated on trees , not on forests of trees . to a forest @xmath20 in our terminology , they add a ( proper ) root to turn it into a tree . the definition of the gl product in @xcite is modified accordingly . the gl product is an associative @xmath39-graded product , @xmath76 , satisfying for all @xmath77 , @xmath78 , and @xmath79 & = b^-(a[b^+_i(b)])[c]\\ & = b^-(b^+_i(a[b])+b^+_i(ab))[c]\\ & = ( a[b]+(ab))[c]\\ & = a[b[{c } ] ] . \end{aligned}\ ] ] @xmath80 & \omega\opr\tau \\[2 mm ] \hline & & \\[-1 mm ] \ab\tpr\ab & \aabb & \ab\ab+\aabb \\[2 mm ] \ab\ab\tpr\ab & \aababb & \ab\ab\ab+2\ab\aabb+\aababb \\[2 mm ] \ab\tpr\ab\ab & \aabb\ab+\ab\aabb & \ab\ab\ab+\aabb\ab+\ab\aabb \\[2 mm ] \aabb\tpr\ab & \aaabbb & \aabb\ab+\aaabbb \\[2 mm ] \ab\tpr\aabb & \aababb+\aaabbb & \ab\aabb+\aababb+\aaabbb \\[2 mm ] \ab\tpr\ab\ab\ab & \aabb\ab\ab+\ab\aabb\ab+\ab\ab\aabb & \ab\ab\ab\ab+\aabb\ab\ab+\ab\aabb\ab+\ab\ab\aabb \\[2 mm ] \ab\ab\ab\tpr\ab & \aabababb & \ab\ab\ab\ab+3\ab\ab\aabb+3\ab\aababb+\aabababb \\[2 mm ] \ab\aabb\tpr\ab & \aabaabbb & \ab\aabb\ab+\ab\aaabbb+\aabb\aabb+\aabaabbb \\[2 mm ] \aabb\ab\tpr\ab & \aaabbabb & \aabb\ab\ab+\aabb\aabb+\ab\aaabbb+\aaabbabb \\[2 mm ] \ab\tpr\ab\aabb & \aabb\aabb+\ab\aababb+\ab\aaabbb & \ab\ab\aabb+\aabb\aabb+\ab\aababb+\ab\aaabbb \\[2 mm ] \ab\tpr\aabb\ab & \aababb\ab+\aaabbb\ab+\aabb\aabb & \ab\aabb\ab+\aababb\ab+\aaabbb\ab+\aabb\aabb \\[2 mm ] \ab\tpr\aaabbb & \aabaabbb+\aaababbb+\aaaabbbb & \ab\aaabbb+\aabaabbb+\aaababbb+\aaaabbbb \\[2 mm ] \aaabbb\tpr\ab & \aaaabbbb & \aaabbb\ab+\aaaabbbb \\[2 mm ] \ab\tpr\aababb & \aabababb+\aaabbabb+\aabaabbb & \ab\aababb+\aabababb+\aaabbabb+\aabaabbb \\[2 mm ] \aababb\tpr\ab & \aaababbb & \aababb\ab+\aaababbb \\[2 mm ] \ab\ab\tpr\ab\ab & 2\aabb\aabb+\aababb\ab+\ab\aababb & \ab\ab\ab\ab+2\ab\aabb\ab+2\ab\ab\aabb+2\aabb\aabb+ \aababb\ab+\ab\aababb \\[2 mm ] \ab\ab\tpr\aabb & \aabababb+2\aabaabbb+\aaababbb & \ab\ab\aabb+2\ab\aababb+2\ab\aaabbb+\aabababb+2\aabaabbb+\aaababbb \\[2 mm ] \aabb\tpr\ab\ab & \aaabbb\ab+\ab\aaabbb & \aabb\ab\ab+\aaabbb\ab+\ab\aaabbb \\[2 mm ] \aabb\tpr\aabb & \aaabbabb+\aaaabbbb & \aabb\aabb+\aaabbabb+\aaaabbbb \\[2 mm ] \hline \end{array}\ ] ] let @xmath81 be an associative @xmath82-graded algebra @xmath83 with associative product @xmath84 , a unit @xmath85 and grading @xmath86 such that @xmath87 . we call @xmath81 a _ d - algebra _ if it is also equipped with a linear derivation @xmath88}{d\tpr d}{d}$ ] such that and hold for any @xmath89 and any @xmath90 . define a d - algebra homomorphism as a linear degree preserving map @xmath91 between d - algebras satisfying for any @xmath92 : @xmath93 ) & = \f(a)[\f(b)].\end{aligned}\ ] ] let @xmath3 be the algebra of forests colored with a set @xmath9 . for any d - algebra @xmath81 and any map @xmath94 , there exists a unique homomorphism @xmath95 such that @xmath96 . from and we find @xmath97 $ ] , for any @xmath18 . together with and linearity , this shows that by recursion , we can extend @xmath91 to a uniquely defined homomorphism defined on all of @xmath3 . @xmath98 this shows that @xmath3 is a universal object , free over the set @xmath9 , in the category of @xmath81-algebras . as an example of a d - algebra , we consider an algebra related to the numerical lie group integrators . let @xmath100 be a lie algebra of vector fields on a manifold @xmath101 and let @xmath102 , denote the flow operator . a basic assumption of numerical lie group integrators @xcite is the existence of a @xmath100 which is transitive ( i.e. spans all tangent directions in any point on @xmath103 ) , and for which the exponential map can be computed efficiently and exactly . transitivity implies that _ any _ vector field can be written in terms of a function @xmath104 . the goal of numerical lie group integrators is to approximate the flow of a general differential equation @xmath105 by composing exponentials of elements in @xmath100 . the study of order conditions for lie group integrators leads to a need for understanding the algebraic structure of non - commuting vector fields on @xmath101 , generated from @xmath106 . elements @xmath107 are often called invariant or ` frozen ' vector fields on @xmath101 . these define first - degree invariant differential operators through the lie derivative . let @xmath108 be any normed vector space and denote @xmath109 the set of all smooth functions from @xmath101 to @xmath108 , called the space of _ @xmath108-sections_. for @xmath110 and @xmath111 , the lie derivative , @xmath112\in \left(\m\rightarrow \mathcal{v}\right)$ ] , is defined as @xmath113(p ) = \ddto\psi\left(\exp(tv)(p)\right ) , \qquad\mbox{for any point $ p\in \m$}.\ ] ] for two elements @xmath114 we iterate this definition and define the concatenation @xmath115 as the second degree invariant differential operator @xmath116 = v[w[\psi]]$ ] . the linear space spanned by the @xmath117-degree identity operator @xmath118 = \psi$ ] and all higher degree invariant derivations is called the _ universal enveloping algebra _ of @xmath100 , denoted @xmath99 . this is a graded algebra with the concatenation product and degree @xmath119 , @xmath120 and @xmath121 . given a norm on the vector space @xmath99 , we consider the space of @xmath99 sections is a trivial vector bundle over @xmath101 , @xmath100 a trivial sub - bundle and the tangent bundle @xmath122 is a non - trivial sub - bundle of @xmath100 . ] for two sections @xmath124 we define @xmath125\in\left(\m\rightarrow\ge\right)$ ] pointwise from the lie derivative as @xmath126(p ) = \left(f(p)[g]\right)(p ) , \qquad p\in\m .\ ] ] similarly , the concatenation on @xmath99 is extended pointwise to a concatenation @xmath127 as @xmath128 from these definitions we find : [ lem : dalgebra ] let @xmath129 and @xmath130 . then : @xmath131 & = f[g]h + gf[h ] , \\ ( f\opr g)[h ] \equiv f[g[h ] ] & = fg[h ] + f[g][h ] . \end{aligned}\ ] ] for @xmath132 let @xmath133 . then @xmath131(p ) & = \ddto ( gh)(\exp(tv)(p))\\ & = \ddto g(\exp(tv)(p))h(\exp(tv)(p))\\ & = \ddto g(\exp(tv)(p))h(p ) + g(p ) h(\exp(tv)(p))\\ & = \left(f[g]h + gf[h]\right)(p),\\[2 mm ] ( f[g[h]])(p ) & = \ddto(g(\exp(tf)(p))[h](\exp(tf)(p))\\ & = \ddto g(p)[h](\exp(tf)(p))+g(\exp(tf)(p))[h](p)\\ & = ( fg)[h](p ) + ( f[g])[h](p ) . \end{aligned}\ ] ] @xmath98 note the difference between @xmath134 and @xmath135 . in the concatenation the value of @xmath136 is frozen to @xmath137 before the differentiation with @xmath106 is done , whereas in the latter case the spatial variation of @xmath136 is seen by the differentiation using @xmath106 . lemma [ lem : dalgebra ] shows that @xmath138 is a @xmath81-algebra . thus if we , for every @xmath11 , pick a vector field @xmath139 then there exists a unique homomorphism @xmath140 such that @xmath141 . the images of the trees @xmath142 , for @xmath46 , are called the _ elementary differentials _ in butcher s theory ( see @xcite ) and the images of the forests @xmath143 , for @xmath18 , are called _ elementary differential operators _ in merson s theory ( see @xcite ) . to motivate the algebraic structures of the next section , we briefly introduce some elements of lie butcher theory . this theory is the non - commutative generalization of the classical butcher theory and is the general foundation behind the construction of order conditions for lie group integrators . various aspects of this theory have been developed in @xcite . a comprehensive treatment is given in @xcite . with the vector space @xmath3 being defined as the finite @xmath25-linear combinations of @xmath5 , we now let @xmath31 denote the space of infinite @xmath25-linear combinations ( sequences ) , or the algebraic dual space of @xmath3 . all the operations of the previous paragraphs extend to @xmath31 by local finiteness , see @xcite . for example , the inner - product in is extended to a dual pairing @xmath144 , where the computation @xmath145 is always finite , since @xmath146 is required to be finite . consider the homomorphism @xmath91 introduced in section [ sect : gsection ] extended to a homomorphism of infinite series @xmath147 , where @xmath148 should now be understood as a space of formal series . the series might not converge , but all definitions make sense termwise , and any finite truncation yields a proper @xmath99-section . in classical ( commutative ) butcher theory the image of @xmath34 is called an s - series , see murua @xcite . similarly , we define an ls - series as an infinite formal series in @xmath149 , given by on the chosen normalization . ] @xmath150 classical lie series on manifolds is a generalization of taylor series , where the fundamental result is the following ` pull - back formula ' : let @xmath151 , be a vector field and @xmath152 , be the @xmath153-flow . for any @xmath154 we have ( see @xcite ) that @xmath155(p)\equiv \operatorname{exp}(f)[g](p),\ ] ] where @xmath156 and @xmath157 = f[\cdots f[f[g]]\cdots ] = ( f\opr\cdots\opr f)[g]$ ] . note that if @xmath158 , then @xmath159 , thus the operator exponential @xmath160 is a ls - series . two special cases of ls series are of particular importance : a ls - series @xmath161 is called _ logarithmic _ or _ algebra - like _ if @xmath162 represents a vector field , and it is is called _ exponential _ or _ group - like _ if @xmath161 is the ( formal ) operator exponential of a logarithmic series . a logarithmic ls - series is the non - commutative generalization of a b - series , named the _ lie butcher series _ @xcite . note that if @xmath53 , then @xmath163 is a logarithmic series since it represents the commutator of two vector fields . more generally , a series @xmath34 is logarithmic , if and only if , all its finite components belong to the free lie algebra generated by @xmath4 . a hall basis for this space is characterized in @xcite . reutenauer @xcite presents several alternative characterizations of logarithmic and exponential series . we find the characterization in terms of _ shuffle products _ particularly useful . the shuffle product @xmath164 is defined for two forests as the summation over all permutations of the trees in the forests while preserving the ordering of the trees in each of the initial forests , and is extended to @xmath165 by linearity . it can also be recursively defined in the asymmetric way @xmath166 for any forest @xmath18 , and if @xmath167 and @xmath168 for @xmath53 and @xmath169 , then @xmath170 the shuffle product is associative and commutative , for all @xmath171 we have @xmath172 table [ tab : product ] gives some simple , but nontrivial , examples of the shuffle product . @xmath173 \hline & \\[-1 mm ] \ab\tpr\ab & 2\ab\ab \\[2 mm ] \ab\tpr\ab\ab & 3\ab\ab\ab \\[2 mm ] \ab\tpr\aabb & \aabb\ab+\ab\aabb \\[2 mm ] \ab\tpr\ab\ab\ab & 4\ab\ab\ab\ab \\[2 mm ] \ab\tpr\ab\aabb & 2\ab\ab\aabb+\ab\aabb\ab \\[2 mm ] \ab\tpr\aabb\ab & 2\aabb\ab\ab+\ab\aabb\ab \\[2 mm ] \ab\tpr\aaabbb & \ab\aaabbb+\aaabbb\ab \\[2 mm ] \ab\tpr\aababb & \ab\aababb+\aababb\ab \\[2 mm ] \ab\ab\tpr\ab\ab & 6\ab\ab\ab\ab \\[2 mm ] \ab\ab\tpr\aabb & \ab\ab\aabb+\ab\aabb\ab+\aabb\ab\ab \\[2 mm ] \aabb\tpr\aabb & 2\aabb\aabb \\[2 mm ] \hline \end{array}\ ] ] @xcite [ lem : log_exp ] a series @xmath34 is logarithmic if and only if @xmath174 a series @xmath34 is exponential if and only if @xmath175 the ls - series of an exponential series @xmath34 represents pull - backs , or finite motions on @xmath101 . they form a group under composition with the grossman larson product , which is the generalization of the _ butcher group _ to the case of non - commutative group actions . consider the gl product as a linear operator @xmath176 defined by @xmath177 . to compute the composition @xmath178 for general @xmath179 , it is useful to introduce the dual of @xmath180 , the linear map @xmath181 to be defined in section [ sect : recur ] . using corollary [ cor : dual ] , we find @xmath182 as an illustrative example , we read from table [ tab : coproduct ] that @xmath183 the inverse in the group is found from the _ antipode _ , a linear map @xmath184 defined in section [ sect : hopf ] . it can be shown that the dual of the antipode @xmath185 also defines the inverse in the group : @xmath186 for all exponential @xmath34 . in this section we will study a commutative graded hopf algebra @xmath0 of ordered trees . the coproduct @xmath187 in the algebra is defined by recursion formulae , and at a later stage we will show that this @xmath187 is the dual of the gl product thereby establishing the connection between this hopf algebra and the algebra of the butcher group . we begin by briefly reviewing the definition of a hopf algebra , see @xcite for details . a real associative _ algebra _ @xmath188 is a real vector space with an associative product @xmath189 and a unit @xmath190 such that @xmath191 for all @xmath192 . the dual of an algebra is called a _ coalgebra _ , @xmath193 , which is a vector space equipped with a coassociative coproduct @xmath194 and counit @xmath195 . a _ bialgebra _ @xmath196 is a linear space which is both an algebra and also a coalgebra structure such that the coproduct and the counit are compatible with the product , in the sense that @xmath197 where @xmath198 is the twist map . let @xmath199 denote all linear maps from @xmath196 to itself . we define the _ convolution _ @xmath200 as @xmath201 let @xmath202 denote the identity matrix . an _ antipode _ is a linear map @xmath203 , which is the two - sided inverse of the identity matrix under convolution , with the antipode satisfying @xmath204 a hopf algebra @xmath205 is a bialgebra equipped with an antipode . we will study a particular hopf algebra based on the vector space of ordered forests @xmath24 , where the coproduct is defined by the following recursion . [ def : coprod ] let @xmath181 be defined by linearity and the recursion @xmath206 where @xmath207 and @xmath47 . the linear operation + @xmath208 is a shuffle on the left and concatenation on the right , satisfying @xmath209 note that letting @xmath210 yields the special recursion formula for a tree @xmath51 : @xmath211 [ thm : hopf ] let @xmath0 be the vector space @xmath24 with the operations @xmath212 then @xmath0 is a hopf algebra with an antipode @xmath213 given by the recursion @xmath214 where @xmath207 and @xmath47 . in particular @xmath215 the coassociativity of the coalgebra will be established once we have shown that the algebra and coalgebra are compatible . from the fact that @xmath216 for all @xmath217 and that the shuffle product of two scalars is just standard multiplication , we immediately get . to show we find it convenient to introduce the linear operation @xmath218 with the shuffle product both on the left and the right , satisfying @xmath219 the compatibility condition is now equivalent to @xmath220 to simplify the notation we use the fact that @xmath221 let @xmath167 and @xmath168 for @xmath53 and @xmath169 , now using the recursive definition of the shuffle product and substituting the expression for the coproduct of an ordered forest yields @xmath222 we now have the tools needed to show that the coalgebra is coassociative , which follows from the fact that the coproduct @xmath187 satisfies @xmath223 thus we have established the structure of a bialgebra . substituting definition [ def : coprod ] in yields the recursion for the antipode . @xmath98 both the definition of the coproduct @xmath187 and thus the antipode @xmath213 are recursive and difficult to use in practice . to develop non - recursive formulae for these , it is first necessary to define certain cutting operations . for a given forest @xmath18 , a _ parent _ is any node @xmath224 with at least one branch growing from that node and the _ children _ are the nodes branching from @xmath224 . let @xmath225 denote the number of children of @xmath224 . cutting off a child node equates to removing the edge connecting the child to its parent . * a _ nodal left cut _ of degree @xmath226 is a cut where the @xmath226 leftmost children of a given parent node @xmath224 are cut off . we can write a nodal left cut as @xmath227 where @xmath228 . the cut splits a forest @xmath20 into two sub - forests , @xmath229 and @xmath230 , where @xmath231 is the part cut off , with the forest containing the @xmath226 leftmost children of @xmath224 as root nodes and @xmath232 is the remaining bottom part of @xmath20 . * a _ left cut _ is a collection of @xmath233 nodal left cuts @xmath234 , where @xmath235 are distinct nodes of @xmath20 . this splits @xmath20 in @xmath236 cut - off forests @xmath237 and a remaining forest @xmath238 , where @xmath239 is the forest containing the nodes connected to the @xmath240 leftmost children of @xmath241 and @xmath238 is the forest of the nodes connected to the original root nodes . we define @xmath242 as @xmath243 note that the definition of a left cut includes the case @xmath244 , called the _ empty cut _ , where @xmath245 and we define @xmath246 . * an _ admissible left cut _ is a left cut , containing the restriction that any path from a node in @xmath20 to the corresponding root is cut no more than once . we denote by lc , nlc and alc the set of all left cuts , nodal left cuts and admissible left cuts . to define the coproduct we need to slightly extend the definition of an admissible left cut , which we choose to call a _ full admissible left cut_. the full admissible left cuts of @xmath18 are obtained by adding an ( invisible ) root node to form the tree @xmath247 , applying an admissible left cut on @xmath51 , and finally removing the invisible root node again . we denote by falc the set of all full admissible left cuts . thus @xmath248 and for any @xmath249 we have @xmath250 and @xmath251 . note that @xmath252 contains the ` cut everything ' , where @xmath253 , @xmath254 and @xmath255 , as well as the empty cut with @xmath244 , @xmath246 and @xmath256 . it is useful to note that the order in which the cuts are performed does not affect @xmath257 or @xmath238 . the order of the cuts is taken care of by the use of the shuffle product in the definition of @xmath257 given by equation . as an example , we list all the cuttings of an example tree in table [ tab : cutexample ] . @xmath258 \hline & & & & & & & & \\[-1 mm ] 0 & \trxl & 0 & \one & \trxl & 7 & \trxg & 2 & \aababb\sqcup\ab & \ab \\[2 mm ] 1 & \trxa & 1 & \ab & \aaababbb & 8 & \trxh & 2 & \aaabbb\sqcup\ab & \ab \\[2 mm ] 2 & \trxb & 1 & \ab & \aaaabbbb & 9 & \trxi & 2 & \ab\sqcup\ab\aabb & \ab \\[2 mm ] 3 & \trxc & 1 & \ab\aabb & \aabb & 10 & \trxj & 3 & \aabb\sqcup\ab\sqcup\ab & \ab \\[2 mm ] 4 & \trxd & 1 & \aabaabbb & \ab & 11 & \trxk & 3 & \ab\sqcup\ab\ab\sqcup\ab & \ab \\[2 mm ] 5 & \trxe & 2 & \ab\sqcup\ab & \aaabbb & 12 & \trxl & 1 & \aaabaabbbb & \one \\[2 mm ] 6 & \trxf & 2 & \ab\ab\sqcup\ab & \aabb & & & & & \\[2 mm ] \hline \end{array}\ ] ] the coproduct @xmath187 of @xmath0 is non - recursively defined as @xmath259 to prove that the recursive definition and the non - recursive definition of the coproduct are identical , an induction argument on the number of vertices is used . first recall that @xmath260 , and for any @xmath261 we have @xmath262 and @xmath263 , this implies that @xmath264 using this fact , the coproduct now takes the form @xmath265 the last equality is true because the sum over @xmath249 and @xmath266 is equivalent to the sum over @xmath267 except for the ` cut everything ' cut which is equal to the term @xmath268 . @xmath98 [ cor : dual ] the dual of the coproduct @xmath187 is @xmath180 the grossman larson product , that is for any @xmath269 and @xmath270 we have @xmath271 if the sum in had been over alc instead of falc , then the dual would have been the left grafting . to see this , we use the characterization of left grafting in lemma [ lem:1 ] , and observe that the nodal left cut corresponds to the dual operation of attaching a number of trees in a given order to a common node , while the shuffles in @xmath257 corresponds to the dual operation of attaching the forests in all possible ways to different nodes . from definition [ defn : gl ] , we see that when the sum is extended from alc to falc , then we obtain the dual of the grossman larson product . @xmath98 to present a non - recursive definition of the antipode , we define the reversal map @xmath272 as @xmath273 extended to @xmath3 by linearity . thus @xmath274 is the unique anti - automorphism of the concatenation algebra which sends @xmath275 . [ prop : antipode_nr ] the antipode @xmath213 of @xmath0 is non - recursively defined as @xmath276 in order to prove this result , we need some results about a hopf algebraic structure of the free associative algebra ( faa ) @xcite . given an alphabet @xmath277 , faa is the vector space formed by taking all finite linear combinations of words over @xmath277 . in our case , the alphabet is @xmath4 , the words are @xmath5 and the vector space is @xmath3 . a hopf algebraic structure @xmath2 is obtained by taking the product @xmath278 as the shuffle product and the coproduct @xmath279 defined as the dual of the concatenation product . the antipode is the map @xmath280 defined in . we need a characterization of @xmath279 and @xmath280 in terms of cutting operations . for a @xmath281 let the set of word cuts ( wc ) be a simple cut @xmath282 which splits a word @xmath20 into two parts @xmath283 and @xmath284 such that @xmath285 . wc contains both the empty cut where @xmath246 , @xmath245 and cut everything where @xmath254 , @xmath286 . note that the difference between alc and falc is that falc may contain a nonempty cut from wc . a direct definition of @xmath279 is @xmath287 from we find for @xmath217 that @xmath288 thus we find a recursive definition of the antipode @xmath280 @xmath289 where _ c.e . _ denotes _ cut everything_. now we repeat the same computation for @xmath213 , using . this gives the recursive definition of the antipode @xmath213 @xmath290 we prove by induction on the number of nodes . plugging into , we find for @xmath217 that @xmath291 we have used the relation @xmath292 see corollary [ coro : connect ] and the recursion , as well as a careful replacement of the summations over @xmath249 and @xmath293 with an equivalent sum over @xmath294 and @xmath295 . @xmath98 as an example , we compute @xmath187 and @xmath213 for the word @xmath296 . the cuts lc and falc are shown in table [ tab : directex ] . from the direct formulae we find @xmath297 and @xmath298 as listed in table [ tab : coproduct ] and table [ tab : antipode ] . @xmath299 \hline & & & & & & & & & \\[-1 mm ] 0 & \ab\aaabbb & 0 & \one & \ab\aaabbb & 4 & \ab\aaabbb & 1 & \ab & \aaabbb \\[2 mm ] 1 & \ab\aaabbbz & 1 & \ab & \ab\aabb & 5 & \ab\aaabbbz & 2 & \ab\sqcup\ab & \aabb \\[2 mm ] 2 & \ab\aaabbby & 1 & \aabb & \ab\ab & 6 & \ab\aaabbby & 2 & \ab\sqcup\aabb & \ab \\ [ 2 mm ] 3 & \ab\aaabbbx & 2 & \ab\sqcup\ab & \ab\ab & 7 & \ab\aaabbb & 1 & \ab\aaabbb & \one \\[2 mm ] \hline \end{array}\ ] ] we complete this section by listing some well known but very useful relations of hopf algebras , see sweedler @xcite for further details . [ coro : connect ] @xcite given @xmath0 a hopf algebra , with product @xmath300 , coproduct @xmath187 and antipode @xmath213 , then for all @xmath301 @xmath302 if @xmath0 is either commutative or cocommutative , then @xmath303 for all @xmath18 . there are two interesting commutative graded hopf sub - algebras of @xmath0 obtained by restricting from the set of ordered rooted trees to either the set of tall trees ( that is trees where each parent has one child ) or bushy trees ( that is trees where there is only one parent ) . these hopf sub - algebras are useful respectively for determining the order conditions for the problem , when @xmath304 is constant , or when the numerical scheme has high stage order . in this section we will also show that the hopf algebra @xmath1 of butcher , based on unordered trees , can be identified as a sub - algebra of @xmath0 . finally we find that the hopf algebra @xmath2 of the free associative algebra is related to @xmath0 through the operation of _ freezing _ vector fields , which can be defined as a quotient construction on @xmath0 . let @xmath305 denote all unordered trees and @xmath306 denote all unordered forests , defined as the set of all empty or non - empty unordered words over the alphabet @xmath305 . recall from @xcite the following definition . given the real vector space @xmath307 , denote the hopf algebra of unordered forests as @xmath308 . the product @xmath309 is defined as the ( commutative ) concatenation @xmath310 the unit element @xmath311 , is given by @xmath312 . the coproduct @xmath313 is defined by linearity and for any @xmath314 and @xmath315 by the recursion @xmath316 the counit @xmath317 is defined by @xmath318 and @xmath216 for @xmath319 . the antipode @xmath320 is , as usual , the two - sided inverse of the convolution in @xmath1 , see @xcite for details . the main tool used to provide the relationship between the hopf algebras of ordered and unordered forests is the symmetrization operator defined below . [ defn : symm ] the symmetrization operator @xmath321 is a map defined by linearity and the relations @xmath322 the shuffle product permutes the trees in a forest in all possible ways , and the symmetrization of a tree is a recursive splitting in sums over all permutations of the branches . the symmetrization defines an equivalence relation on @xmath5 , that is , if @xmath323 thus @xmath324 if and only if @xmath325 can be obtained from @xmath326 by permuting the order of the trees in the forest and the order of the branches of the trees . we see that an alternative characterization of @xmath327 is @xmath328 the integer @xmath329 is the classical symmetry coefficient , defined for trees and forests as @xmath330 where the integers @xmath331 count the number of equivalent trees among @xmath332 . in other words , if we consider the full group of all possible permutations of trees and branches acting on a forest @xmath281 , then @xmath329 is the size of the isotropy subgroup i.e. the number of permutations leaving @xmath20 invariant . the total number of permutations acting on a given forest @xmath20 is given by the integer @xmath333 defined as @xmath334 note that once a forest @xmath20 is symmetrized , then another application of the symmetrization yields the scaling @xmath335 let @xmath306 be the unordered forests . clearly , there is a 11 correspondence between unordered forests and equivalence classes of ordered forests , thus there is a natural isomorphism @xmath336 . through this identification , we can interpret @xmath327 as an injection @xmath337 where @xmath307 and @xmath24 . from we see that the map @xmath338 defined as @xmath339 where @xmath340 is the natural identification of an ordered forest with the corresponding unordered forest , defines a left - sided inverse @xmath341 for all @xmath342 . [ thm : hopfhomo ] the symmetrization operator @xmath337 defines an injective hopf algebra homomorphism from the hopf algebra @xmath1 of unordered forests into the hopf algebra @xmath0 of ordered forests . a hopf algebra homomorphism is a bialgebra homomorphism , which is a linear map that is both an algebra and a coalgebra homomorphism . @xmath343 is an algebra homomorphism if @xmath344 these conditions are automatically satisfied by definition [ defn : symm ] . @xmath343 is a coalgebra homomorphism if @xmath345 the first condition follows immediately given that the counits are only non - zero when the argument is the empty forest . the second relation follows using an induction argument . first we need to establish a useful relationship between @xmath327 and @xmath346 . using the _ sumless sweedler notation _ @xmath347 we find : @xmath348 now we prove by induction . for a forest we find using and @xmath349 for a tree @xmath350 and @xmath351 we find using the definition of @xmath327 and the recursion formulas and that @xmath352 the proof extends to a general @xmath27 by linearity . @xmath98 a consequence of the above theorem is a connection between the antipodes of the corresponding hopf algebras . given @xmath353 is a bialgebra homomorphism then for @xmath354 , it follows that @xmath355 note that symmetrization operator @xmath327 is invertible , so expressions for the product , coproduct and antipode of the hopf algebra of unordered forests can be directly expressed in terms of the corresponding functions in the hopf algebra of ordered forests , they are @xmath356 in the final part of this section we will elaborate on the connections between the ls - series built on ordered trees , and their commutative counterpart the s - series . these series belong respectively to the dual spaces @xmath31 and @xmath357 , are naturally associated through the dual map @xmath358 taking the series of ordered forests to unordered forests . if @xmath359 and @xmath360 , we find @xmath361 on a manifold with a commutative lie group action the elementary differential operators @xmath143 do not depend on the ordering . thus we find that the s - series of @xmath362 as defined in @xcite equals the ls - series of @xmath363 as given in , @xmath364 this shows that the normalization @xmath365 in the commutative case is compatible with our normalization in the non - commutative case . it is interesting to characterize the image of the _ logarithmic _ and _ exponential _ series under @xmath366 . if @xmath363 is logarithmic ( lemma [ lem : log_exp ] ) then @xmath367 thus @xmath362 is non - zero only on trees . if @xmath363 is exponential then @xmath368 this is a well known condition in the composition of b - series , see @xcite . in the proof of proposition [ prop : antipode_nr ] , we defined the hopf algebraic structure @xmath2 as the free associative algebra built of words over an alphabet @xmath277 , where @xmath369 is the shuffle and @xmath279 the dual of the concatenation product . we will briefly comment upon the connection between @xmath0 and @xmath2 in the context of lie group integrators . in the theory of lie group integrators , it is common to call constant sections @xmath370 _ frozen vector fields_. if @xmath106 is frozen , then the lie derivative @xmath371 = 0 $ ] for all vector fields @xmath136 . on the algebraic side , a tree @xmath372 represents a frozen vector field if the left grafting of anything non - constant to the tree is zero , @xmath373 = 0 , \qquad\mbox{for all $ \widetilde{\tau}\in \ot\backslash\{\one\}$}.\ ] ] in this case we see that the grossman larson product becomes just the concatenation product @xmath374 . the freezing of certain vector fields can be understood as the quotient @xmath375 , where @xmath376 is the linear span of any @xmath377 grafted to a frozen vector field . as a special example , we consider the case where all single - node trees are frozen , so that taller trees can not be produced . letting @xmath378 be the alphabet of all single - node trees , we find from the following well known recursion for the coproduct @xmath379 we have in this paper investigated the algebraic structure of the hopf algebra underlying numerical lie group integrators . we have developed both recursive and direct formulae for the coproduct and the antipode , and we have in particular emphasized the connection to the hopf algebra of classical butcher theory and to the free associative hopf algebra . we believe that this work is of particular interest for the construction of symbolic software packages dealing with computations involving algebras of non - commutative derivations . the algebraic structure of @xmath0 is of a universal nature and should be of interest also outside the field of numerical integration , for example , in the renormalization of quantum field theory and the chen fliess theory for optimal control . @xmath380 \hline \\[-1 mm ] \one & \one\tpr\one \\[1 mm ] \ab & \ab\tpr\one+\one\tpr\ab \\[2 mm ] \aabb & \aabb\tpr\one+\ab\tpr\ab+\one\tpr\aabb \\[2 mm ] \ab\ab & \ab\ab\tpr\one+\ab\tpr\ab+\one\tpr\ab\ab \\[2 mm ] \aaabbb & \aaabbb\tpr\one+\ab\tpr\aabb+\aabb\tpr\ab+ \one\tpr\aaabbb \\[2.5 mm ] \aababb & \aababb\tpr\one+\ab\ab\tpr\ab+\ab\tpr\aabb+ \one\tpr\aababb \\[2.5 mm ] \ab\aabb & \ab\aabb\tpr\one+2\ab\ab\tpr\ab + \ab\tpr\aabb+\ab\tpr\ab\ab+ \one\tpr\ab\aabb \\[2.5 mm ] \aabb\ab & \aabb\ab\tpr\one+\aabb\tpr\ab+ \ab\tpr\ab\ab+\one\tpr\aabb\ab \\[2 mm ] \ab\ab\ab & \ab\ab\ab\tpr\one+\ab\ab\tpr \ab+\ab\tpr\ab\ab+\one\tpr\ab\ab\ab \\[2 mm ] \aaaabbbb & \aaaabbbb\tpr\one+\aaabbb\tpr\ab+\aabb\tpr\aabb+ \ab\tpr\aaabbb+\one\tpr\aaaabbbb \\[2.5 mm ] \aaababbb & \aaababbb\tpr\one+\aababb\tpr\ab+\ab\ab\tpr\aabb+ \ab\tpr\aaabbb+\one\tpr\aaababbb \\[2.5 mm ] \aabaabbb & \aabaabbb\tpr\one+\ab\aabb\tpr\ab+2\ab\ab\tpr \aabb+\ab\tpr\aaabbb+\ab\tpr\aababb+\one\tpr\aabaabbb \\[2.5 mm ] \aaabbabb & \aaabbabb\tpr\one+\aabb\ab\tpr\ab+\aabb\tpr \aabb+\ab\tpr\aababb+\one\tpr\aaabbabb \\[2.5 mm ] \aabababb & \aabababb\tpr\one+\ab\ab\ab\tpr\ab+\ab\ab\tpr \aabb+\ab\tpr\aababb+\one\tpr\aabababb \\[2.5 mm ] \ab\aaabbb & \ab\aaabbb\tpr\one+\ab\aabb\tpr\ab+\aabb\ab\tpr\ab+\aabb\tpr \ab\ab+2\ab\ab\tpr\aabb+\ab\tpr\ab\aabb+ \ab\tpr\aaabbb+\one\tpr\ab\aaabbb \\[2.5 mm ] \aaabbb\ab & \aaabbb\ab\tpr\one+\aaabbb\tpr\ab+\aabb\tpr \ab\ab+\ab\tpr\aabb\ab+\one\tpr\aaabbb\ab \\[2.5 mm ] \ab\aababb & \ab\aababb\tpr\one+3\ab\ab\ab\tpr\ab+ \ab\ab\tpr\ab\ab+2\ab\ab\tpr\aabb+ \ab\tpr\ab\aabb+\ab\tpr\aababb+ \one\tpr\ab\aababb \\[2.5 mm ] \aababb\ab & \aababb\ab\tpr\one+\aababb\tpr\ab+\ab\ab\tpr \ab\ab+\ab\tpr\aabb\ab+\one\tpr\aababb\ab \\[2.5 mm ] \aabb\aabb & \aabb\aabb\tpr\one+\aabb\ab\tpr\ab+\ab\aabb\tpr\ab + \aabb\tpr\aabb+2\ab\ab\tpr\ab\ab+ \ab\tpr\ab\aabb+\ab\tpr\aabb\ab+ \one\tpr\aabb\aabb \\[2.5 mm ] \ab\ab\aabb & \ab\ab\aabb\tpr\one+3\ab\ab\ab\tpr\ab+ 2\ab\ab\tpr\ab\ab+\ab\ab\tpr\aabb+ \ab\tpr\ab\ab\ab+\ab\tpr\ab\aabb+ \one\tpr\ab\ab\aabb \\[2.5 mm ] \ab\aabb\ab & \ab\aabb\ab\tpr\one+\ab\aabb\tpr\ab+ 2\ab\ab\tpr\ab\ab+\ab\tpr\ab\ab\ab+ \ab\tpr\aabb\ab+\one\tpr\ab\aabb\ab \\[2.5 mm ] \aabb\ab\ab & \aabb\ab\ab\tpr\one+\aabb\ab\tpr\ab+ \aabb\tpr\ab\ab+\ab\tpr\ab\ab\ab+ \one\tpr\aabb\ab\ab \\[2 mm ] \ab\ab\ab\ab & \ab\ab\ab\ab\tpr\one+\ab\ab\ab\tpr\ab+ \ab\ab\tpr\ab\ab+\ab\tpr\ab\ab\ab+ \one\tpr\ab\ab\ab\ab \\[2 mm ] \hline \end{array}\ ] ] @xmath381 \hline \\[-1 mm ] \one & \one \\ \ab & -\ab \\[2 mm ] \aabb & -\aabb+2\ab\ab \\[1.5 mm ] \ab\ab & \ab\ab \\[2 mm ] \aaabbb & -\aaabbb+2\ab\aabb+2\aabb\ab-6\ab\ab\ab \\[2.5 mm ] \aababb & -\aababb+\ab\aabb+\aabb\ab-3\ab\ab\ab \\[2.5 mm ] \ab\aabb & \aabb\ab-3\ab\ab\ab \\[2.5 mm ] \aabb\ab & \ab\aabb-3\ab\ab\ab \\[1.5 mm ] \ab\ab\ab & -\ab\ab\ab \\[2 mm ] \aaaabbbb & -\aaaabbbb+2\ab\aaabbb+2\aaabbb\ab+2\aabb\aabb-6\ab\ab\aabb-6\ab\aabb\ab-6\aabb\ab\ab+24\ab\ab\ab\ab \\[2.5 mm ] \aaababbb & -\aaababbb+\ab\aaabbb+\aaabbb\ab+\ab\aababb+\aababb\ab-3\ab\ab\aabb-3\ab\aabb\ab-3\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \aabaabbb & -\aabaabbb+\ab\aaabbb+\aaabbb\ab+\ab\aababb+\aababb\ab-2\ab\ab\aabb-3\ab\aabb\ab-4\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \aaabbabb & -\aaabbabb+\ab\aababb+\aababb\ab+2\aabb\aabb-4\ab\ab\aabb-3\ab\aabb\ab-2\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \aabababb & -\aabababb+\ab\aababb+\aababb\ab-\ab\ab\aabb-\ab\aabb\ab-\aabb\ab\ab+4\ab\ab\ab\ab \\[2.5 mm ] \ab\aaabbb & \aaabbb\ab-\ab\ab\aabb-2\ab\aabb\ab-3\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \aaabbb\ab & \ab\aaabbb-3\ab\ab\aabb-2\ab\aabb\ab-\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \ab\aababb & \aababb\ab-\ab\aabb\ab-2\aabb\ab\ab+6\ab\ab\ab\ab\\[2.5 mm ] \aababb\ab & \ab\aababb-2\ab\ab\aabb-\ab\aabb\ab+6\ab\ab\ab\ab \\[2.5 mm ] \aabb\aabb & \aabb\aabb-2\ab\ab\aabb-2\ab\aabb\ab-2\aabb\ab\ab+12\ab\ab\ab\ab \\[2.5 mm ] \ab\ab\aabb & -\aabb\ab\ab+4\ab\ab\ab\ab \\[2.5 mm ] \ab\aabb\ab & -\ab\aabb\ab+4\ab\ab\ab\ab \\[2.5 mm ] \aabb\ab\ab & -\ab\ab\aabb+4\ab\ab\ab\ab \\[2 mm ] \ab\ab\ab\ab & \ab\ab\ab\ab \\[2 mm ] \hline \end{array}\ ] ] h. berland and b. owren . algebraic structures on ordered rooted trees and their significance to lie group integrators . in _ group theory and numerical analysis _ , volume 39 of _ crm proc . lecture notes _ , pages 4963 . soc . , 2005 . a. connes and d. kreimer . renormalization in quantum field theory and the riemann hilbert problem . i. the hopf algebra structure of graphs and the main theorem . , 210(1):249273 , 2000 , arxiv : hep - th/9912092 . h. z. munthe - kaas . high order runge kutta methods on manifolds . in _ proceedings of the nsf / cbms regional conference on numerical analysis of hamiltonian differential equations ( golden , co , 1997 ) _ , volume 29,1 , pages 115127 , 1999 .

in this paper , we provide a novel matrix - analytic approach for studying doubly exponential solutions of randomized load balancing models ( also known as _ supermarket models _ ) with markovian arrival processes ( maps ) and phase - type ( ph ) service times . we describe the supermarket model as a system of differential vector equations by means of density dependent jump markov processes , and obtain a closed - form solution with a doubly exponential structure to the fixed point of the system of differential vector equations . based on this , we show that the fixed point can be decomposed into the product of two factors inflecting arrival information and service information , and further find that the doubly exponential solution to the fixed point is not always unique for more general supermarket models . furthermore , we analyze the exponential convergence of the current location of the supermarket model to its fixed point , and apply the kurtz theorem to study density dependent jump markov process given in the supermarket model with maps and ph service times , which leads to the lipschitz condition under which the fraction measure of the supermarket model weakly converges the system of differential vector equations . this paper gains a new understanding of how workload probing can help in load balancing jobs with non - poisson arrivals and non - exponential service times . 0.5 cm * keywords : * randomized load balancing , supermarket model , matrix - analytic approach , doubly exponential solution , density dependent jump markov process , markovian arrival process ( map ) , phase - type ( ph ) distribution , fixed point , exponential convergence , lipschitz condition .

1009.4970

randomized load balancing , where a job is assigned to a server from a small subset of randomly chosen servers , is very simple to implement , and can surprisingly deliver better performance ( for example reducing collisions , waiting times , backlogs ) in a number of applications , such as data centers , hash tables , distributed memory machines , path selection in networks , and task assignment at web servers . one useful model extensively used to study randomized load balancing schemes is the supermarket model . in the supermarket model , a key result by vvedenskaya , dobrushin and karpelevich @xcite indicated that when each poisson arriving job is assigned to the shortest one of @xmath0 randomly chosen queues with exponential service times , the equilibrium queue length can decay doubly exponentially in the limit as the population size @xmath1 , and the stationary fraction of queues with at least @xmath2 customers is @xmath3 , which indicates a substantially exponential improvement over the case for @xmath4 , where the tail of stationary queue length in the corresponding m / m/1 queue is @xmath5 . at the same time , the exponential improvement is also illustrated by another key work in which luczak and mcdiarmid @xcite studied the maximum queue length in the supermarket model with poisson arrivals and exponential service times . the distributed load balancing strategies in which individual job decisions are based on information on a limited number of other processors , have been studied by analytical methods in eager , lazokwska and zahorjan @xcite and by trace - driven simulations in zhou @xcite . based on this , the supermarket models can be developed by using either queueing theory or markov processes . most of recent research deals with a simple supermarket model with poisson arrivals and exponential service times by means of density dependent jump markov processes . the methods used in the recent literature are based on determining the behavior of the supermarket model as its population size grows to infinity , and its behavior is naturally described as a system of differential equations whose fixed point leads to a closed - form solution with a doubly exponential structure . readers may refer to , such as , azar , broder , karlin and upfal @xcite , vvedenskaya , dobrushin and karpelevich @xcite and mitzenmacher @xcite . certain generalizations of the supermarket models have been explored , for example , in studying simple variations by mitzenmacher and vcking @xcite , mitzenmacher @xcite , vcking @xcite , mitzenmacher , richa , and sitaraman @xcite and vvedenskaya and suhov @xcite ; in discussing load information by mirchandaney , towsley , and stankovic @xcite , dahlin @xcite and mitzenmacher @xcite ; and in mathematical analysis by graham @xcite , luczak and norris @xcite and luczak and mcdiarmid @xcite . using fast jackson networks , martin and suhov @xcite , martin @xcite , suhov and vvedenskaya @xcite studied supermarket mall models , where each node in a jackson network is replaced by @xmath6 parallel servers , and a job joins the shortest of @xmath7 randomly chosen queues at the node to which it is directed . for non - poisson arrivals or for non - exponential service times , li , lui and wang @xcite discussed the supermarket model with poisson arrivals and ph service times , and indicated that the fixed point decreases doubly exponentially , where the stationary phase - type environment is shown to be a crucial factor . bramson , lu and prabhakar @xcite provided a modularized program based on ansatz for treating the supermarket model with poisson arrivals and general service times , and li @xcite further discussed this supermarket model by means of a system of integral - differential equations , and illustrated that the fixed point decreases doubly exponentially and that the heavy - tailed service times do not change the doubly exponential solution to the fixed point . for the ph distribution , readers may refer to neuts @xcite and li @xcite . the map is a useful mathematical model , for example , for describing bursty traffic , self similarity and long - range dependence in modern computer networks , e.g. , see adler , feldman and taqqu @xcite . for detail information of the map , readers may refer to chapter 5 in neuts @xcite , lucantoni @xcite , chapter 1 in li @xcite , and three excellent overviews by neuts @xcite , chakravarthy @xcite and cordeiro and kharoufeh @xcite . in computer networks , andersen and nielsen @xcite applied the map to describe long - range dependence , and yoshihara , kasahara and takahashi @xcite analyzed self - similar traffic by means of a markov - modulated poisson process . it is interesting to answer whether or how non - poisson arrivals or non - exponential service times can disrupt doubly exponential solutions to the fixed points in supermarket models . to that end , this paper studies a supermarket model with maps and ph service times , and shows that there still exists a doubly exponential solution to the fixed point . the main contributions of the paper are threefold . the first one is to provide a novel matrix - analytic approach to study the supermarket model with maps and ph service times . based on density dependent jump markov processes , the supermarket model is described as a system of differential vector equations whose fixed point has a closed - form solution with a doubly exponential structure . the second one is to obtain a crucial result that the fixed point can be decomposed into the product of two factors inflecting arrival information and service information , which indicates that the doubly exponential solution to the fixed point can exist extensively , but it is not always unique for more general supermarket models . the third one is to analyze exponential convergence of the current location of the supermarket model to its fixed point . not only does the exponential convergence indicate the existence of the fixed point , but it also shows that such a convergent process is very fast . to study the limit behavior of the supermarket model as its population size goes to infinity , this paper applies the kurtz theorem to study density dependent jump markov process given in the supermarket model with maps and ph service times , which leads to the lipschitz condition under which the fraction measure of the supermarket model weakly converges the system of differential vector equations . the remainder of this paper is organized as follows . in section 2 , we first describe a supermarket model with maps and ph service times . then the supermarket model is described as a systems of differential vector equations in terms of density dependent jump markov processes . in section 3 , we first introduce a fixed point of the system of differential vector equations , and set up a system of nonlinear equations satisfied by the fixed point . then we provide a closed - form solution with a doubly exponential structure to the fixed point , and show that the fixed point can be decomposed into the product of two factors inflecting arrival information and service information . in section 4 , we provide an important observation in which the doubly exponential solution to the fixed point is not always unique for more general supermarket models . in section 5 , we study exponential convergence of the current location of the supermarket model to its fixed point . in section 6 , we apply the kurtz theorem to study density dependent jump markov process given in the supermarket model with maps and ph service times , which leads to the lipschitz condition under which the fraction measure of the supermarket model weakly converges the system of differential vector equations . some concluding remarks are given in section 7 . in this section , we first provide a supermarket model with maps and ph service times . then the supermarket model is described as a system of differential vector equations based on density dependent jump markov processes . we first introduce some notation as follows . let @xmath8 be the _ kronecker product _ of two matrices @xmath9 and @xmath10 , that is , @xmath11 ; @xmath12 the _ kronecker sum _ of @xmath13 and @xmath14 , that is , @xmath15 . we denote by @xmath16 the _ hadamard product _ of @xmath13 and @xmath14 as follows : @xmath17 specifically , for @xmath18 , we have@xmath19 for a vector @xmath20 , we write@xmath21 now , we describe the supermarket model , which is abstracted as a multi - server multi - queue queueing system . customers arrive at a queueing system of @xmath22 servers as a map with an irreducible matrix descriptor @xmath23 of size @xmath24 . let @xmath25 be the stationary probability vector of the irreducible markov chain @xmath26 . then the stationary arrival rate of the map is given by @xmath27 , where @xmath28 is a column vector of ones with a suitable size . the service time of each customer is of phase type with an irreducible representation @xmath29 of order @xmath30 , where the row vector @xmath31 is a probability vector whose @xmath32th entry is the probability that a service begins in phase @xmath32 for @xmath33 ; @xmath34 is an @xmath35 matrix whose @xmath36 entry is denoted by @xmath37 with @xmath38 for @xmath39 , and @xmath40 for @xmath41 and @xmath42 . let @xmath43 . when a ph service time is in phase @xmath44 , the transition rate from phase @xmath44 to phase @xmath32 is @xmath37 , the service completion rate is @xmath45 , and the output rate from phase @xmath44 is @xmath46 . at the same time , the expected service time is given by @xmath47 . each arriving customer chooses @xmath48 servers independently and uniformly at random from the @xmath49 servers , and waits for service at the server which currently contains the fewest number of customers . if there is a tie , servers with the fewest number of customers will be chosen randomly . all customers in every server will be served in the first - come - first service ( fcfs ) manner . we assume that all the random variables defined above are independent , and that this system is operating in the region @xmath50 . please see figure 1 for an illustration of such a supermarket model . servers , width=377 ] the following lemma , which is stated without proof , provides an intuitively sufficient condition under which the supermarket model is stable . note that this proof can be given by a simple comparison argument with the queueing system in which each customer queues at a random server ( i.e. , where @xmath4 ) . when @xmath4 , each server acts like a map / ph/1 queue which is stable if @xmath50 , see chapter 5 in neuts @xcite . the comparison argument is similar to those in winston @xcite and weber @xcite , thus we can obtain two useful results : ( 1 ) the shortest queue is optimal due to the assumptions on maps and ph service times ; and ( 2 ) the size of the longest queue in the supermarket model is stochastically dominated by the size of the longest queue in a set of @xmath49 independent map / ph/1 queues . the supermarket model with maps and ph service times is stable if @xmath51 we define @xmath52 as the number of queues with at least @xmath2 customers who include the customer in service , the map in phase @xmath44 and the ph service time in phase @xmath32 at time @xmath53 . clearly , @xmath54 for @xmath55 , @xmath33 and @xmath56 . let@xmath57 and for @xmath58@xmath59 which is the fraction of queues with at least @xmath2 customers , the map in phase @xmath44 and the ph service time in phase @xmath32 at time @xmath53 . using the lexicographic order we write@xmath60 and for @xmath58@xmath61@xmath62 the state of the supermarket model may be described by the vector @xmath63 for @xmath53 . since the arrival process to the queueing system is a map and the service time of each customer is of phase type , the stochastic process @xmath64 describing the state of the supermarket model is a markov process whose state space is given by@xmath65 let@xmath66\ ] ] and @xmath58@xmath67 .\ ] ] using the lexicographic order we write@xmath68 and for @xmath58@xmath69@xmath70 as shown in martin and suhov @xcite and luczak and mcdiarmid @xcite , the markov process @xmath71 is asymptotically deterministic as @xmath1 . thus the limits @xmath72 $ ] and @xmath73 $ ] always exist by means of the law of large numbers . based on this , we write@xmath74 for @xmath58@xmath75 and@xmath76 note that @xmath77 and @xmath78 are two row vectors of order @xmath24 and @xmath79 , respectively . let @xmath80 . then it is easy to see from the maps and the ph service times that @xmath81 is also a markov process whose state space is given by@xmath82 if the initial distribution of the markov process @xmath83 approaches the dirac delta - measure concentrated at a point @xmath84 @xmath85 , then the limit @xmath86 is concentrated on the trajectory @xmath87 . this indicates a law of large numbers for the time evolution of the fraction of queues of different lengths . furthermore , the markov process @xmath88 converges weakly to the fraction vector @xmath89 as @xmath1 , or for a sufficiently small @xmath90,@xmath91 where @xmath92 is the @xmath93-norm of vector @xmath94 . the following proposition shows that the sequence @xmath95 is monotone decreasing , while its proof is easy by means of the definition of @xmath96 . for @xmath97@xmath98 and@xmath99 in what follows we set up a system of differential vector equations satisfied by the fraction vector @xmath96 by means of density dependent jump markov processes . we first provide an example to indicate how to derive the differential vector equations . consider the supermarket model with @xmath49 servers , and determine the expected change in the number of queues with at least @xmath2 customers over a small time period of length d@xmath100 . the probability vector that an arriving customer joins a queue of size @xmath101 during this time period is given by@xmath102 \cdot n\text{d}t,\ ] ] since each arriving customer chooses @xmath48 servers independently and uniformly at random from the @xmath49 servers , and waits for service at the server which currently contains the fewest number of customers . similarly , the probability vector that a customer leaves a server queued by @xmath2 customers during this time period is given by@xmath103 \cdot n\text{d}t.\ ] ] therefore , we can obtain@xmath104 = & \left [ s_{k-1}^{\odot d}\left ( n , t\right ) \left ( d\otimes i\right ) + s_{k}^{\odot d}\left ( n , t\right ) \left ( c\otimes i\right ) \right ] \cdot n\text{d}t\\ & + \left [ s_{k}\left ( n , t\right ) \left ( i\otimes t\right ) + ns_{k+1}\left ( n , t\right ) \left ( i\otimes t^{0}\alpha\right ) \right ] \cdot n\text{d}t,\end{aligned}\ ] ] which leads to@xmath105 using a similar analysis for equation ( [ equ1 ] ) , we can obtain a system of differential vector equations for the fraction vector @xmath106 as follows:@xmath107@xmath108@xmath109 and for @xmath18@xmath110 noting that the limit @xmath111 exists for @xmath56 and taking @xmath1 in the both sides of the system of differential vector equations ( [ e1 ] ) to ( [ e4 ] ) , we can easily obtain a system of differential vector equations for the fraction vector @xmath89 as follows:@xmath112@xmath113@xmath114 and for @xmath18,@xmath115 \(a ) for the supermarket model , many papers , such as mitzenmacher @xcite and luczak and mcdiarmid @xcite , assumed that the arrival process is poisson with rate @xmath116 . as a direct generalization of the poisson arrivals with rate @xmath116 , this paper uses a map with an irreducible matrix descriptor @xmath117 of size @xmath24 whose stationary arrival rate is given by @xmath27 . \(b ) when there are @xmath49 servers in the supermarket model , we may use a more general map with an irreducible matrix descriptor @xmath118 of size @xmath24 , where@xmath119 and @xmath120 is also the irreducible matrix descriptor of a map . it is easy to see from the above analysis that we can also obtain the system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) with respect to the more general map . in this section , we provide a novel matrix - analytic approach for computing the fixed point of the system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) , and give a closed - form solution with a doubly exponential structure to the fixed point . a row vector @xmath121 is called a fixed point of the fraction vector @xmath96 if @xmath122 . in this case , it is easy to see that@xmath123 = 0.\ ] ] therefore , as @xmath124 the system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) can be simplified as a system of nonlinear equations as follows:@xmath125@xmath126@xmath127 and for @xmath18,@xmath128 it is very challenging to solve the system of nonlinear equations ( [ equ7 ] ) to ( [ equ10 ] ) . here , our goal is to derive a closed - form solution with a doubly exponential structure to the fixed point @xmath129 through a novel matrix - analytic approach . it follows from equations ( [ equ9 ] ) and ( [ equ10 ] ) that @xmath130{ccccc}c\otimes i & d\otimes i & & & \\ & c\otimes i & d\otimes i & & \\ & & c\otimes i & d\otimes i & \\ & & & \ddots & \ddots \end{array } \right ) \nonumber\\ + & \left ( \pi_{1},\pi_{2},\pi_{3},\ldots\right ) \left ( \begin{array } [ c]{cccc}i\otimes t & & & \\ i\otimes t^{0}\alpha & i\otimes t & & \\ & i\otimes t^{0}\alpha & i\otimes t & \\ & & \ddots & \ddots \end{array } \right ) \nonumber\\ & = -\left ( \pi_{0}^{\odot d}\left ( d\otimes\alpha\right ) , 0,0,\ldots \right ) . \label{equ11}\ ] ] let@xmath131{cccc}i\otimes t & & & \\ i\otimes t^{0}\alpha & i\otimes t & & \\ & i\otimes t^{0}\alpha & i\otimes t & \\ & & \ddots & \ddots \end{array } \right ) .\ ] ] then it is easy to check that the matrix @xmath13 is invertible , and@xmath132{ccccc}i\otimes\left ( -t\right ) ^{-1 } & & & & \\ i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left ( -t\right ) ^{-1 } & & & \\ i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left ( -t\right ) ^{-1 } & & \\ i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left [ e\alpha\left ( -t\right ) ^{-1}\right ] & i\otimes\left ( -t\right ) ^{-1 } & \\ \vdots & \vdots & \vdots & \vdots & \ddots \end{array } \right ) , \]]@xmath133{ccccc}c\otimes i & d\otimes i & & & \\ & c\otimes i & d\otimes i & & \\ & & c\otimes i & d\otimes i & \\ & & & \ddots & \ddots \end{array } \right ) \left ( -a^{-1}\right ) = \left ( \begin{array } [ c]{ccccc}r & v & & & \\ w & r & v & & \\ w & w & r & v & \\ \vdots & \vdots & \vdots & \vdots & \ddots \end{array } \right ) , \ ] ] and@xmath134 , 0,0,\ldots\right ) , \ ] ] where@xmath135@xmath136\ ] ] and@xmath137 .\ ] ] thus it follows from ( [ equ11 ] ) that@xmath138{ccccc}r & v & & & \\ w & r & v & & \\ w & w & r & v & \\ \vdots & \vdots & \vdots & \vdots & \ddots \end{array } \right ) \nonumber\\ & + \left ( \pi_{0}^{\odot d}d\otimes\left [ \alpha\left ( -t\right ) ^{-1}\right ] , 0,0,\ldots\right ) , \label{equ12}\ ] ] which leads to a new system of nonlinear equations as follows:@xmath139 + \pi_{1}^{\odot d}r+\sum_{j=2}^{\infty}\pi_{j}^{\odot d}w\label{equ13}\ ] ] and for @xmath18,@xmath140 now , we need to omit the two terms @xmath141 for @xmath142 and @xmath143 for @xmath18 in equations ( [ equ13 ] ) and ( [ equ14 ] ) . note that the markov chain @xmath26 is positive recurrent , we assume that the system of nonlinear equations ( [ equ13 ] ) and ( [ equ14 ] ) has a closed - form solution@xmath144 and for @xmath58@xmath145 where @xmath146 , and @xmath147 is a positive constant for @xmath58 . then it follows from ( [ equ13 ] ) , ( [ equ14 ] ) and ( [ equ15 ] ) that@xmath148 + r^{d}\left ( 1\right ) \left ( \gamma\otimes\alpha\right ) r\nonumber\\ & + \sum_{j=2}^{\infty}r^{d}\left ( j\right ) \left ( \gamma\otimes \alpha\right ) w;\label{equ16}\ ] ] and for @xmath18,@xmath149 note that@xmath150 \right\ } \\ & = \gamma\left ( c+d\right ) \otimes\alpha\left [ e\alpha\left ( -t\right ) ^{-1}\right]\end{aligned}\ ] ] and@xmath151 it is clear that@xmath152 similarly , we can compute@xmath153 \right\ } \\ & = \gamma\left ( c+d\right ) \otimes\alpha\left ( -t\right ) ^{-1}=0.\end{aligned}\ ] ] it follows from ( [ equ16 ] ) and ( [ equ17 ] ) that@xmath139 \label{equ17 - 1}\ ] ] or@xmath154 ; \label{equ18}\ ] ] and for @xmath18@xmath155 or@xmath156 let @xmath157 . then @xmath158 due to @xmath159 and @xmath160 . note that @xmath161 , @xmath162 , @xmath163 and @xmath164 , it follows from ( [ equ18 ] ) and ( [ equ19 ] ) that@xmath165 and for @xmath18@xmath166 ^{d^{k-1}}\left ( \theta\omega \rho\right ) ^{d^{k-2}+d^{k-3}+\cdots+1}\nonumber\\ & = \theta^{d^{k}}\left ( \theta\omega\rho\right ) ^{d^{k-1}+d^{k-2}+\cdots + 1}\nonumber\\ & = \theta^{d^{k}}\left ( \theta\omega\rho\right ) ^{\frac{d^{k}-1}{d-1}}.\label{equ21}\ ] ] it is easy to see from ( [ equ14 - 1 ] ) , ( [ equ15 ] ) and ( [ equ21 ] ) that@xmath167 and for @xmath58@xmath168 now , we use ( [ equ22 ] ) and ( [ equ23 ] ) to check equations ( [ equ8 ] ) and ( [ equ17 - 1 ] ) that@xmath169{l}\pi_{0}^{\odot d}c+\pi_{1}\left ( i\otimes t^{0}\right ) = 0,\\ \pi_{1}=\pi_{0}^{\odot d}d\otimes\left [ \alpha\left ( -t\right ) ^{-1}\right ] , \end{array } \right.\ ] ] which leads to@xmath170 obviously , @xmath171 is a non - zero nonnegative solution to equation ( [ equ24 ] ) , and @xmath172 . summarizing the above analysis , the following theorem describes a closed - form solution with a doubly exponential structure to the fixed point . [ the : fixedp]the fixed point @xmath121 is given by@xmath173 and for @xmath174@xmath175 the following corollary indicates that the fixed point can be decomposed into the product of two factors inflecting arrival information and service information . based on this , it is easy to see the role played by the arrival and service processes in the fixed point . the fixed point @xmath121 can be decomposed into the product of two factors inflecting arrival information and service information @xmath176 we consider a supermarket model with poisson arrivals with rate @xmath177 and exponential service times with rate @xmath178 , which has been extensively analyzed in the literature . obviously , @xmath179 . it follows from ( [ equ11 ] ) that@xmath130{ccccc}-\lambda & \lambda & & & \\ & -\lambda & \lambda & & \\ & & -\lambda & \lambda & \\ & & & \ddots & \ddots \end{array } \right ) \newline \\ + & \left ( \pi_{1},\pi_{2},\pi_{3},\ldots\right ) \left ( \begin{array } [ c]{cccc}-\mu & & & \\ \mu & -\mu & & \\ & \mu & -\mu & \\ & & \ddots & \ddots \end{array } \right ) = -\left ( \lambda,0,0,\ldots\right ) , \end{aligned}\ ] ] which leads to@xmath180{ccccc}0 & \rho & & & \\ & 0 & \rho & & \\ & & 0 & \rho & \\ & & & \ddots & \ddots \end{array } \right ) + \left ( \rho,0,0,\ldots\right ) .\ ] ] thus we obtain@xmath181 and for @xmath18,@xmath182 which is the same as lemma 3.2 in mitzenmacher @xcite . based on theorem [ the : fixedp ] , we now compute the expected sojourn time @xmath183 that a tagged arriving customer spends in the supermarket model . for the ph service time @xmath184 with an irreducible representation @xmath185 , the residual time @xmath186 of @xmath184 is also of phase type with an irreducible representation @xmath187 , where @xmath188 is the stationary probability vector of the markov chain @xmath189 . thus , we have@xmath190 = \alpha\left ( -t\right ) ^{-1}e=\frac{1}{\mu},\text { \ } e\left [ x_{r}\right ] = \tau\left ( -t\right ) ^{-1}e.\ ] ] for the ph service times , a tagged arriving customer is the @xmath2th customer in the corresponding queue with probability @xmath191 . thus it is easy to see that the expected sojourn time of the tagged arriving customer is given by@xmath192 & = \left ( \pi_{0}^{\odot d}e-\pi_{1}^{\odot d}e\right ) e\left [ x\right ] + \sum_{k=1}^{\infty}\left ( \pi_{k}^{\odot d}e-\pi_{k+1}^{\odot d}e\right ) \left [ e\left [ x_{r}\right ] + ke\left [ x\right ] \right ] \\ & = \left\ { e\left [ x_{r}\right ] -e\left [ x\right ] \right\ } \pi _ { 1}^{\odot d}e+e\left [ x\right ] \sum_{k=0}^{\infty}\pi_{k}^{\odot d}e\\ & = \theta^{d^{2}}\left ( \theta\omega\rho\right ) ^{d}\left ( \tau -\alpha\right ) \left ( -t\right ) ^{-1}e+\frac{1}{\mu}\sum_{k=0}^{\infty } \theta^{d^{k+1}}\left ( \theta\omega\rho\right ) ^{\frac{d^{k+1}-d}{d-1}}.\end{aligned}\ ] ] when the arrival process and the service time distribution are poisson and exponential , respectively , it is clear that @xmath193 and @xmath194 , thus we have@xmath195 = \frac{1}{\mu}\sum_{k=0}^{\infty}\rho^{\frac{d^{k+1}-d}{d-1}},\ ] ] which is the same as corollary 3.8 in mitzenmacher @xcite . in what follows we provide an example to indicate how the expected sojourn time @xmath196 $ ] depends on the choice number @xmath7 . we assume that @xmath197 and @xmath198{cc}-10 & 7\\ 4 & -9 \end{array } \right ) , \ \ d=\left ( \begin{array } [ c]{cc}1 & 2\\ 3 & 2 \end{array } \right ) , \ ] ] and the service times are exponential with service rate @xmath199 , respectively . it is seen from figure 2 that the expected sojourn time @xmath196 $ ] decreases very fast as the choice number @xmath7 increases . $ ] vs @xmath7 for the map , width=377 ] in this section , we analyze a special supermarket model with poisson arrivals an ph service times , and obtain two different doubly exponential solutions to the fixed point . based on this , we give an important observation , namely that the doubly exponential solution to the fixed point is not always unique for more general supermarket models . when the arrival process is poisson , it follows from ( [ equ7 ] ) to ( [ equ10 ] ) that@xmath200@xmath201@xmath202 and for @xmath18,@xmath203 for the system of nonlinear equations ( [ equ25 ] ) to ( [ equ28 ] ) , we can provide two different doubly exponential solutions to the fixed point @xmath121 . the first doubly exponential solution has been given in section 3 . here , we simply list the crucial derivations for the special supermarket model . it follows from ( [ equ11 ] ) that @xmath130{ccccc}-\lambda & \lambda & & & \\ & -\lambda & \lambda & & \\ & & -\lambda & \lambda & \\ & & & \ddots & \ddots \end{array } \right ) + \left ( \pi_{1},\pi_{2},\pi_{3},\ldots\right ) \left ( \begin{array } [ c]{cccc}t & & & \\ t^{0}\alpha & t & & \\ & t^{0}\alpha & t & \\ & & \ddots & \ddots \end{array } \right ) \\ & = -\left ( \lambda\alpha,0,0,\ldots\right ) , \end{aligned}\ ] ] which leads to@xmath204{ccccc}r & v & & & \\ & r & v & & \\ & & r & v & \\ & & & \ddots & \ddots \end{array } \right ) \\ & + \left ( \lambda\alpha\left ( -t\right ) ^{-1},0,0,\ldots\right ) , \end{aligned}\ ] ] where@xmath205 and@xmath206 thus we obtain@xmath207 \label{equ29}\ ] ] and for @xmath18@xmath208 + \pi_{k}^{\odot d}\left [ \lambda\left ( -i+e\alpha\right ) \left ( -t\right ) ^{-1}\right ] . \label{equ30}\ ] ] to omit the term @xmath209 $ ] for @xmath58 , we assume that @xmath210 has the following expression@xmath211 in this case , we have@xmath212 = r^{d}\left ( k\right ) \alpha\left [ \lambda\left ( -i+e\alpha\right ) \left ( -t\right ) ^{-1}\right ] = 0,\ ] ] thus it follows from ( [ equ29 ] ) and ( [ equ30 ] ) that@xmath213 and for @xmath18@xmath208 . \label{equ32}\ ] ] it follows from ( [ equ31 ] ) that@xmath214 which follows that@xmath215 it follows from ( [ equ32 ] ) that@xmath216\ ] ] which follows that@xmath217 therefore , we can obtain@xmath218 and for @xmath58@xmath219 the second doubly exponential solution was given in li , lui and wang @xcite , thus we provide some crucial computational steps . it follows from ( [ equ26 ] ) that@xmath220 to solve equation ( [ equ34 ] ) , we denote by @xmath188 the stationary probability vector of the irreducible markov chain @xmath189 . obviously , we have@xmath221@xmath222 thus , we obtain@xmath223 using @xmath179 and @xmath224 , it follows from equation ( [ equ27 ] ) that@xmath225 which leads to@xmath226 note that @xmath227 and @xmath164 , we obtain@xmath228 let @xmath229 . then it is easy to see that @xmath230 , and@xmath231 using similar analysis on equation ( [ equ34 ] ) , we have@xmath232 based on @xmath224 and @xmath233 , it follows from equation ( [ equ28 ] ) that for @xmath234 , @xmath235 which leads to@xmath236 thus we obtain@xmath237 using a similar analysis on equation ( [ equ34 ] ) , we have@xmath238 now , we assume that @xmath239 is correct for the cases with @xmath240 . then for @xmath241 we have @xmath242 which leads to@xmath243 thus we obtain@xmath244 by a similar analysis to ( [ equ34 ] ) and ( [ equ35 ] ) , we have@xmath245 therefore , by induction we can obtain@xmath246 and for @xmath58@xmath247 now , we have given two expressions ( [ equ33 ] ) and ( [ equ36 ] ) for the fixed point . in what follows we provide some examples to indicate that the two expressions may be different from each other . * example one : * when the ph service time is exponential , it is easy to see that @xmath193 , which leads to that @xmath248 . thus the fixed point is given by@xmath249 in this case , ( [ equ33 ] ) is the same as ( [ equ36 ] ) . * example two : * when the service time is an @xmath250-order erlang distribution with an irreducible representation @xmath251 , where@xmath252 and@xmath253{ccccc}-\eta & \eta & & & \\ & -\eta & \eta & & \\ & & \ddots & \ddots & \\ & & & -\eta & \eta\\ & & & & -\eta \end{array } \right ) , \text { \ \ } t^{0}=\left ( \begin{array } [ c]{c}0\\ 0\\ \vdots\\ 0\\ \eta \end{array } \right ) .\ ] ] we have@xmath254 and@xmath255 thus the first doubly exponential solution is given by@xmath256 it is clear that@xmath257{ccccc}-\eta & \eta & & & \\ & -\eta & \eta & & \\ & & \ddots & \ddots & \\ & & & -\eta & \eta\\ \eta & & & & -\eta \end{array } \right ) , \ ] ] which leads to the stationary probability vector of the markov chain @xmath189 as follows : @xmath258@xmath259@xmath260 and@xmath261 thus the second doubly exponential solution is given by @xmath262 it is clear that ( [ equ37 ] ) and ( [ equ38 ] ) are different from each other for @xmath263 , and@xmath264 it is clear that @xmath265 for @xmath263 . for the supermarket model with poisson arrivals and ph service times , we have obtained two different doubly exponential solutions to the fixed point . it is interesting but difficult to be able to find another new doubly exponential solution to the fixed point . furthermore , we believe that it is an open problem how to give all doubly exponential solutions to the fixed point for more general supermarket models . in this section , we provide an upper bound for the current location @xmath266 of the supermarket model , and study exponential convergence of the current location @xmath96 to its fixed point @xmath267 . for the supermarket model , the initial point @xmath268 can affect the current location @xmath96 for each @xmath269 , since the arrival and service processes are under a unified structure through a sample path comparison . to explain this , it is necessary to provide some notation for comparison of two vectors . let @xmath270 and @xmath271 . we write @xmath272 if @xmath273 for some @xmath58 and @xmath274 for @xmath275 ; and @xmath276 if @xmath277 for all @xmath58 . now , we can easily obtain the following useful proposition , while the proof is clear by means of a sample path analysis , and thus is omitted here . [ prop1]if @xmath278 , then @xmath279 for @xmath269 . based on proposition [ prop1 ] , the following theorem shows that the fixed point @xmath267 is an upper bound of the current location @xmath96 for all @xmath53 . [ the : bound]for the supermarket model , if there exists some @xmath2 such that @xmath280 , then the sequence @xmath281 for all @xmath53 has an upper bound sequence @xmath282 which decreases doubly exponentially , that is , @xmath283 for all @xmath53 . * proof : * let@xmath284 then for each @xmath56 , @xmath285 for all @xmath53 , since@xmath286 is a fixed point for the supermarket model . if @xmath280 for some @xmath2 , then @xmath287 . again , if @xmath288 for all @xmath289 , then @xmath290 . it is easy to see from proposition [ prop1 ] that @xmath291 for all @xmath56 and @xmath53 . thus we obtain that for all @xmath56 and @xmath53 @xmath292 since @xmath293 , @xmath294 decreases doubly exponentially . this completes the proof . ' '' '' to show exponential convergence , we define a lyapunov function @xmath295 as@xmath296 where @xmath297 is a positive scalar sequence with @xmath298 for @xmath18 . the following theorem measures the distance of the current location @xmath266 to the fixed point @xmath267 for @xmath53 , and illustrates that this distance will quickly come close to zero with exponential convergence . hence , it shows that from any suitable starting point , the supermarket model can be quickly close to the fixed point , that is , there always exists a fixed point in the supermarket model . for @xmath53,@xmath299 where @xmath300 and @xmath301 are two positive constants . in this case , the lyapunov function @xmath302 is exponentially convergent . * proof : * it is seen from ( [ equ39 ] ) that @xmath303 it follows from equations ( [ equ3 ] ) to ( [ equ6 ] ) that@xmath304 e\\ & -w_{1}[s_{0}^{\odot d}\left ( t\right ) \left ( d\otimes\alpha\right ) + s_{1}^{\odot d}\left ( t\right ) \left ( c\otimes i\right ) \\ & + s_{1}\left ( t\right ) \left ( i\otimes t\right ) + s_{2}\left ( t\right ) \left ( i\otimes t^{0}\alpha\right ) ] e\\ & -\sum_{k=2}^{\infty}w_{k}[s_{k-1}^{\odot d}\left ( t\right ) \left ( d\otimes i\right ) + s_{k}^{\odot d}\left ( t\right ) \left ( c\otimes i\right ) \\ & + s_{k}\left ( t\right ) \left ( i\otimes t\right ) + s_{k+1}\left ( t\right ) \left ( i\otimes t^{0}\alpha\right ) ] e.\end{aligned}\ ] ] by means of @xmath305 and @xmath306 , we can obtain@xmath307 \nonumber\\ & -w_{1}[s_{0}^{\odot d}\left ( t\right ) \left ( de\right ) + s_{1}^{\odot d}\left ( t\right ) \left ( \left ( -de\right ) \otimes e\right ) \nonumber\\ & + s_{1}\left ( t\right ) \left ( e\otimes\left ( -t^{0}\right ) \right ) + s_{2}\left ( t\right ) \left ( e\otimes t^{0}\right ) ] \nonumber\\ & -\sum_{k=2}^{\infty}w_{k}[s_{k-1}^{\odot d}\left ( t\right ) \left ( \left ( de\right ) \otimes e\right ) + s_{k}^{\odot d}\left ( t\right ) \left ( \left ( -de\right ) \otimes e\right ) \nonumber\\ & + s_{k}\left ( t\right ) \left ( e\otimes\left ( -t^{0}\right ) \right ) + s_{k+1}\left ( t\right ) \left ( e\otimes t^{0}\right ) ] . \label{equ40}\ ] ] let@xmath308 e,\ ] ] for @xmath58@xmath309 e\ ] ] and@xmath310 e.\ ] ] then it follows from ( [ equ40 ] ) that@xmath311 e\\ & -\sum_{k=1}^{\infty}\left [ w_{k+1}c_{k}\left ( t\right ) -w_{k}\left ( c_{k}\left ( t\right ) + d_{k}\left ( t\right ) \right ) + w_{k-1}d_{k}\left ( t\right ) \right ] \cdot\left [ \pi_{k}-s_{k}\left ( t\right ) \right ] e.\end{aligned}\ ] ] let@xmath312@xmath313 and@xmath314 then@xmath315@xmath316 and for @xmath18@xmath317 thus we have @xmath318 e.\ ] ] it follows from ( [ equ39 ] ) that@xmath319 which can leads to@xmath320 this completes the proof . ' '' '' we have provided an algorithm for computing the positive scalar sequence @xmath297 with @xmath321 for @xmath18 as follows : step one:@xmath322 step two:@xmath323 and@xmath324 step three : for @xmath18@xmath325 this illustrates that @xmath326 is a function of time @xmath100 for @xmath58 . note that @xmath327 , it is clear that for @xmath18@xmath328 in this section , we apply the kurtz theorem to study density dependent jump markov process given in the supermarket model with maps and ph service times , which leads to the lipschitz condition under which the fraction measure of the supermarket model weakly converges the system of differential vector equations . the supermarket model can be analyzed by a density dependent jump markov process , where the density dependent jump markov process is a markov process with a single parameter @xmath49 which corresponds to the population size . kurtz s work provides a basis for density dependent jump markov processes in order to relate infinite - size systems of differential equations to corresponding finite - size systems of differential equations . readers may refer to kurtz @xcite for more details . in the supermarket model , when the population size is @xmath49 , we write@xmath329 and for @xmath58@xmath330@xmath331 in the state space @xmath332 , the density dependent jump markov process for the supermarket model with maps and ph service times contains four classes of state transitions as follows : class one @xmath333 @xmath334 @xmath333 : @xmath335 or @xmath336 , where @xmath337 ; class two @xmath333 @xmath338 @xmath333 : @xmath339 , where @xmath340 ; class three @xmath333 @xmath334 @xmath341 : @xmath342 or @xmath343 ; and class four @xmath333 @xmath338 @xmath344 : @xmath345 or @xmath346 . note that the transitions @xmath334 and @xmath338 express arrival transition and service transition , respectively . we write@xmath347 @xmath348 and for @xmath58@xmath349 and@xmath350 note that the states of the density dependent jump markov process can be normalized and interpreted as measuring population densities@xmath351 the transition rates of the markov process depend only on these densities . let @xmath352 be a density dependent jump markov process on the state space @xmath332 whose transition rates corresponding to the above four cases are given by@xmath353@xmath354@xmath355@xmath356@xmath357@xmath358@xmath359 let@xmath360@xmath361@xmath362@xmath363@xmath364@xmath365@xmath366 in the supermarket model , @xmath367 is an unscaled process which records the number of servers with at least @xmath2 customers for @xmath368 . we write@xmath369{ccccc}q_{0}^{\left ( n\right ) } & q_{0}^{\left ( n\right ) } \left ( a\right ) & & & \\ q_{1}^{\left ( n\right ) } \left ( s\right ) & q_{a,1}^{\left ( n\right ) } + q_{s,1}^{\left ( n\right ) } & q_{1}^{\left ( n\right ) } \left ( a\right ) & & \\ & q_{2}^{\left ( n\right ) } \left ( s\right ) & q_{a,2}^{\left ( n\right ) } + q_{s,2}^{\left ( n\right ) } & q_{2}^{\left ( n\right ) } \left ( a\right ) & \\ & & \ddots & \ddots & \ddots \end{array } \right ) , \ ] ] where @xmath370 ^{\odot d}c,\]]@xmath371 ^{\odot d}\left ( c\otimes i\right ) ; \]]@xmath372@xmath373 ^{\odot d}\left ( d\otimes\alpha\right ) , \]]@xmath374 ^{\odot d}\left ( d\otimes i\right ) ; \]]@xmath375@xmath376 using chapter 7 in kurtz @xcite or subsection 3.4.1 in mitzenmacher @xcite , the markov process @xmath377 with transition rate matrix @xmath378 @xmath379 is given by@xmath380 where @xmath381 for @xmath382 are independent standard poisson processes , @xmath383 is a positive integer with @xmath384 , and@xmath385 clearly , the jump markov process in equation ( [ equ6.1 ] ) at time @xmath100 is determined by the starting point and the transition rates which are integrated over its history . let@xmath386 where@xmath387 taking @xmath388 which is an appropriate scaled process , we have@xmath389 where @xmath390 is a poisson process centered at its expectation . let @xmath391 and @xmath392 , we obtain@xmath393 due to the fact that@xmath394 by means of the law of large numbers . in the supermarket model , the deterministic and continuous process @xmath395 is described by the infinite - size system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) , or simply,@xmath396 with the initial condition@xmath397 now , we consider the uniqueness of the limiting deterministic process @xmath398 with ( [ equ6.3 ] ) and ( [ equ6.4 ] ) , or the uniqueness of the solution to the infinite - size system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) . to that end , a sufficient condition is lipschitz , that is , for some constant @xmath399@xmath400 in general , the lipschitz condition is standard and sufficient for the uniqueness of the solution to the finite - size system of differential vector equations ; while for the countable infinite - size case , readers may refer to theorem 3.2 in deimling @xcite and subsection 3.4.1 in mitzenmacher @xcite for some useful generalization . to check the lipschitz condition , by means of the law of large numbers we have@xmath401 which leads to@xmath402{ccccc}q_{0 } & q_{0}\left ( a\right ) & & & \\ q_{1}\left ( s\right ) & q_{a,1}+q_{s,1 } & q_{1}\left ( a\right ) & & \\ & q_{2}\left ( s\right ) & q_{a,2}+q_{s,2 } & q_{2}\left ( a\right ) & \\ & & \ddots & \ddots & \ddots \end{array } \right ) , \label{equ6.5}\ ] ] where @xmath403@xmath404@xmath405@xmath406@xmath407@xmath408@xmath409 let@xmath410 and for @xmath58@xmath411@xmath412 then @xmath413 for @xmath58 the following theorem shows that the supermarket model with maps and ph service times satisfies the lipschitz condition for analyzing the uniqueness of the solution to the infinite - size system of differential vector equations ( [ equ3 ] ) to ( [ equ6 ] ) . [ the : lip]the supermarket model with maps and ph service times satisfies the lipschitz condition . * proof * let the state space of the markov process @xmath81 be@xmath414 for two arbitrary entries @xmath415 , we have@xmath416 note that @xmath94 expresses either an arrival transition or a service transition in the above four cases . when @xmath94 expresses an arrival transition , we can analyze the function @xmath417 from the two cases of arrival transitions ; while when @xmath418 expresses a service transition , the function @xmath419 can similarly be dealt with from the two cases of service transitions . when @xmath94 expresses an arrival transition , we analyze the function @xmath420 based on @xmath421 from the following two cases . case one : @xmath422 . in this case , we have@xmath423 taking@xmath424 it is clear that@xmath425 note that @xmath426 and @xmath427 are two row vectors of sizes @xmath24 and @xmath79 , respectively , in this case we write@xmath428 thus we have@xmath429 case two : @xmath430 for @xmath18 . in this case , we have@xmath431 let@xmath432 then@xmath433 based on the above two cases , taking@xmath434 we obtain that for two arbitrary entries @xmath435@xmath436 similarly , when @xmath418 expresses a service transition , we can choose a positive number @xmath437 such that for two arbitrary entries @xmath435@xmath438 let @xmath439 . then it follows from ( [ equ6.6 ] ) and ( [ equ6.7 ] ) that for two arbitrary entries @xmath440@xmath400 this completes the proof . based on theorem [ the : lip ] , the following theorem easily follows from theorem 3.13 in mitzenmacher @xcite . [ the : kurtz]in the supermarket model with maps and ph service times , @xmath441 and @xmath442 are respectively given by ( [ equ6.2 - 1 ] ) and ( [ equ6.2 - 2 ] ) , we have@xmath443 * proof * it has been shown that in the supermarket model with maps and ph service times , the function @xmath444 for @xmath445 satisfies the lipschitz condition . at the same time , it is easy to take a subset @xmath446 such that@xmath447 and@xmath448 where @xmath94 and @xmath418 express an arrival transition and a service transition , respectively . thus , this proof can easily be completed by means of theorem 3.13 in mitzenmacher @xcite . this completes the proof . using theorem 3.11 in mitzenmacher @xcite and theorem [ the : kurtz ] , the following theorem for the expected sojourn time that an arriving tagged customer spends in an initially empty supermarket model with maps and ph service times over the time interval @xmath449 $ ] . in the supermarket model with maps and ph service times , the expected sojourn time that an arriving tagged customer spends in an initially empty system over the time interval @xmath449 $ ] is bounded above by@xmath450 where @xmath451 is understood as @xmath1 . in this paper , we provide a novel matrix - analytic approach for studying doubly exponential solutions of the supermarket models with maps and ph service times . we describe the supermarket model as a system of differential vector equations , and obtain a closed - form solution with a doubly exponential structure to the fixed point of the system of differential vector equations . based on this , we shows that the fixed point can be decomposed into the product of two factors inflecting arrival information and service information , and indicate that the doubly exponential solution to the fixed point is not always unique for more general supermarket models . furthermore , we analyze the exponential convergence of the current location of the supermarket model to its fixed point , and apply the kurtz theorem to study density dependent jump markov process given in the supermarket model with maps and ph service times , which leads to the lipschitz condition under which the fraction measure of the supermarket model weakly converges the system of differential vector equations . therefore , we gain a new and crucial understanding of how the workload probing can help in load balancing jobs with either non - poisson arrivals or non - exponential service times . our approach given in this paper is useful in the study of load balancing in data centers and multi - core servers systems . we expect that this approach will be applicable to the study of other randomized load balancing schemes , for example , analyzing a renewal arrival process or a general service time distribution , discussing retrial service discipline and processor - sharing discipline , and studying supermarket networks . the author are very grateful to professors ke blomqvist and juan eloy ruiz - castro whose comments have greatly improved the presentation of this paper . lui was supported by the rgc grant . the work of q.l . li was supported by the national science foundation of china under grant no . 10871114 and the national grand fundamental research 973 program of china under grant no . 2006cb805901 . m. bramson , y. lu and b. prabhakar ( 2010 ) . randomized load balancing with general service time distributions . in _ proceedings of the acm sigmetrics international conference on measurement and modeling of computer systems _ , pages 275286 . chakravarthy ( 2000 ) . the batch markovian arrival process : a review and future work . in _ advances in probability theory and stochastic processes _ , a. krishnamoorthy , n. raju and v. ramaswami ( eds ) , notable publications : new jersey , pages 2139 . li , john c.s . lui and y. wang ( 2010 ) . a matrix - analytic solution for randomized load balancing models with phase - type service times . in _ international workshop on performance evaluation of computer and communication systems _ , lecture notes of computer science , w. gansterer , h. hlavacs and k.a . hummel ( eds ) , springer . m. mitzenmacher , a. richa , and r. sitaraman ( 2001 ) . the power of two random choices : a survey of techniques and results . in _ handbook of randomized computing : volume 1 _ , p. pardalos , s. rajasekaran and j. rolim ( eds ) , pages 255312 . m. mitzenmacher and b. vcking ( 1998 ) . the asymptotics of selecting the shortest of two , improved . in _ proceedings of the 37th annual allerton conference on communication , control , and computing _ , pages 326327 .

efforts to achieve better accuracy in numerical relativity have so far focused either on implementing second order accurate adaptive mesh refinement or on defining higher order accurate differences and update schemes . here , we argue for the combination , that is a higher order accurate adaptive scheme . this combines the power that adaptive gridding techniques provide to resolve fine scales ( in addition to a more efficient use of resources ) together with the higher accuracy furnished by higher order schemes when the solution is adequately resolved . to define a convenient higher order adaptive mesh refinement scheme , we discuss a few different modifications of the standard , second order accurate approach of berger and oliger . applying each of these methods to a simple model problem , we find these options have unstable modes . however , a novel approach to dealing with the grid boundaries introduced by the adaptivity appears stable and quite promising for the use of high order operators within an adaptive framework .

gr-qc0510111

there is growing concern in the numerical relativity community about the accuracy that can be achieved as robust implementations of the einstein equations become available . as has been discussed a number of times , the simulation of strongly gravitating astrophysical systems in 3d requires the stable and accurate handling of a large range of time and length scales @xcite . in the context of finite difference schemes , increasing resolution by using more grid points in a uniform grid quickly becomes prohibitively expensive , particularly with multiple spatial dimensions . two ways of achieving better accuracy include ( i ) the use of higher order methods for derivative approximations and time integration , and ( ii ) the use of a grid hierarchy which adapts to provide resolution only where and when needed , so called _ adaptive mesh refinement _ ( amr ) . both strategies have been adopted by the community largely following two independent paths . on one hand higher derivative operators are being introduced @xcite and on the other amr techniques within a second order accurate implementation have been adopted for over a decade @xcite . ( for recent efforts implementing second order amr / fmr see for instance @xcite/@xcite ) one key reason behind these independent developments lies in the complications encountered when implementing amr techniques , especially related to the way the ( artificial ) interface boundaries are treated . special care must be taken when dealing with the conditions at these boundaries , in particular how derivative operators should be modified near them . in the case of second order accurate implementations , there is little ambiguity on what these modifications can be as they need only be done at interface points . for higher order implementations , the strategy is far less clear . however , there are strong reasons to combine amr and implementations of higher orders . for instance , although higher order operators ( be they within finite difference , finite elements or spectral approximations ) do in principle provide an approximation with errors which are considerably smaller , they do so only if the underlying solution involved is well resolved by the adopted grid . certainly , the required grid can be anticipated for linear problems but this is hardly the case in non - linear ones where finer features induced by the non - linear nature of the problem can appear during the evolution . hence , an adaptive gridding mechanism which reacts to the solution itself is crucial . the combination of amr with higher order operators then provides lower errors for a given grid while being able to react to unanticipated features in the solution in addition to providing a more efficient use of computational resources. in order to combine these techniques , one must face how to treat the derivative operators near boundaries , and how to deal with the artificial boundaries themselves . a clean way to approach this problem is to consider operators satisfying summation by parts ( sbp ) which have recently been incorporated successfully in strongly / symmetric hyperbolic systems in numerical relativity @xcite . these techniques guarantee that a large set of problems can be implemented stably . having stability of the unigrid evolution , however , is not sufficient to guarantee a fully discrete energy estimate for an amr implementation . the analysis of the time integration , in particular how different grids communicate with each other to provide boundary conditions , becomes quite involved . nevertheless , one can argue strongly for adopting as a starting point an implementation in which a semi - discrete estimate is guaranteed since a necessary condition for an amr implementation to be stable is that its unigrid version be stable . as discussed in the literature , the latter is guaranteed when employing runge kutta operators of orders higher than 2 @xcite ( in conjunction with suitable boundary conditions ) . we thus adopt a strategy where spatial derivative operators satisfying sbp , along with runge kutta time integrators , are employed on all levels of the amr hierarchy . this choice , of course , does not dictate the treatment of the interface boundaries , and we consider various options for this treatment . for instance , in the standard berger oliger strategy , the solution obtained at a parent grid is employed to define boundary conditions for the child grid . leaving the stability issue aside for a moment , an immediate observation is that when employing a two - level implementation , this strategy can not generically give better accuracy than second order if boundary values are obtained by linear interpolation in time of the parent grid values . additionally , the definition of all field values via interpolation without regard for their propagation features is a likely source of inconsistencies making stability a delicate issue . in this work we examine alternative treatments of grid boundaries and derivative operators to deal with these issues . we pay particular attention to the stability of the system by effectively implementing a toy model problem within a fixed two - grid hierarchy and analyzing the spectrum of the overall update operator . these studies demonstrate that implementations with accuracy orders greater than two , employing the standard berger oliger scheme have unstable modes . furthermore , related schemes share this unfortunate feature . this instability can be remedied , with some caveats , with the addition of dissipation though at the expense of larger errors in the approximation . additionally , we observe that a simple modification of the standard strategy yields a stable scheme without requiring the addition of dissipation , albeit at additional computational cost . in this new strategy , a child grid effectively never requires boundary points from a parent . this can be achieved if the solution at level @xmath0 is obtained by integrating using the past domain of dependence of the child grid only . at first sight this is more computationally demanding as it requires the area of the child grid at level @xmath1 to be larger than that at @xmath0 so that the intersection of the numerical domain of dependence of the solution at the level @xmath0 is completely contained in the refined region at level @xmath1 . this is achieved simply by defining an ` extended ' child grid at level @xmath1 and discarding a buffer zone at level @xmath0 . as we will see later in appendix [ sec : cost ] , this extra cost need not be relevant at all , as for a given target error , a coarser base grid treated this way might be comparable to a finer base grid treated with the standard berger oliger scheme plus the addition of dissipation . this work is divided as follows . in section [ sec : technique ] , we detail different boundary treatments and describe our method to assess their stability properties . in section [ sec : tests ] , we present one and three dimensional tests which illustrate the behavior observed . section [ sec : conclusion ] concludes with some final comments , and we defer to appendix [ sec : cost ] the discussion of the cost associated with 2nd vs. 4th order implementations . in searching for a stable amr implementation , a basic requirement is to have a stable implementation of the equations in the uniform grid case . this requirement is perhaps the most important building block , without which an amr implementation will suffer even more in terms of stability . unfortunately , even when ensuring the unigrid scheme is stable , no results are available on what guarantees the stability of an amr / fmr implementation with ` sub - cycling ' for arbitrary accuracy orders . such a scheme which advances a grid with a time step defined by its grid spacing stands in contrast to one which advances all points by a global time step determined by the finest resolution in use . the latter is considerably less efficient than the former ( becoming more so with more refinement levels ) but is simpler to analyze . we concentrate here on the former case , i.e. one that involves sub - cycling , as this is the method one would like to employ . in order to develop a stable amr / fmr scheme , we analyze different natural options from their fully - discrete point of view and study their stability properties . we begin by adopting a strategy that ensures stability for a uniform grid case for first order ( non - constant coefficient ) symmetric hyperbolic linear systems . this strategy calls for employing derivative operators satisfying `` summation by parts '' ( sbp ) together with an appropriate time integrator such as 3rd/4th order runge kutta schemes@xcite . the first ingredient allows one to derive a semidiscrete energy estimate while the second one guarantees this estimate will be preserved in the fully discrete case . following this strategy , we employ such operators here as our main building blocks ; as mentioned above , they provide a strong platform for developing a useful amr / fmr prescription . having defined the constituents of the basic update routine , we now turn our attention to the creation and update of a child grid and how updated field values on this grid are to affect those in the parent one . our goal is to obtain a recipe that will allow us to distinguish among different possibilities , the ones that guarantee stability for a simple model problem . as we will show , even when applied to a simple model problem , the analysis we present does well at predicting either stability or instability , suggesting applicability for more general problems . although our analysis applies to a simple model problem , the scheme it reveals to be stable is one that will naturally yield a stable scheme if it is so for a unigrid implementation . since the adoption of operators satisfying sbp together with an appropriate runge kutta update scheme guarantees stability for generic non - constant coefficients first order linear systems@xcite , the higher order amr scheme will preserve this property . to fix ideas and to simplify the presentation of our analysis , we restrict to the case where second and fourth order accurate schemes are employed . this is sufficient to illustrate the salient aspects of the analysis and main results ; the conclusion can be trivially extended to arbitrary orders of accuracy . in order to explore different possible ways to implement an amr strategy we analyze what a full discrete step involves for the simple equation @xmath2 on the region @xmath3 $ ] with periodic boundary conditions , and where a refined region ( the child grid ) is included within this domain . this can be assumed without loss of generality as the operators employed ( satisfying sbp ) and a suitable runge kutta time update ( together with suitable boundary conditions ) guarantee the stability of the implementation . thus , to avoid unnecessary clutter in the analysis we examine the periodic case for the coarsest grid involved . for simplicity , we restrict ourselves to the case of _ fixed mesh refinement _ with the child grid @xmath4 refined with respect to the parent grid . we then compute the eigenvalues of the full amplification matrix ( @xmath5 ) corresponding to the update scheme of field values on the parent level . clearly , if all the absolute values of this matrix are bounded by unity , amplification at each step is bounded by one ( i.e. modes are either kept constant or damped ) and stability follows for the problem under consideration if the matrix can be diagonalized . on the other hand , if there is an eigenvalue with absolute value larger than unity , the amplification per step is unbounded and the algorithm is therefore unstable . since this applies already to the simplest first order problem , the conclusion that the system is unstable will naturally hold for generic ones . note that it is sufficient to consider a single refinement level ( and the corresponding coarse grid ) in this analysis , as in a generic case consisting of several child grids the total amplification matrix will be a composition of several @xmath5 matrices . the specific form of @xmath5 depends on several different ingredients . the cleanest way to construct @xmath5 is to express it as a composition of different matrices , each representing the action of an operator performing a specific task involved in the overall update strategy . for the problem under consideration , we need to consider the matrices * @xmath6 which defines the grid values on the child level by knowing those at the coarse level , usually referred to as prolongation or interpolation . * @xmath7 and @xmath8 the derivative operators used on the parent and child grids respectively . * @xmath9 , the specific time stepping algorithm . * @xmath10 which restricts the values obtained on the child grid at the new coarse time level at appropriate points of the coarse grid ; usually referred to as restriction or injection . * @xmath11 a restriction operator which , when acting on an overall update scheme , restricts its values to the region in the coarse grid which needs no refinement . hence , @xmath12 restricts the action of any operator to the points in the coarse level which do require refinement . * @xmath13 , boundary points definition at intermediate child grid time levels which , depending on the strategy adopted , one might have to consider . an update step in a two - level scheme which takes values on the @xmath1-th level of the coarse grid ( @xmath14 ) and yields values at the @xmath0-th level ( @xmath15 ) can be expressed as @xmath16 , with : @xmath17 with @xmath18 and @xmath19 the update steps on the parent and child grids . for instance , for the advection equation above , and employing a 3rd order runge kutta scheme , @xmath18 becomes ( for a periodic boundary problem ) @xmath20 the expression for @xmath19 is more delicate and depends crucially on the scheme adopted . for instance , when employing @xmath21 order derivative operators , and the standard berger - oliger strategy for treating interface boundaries , one schematically has @xmath22 with @xmath23 denoting the intermediate steps of the runge - kutta algorithm , @xmath13 the identity operator at the interior points of the child grid , but zero at interface boundaries ; @xmath24 the interpolation of field values at points on the coarse grid ( at interface boundary points ) providing values at the @xmath25-th time level required by the time stepping algorithm . the above is just an example of what a typical procedure entails as there certainly are many alternatives for the injection / prolongation , update , derivatives operators and boundary definition that will be part of the amr / fmr strategy . a partial restriction is introduced when choosing the accuracy order desired as this fixes the derivative operators at interior points ( by adopting the standard centered operators ) . however this still leaves considerable freedom related to the way derivatives are to be defined at points at , and close to , the boundaries , in addition to the definition of the interpolating operators @xmath26 to be used ( if so desired ) in defining interface boundary points , the injection and prolongation operators , etc . the analytical investigation of every option is an involved task . to reduce the number of possibilities we adopt some simplifying , yet natural and commonly used , assumptions . throughout this work we concentrate on two - level schemes . we consider interpolating operators acting only at interface boundary points and , at most , a few additional points at each side of the refined grid . as much as possible we adopt the same basic integration step for all grids to ensure that a recursive type algorithm can be easily employed . finally we concentrate on vertex - centered schemes and child grids having points lying on a coarse grid points . even with these simplifications , obtaining the eigenvalues of the amplification matrix is a cumbersome task . for this reason , we perform our investigations with the aid of a computer algebra package ( maple ) to numerically calculate the eigenvalues . to do this , we define a grid composed of @xmath27 points describing the coarse level and where @xmath28 points will be refined ( the child grid has @xmath29 points ) . we construct the matrices defined above , obtain the amplification matrix and numerically calculate the eigenvalues . since we perform these calculations for a fixed number of grid points ( though we have varied @xmath27 and @xmath30 over several values ) , a negative result ( all absolute values less or equal than unity ) does not constitute a stability proof , but certainly a positive result , ( at least one eigenvalue with larger than unity absolute value ) shows that the proposed algorithm is unstable . we can thus substantially narrow the search for likely stable ones and the tests considered in section [ sec : tests ] lend additional strength to our analysis . as we will see , our analysis indicates a preferred option which , intuition suggests , will be stable for any problem where the unigrid case is stable . the construction of the amplification matrix proceeds in the following way . first a whole time step is taken in the coarser grid , so that we can use , if so desired , the values at the next time step to impose boundary conditions for the intermediate time steps taken on the finer grid . once the recipe to provide these boundary conditions is given and the difference operator on the finer grids chosen , the algorithm is pretty much fixed . other alternatives include not providing boundary conditions at all but rather to taper off the child grid as subsequent intermediate child time levels are obtained . additionally , one might consider other variants at the level of the interpolation order on @xmath6 , a possible averaging at the injection level , @xmath10 , or at the coarser grid , etc . we analyzed several of these options and report on their results in the following subsections . finally , as mentioned , we limit ourselves to runge - kutta time integrators of order larger than two as these are known to lead to stable schemes if the semi - discrete system is stable @xcite . here we use a third order runge - kutta time integrator ( rk3 ) as the second - order runge - kutta method is not known to guarantee the fully - discrete stable scheme if the semi - discrete scheme is stable . the population of the finer grid , given by the operator @xmath6 , is carried out using four - point stencil interpolation ( having an @xmath31 error ) on the child grid points lying in between coarse ones and using a direct copy of field values where the grid points coincide ( see appendix b for the interpolator s specific form ) . we consider the following cases : * * berger oliger : ( @xcite ) * here we reproduce the `` standard '' berger - oliger scheme and check its stability when used with rk3 . figure [ fig:1 ] illustrates the points involved in this scheme . here the values at the diamond points , needed for the intermediate rk steps of the finer grid ( squares ) , are obtained via linear interpolation in time employing values at the top and bottom points of the coarse grid ( circles ) . the coarse grid points are already known at the advanced time step because it takes a full time step before the fine grid is advanced . the dotted lines represent the rk steps intermediate to a full time step of the child grid . on the fine grid , only the square points are evolved , using centered difference operators , and at the end of the cycle the values of the fine grid coincident with coarser points are _ directly injected _ into the corresponding coarse grid points . * * penalty method : * in this case , in contrast to the previous one , we pay close attention to the propagation modes defined by the equation of motion . in this simple case , the single characteristic propagates from right to left ( see figure [ fig:2 ] ) . thus , only boundary values at the right end of the fine grid are required . these are provided by linearly interpolating coarse field values at the top and bottom levels and used as boundary conditions in a scheme enforcing these via penalty terms as described in @xcite . in this technique , boundary values are incorporated by modifying slightly the right hand side of the ( first order ) equations . for our test problem this results in @xmath32 where @xmath33 is the discrete derivative operator ( satisfying sbp ) employed ; @xmath34 for @xmath35 and @xmath36 at @xmath37 ( i.e. only non - zero at the right interface boundary point ) ; @xmath38 is a free parameter and @xmath39 is related to the norm / derivative operator employed in ensuring sbp is satisfied ( see @xcite for a detailed description ) . + the penalty technique provides a clean and efficient method to provide boundary values which , coupled with the use of operators satisfying sbp , guarantees the stability of the implementation . furthermore , within this approach , all points are updated and the boundary conditions are imposed via a simple addition of suitable penalty terms to the right hand side at boundary points . for these reasons it represents , at first sight , a convenient approach to handle characteristic modes within an amr / fmr implementation . as with the berger - oliger scheme , after the child grid update to the same time as the parent , field values coincident with the parent are directly copied . * * tapered boundary method : * in this case , we adopt restriction operators which effectively remove points that have been influenced by the interface boundary . in practical terms , this means that if the child grid at the advanced @xmath0 level is defined as @xmath40 \}$ ] , the update from level @xmath1 must take place in the region @xmath41 \}$ ] with @xmath42 given by a sufficient number of points to insure that the numerical domain of dependence of @xmath43 $ ] starting at @xmath0 is _ completely _ contained in @xmath44 $ ] . this is illustrated in fig [ fig:3 ] . the size of @xmath45 is determined by the order of accuracy of the discrete operators involved and the maximum speed of the hyperbolic problem considered . + incidentally , note that this is not the same as an alternative approach where ghost zones populating a region around the refined region ( with values obtained via linear interpolation from the coarse values at levels @xmath1 and @xmath0 ) are defined . that approach is relatively straightforward to implement and reduces the complexity of the derivative operators employed , but does not satisfy the requirement that what is done at the boundary has no influence on the final injected point . as we will see later , this scheme also has unstable modes . the analysis of the eigenvalues of the amplification matrices corresponding to each of the three cases discussed above must be examined carefully as we are considering the fully discrete case . here a competition among the unstable modes and the possible inherent dissipation of the update scheme determines whether the eigenvalues magnitude of the full amplification matrix are indeed larger than unity . in particular , for large cfl values , the inherent dissipation grows and the eigenvalues might be bounded by unity while for smaller cfl values this might not be the case . as we are interested in establishing stability with at most an upper bound for the cfl value we have examined several possible values of cfl ( ranging from @xmath46 to @xmath47 ) and observe when unstable modes arise . for instance , for a cfl value of @xmath46 all eigenvalues for the three approaches lie within the unit circle as illustrated in fig . [ bo2_eigenv ] . however , as this value is decreased unstable modes do appear for the penalty case while not for the other two options . one can exploit the free parameter @xmath38 to alleviate the instabilities in the penalty case ( in the sense that the magnitude of the eigenvalues gets closer to unity ) but not completely remove them with the addition of artificial dissipation . additionally , it is instructive to investigate the behavior of the error as a function of position to obtain some indication of the effect the artificial boundaries have . to this end , we compute the error in the evolution of @xmath48 after a single step for each case . here the solution after a step is @xmath49 and a point - wise error calculation is straightforward to compute . the results are indicated in fig . [ o2_error ] . the differences in the errors among the three schemes are clear . the b - o and tapered approaches have similar errors while those of the penalty scheme are larger due to the derivatives at the boundaries being calculated to first order accuracy . in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] here we use a fourth order runge - kutta time integrator and analyze the stability properties of the obtained schemes under the cases considered above . in this case , we will find that only the tapered boundary approach yields a stable scheme unless a carefully tuned amount of dissipation is added to the problem . we have found the same holds true when adopting a third order runge - kutta operator . we adopt _ fourth order accurate _ centered difference operators ( in the case of berger - oliger type schemes ) or those satisfying the sbp property , in the case of the penalty approach . for the tapered boundary approach it does not matter which one is employed as points affected by ` special ' boundary treatments are to be discarded . the population of child grid points lying between coarse ones uses a six - point stencil interpolation operator ( having an error of @xmath51 ) to ensure that the discontinuity in the truncation error in between different grids does not affect the order of the scheme ( see appendix b for the specific expressions of the interpolators ) . for a first order equation , employing fourth order derivative operators requires at least a fifth order interpolation stencil . we adopt sixth order to avoid spoiling the symmetry of the implementation . we consider the following cases : * * berger - oliger method : * + here the setting is as in the second order case , but some modifications must be introduced at the level of the derivative operators employed . one option is to modify them near the boundaries in a side - ways manner so as to preserve the order of the approximation . the other is to define extra ghost zones around the refined region and keep the derivative operator in its standard centered expression . we have analyzed both cases with similar qualitative results . we discuss here the latter case in some detail . + for a fourth order accurate approximation to the derivative operator only two ghost zones on each side of the grid are needed . one is located at the interface boundary while the other is placed next to it outside of the refined region . as before , the field values at these points are to be defined via interpolation from those known at the coarse level on levels @xmath1 and @xmath0 . we examined both linear interpolation with error @xmath52 and quadratic interpolation with error @xmath53 . the latter option employs the value of the field at the @xmath1 and @xmath0 level ( as in the linear case ) together with the time derivative of the fields at level @xmath1 . the qualitative features of the amplification matrix is the same in both cases , namely that there exists eigenvalues with magnitude greater than one . we did not go beyond this order as it would involve taking further derivatives of the right hand side , which in a general case would involve derivatives of both coefficients and variables * * penalty method : * here again the setting is similar to the second order one , the only difference is that at the boundary the interpolation is done at third order , again using the computed right - hand - side of the equations at the initial time . this is only needed at boundaries where there are incoming characteristics as at the other boundaries field values will be defined solely by the update within the child grid itself . additionally , no extra ghost zones are required as one must use sbp operators of the appropriate order ( see for instance , @xcite ) . * * tapered boundary method : * this case is analogous to the second order case . the only significant difference is the size of @xmath54 , which must be enlarged so as to ensure the intersection of the past domain of dependence of the region of interest at the @xmath0 level is completely contained inside the refined region at level @xmath1 . the analysis of the eigenvalues of the amplification matrices corresponding to each of the three cases discussed above reveals that there exists eigenvalues with magnitude larger than unity for the b - o and penalty approaches while none for the tapered one . clearly the stability of the tapered approach follows from the stability of the unigrid problem . here again , the amount by which the eigenvalues of the b - o and penalty approaches are above unity is not too large and depends strongly on the cfl factor employed . however , from our analysis it is expected that as the base grid is refined and smaller cfl factors are used ( or a larger number of refinement levels are employed ) instabilities associated with these modes will appear . beyond this point however , the tapered approach also yields evolutions with much smaller errors . to illustrate this we again compute the error in the evolution of @xmath48 . the results are indicated in figure [ o4_error ] ; again , the effects of the boundaries can be clearly appreciated but this time the errors in the tapered grid are 40 times smaller . in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and no artificial dissipation added.,title="fig:",width=207,height=264 ] these results make it clear that straight forward extensions of berger - oliger to higher order , as well as the use of a penalty method which takes into account the characteristic structure of the problem , suffer from unstable modes in a higher order amr / fmr scheme . the only option we have found to be free of this problem is that given by the tapered boundary approach , which a priori would appear more computationally demanding as points are evolved which will be discarded . for sufficiently high order schemes , this can be a significant fraction of the grid . although this extra cost might not be as significant as it appears ( see appendix [ sec : cost ] ) , one might prefer to maintain the standard berger - oliger prescription and deal with the unstable modes by introducing some amount of dissipation . therefore we revisit the analysis above but now with the addition of dissipation . to anticipate the results obtained , we indeed find that dissipation can control the unstable modes but at the expense of increasing the overall error and needing to fine tune it to ensure a desired degree of smoothness in the solution . as noted , the update scheme obtained for the bo and penalty schemes are unstable for the higher order case ; this can be remedied or alleviated by the introduction of some amount of dissipation . we consider the addition in two ways , one as a dissipative operator modifying the right hand side of the equation ( and hence involved in all time levels ) , or as a high - pass filter only at parent levels . the latter is amenable to be analyzed through the method outlined in section [ sec : analytical ] while the same for the former becomes too cumbersome . the analysis with the addition of the filter shows that the addition of dissipation can render the scheme stable and experimental tests ( in the tests presented in the next section ) indicates the same holds true with the first option . here we present results of the errors obtained after a single step in each of the approaches considered , illustrated in figures [ b4_error_diss0 ] and [ b4_error_diss1 ] . these figures show the errors obtained after a single evolution step of a ( well resolved ) smooth initial data given by @xmath55 . figure [ b4_error_diss0 ] displays the observed behavior when no dissipation is employed . the errors obtained with the b - o and penalty approaches are about two orders of magnitude above those seen when employing the tapered approach after a single step . the addition of dissipation alleviates this issue though the amount of the dissipation has to be tuned to yield smooth errors while not affecting their magnitude . figure [ b4_error_diss1 ] show the obtained behavior with @xmath56 which is not yet enough to give smooth errors for the b - o approach and even less for the penalty method while quite suitable for the tapered approach . in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation of the kreiss - oliger form has been added with @xmath57 . this value of @xmath58 is not yet enough to smooth - out the errors in the b - o and penalty approaches , though in the first case makes the errors about the same as those obtained with the tapered approach.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation of the kreiss - oliger form has been added with @xmath57 . this value of @xmath58 is not yet enough to smooth - out the errors in the b - o and penalty approaches , though in the first case makes the errors about the same as those obtained with the tapered approach.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation of the kreiss - oliger form has been added with @xmath57 . this value of @xmath58 is not yet enough to smooth - out the errors in the b - o and penalty approaches , though in the first case makes the errors about the same as those obtained with the tapered approach.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation kreiss - oliger type dissipation has been added with @xmath59 . this value of @xmath58 is too small to smooth - out the errors in the b - o and penalty approaches while is still reasonable for the tapered approach . furthermore , the errors in the b-0 approach are now about @xmath60 times larger than in the tapered one.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation kreiss - oliger type dissipation has been added with @xmath59 . this value of @xmath58 is too small to smooth - out the errors in the b - o and penalty approaches while is still reasonable for the tapered approach . furthermore , the errors in the b-0 approach are now about @xmath60 times larger than in the tapered one.,title="fig:",width=207,height=264 ] in the domain @xmath50 $ ] under the three schemes considered from left to right : the b - o , penalty and tapered schemes. a cfl factor of 0.25 was employed and artificial dissipation kreiss - oliger type dissipation has been added with @xmath59 . this value of @xmath58 is too small to smooth - out the errors in the b - o and penalty approaches while is still reasonable for the tapered approach . furthermore , the errors in the b-0 approach are now about @xmath60 times larger than in the tapered one.,title="fig:",width=207,height=264 ] in the following one - dimensional series of tests , we consider four natural alternatives . these differ in their handling of refinement boundaries , in particular how they compute derivatives there and what conditions are imposed . derivatives at interior points are defined by the standard centered ones . these alternatives are : ( i ) sideways derivatives ( constructed so that their approximation order is equal to that of the centered derivative operator employed at the interior points ) ; ( ii ) derivatives of lower order near the boundaries , which , in the absence of a time interpolation , would satisfy summation by parts ; ( iii ) centered derivatives throughout , with extra ghost zones on each side of the child grid as needed for applying the standard centered derivative . in the above three cases , boundary ( and guard ) data as the integration proceeds is obtained via interpolation from the parent grid . the last option ( iv ) is the `` tapered boundary approach '' described in section [ sec : second ] . we study the long term evolution of several systems implementing an fmr scheme . since in this case the grid - structure is fixed it does not correspond to a well suited adaptive simulation , as the adaptivity would sequentially create grids to follow the traveling features in the solution ( assuming , of course , an appropriately defined refinement criterion is in place ) . however , this arrangement is well suited for our task here to assess the stability of the simulation and the interaction of features in the solution propagating in and out of the finer refined regions . our tests consist of : * the first test is restricted to the simplest possible case provided by the advection equation @xmath61 . this is a hyperbolic system with constant in time and space coefficients . for such a simple case , it is observed that higher order accuracy can be achieved with a rather simple interface boundary treatment . * the second test corresponds to a linearization of the einstein equations with respect to a time / space dependent background . * the last test corresponds to the full einstein equations restricted to a spherically symmetric case . in this case , excision techniques are also implemented . we here concentrate on the advection equation @xmath61 discretized in the domain @xmath62 \}$ ] with two levels of refinement @xmath63 given by @xmath64 \}$ ] ( @xmath65 ) . periodic boundary conditions in the @xmath66 grid are enforced directly at the difference operators level while interface boundaries @xmath67 are handled via the standard b - o method or the tapered boundary approach . we implemented both second and fourth order accurate derivative operators with a fourth order runge kutta operator for the time integration . the initial data is given by a compact support pulse given by @xmath68 $ } \\ 0 & \mbox{otherwise \ , , } \end{array } \right.\end{aligned}\ ] ] with @xmath69 , @xmath70 and @xmath71 . these limits are arbitrarily defined ensuring the pulse lies in the middle grid . as time progresses , the pulse will move across all grids . for the second order case , we have seen that all options discussed give a stable implementation . we corroborated this by monitoring the error in the solution obtained with both the bo and tapered boundary approaches . to this end , we evolved the solution on a uniform grid with @xmath72 points and adopted it as our `` analytical solution '' @xmath73 ( recall that the unigrid problem is guaranteed to be stable ) . we then run with both implementations using a coarse grid consisting of @xmath74 and @xmath75 points and calculate the error for each numerical solution @xmath76 as @xmath77 . figure [ botap_2nd_error ] illustrates the observed behavior which indicates both approaches give practically the same solutions . next we calculate the convergence rate with these obtaining values consistent with second order accuracy as shown in figure [ botap_2nd_conv ] . norm of the solution obtained with the b - o and tapered boundary approaches . a cfl @xmath78 was employed and no dissipation . note that the vertical axis has not been rescaled the error in the solution indeed becomes very large after several crossing times manifested in large phase / amplitude errors.,height=302 ] . [ botap_2nd_error ] . [ botap_2nd_conv ] when employing fourth order accurate derivative operators as discussed in section [ sec : fourth ] , boundary conditions are required in approaches ( i ) through ( iii ) . we consider both linear and cubic interpolation employing in the first case the values of the function at the coarser time levels and in the second case we include the time derivative to obtain a cubical interpolant . these interpolations provide boundary data at points @xmath67 and @xmath79 . notice that when employing cubical interpolation in time and linear interpolation in space , no significant qualitative differences were observed if a higher order interpolation in space was employed . as discussed previously , without the addition of dissipation , options ( i ) through ( iii ) yield unstable implementations , and we have confirmed this by monitoring the @xmath80 norm of the solution vs. time . by running for several crossing times for different resolutions , it is seen that the norm grows faster for finer refined grids . next , in order to control to some extent the instability , we add dissipation to the problem . this can be done in two forms , one is by modifying the right hand side of the equations by introducing a dissipative term and the other is to apply a filter to the updated variables . in the second option the filter applied is , in a sense , equivalent to introducing an integration on a fictitious time after each step ( with respect to the coarsest grid ) of the equation @xmath81 . this results in the replacement @xmath82 as a comparison of what is observed , figure [ bo_tap_4th ] illustrates the the @xmath80 norm of the difference between the solution obtained with @xmath83 and @xmath84 points vs. time ( with @xmath85 ) . the figure illustrates the behavior of this quantity for the solutions obtained with the b - o approach obtained with two different dissipation and cfl parameters . these solutions are obtained with strategy iii ( namely with two ghost zones on each side of the child grids to employ the standard centered difference operator ) . we additionally show the corresponding result for the tapered boundary approach . as illustrated in the figure , while the dissipation manages to control the exponential growth of the solution , convergence is severely affected . the solution obtained with the tapered boundary approach on the other hand , behaves stably and the error scales as expected . figure [ bo_tap_4thconv ] shows the convergence factors calculated with these solutions , while the tapered boundary approach yields a value quite close to the expected one of @xmath60 , the one obtained with strategy iii quickly deteriorates . . [ bo_tap_4th ] . [ bo_tap_4thconv ] finally , as was discussed in section [ sec : fourth ] , unless both the cfl and dissipation parameter are chosen carefully , this filtering might not be enough to control the instability . for instance , for a cfl factor @xmath86 and @xmath87 , the growth of the instability is less severe and might not be visible for relatively coarse resolutions for long times though as the grid is refined the instability might become evident . we illustrate this behavior , and contrast it with the one observed with the tapered grid approach , by monitoring the @xmath80 norm of the error @xmath88 defined as in the second order case ( i.e. with respect to the solution obtained with a unigrid evolution on a grid consisting of @xmath89 points ) . figure [ bo_tap_4therror ] shows the results obtained for the coarse grids having @xmath90 and @xmath75 points . clearly , while the tapered boundary approach gives the expected behavior , strategy ( iii ) with the addition of dissipation fails to give good answers . notice that while the solution obtained with @xmath90 points does not exhibit an instability the one with @xmath75 points does so clearly . the reason behind this is that the instability for the coarser case has not yet grown above the truncation error in the solution . norm of the solution obtained with the b - o and tapered boundary approaches . a cfl @xmath86 was employed and @xmath87 in the b - o case.,height=302 ] . [ bo_tap_4therror ] the second test is defined by considering the linearized einstein equations off the `` gauge - wave '' solution . this solution describes flat - spacetime in a coordinate system that introduces an explicit time - position dependence of the metric as @xmath91 we then write the linearized einstein equations off this spacetime in a first order symmetric hyperbolic form as discussed in @xcite following the formulation introduced in @xcite which includes a dynamical lapse . by restricting to perturbations in the @xmath92 directions , solely depending on these coordinates , one obtains a reduced system of non - trivial equations for @xmath93 . we solve the resulting equations in the domain @xmath94 $ ] , with @xmath95 and compare the results obtained with 2nd and 4th order accurate spatial derivative operators while maintaining @xmath96 order accurate integration in time via a runge kutta algorithm . we first run a series of tests employing a @xmath21 order accurate spatial derivative operators , a 4th order interpolation to create child grids , a @xmath96 order runge - kutta time integrator and a direct injection from child to parent grid points . we concentrate on the accuracy and convergence rate observed for cases ( i - iii ) and ( iv ) above . ie , in the first case we employ second order derivative operators at interior points while first order one at the boundaries . since boundary points in this case are over - written by the time - interpolation cases ( i - iii ) are identical . for case ( iv ) however , we define extra child points via interpolation from parent to child at level @xmath97 . these extra points are discarded at the @xmath98 level and are solely used to remove any dependence on the refinement boundaries that child points , at the region of interest , have at the @xmath97 level . first we confirmed that the unigrid code is stable and convergent , and pick the solution obtained with the highest resolution ( corresponding to a grid with 3201 points ) and consider it the ` analytically ' expected solution @xmath73 which we use to compare the results obtained when using fixed mesh refinement . for these tests we employ a base grid of @xmath90 and @xmath99 points , each with 2 levels of refinement ( with boundaries at @xmath100 $ ] with @xmath101 ) and no dissipation is included . figure [ 2ndordercomparegw ] illustrates the ( log of the ) @xmath80 norm of the difference between the solution obtained at the base grid @xmath76 and @xmath73 with @xmath102 . the obtained results with the standard fmr approach and the tapered boundary one are practically indistinguishable from each other . this also implies the convergence of both should be practically the same , ie . second order , a fact that is shown in figure [ 2ndorderconvergegw ] . . [ 2ndorderconvergegw ] in the @xmath96 order accurate case , as expected from the analysis presented in section ii , differences do arise . for these tests we employ 4th order derivative operators at interior points suitably modified near the boundaries to cover cases ( i ) and ( ii ) or ghost zone points suitable added for case ( iii) ; a 6th order interpolation from parent points to define child grid points ; a 4th order rk time integrator and a direct injection from child to parent grid points . as before , we first run a series of unigrid tests to confirm that a stable evolution is obtained as expected and choose the solution obtained with the highest resolution run ( obtained with a grid consisting of 3201 points ) and consider it the ` analytically ' expected solution @xmath73 . again we consider a base grid of @xmath90 and @xmath99 points , each with 2 levels of refinement ( with boundaries at @xmath100 $ ] with @xmath101 ) and as above no dissipation is included . figure [ 4thordercomparegw ] illustrates the ( log of the ) @xmath80 norm of the difference between the `` analytical '' solution and those obtained with a coarse grid ( plus the two refinement level ) @xmath76 with @xmath102 . clearly the errors associated with the implementations ( i - iii ) grow considerably faster ( indeed exponentially ) than those obtained with the tapered boundary approach ( linearly ) . in fact , after about @xmath103 crossing times the error associated with the tapered boundary approach using @xmath90 base points is approximately the same as that with the standard approach but with a base grid of @xmath99 points . figure [ 4thorderconvergegw ] illustrate the measured order of convergence for these methods , while the value obtained with the `` standard approaches '' significantly reduces as time proceeds the corresponding one to the tapered boundary approach remains near @xmath60 decreasing just slightly as errors accumulate over time . . [ 4thordercomparegw ] . [ 4thorderconvergegw ] last , for illustrative purposes , we show in figure [ 4th2ndnew ] the errors associated with the different solutions obtained with the tapered boundary approach for @xmath21 and @xmath96 order accurate spatial operators . the accuracy gained by employing a higher order operator is evident . . [ 4th2ndnew ] in the last one - dimensional test we consider einstein equations restricted to spherically symmetric spacetimes in vacuum . the solution for a black hole initial data must be static and hence field variables have only non - trivial radial dependence . we again employ a grid structure composed of a base grid given by @xmath104 \}$ ] with two levels of refinement @xmath63 given by @xmath105 \}$ ] ( @xmath65 ) . thus , while the boundaries at the right boundaries of each domain are sequentially located at smaller radii , the left boundaries all coincide at @xmath106 . at these boundaries , for the tapered - boundary approach one takes advantage of the fact that all characteristic variables are inflow , and hence the values at each grid at and near these boundaries can be updated using solely grid values within that grid . we implemented einstein equations in 1 + 1 dimensions as described in @xcite with fourth order accurate derivative operators at interior points ( suitably modified near boundary points as dictated by the sbp requirement ) and a fourth order runge - kutta update scheme in time . we adopt unperturbed initial data corresponding to schwarzschild in ingoing eddington finkelstein coordinates with an analytically given shift and the free part of the densitized lapse function . at the outer boundary of the coarse grid , constraint preserving boundary conditions are imposed as described in @xcite while the artificial boundaries are either handled via the standard b - o scheme ( interpolating two ghost - zone points ) or via the tapered grid approach . a small amount of dissipation is added in all tests with a fifth order kreiss - oliger style operator @xmath107 suitably modified near the boundaries as explained in @xcite ( in our runs @xmath59 ) . since the solution should converge to the analytical value , we can compute the error with respect to it . figure [ conv_grr ] illustrates the convergence factor obtained from the variable @xmath108 ( for @xmath109 and @xmath110 points ) in the coarse grid . for both cases the obtained convergence factor is consistent with the expected value of @xmath60 . for both the bo and tapered boundary approaches . in both cases , a behavior consistent with fourth order accuracy is obtained.,height=321 ] . [ conv_grr ] last , we examine the behavior of the errors themselves for the resolutions employed . figure [ errors_grr ] display the @xmath80 norm of the errors @xmath111 with @xmath112 the analytical and numerical solution ( the latter obtained with @xmath1 points in the coarse grid ) . notice that although both schemes yield a stable evolution , the errors associated with the tapered grid approach are about an order of magnitude smaller than those obtained with the bo scheme . for both the bo and tapered boundary approaches . in both cases , a behavior consistent with fourth order accuracy is obtained.,height=321 ] . [ errors_grr ] to test these ideas in three dimensions , we use an implementation of the semilinear wave equation . a scalar field @xmath113 obeys the wave equation @xmath114 for odd @xmath115 and where @xmath116 serves to parameterize the nonlinearity . for @xmath117 the linear case is obtained , whereas for @xmath118 the model is focusing ( studied in @xcite ) and for @xmath119 the model is defocusing . choosing cartesian coordinates and defining as the fundamental fields in order to cast the problem in first order differential form : @xmath120 the equation of motion eq . ( [ sll : eom ] ) takes the form of the following system of equations @xmath121 solutions of this system are found by incorporating appropriate finite differences within a distributed amr infrastructure . using fourth order accurate finite difference approximations of spatial derivatives along with a runge - kutta time integrator accurate to third order , convergence to third order is expected as the resolution is increased . this serves also to illustrate that the tapered grid approach allows for decoupling the order of the time - integrator and the derivative operators . an example of such a convergence test is shown in fig . [ sll : conv ] . a coarse grid with bounds in each direction of @xmath122 is established with a ( fixed mesh ) refined ( 2:1 refinement ) region established with bounds of @xmath123 . evolutions were carried out for increasing resolution and the results compared . the lines shown indicate the rescaled differences @xmath124 where @xmath125 and @xmath126 indicate the number of points in each dimension , @xmath127 , @xmath128 , and @xmath129 . the rescaling factor @xmath130 is computed as @xmath131 / [ 1/(a-1)^3 - 1/(b-1)^3 ] $ ] which is how one expects these differences to scale if indeed the code is third order convergent . as the results in the figure show , these differences do decrease as expected . furthermore , although the size of the differences increase in absolute terms as the pulse propagates across the refinement boundary , convergence and smoothness is not disturbed . similar results are obtained for a nonlinear example , as shown in fig . [ sll : conv2 ] , where a nonlinear case ( @xmath132 ) is employed . the initial data describes and incoming pulse which display similar behavior at @xmath133 . our simulations show that even in this highly non - linear regime , the solutions converge to the expected order . . the initial data consists of an initially static gaussian pulse centered about the origin for the case where @xmath117 . that the rescaled differences agree , even as the pulse traverses the boundary of the refined region ( the green vertical lines ) , suggests that third order convergence is maintained : red , dotted line ( @xmath134 - @xmath135 ) , @xmath136 , blue , solid line ( @xmath135 - @xmath137 ) , @xmath138 , magenta , short dashed line ( @xmath137 - @xmath139 ) , @xmath140 , yellow , long dashed line ( @xmath139 - @xmath141 ) , @xmath142 , and cyan , dot - short dashed line ( @xmath141 - @xmath143 ) , @xmath144 . , width=491 ] , @xmath145 ) of an ingoing wave pulse . the resolutions are the same as those shown in fig . [ sll : conv ] , and similar agreement among them suggests third order convergence . the evolutions demonstrated singular behavior soon after the last frame , placing this evolution confidently within the nonlinear regime of the model . , width=491 ] in the present work we have presented an analysis of different alternatives aiming to yield a stable fmr / amr implementation . it is observed that the straightforward application of the standard b - o scheme has unstable modes and that natural variations of this scheme share this problem . on the other hand , the tapered boundary approach , in which points that have been affected by the interface boundary treatment are discarded , is clearly stable . in addition to being free of unstable modes , by discarding these points the reflections at the interface between child and parent grids are significantly reduced . the unstable modes found are represented by eigenvalues of the amplification matrix with norm larger than @xmath46 . an inspection of these reveal that they differ from unity by a small amount ( for the toy model considered , and typical grid / cfl values are @xmath146 ) . therefore , the instabilities associated with these modes will grow slowly for few refinement levels and might not become too evident for evolutions of moderate length . it is thus not surprising that in common applications in numerical relativity where typical cfl factors employed are of order unity , relatively coarse resolutions are employed and the update scheme used is the so - called iterative crank nicholson which is inherently much more dissipative than the runge - kutta method considered here simple test - beds with few levels of refinement do not display unstable modes . for several refinement levels , the instabilities will play a role unless suitably dissipated away . however , we stress again that even if the instabilities might be controlled this way , the errors obtained in all our tests with `` standard '' approaches are significantly larger than those seen when employing the tapered grid approach . other schemes have been recently introduced to reduce the interface reflections for second order methods@xcite . this involves suitably modifying the derivative operators to preserve smoothness in the error of the advection equation at the interfaces . although the application of these operators give improved results in black hole spacetimes , the results presented did not include sub - cycling and higher derivative schemes have not yet been worked out . finally , although we have noticed that the penalty approach does not help in developing a stable amr strategy , it does play a key role in the stable handling of multiple grid structures that abut @xcite . thus an obvious strategy to implementing amr techniques within a multiple grid structure is to combine both the penalty technique with the tapered approach . the penalty technique would be used to deal with different grid interfaces while the tapered approach to handle artificial interface boundaries as presented in this work . throughout our work we have restricted ourselves to fully first order equations because , in this case , a number of results exist on how to ensure stability of the fully discrete implementation . such results are not available for fully second order or first order in time / second in space systems , though work in this direction is underway@xcite . certainly , the tapered approach can still be used in both these types of systems as it only relies on having a stable unigrid implementation and ensuring enough fine grid points in the numerical past domain of dependence at level @xmath1 to define the solution at level @xmath0 . in summary , the tapered grid approach provides a safe and natural way to build a stable amr / fmr scheme of arbitrary order . it has a non - trivial overhead cost , but this is offset by the significantly smaller errors and spurious reflections the solutions obtained with it has when compared to existing alternatives . note : while completing this work a scheme , similar in spirit , for a fourth order implementation has been presented in @xcite though in this case time and space interpolations are used to fill needed points while the boundary is tapered off . we here compare two stable options for the @xmath61 problem in their associated computational cost . we concentrate on the relatively inexpensive bo 2nd order scheme and a tapered boundary approach implemented with 4th order accuracy . the particular issue we concentrate on is whether the additional expense introduced by the 4th order tapered boundary approach renders it too costly to reach a target error when compared with a 2nd order implementation . to obtain a conservative estimate we assume a base grid with @xmath147 points covering a cubical domain of size @xmath148 and so we define @xmath149 . we assume that within this domain , a further refined grid is needed which has @xmath150 points . now , suppose one has the option of adopting a 2nd order or a 4th order accurate overall stable scheme . we want to estimate the cost that each one will have associated to it . the computational cost will be determined by the cost of the coarse and fine grid updates . for the coarse grid one needs to take second or fourth order derivatives ( which involves two and six operations per point respectively ) and the update step ( which we take as runge - kutta 3rd and 4th order respectively ) . the cost to update the coarse grids is simply governed by the derivative calculation and the runge - kutta update . a simple calculation indicates the cost is @xmath151 ( @xmath152 ) floating point operations per point for the second ( fourth ) order derivative operators plus runge kutta 3rd order ( 4th order ) . these result from adding the cost associated to computing the different runge - kutta steps ( ignoring the storage cost ) , to that of the final update . the algorithm to update the child grid has four basic steps : * a * creation of the child grid , * b * computation of derivatives , * c * update and * d * injection from child to parent grid . the cost associated with the child grids for * b * and * c * , per point , are the same as those in the parent grid . however , since for the 4th order case the child grid must be enlarged , we will here consider that the number of child grid points doubles ( hence there will be @xmath83 and @xmath84 points associated to the second and fourth order accurate schemes . this is certainly an exaggeration but serves as a rough upper bound estimate ) . in addition to the cost associated with the update of the child grid , its creation , step * a * , via fourth ( sixth ) order interpolation adds a cost of five ( eight ) operations per point . since we adopt a direct injection ( i.e. just copying the child grid value on the parent s at the corresponding point ) , the floating point operation associated with step * d*. hence , the cost of @xmath153 is @xmath154 and @xmath155 . combining these numbers one has a total cost of for the second order scheme of @xmath156 and @xmath157 . assuming @xmath158 we have for the @xmath159 problem @xmath160 and @xmath161 . thus , @xmath162 . therefore , the 2nd order update scheme could be employed ( in this simple and conservative estimate ) in a grid that is @xmath163 times more refined than that of the fourth order scheme with comparable cost . this however does not take into account the errors in the solution obtained with both methods . to obtain an estimate of this , we can make use of the analysis in @xcite where a comparison of the number of points needed to reach a tolerance error per wavelength is presented . as can be seen in the quoted numbers in table [ table ] , for a target error @xmath164 the fourth order accurate scheme is more efficient than the second order one . we here quote some relevant numbers from @xcite , + .number of points required per wavelength to achieve a target error @xmath165 with second , fourth and sixth order accurate operators . [ cols="^,^,^,^",options="header " , ] [ table ] hence , the associated extra - cost of a higher order accurate scheme done via the tapered grid approach which discards a non - trivial number of points is offset by the gains in accuracy , and of course , stability of the scheme . points in the child grid lying ( as seen from the parent grid ) at @xmath166 ( @xmath167 ) are directly injected from the parent value . the points lying in between coarse points at @xmath168 ( @xmath35 ) are obtained by the interpolating polynomial , @xmath169 when excision is employed , the stencil for the second second child grid point must be modified as there are no enough points to one side of it . in this case we employ , @xmath170 as in the previous case , points in the child grid lying ( as seen from the parent grid ) at @xmath166 ( @xmath167 ) are directly injected from the parent value . the points lying in between coarse points at @xmath168 ( @xmath35 ) are obtained by the interpolating polynomial , @xmath171 when excision is employed , the stencil for the second and fourth child grids point must be modified as there are no enough points to one side of it . in these cases we employ , @xmath172 this research was supported in part by the nsf under grants no : phy0244335 , phy0326311 , int0204937 to louisiana state university and nsf grant no . phy0325224 to long island university ; conicet and secyt - unc . many of these computations were carried on npaci computer resources under allocation awards phy040021 and phy040027 . l. l. was partially supported by the alfred p. sloan foundation . for hospitality while completing parts of this work we wish to thank the louisiana state university ( o.r & s.l . ) , long island university ( l.l & o.r . ) and the isaac newton institute for mathematical sciences ( l.l & o.r . ) . m. miller , arxiv : gr - qc/0502087 . l. lehner , o. reula and m. tiglio , `` multi - block simulations in general relativity : high order discretizations , numerical stability , and applications , '' to appear in class . quant . arxiv : gr - qc/0507004 . m. w. choptuik , e. w. hirschmann , s. l. liebling and f. pretorius , phys . lett . * 93 * , 131101 ( 2004 ) . f. pretorius and l. lehner , j. comput . phys . * 198 * , 10 ( 2004 ) s. l. liebling , phys . d * 71 * , 044019 ( 2005 ) . f. pretorius , class . * 22 * , 425 ( 2005 ) . e. evans , s. iyer , e. schnetter , w. m. suen , j. tao , r. wolfmeyer and h. m. zhang , phys . d * 71 * , 081301 ( 2005 ) f. pretorius and m. w. choptuik , arxiv : gr - qc/0508110 . ( 2005 ) . b. zink , n. stergioulas , i. hawke , c. d. ott , e. schnetter and e. muller , arxiv : gr - qc/0501080 ( 2005 ) . u. sperhake , b. kelly , p. laguna , k. l. smith and e. schnetter , phys . d * 71 * , 124042 ( 2005 ) . b. bruegmann , w. tichy and n. jansen , phys . lett . * 92 * , 211101 ( 2004 ) . e. schnetter , s. h. hawley and i. hawke , class . * 21 * , 1465 ( 2003 ) . j. ray , c.a . kennedy , s. lefanty and h.n . " high order spatial discretizations and extended stability methods for reacting flows on structured adaptively refined meshes . proceedings of the third joint meeting of the u.s . section of the combustions institute . ( 2003 ) .

we present the results from _ suzaku _ observations of the perseus galaxy cluster , which is relatively close , the brightest in the x - ray sky and a relaxed object with a cool core . a number of exposures of central regions and offset pointing with the x - ray imaging spectrometer cover a region within radii of . the central data are used to evaluate the instrumental energy - scale calibration with accuracy confirmed to within around 300 km s by the spatial and temporal variation of the instruments . these deep and well - calibrated data are used to measure x - ray redshifts of the intracluster medium . a hint of gas bulk motion , with radial velocity of about km s relative to the main system was found at arcmin ( 4590 kpc ) west of the cluster center , where an x - ray excess and a cold front were found previously . no other velocity structure was discovered . over spatial scales of 50100 kpc and within 200 kpc radii of the center , the gas - radial - velocity variation is below 300 km s , while over scales of 400 kpc within 600 kpc radii , the variation is below 600 km s . these x - ray redshift distributions are compared spatially with those of optical member galaxies for the first time in galaxy clusters . based on x - ray line widths gas turbulent velocities within these regions are also constrained within 10003000 km s . these results of gas dynamics in the core and larger scales in association with cluster merger activities are discussed and future potential of high - energy resolution spectroscopy with astro - h is considered .

1312.4526

galaxy clusters are the largest and youngest gravitationally - bound cosmic structure and form hierarchically through collisions and mergers of smaller systems . spatial and radial velocity distributions of member galaxies and x - ray observations of the intracluster medium ( icm ) have revealed that some systems are still forming and unrelaxed . along with gravitational lensing observations , these studies have provided dynamical measurements of dark matter in clusters . as the structure forms , gravitational potential governed by dark matter pulls and thus heats the icm through shocks . sharp x - ray images obtained by _ chandra _ revealed shocks and density discontinuities ( so - called `` cold fronts '' ) suggesting supersonic or transonic gas motions , respectively ( see * ? ? ? * for a review ) . these heating processes would then develop gas turbulence , and accelerate particles , which in turn generate diffuse radio halos and relics . the activities of central massive black holes also disturb and warm up the gas around cluster cores . the motion of gas should be measured to facilitate direct understanding of these energy flows . constraining total energy distribution in a system allows the total gravitational mass to be measured , as required for precise cosmology . energies in gas random or ordered motions could be the key uncertainty to calibrate total mass distribution , as suggested by numerical simulations ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? . gas bulk or turbulent motions can be measured directly using the doppler shift or by broadening of x - ray line emissions . however , the limited energy resolutions of current x - ray instruments continues to hinder such measurements . for example typical ccd energy resolution is about 130 ev ( in full width half maximum ; fwhm ) , compared with a possible energy shift of @xmath4 ev corresponding to a radial velocity of 1000 km s@xmath1 , all at the fe - k line energy . the _ suzaku _ xis @xcite is currently the optimal x - ray spectrometer for gas motion search with fe - k lines , as was demonstrated when gas bulk motion was detected in a merging cluster , abell 2256 , by @xcite . in addition , xis data imposed tight constraints on gas motion in a number of clusters , as summarized in table [ tbl : suzaku - res ] . lll centaurus & @xmath5 km s@xmath1 & 1 + oph . & @xmath6 km s@xmath1 & 2 + amw 7 & @xmath7 km s@xmath1 & 3 + a 2319 & @xmath8 km s@xmath1 & 4 + a 2256 & @xmath9 km s@xmath1 & 5 + coma & @xmath8 km s@xmath1 & 6 + a 3627 & @xmath10 km s@xmath1 & 7 + numerical studies have striven to clarify cluster formation histories and the associated origins , development , and dissipation of various types of gas motions . however , by nature cluster evolution is governed by random processes , which means systematic measurements of systems in a wide range of evolution stages are also required as well as these theoretical efforts . the geometrically - complex nature of gas dynamics should be spatially resolved . for example , gas dynamics in apparently relaxed clusters should be studied as demonstrated by ( * ? ? ? * hereafter ota07 ) using early _ suzaku _ observations of the centaurus cluster . to improve the limit of gas bulk motion in a relaxed system , we analyzed a large set of _ suzaku _ data of the perseus cluster . the perseus cluster , the brightest in x - ray , is a prototype of a nearby and relaxed object . at the same time , it exhibits a number of past and current activities at core and larger scales , while at the cluster center , the very peculiar central galaxy ngc 1275 displays a complex network of emission - line nebulosity . based on the thread - like structure of filaments resolved by the hubble space telescope , @xcite suggested the magnetic support of the structure and ordered gas motions not highly turbulent over scales of @xmath11 kpc around the galaxy . the cluster x - ray emission peaks sharply toward the galaxy and its central active galactic nucleus ( agn ) , 3c 84 . the agn also develops jets and radio lobes where high - energy particles and x - ray - emitting - hot gas interact , as discovered by @xcite . associated with these current and possibly past agn activities , fine gas structures such as bubbles , ripples , and weak shock fronts were revealed by deep _ _ imaging ( * ? ? ? * and references therein ) . these processes dissipate energies into various types of gas motions . independently , based on the lack of resonant scattering of fe - k line emissions , @xcite suggested gas motions in the core of a velocity of at least half the speed of sound . west of ngc 1275 , a chain of galaxies is distributed toward another radio galaxy ic 310 , while other member galaxies and clusters virtually aligned with this chain form the pisces - perseus supercluster . on the other hand , the icm x - ray emission has been traced up to at least 2 mpc with an elongation in the east - west direction . based on the x - ray morphology of early observations , @xcite hypothesized that the gas and galaxies are rotating with a velocity of close to 3000 km s@xmath1 . @xcite found asymmetrical icm temperature distribution and suggested a past merger in the direction parallel to the line of sight . using _ xmm - newton _ @xcite further studied the asymmetric structures around the core ( @xmath12 ) and revealed a hot `` horseshoe '' surrounding the cool core . they also assumed a minor merger along the east - west direction in the sky plane . @xcite studied the gas distribution over wider spatial scale ranges and suggested mergers and associated swirling gas motions . the cluster center has been observed twice - yearly as a calibration target of the xis and we used all available data . the integrated deep exposure of this unique object along with the high sensitivity of the xis provides one of the best quality x - ray spectra from clusters . we also used some offset region ( @xmath13 ) data and derived larger scale icm emission properties . parts of the _ suzaku _ perseus data set were used for a number of measurements . @xcite reported the first detection of cr and mn x - ray lines from clusters and icm elemental abundance measurements , while @xcite and @xcite reported abundance measurements out to the cluster outskirt . using offset xis data along with the hxd data , @xcite reported an upper temperature limit in the outer region . @xcite measured large - scale gas thermal distribution over the viral radius . in the next section , we describe observations and data reduction , while in section [ cal ] we evaluate the accuracy of the xis energy response , which is crucial to calibrate our measurement . in sections [ sect : ana - center ] and [ ana - large ] we measure gas bulk velocities within the central pointing and over a larger scale respectively , for the first time in this system , which enables one of the most accurate measurements of gas bulk motion in relaxed clusters . finally these results are summarized and discussed in the last section . throughout this paper , we assume the flowing cosmological parameters : @xmath14 km s@xmath1mpc@xmath1 , @xmath15 , and @xmath16 . at the cluster redshift of 0.0183 , one arcmin corresponds to 22.2 kpc . we use the 68% ( @xmath17 ) confidence level for errors , unless otherwise stated . the perseus cluster center is a calibration target and the central galaxy , ngc 1275 , was always located at the ccd center . we call these observations _ central _ pointings . here we use the data obtained from 2006 to 2013 taken in the normal window mode and table [ obs : obs - center ] shows an observation log . the total exposure time is 576 ks . the xis 2 sensor was only available until november 2006 . these data were used in the instrument calibration reported by @xcite , @xcite , and @xcite . the off - center regions were also observed many times . we call these _ offset _ pointings and use the data given in table . [ obs - offset ] . the total exposure time is 327 ks . [ obs : rosat2 ] shows the field of views of the observations . detailed descriptions of the _ suzaku _ observatory , xis instrument , and x - ray telescope are found in @xcite , @xcite , and @xcite , respectively . the xis field of view is a square of @xmath18 with @xmath19 pixels . accordingly , 1 pixel @xmath20 1.04 arcsec . the energy resolution is about 50 ev at 1 kev and 130 ev at 6 kev in fwhm . initially the xis are operated normally in the spaced - row charge injection ( sci ) off mode . since 2006 october , the sci on has been used in normal mode . we started the analysis from archived cleaned event files , which were filtered with standard selection criteria , and used the latest calibration file as of 20012 november 6 . we examined light curves in the 0.32.0 kev band excluding the central bright region events ( @xmath21 ) , for stable - background periods . none of the data showed any flaring event . the xis energy data have an original energy channel width of 3.65 ev . we grouped these by common bin sizes of about one - third of the fwhm of the energy resolution . around the fe - k line , the width is 8 channels or about 30 ev . to establish the xis energy response function , we used xisrmfgen software alongside makepi ( version-20121009 ) to assign pi values . the energy bin size is 1 ev ( @xmath22 of the fe - k line energy ) , which means we can resolve the change of energy at less than the systematic limit of about @xmath23 . for spectra fitting , we use xspec ( version-12.8 ; * ? ? ? * ) and maximum likelihood ( @xmath24 ) statistics implemented in the same . note that the above binning gives about 15 counts per bin in the lowest - count spectrum and more than 20 counts per bin in other spectra analyzed below . llllcc cen-0602 & 2006 feb & 800010010 & off & 43.7 & 260.2 + cen-0608 & 2006 aug & 101012010 & off / on & 92.0 & 66.0 + cen-0702 & 2007 feb & 101012020 & on & 40.0 & 258.7 + cen-0708 & 2007 aug & 102011010 & on & 35.1 & 83.4 + cen-0802 & 2008 feb & 102012010 & on & 34.9 & 255.2 + cen-0808 & 2008 aug & 103004010 & on & 34.1 & 86.8 + cen-0902 & 2009 feb & 103004020 & on & 46.3 & 256.1 + cen-0908 & 2009 aug & 104018010 & on & 34.2 & 67.0 + cen-1002 & 2010 feb & 104019010 & on & 33.6 & 277.3 + cen-1008 & 2010 aug & 105009010 & on & 29.6 & 66.6 + cen-1102 & 2011 feb & 105009020 & on & 32.9 & 259.7 + cen-1108 & 2011 aug & 106005010 & on & 34.1 & 83.8 + cen-1202 & 2012 feb & 106005020 & on & 41.1 & 262.0 + cen-1208 & 2012 aug & 107005010 & on & 33.2 & 72.6 + cen-1302 & 2013 feb & 107005020 & on & 41.2 & 256.3 + lllllrrr a & 2006 sep & 801049010 & off & 25 & 36 & 345 + b & 2006 sep & 801049020 & off & 27 & 31 & 260 + c & 2006 sep & 801049030 & off & 31 & 28 & 140 + d & 2006 sep & 801049040 & off & 8 & 33 & 56 + e & 2011 feb & 805045010 & on & 27 & 18 & 180 + f & 2011 feb & 805046010 & on & 18 & 16 & 0 + g & 2011 feb & 805047010 & on & 17 & 14 & 110 + h & 2011 feb & 805048010 & on & 15 & 17 & 58 + e1 & 2009 jul & 804056010 & on & 7 & 16 & 90 + n1 & 2009 aug & 804063010 & on & 14 & 15 & 330 + s1 & 2010 aug & 805096010 & on & 8 & 16 & 200 + w1 & 2010 aug & 805103010 & on & 6 & 16 & 280 + se1 & 2011 aug & 806099010 & on & 11 & 15 & 140 + sw1 & 2011 aug & 806106010 & on & 12 & 14 & 225 + nne1 & 2011 aug & 806113010 & on & 9 & 16 & 10 + f_2 & 2013 feb & 807019010 & on & 14 & 16 & 0 + s1 _ 2 & 2013 feb & 807020010 & on & 23 & 16 & 200 + h_2 & 2013 feb & 807021010 & on & 18 & 19 & 58 + w1_2 & 2013 feb & 807022010 & on & 23 & 16 & 280 + g_2 & 2013 feb & 807023010 & on & 14 & 19 & 110 + the main goal of this paper is to measure doppler shifts of cluster k - shell iron lines in x - rays . these astronomical shifts should be separated from instrumental effects on the energy scale ( i.e. recorded pulse - height vs. energy ) . the instrumental effects include ( i ) ccd - to - ccd variation , ( ii ) intra - chip _ spatial _ variations attributable to charge - transfer inefficiency ( cti ) and by ccd segment - to - segment qualities , ( iii ) observation - to - observation _ temporal _ variation , ( iv ) absolute energy determination , and unknown systematics . these instrumental characteristics have been calibrated and extensively evaluated by the instrument team , which is an advantage of the _ suzaku _ xis over other ccd - type spectrometers . these results have also been demonstrated ; not only in the cluster redshift analysis stated in section 1 but also in x - ray spectroscopy in the galactic center and ridge ( e.g. * ? ? ? * ) and in supernova remnants ( e.g. * ? ? ? we summarize the latest calibration status in subsection [ sect : cal - status ] . using the latest calibration and the perseus data , we further evaluate accuracy in subsection [ sect : cal - cal ] . among the above instrumental variations , factor ( i ) can be checked by comparing observed data from different ccds in any observations . the built - in calibration source lines are used to evaluate factors ( iv ) for each observation . when analyzing the perseus central data ( section [ sect : ana - center ] ) , factor ( ii ) is crucial and ( iii ) can be internally checked against the data . for the offset data analysis ( section [ ana - large ] ) , ( iii ) is important . moreover , ( ii ) also affects the result because each pointing observation includes different distributions of events over the detector position . @xcite estimated the systematic uncertainty of absolute energy in the fe - k band to be within @xmath25 and @xmath26% , based on fe lines observed from the galactic center alongside mn k@xmath27 and k@xmath28 lines ( at 5895 and 6490 ev respectively ) from the built - in calibration source ( @xmath29fe ) . independently , ota07 investigated the xis data and evaluated the energy - scale calibration in detail . using early calibration and early perseus cluster data ( sequence 800010010 ) , they estimated the systematic error of the _ spatial _ gain non - uniformity ( ii ) to be @xmath30% ( 68% confidence level ) . a similar analysis was performed by @xcite using two perseus cluster datasets with a new calibration and the accuracy was confirmed . furthermore , @xcite systematically examined the xis data obtained from the start of the operation in july 2005 until december 2006 ( sci off mode data ) . they reported that the _ spatial _ dependence of the energy scale ( ii ) was well corrected for charge - transfer inefficiency and cited a time - averaged uncertainty of absolute energy of @xmath31 0.1% . @xcite reported the energy scale calibration for sci - on mode . @xcite reported an improved cti correction . they archived energy accuracies of @xmath32% and @xmath33% at 1 kev and 7 kev , respectively , over the entire ccd position in the sci - on mode data . we based on these calibration methods in the latest calibration files . the energy resolution changed over time and was measured at 140 ev ( july 2005 ; sci off ) , 200 ev ( april 2007 ; sci off ) , and 160 ev ( march 2008 ; sci on ) for the fi ccd at mn k@xmath34 in fwhm . this gradual change in energy resolution was also calibrated ; the typical uncertainty of resolutions is 1020 ev in fwhm . to evaluate the absolute energy scale in different ccd segments ( a and d in xis-0,1,3 ) we use the perseus central pointing data ( table [ obs : obs - center ] ) and extracted spectra of calibration sources which illuminate two corners of each ccd . these spectra within the energy range 5.37.0 kev are fitted with two gaussian lines ( zero intrinsic width ) for the mn k@xmath35 and k@xmath36 along with a bremsstrahlung continuum component . we parametrize the energy scale using a common `` redshift '' deviated from the expected value of 5895 ev ( mn k@xmath35 ) and 6490 ev ( mn k@xmath35 ) . spectra from different observations , ccds , and segments were fitted separately . as shown in table [ cal : cal - var ] , the average redshifts are all below @xmath23 , which indicates that the absolute energy scale is calibrated accurately if averaged over observations . the variations within each segment are comparable with statistical errors , which suggests that systematic errors in instrumental variations are below statistical ones . following ota07 and based on the obtained distribution of redshifts we measure residual variation by calculating @xmath37 such that @xmath38 equals the degree of freedom . here @xmath39 , and @xmath40 are the obtained redshift , its average , and statistical errors , respectively . we found @xmath37 to be @xmath41 . this variation includes ccd - to - ccd ( i ) , segment - to - segment ( ii ) , and temporal ( iii ) factors . the calibration source is attached to the opposite side of read - out nodes of each ccd ; hence the above calibration source accuracy energy indicates the precision of the cti calibration . averaging the four segments of the fi - ccds ( table [ cal : cal - var ] ) , we obtain a redshift close to @xmath42 , while bi redshift is close to @xmath43 . this relatively significant bi spectral deviation , which is still within statistical errors , could introduce systematic error . because of better sensitivity of the fi at higher energy bands , fi is calibrated better than bi at these energy bands . moreover the fi spectra also provide better energy resolution . in view of these properties , we use only the fi data in the following analysis . to evaluate energy resolution calibration , we fitted the mn lines by adding an artificial line width ( @xmath44 ) and found that the best - fit @xmath44 tended to be zero . the 1 @xmath45 upper limit of @xmath44 ranged from 5 to 80 ev depending on the line counts with an average of 25 ev . these confirm the reported energy resolution calibration . we assume this 25 ev @xmath46 km s@xmath1 to be a 1@xmath45 systematic error in line width at around the fe - k energy . lcccc xis0/a & 0.23 & 1.24 & 2.7 & 1.0 + xis0/d & 0.31 & 1.37 & 2.6 & 0.7 + xis3/a & -0.11 & 0.63 & 1.6 & 1.1 + xis3/d & -0.72 & 1.46 & 4.2 & 1.0 + xis1/a & -0.65 & 1.28 & 3.5 & 0.9 + xis1/d & -0.95 & 1.43 & 2.8 & 1.4 + the spatial variation in the xis energy scale [ ( ii ) in subsection [ cal - mot ] ] was calibrated using perseus cluster data as one of the primal sources . here we use the same source but all available data ( table [ obs : obs - center ] ) alongside the latest calibration information and evaluate the accuracy . we should note that the perseus data themselves are used to correct the cti , assuming the cluster has uniform fe - line energy over the central region ( @xmath47 ) . possible deviations from this assumption could add systematic errors . we assume these systematic errors to be insignificant on the followings grounds . firstly , the cti parameters are determined using perseus data observed in different roll - angles , meaning the cti correction is not solely dependent on cluster - intrinsic distribution of fe - line energy . secondly , the cti correction is not tuned for each observation , but based on instrumental characteristics as a function of the number of ccd read - out transfer ( act - y coordinates ) , ccd segments , observation time , and x - ray energy . thirdly , built - in calibration source data along with other astronomical sources such as the galactic center and supernova remnants ( at lower energies ) were used to correct and evaluate the calibration . we check the assumption more quantitatively in this subsection . here we group observations by their roll - angles . one group ( `` winter '' ) data were obtained in february each year with roll - angles of 255277 . the other ( `` summer '' ) data were obtained in august with roll - angles of 6686 . note that pointing centers are close to each other within about 30 arcsec in all data . within each group , different observations share roll angles close to each other , meaning a certain detector position allows a particular sky region to be observed . using these data sets with two different extraction methods , we checked both _ instrumental _ and _ cluster-_intrinsic variations in the energy scale . in the first extraction ( subsection [ cel - per : det ] ) we integrated spectra from different sky regions sorted by detector positions and found no significant variation in the line centers measured among different detector positions nor between the two - roll groups . in the second extraction , given in section [ sect : ana - center ] , we integrated spectra from different detector positions sorted by sky regions . we found no significant variation among different sky regions nor between the two - roll groups . these analyses are not completely independent and we can not completely separate _ instrumental _ and _ cluster _ variations . nevertheless , both indicate that both _ instrumental _ and _ cluster _ variations are small and within certain error level . we divided the xis field of view into @xmath48 cells of size @xmath49 in detector coordinates ( detx and dety ) as illustrated in fig . [ cal : detector ] . we call these cells det00 , det01 and so on up to det33 . based on each cell and observation , a spectrum is extracted , whereupon each spectrum in the 5.77.3 kev band is fitted with two gaussian lines for he - like k@xmath35 ( @xmath50 ev ) and h - like k@xmath35 ( @xmath51 ev ) and a bremsstrahlung continuum model . as stated in subsection [ ana - center - lw ] , since we found no need for additional line width for gaussian lines , we fixed the width at zero henceforth . we used the common redshift of the two lines as a fitting parameter . examples of the spectral fitting are given in appendix ( fig . [ cal - app : sp ] ) . we found that the instrumental non - x - ray background is well - below the source count and subtracting the estimated background does not affect the obtained redshift , so we decided not to subtract the background from the central pointing data henceforth . we confirmed that the two fi ccd ( xis-0 and -3 ) spectra always give consistent line energies and this validates that ccd - to - ccd variation ( i ) is well calibrated , at least within the fi ccds . we therefore combine these two ccd data henceforth . we fitted a set of spectra from individual cells ( det@xmath52 ) from different observations within each roll - angle group using a common redshift . we also fitted all observations using a common redshift . the obtained redshift variations are given in fig . [ cal - per : reg - z ] and table [ cal - per : det - var ] . there are two special cells , det03 and det33 . because these include calibration source areas , they are much smaller in area ( see fig . [ cal : detector ] ) and hence lower in counts than others . these areas are not used here and in the perseus spectral extraction given below . in particular , the count of det33 are lower than 10% of those in others . det33 also shows the largest deviation from average , which could be attributable to instrumental systematic effects but also statistically . det03 shows larger redshifts in both the two groups . in this case , the deviation is more likely caused by an instrumental effect . these deviations in det03 and det33 are as small as @xmath53 . in addition , the sum of these two positions occupies only a small percentage of the whole ccd area , which means any systematic errors have little effect on our measurements . we found that standard deviations , @xmath54 , among the detector positions for each roll - group and for all observations are @xmath55 in redshift ( table [ cal - per : det - var ] ) . if we exclude the det33 result , these are reduced to @xmath56 . these variations are attributable to the combination of _ instrumental _ and _ cluster _ variations . in the det23 cell , two - roll - group spectra reveal the largest difference , @xmath57 , in redshifts , but elsewhere , they give consistent redshifts . noting that different sky regions are observed by both groups in the same detector position , this indicates that _ cluster-_intrinsic variation is below these differences . further analysis of smaller spatial scales is limited by statistics , but we assume that the instrumental energy scale does not change on such scales . based on these analysis , we conclude that the ccd - to - ccd variation [ ( i ) in subsection [ cal - mot ] ] and intra - chip _ spatial _ variation ( ii ) within spatial scales of a few arcminutes is below 0.1% ( @xmath23 in redshift or 300 km s@xmath1 around the fe - k energy ) over the whole ccd position . llccc + winter & 18.2 & 1.6 & 5.9 & 0.39 + summer & 18.1 & 0.78 & 1.7 & 0.33 + all & 18.2 & 1.0 & 3.4 & 0.23 + + winter & 18.5 & 0.66 & 0.91 & 0.28 + summer & 18.2 & 0.59 & 0.41 & 0.25 + all & 18.4 & 0.56 & 0.60 & 0.18 + here we use a set of central pointing data of the perseus cluster and investigate the spatial distribution of spectral properties within the central region ( @xmath58 ) . we divided the central sky region into cells of sizes @xmath59 or @xmath60 , as illustrated in fig . [ ana - center : regions ] . we call inner and smaller cells cs@xmath61 and larger cells cl@xmath61 , with @xmath61 ranging from 0 to 15 , starting from the south - west corner toward the north - east . note that the sum of cl5,cl6,cl9,and cl10 overlap with cs0cs15 . from each cell and each observation one spectrum is extracted . before measuring redshifts , we evaluate spectra by fitting them in the 2.09.0 kev energy band with a single temperature model and the galactic absorption ( hydrogen column density of @xmath62@xmath63 ) and determine the temperature and metal abundance of each region . we found a two - dimensional temperature variations of 4.56.5 kev . conversely , the metallicity increased radially from 0.3 to 0.45 solar toward the center . these results match previous measurements and confirm that these spectra can be used to measure local properties from different sky regions . we measure redshifts based on the fe - k line center and the same model ( two gaussians and a bremsstrahlung continuum ) as in section [ cal - per ] . [ ana - center - z : ex - sp ] shows fitting examples . the redshift variation as a function of observation and hence observation date is shown in fig . [ ana - center - z : seq_z ] . spectra from the first observation ( sequence 800010010 ) show systematically smaller redshift . other than this , however , there is no systematic change over the observation period , confirming that the _ temporal _ variation in the energy scale is effectively corrected . to compensate possible observation - dependence of the energy scale , we measure the redshift relative to that of a bright cell ( cs5 or cl5 ) for each observation and each sky cell , @xmath64 . fig.[ana - center - z : reg_z ] shows @xmath64 averaged over observations . here we measure the error ( @xmath65 ) for each cell based on standard deviation among all observations . these observations differ in terms of the detector orientation ( i.e. roll - angle ) and observation - date . therefore the scope of error includes not only statistical errors but also at least partially systematic errors due to instrumental energy - scale variations . along with average values from all observations , those from summer and winter roll - angle observations are given separately . differences between the two - roll - angle observations enable systematic errors from instrumental effects to be estimated . in all cases , however , the differences are below @xmath66 , confirming that the _ instrumental _ variation is insignificant as described in section [ cal - per ] . the largest @xmath67 is @xmath68 at cl0 , where @xmath65 is about @xmath69 . all other regions have @xmath67 below @xmath70 . averages and standard deviations of the best - fit @xmath64 along with averages of @xmath65 are given in table [ ana - center : rms ] . these results show uniform @xmath64 distribution and hence no gas motion exceeding these variations and errors . the errors for each sky region range from @xmath71 to @xmath72 , corresponding to radial velocities of 150 and 600 km s@xmath1 respectively . within the inner cells , cs1cs3 show lower @xmath64 in the best - fit value . to further search for possible spatial variation in redshifts , we use data differently from the above case . here we fit all spectra for each cell simultaneously with a common redshift , assuming no instrumental and systematic variation in the energy scale . we remove one observation ( sequence 800010010 ) to avoid systematic error as mentioned above . we sorted the result as a function of the position angle of the cell with respect to the cluster center , as shown in fig . [ ana - center - z : pa_z ] . the obtained redshift are basically consistent with the above result in fig . [ ana - center - z : reg_z ] . , although the given error declines because these include only statistical errors . from the inner cells ( left panel fig . [ ana - center - z : pa_z ] ) we found lower redshifts in easterly and westerly directions and higher redshifts toward the south . from the larger cells ( left panel fig . [ ana - center - z : pa_z ] ) we found some variations but more random and comparable to the possible systematic error ( section [ cal - per ] ; @xmath73 in redshift or 300 km s@xmath1 ) . we can not find all variations if we use the data in summer or winter observations separately ( see fig . [ ana - center - z : reg_z ] ) . based on these careful analyses , we suggest a hint of lower redshift regions at @xmath74 ( 4590 kpc ; cs1cs3 ) west of the cluster center . other variations are insignificant and possibly attributable to instrumental effects . in a search for redshift variation on a finer scale , we divided the central @xmath75 region into 9 cells of sized @xmath76 . we found a uniform redshift distribution within about @xmath77 in redshift . lccc cs0cs15 & -0.1 & 0.35 & 0.92 + cl0cl15 & 0.12 & 0.52 & 1.3 + the observed spectra are used to constrain the turbulent doppler broadening of the emission line . following ota07 and similar to our analysis stated in subsection [ sect : cal - cal ] , we model the fe - line spectra by including an additional line width ( @xmath44 ; a gaussian width ) . here we note that the cluster - intrinsic lines are not a single gaussian line , but integrate many lines . for example the emission around 6.7 kev includes not only a dominant he - like resonance line at 6700 ev but also intercombination lines at 6667 - 6682 ev and a forbidden line at 6636 ev along with other weak lines between the resonance and forbidden lines . effective widths for this he - like triplet are about 30 ev for the temperature of 24 kev gas ( ota07 ) . in addition , gas bulk motions unresolved spatially within the spectral extraction region and projected in the line of sight contribute to @xmath44 . for each observation and region , we obtained @xmath44 and its upper limit . we found that in most cases @xmath44 was consistent with zero and that even using a non - zero @xmath44 did not change the best - fit redshift . these are averaged over different observations and presented in fig . [ ana - center - z : lw ] . the best - fit @xmath44 are all below 20 ev , except for that from cl15 where the statistics are significantly lower than others . given the systematic error and intrinsic cluster line width , these @xmath44 are all consistent with no additional broadening due to turbulent motion . statistical upper limits of @xmath44 are 2050 ev , which equates to the upper limits for turbulent motion of 9002200 km s@xmath1 . we may add systematic uncertainty of 1250 km s@xmath1(subsection [ sect : cal - cal ] ) . here we use a set of offset pointing data with offset angles of @xmath78 from the cluster center ( table [ obs - offset ] and fig . [ obs : rosat2 ] ) and study the spatial redshift distribution on a large scale . we extract a spectrum from the entire ccd field of view for each offset observation . here we use the same gaussian model as in subsection [ ana - center - z ] to fit the fi 5.77.0 kev band spectra and measure the redshift distribution . there are some sets of two observations ( x and x_2 ) sharing a pointing direction but with different roll - angles . we found that these data provide consistent results within the set , hence in these pointing sets , spectra are fitted simultaneously with a common model . examples fittings are shown in fig . [ ana - large - z : ex ] . the obtained redshifts are shown in fig . [ ana - large - z : z1 ] . we notice that g and se regions show larger and smaller redshifts , respectively , compared with the average . these two regions , however , are in proximity each other and overlap in the sky regions by about half the area of the entire field of view . therefore , these deviations are likely statistical or systematic errors , and do not originate entirely from the cluster gas redshift variation . when we combine these two regions , a redshift of @xmath79 consistent with other regions , is obtained , as shown in fig . [ ana - large - z : z1 ] ( a star mark ) . region a also shows a relatively large deviation alongside significant statistical errors . including these deviated regions , the error - weighted average , standard deviation , and average statistical error are found to be @xmath80 , @xmath81 , and @xmath82 , respectively . based on the above results , we conclude that gas redshifts are uniform within @xmath83 over the cluster core region ( @xmath84 ) in a spatial scale of about @xmath85 ( 300 kpc ) . in other words , we found no systematic velocity structure exceeding @xmath86 km s@xmath1 . the average and standard deviation of redshifts in the central region given in subsection [ ana - center - z ] ) are @xmath87 and @xmath88 , respectively . the redshift averaged over the offset regions is consistent with that of the central region within these variations and errors . kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] kev@xmath1 are shown . in the lower panels fit residuals in terms of the data to model ratio are shown . , title="fig : " ] ) . ] to constrain line broadening and turbulent velocity , we performed a similar spectral fitting as in subsection [ ana - center - lw ] . however , lower statistics mean the limits obtained are typically weaker than those in the central regions above . we found the obtained @xmath44 to be consistent with zero turbulent velocity and statistical upper limits of @xmath44 of 2070 ev , corresponding to velocities of 9003000 km s@xmath1 . the photon mixing caused by the _ suzaku _ telescopes and projection effects meant the measured velocity variation was diluted compared with the intrinsic three dimensional variation . the point spread function of the telescopes include half - power diameters of @xmath89 and 90% fraction encircled - energy radii of @xmath90 , meaning variations below these scales would be smoothed . it is also difficult to separate the gas velocity of a faint emission component . this effect was examined by a ray - tracing simulation in ota07 for the _ suzaku _ measurements in the centaurus cluster . they found that when @xmath91 cells are used at the core , about 50 - 60% of the events collected by each cell originate from the surrounding eight cells . because the perseus cluster has a central surface brightness distribution similar to that of centaurus , our measurement in the central region with @xmath92 width cells ( cs0 to cs15 ) is similarly affected . for the outer cells with a @xmath93 width ( cl0 to cl15 ) , mixing effects should be much smaller ( @xmath11 % ) . in the large - scale measurements ( section [ ana - large ] ) , photon mixing due to stray x - rays from the bright center could be an additional contamination . we estimate , however , that photons scattered from the central region to offset regions with a typical offset angle of @xmath94 represent a minute percentage of all photons within each offset pointing and we could hence ignore this mixing . @xcite claimed a radial velocity difference in the perseus cluster using the asca gis . they compared spectra collected at the center and offsets separated by more than @xmath95 . these spatial scales exceed our study within radii of @xmath96 , which means these two measurements can not be compared . we summarize the limits we determined for gas motion along with related parameters in table [ dis - com : summary ] . for comparison , results of ota07 are also presented . both results of the _ suzaku _ xis are limited by similar levels of systematic errors from the energy scale calibration . the levels of statistical errors between the two are also comparable . as described in section 1 , gas velocity measurements are important to calibrate the hydrostatic equilibrium ( h.e . ) . ota07 converted velocity variation obtained into rotational motion and approximated the uncertainty of the h.e . mass by gas kinetic to thermal energy ratio . the ratio is proportional to @xmath97 , where @xmath98 , @xmath99 and @xmath100 are the fraction , velocity , and mach number of the moving gas and @xmath101 is the gas sound velocity ( given in table [ dis - com : summary ] ) . the uncertainty of the rotation angle introduces another error . our limits on gas bulk velocities and hence on @xmath100 are comparable to those in ota07 . the wider physical size covered by our observations ( row r in table [ dis - com : summary ] ) allowed us to constrain the gas motion and hence the total mass within a volume significantly exceeding that in ota07 . this is an important step to measure total masses in cluster scales . provided the fe - k line emission of the perseus central region is exclusively brighter than other nearby clusters , this system could represent a unique measurement of the gas motion at the cluster scale via current ccd - type instruments . the decreasing x - ray surface brightness hinders efforts to constrain the motion beyond our achieved scale . it will also remain challenging to reach out beyond cluster core regions even with near - future high energy resolution instruments such as sxs @xcite onboard astro - h @xcite without a large collecting area . llccc redshift & & 0.0183 & & 0.0104 + @xmath102 & ( kpc ) & 22.2 & & 15.3 + @xmath103 & ( kev ) & 4.5 & 6 & 3 + @xmath101 & ( km s@xmath1 ) & 1100 & 1300 & 900 + region size & & @xmath104 & @xmath105 & @xmath106 + & ( kpc ) & 50/100 & 400 & 30 + r & ( kpc ) & 200 & 600 & 150 + @xmath107 & ( km s@xmath1 ) & 300400 & 550 & 700 + & & @xmath108 ( sys . ) & @xmath31 300 ( sys . ) & @xmath31500 ( sys . ) + @xmath109 & ( km s@xmath1 ) & 9002200 & 9003000 & 900 + & ( km s@xmath1 ) & @xmath31 1250 ( sys . ) & @xmath31 1250 ( sys . ) & @xmath31 1250 ( sys . ) + in and around ngc 1275 , complex interactions occur among various gaseous components , high - energy particles from the agn and radiation from stars and agn , and probably magnetic fields . nearby galaxies as well as the central galaxy may have moved around within the growing gravitational potential . in this active environment , the icm should have various types of motions . observationally , the lack of resonant scattering of the fe - k line at the perseus core indicates gas motions of @xmath110 @xcite . it is crucial to measure the energies in these gas motions in our study . we found a hint of gas bulk motions within the cluster core ( fig . [ ana - center - z : pa_z ] ; @xmath111 ) , while the key detection was lower velocities in regions @xmath74 west of center . west of center , clear x - ray enhancement was found as seen in fig . [ ana - center : regions ] . @xcite interpreted this structure as a result of a past minor merger in an east - west direction . the obtained gas motion is consistent with this minor merger model and additionally suggests a velocity component in the line of sight . because we observe a mix of photons from the sub component and main cluster due to the sky projection and the telescope point spread function , the real gas velocity of the sub exceeds the observed figure ( @xmath112 km s@xmath1 ) . if we assume @xmath98 to be the emission fraction from the sub , @xmath113 can be approximated with sub and main velocities ( @xmath114 and @xmath115 ) as @xmath116 . in this case , the relative velocity with respect to @xmath115 can be estimated as @xmath117 . from the relative deviation map of the surface brightness ( fig . [ ana - center : regions ] ) , we estimate @xmath98 to be 0.20.4 . in this case , @xmath118 could be 5001000 km s@xmath1 . interestingly , the second optically - bright member galaxy , ngc 1272 , sitting on a rim of the west sub structure ( @xmath119 wws of ngc 1275 ) , also has a lower radial velocity than that of ngc 1275 by @xmath120 km s@xmath1 . furthermore , the chain of galaxies further west has an average radial velocity lower than the cluster center by about 1500 km s@xmath1 . we suggest the association of the obtained gas velocity structure with these galaxies . we discuss these gas and galaxy relations on a larger scale in the next subsection . using the obtained measurements , we estimate kinetic energy ( @xmath121 ) by gas bulk motion . as stated in section [ dis - com ] , @xmath121 relative to the thermal energy ( @xmath122 ) in the gas can be presented by @xmath123 . using the relation between @xmath124 and @xmath125 given above , @xmath126 becomes @xmath127 . assuming @xmath101 of 1100 km s@xmath1 and @xmath98 of @xmath128 , the observed @xmath125 ( 150300 km s@xmath1 ) gives @xmath129 . @xcite estimated the energy associated with non - radial structure ( @xmath130 ) based on variation in the gas pressure map . this energy should be closely related to @xmath121 estimated here . they estimated @xmath131 to be within a few percent at radii of @xmath132 kpc corresponding to the west sub component region . because both estimations are order of magnitude , we can not compare these quantitatively . these estimations are among the first attempts to measure all the energy components in the cluster gas and understand the energy distribution of clusters . the west system is the largest substructure within the core . our measurement also indicates that it has the largest bulk velocity , at least in a radial direction , which means this structure could have the largest kinetic energy due to bulk motion . we found no other systematic motions around the core with scales @xmath133 kpc and bulk velocities exceeding half the sound velocity ( @xmath134 km s@xmath1 ) . this is consistent with the observation that any other features revealed around the core are below 10 kpc ( @xmath135 ) in size . these structures are also likely to have velocities not exceeding the virial velocity of the central galaxy , or about @xmath136 km s@xmath1 . the lack of heated gas or strong shock @xcite also pointed to no sonic or super sonic motion . we conclude that within much of the cluster core volume , the gas remains in hydrostatic equilibrium . in table [ dis - cd : z ] we present radial velocities of ngc 1275 and the cluster from optical data and this x - ray results . the optical radial velocity of ngc 1275 differs from that of a mean of cluster member by only about 200 km s@xmath1 , demonstrating that the galaxy remains at the bottom of the cluster potential . the absolute radial velocity of the gas at the core is consistent with that of the galaxy in optical within the errors from x - ray data . this indicates that large parts of stars and the gas in and around the galaxy stay together at the bottom of the cluster potential center . the cluster gas on a large scale also has a mean velocity consistent with that in optical , confirming that these two baryonic components beyond galaxy scale remain together at the same potential center . these comparisons in absolute redshifts are among the first attempts involving clusters . gas dynamics in cluster cool cores has been studied by numerical simulations . for example , @xcite simulated off - axis minor mergers and gas sloshing and reproduce cold fronts . they presented evolutions of gas velocity structures . compared with large velocity variations in their representative runs , our measured variations ( figs . [ cal - per : reg - z ] and [ ana - center - z : pa_z ] ) are more uniform spatially and close to their results for late stages ( e.g. 6/4.8 gyr in fig.5/10 of * ? ? ? note that our doppler measurements are not sensitive to motions in the plane of the sky and diluted by projection . in fact in the perseus cluster gas spiral flows in the plane of the sky are suggested from x - ray images ( e.g. * ? ? ? * ; * ? ? ? focusing on agn jet - driven bubbles , @xcite simulated high - resolution x - ray spectroscopy of nearby clusters including the perseus cluster . they found the velocity structure of the bubbles in this core with energy shifts of about 350 km s@xmath1 , which are close to our limit . however , these structures are too small in size ( @xmath11 arcsec ) and too faint in x - ray brightness contrast to resolve by the current data . lccl ngc 1275 & 5250 & 20 & 1 + member galaxy mean & 5470 & 100 & 2 + x - ray gas at center & 5520 & 300 & section [ sect : ana - center ] and [ ana - large ] + x - ray gas at @xmath137 & 5760 & 600 & section [ ana - large ] + we found an absence of gas bulk motion on a large scale . our upper limit for the radial gas velocity is about 600 km s@xmath1 , which implies either that gas motion in the cluster is predominantly in the plane of the sky or subsonic at most . in subsection [ dis - core ] we discussed the potential of the gas motion around the western core being associated with galaxy motions . the associated chain of galaxies is further distributed toward west . to study the gas and galaxies relation in the west and larger areas , we compare our derived gas dynamics with that in galaxies in the cluster . we use the following two catalogs as the deepest collections of galaxies in the perseus cluster . the cfa redshift catalog @xcite includes 94 ( 120 ) galaxies with radial velocities within @xmath96 ( @xmath138 ) of the cluster center , while the 2mass redshift survey @xcite includes 66 ( 103 ) galaxies in the same regions but with complete photometric data . [ dis - large : pa - vh ] shows the radial velocities of galaxies within a radius of @xmath96 of the center as a function of azimuthal angles relative to the center . most galaxies are distributed around the cluster dynamical center at redshift of @xmath139 , corresponding to that of ngc 1275 . this is consistent with early studies of galaxy distributions ( e.g * ? ? ? some galaxies have higher or lower velocities ( @xmath140 km s@xmath1 ) , which may be contamination dynamically unrelated to the main system . we include gas velocities from fig . [ ana - large - z : z1 ] into fig.[dis - large : pa - vh ] , which is the first such comparison . in galaxy velocities , there are some local structures , the clearest of which is a group at an angle of around 260 with lower velocities ( @xmath141 km s@xmath142 km s@xmath1 ) . this is part of the chain of galaxies mentioned in subsection [ dis - core ] , which should be before or after a collision into the main cluster centered on ngc 1275 . based on the x - ray distribution around the core , @xcite suggested that the substructure is after a collision about 0.25 gyr ago . based on comparison between the galaxy and gas velocities around this angle at the cluster west , we notice that the gas is closer to the cluster center than galaxies at this azimuth . note that these galaxies in total exceed the cd galaxy in terms of luminosity and hence mass . if these two baryonic components are associated , what explains this difference ? we note a difference in effective radii between the two wavelength bands . the x - ray emission is weighted toward the inner region , due not only to the positions of spectral extractions ( fig . [ obs : rosat2 ] ) but also the density - squared x - ray emission intensity . considering these effects , the effective radius of the x - ray data can be approximated around @xmath143 . on the other hand , galaxies in the west are uniformly distributed over the @xmath96 radius region . these galaxies are point sources in x - ray without any bright extended excess emission @xcite . in addition , this difference suggests a segregation between the two components during a violent collision . such detachments in the sky position have been observed in some merging systems such as the bullet @xcite . our measurement is a new approach to reveal such actions not only in spatial terms but also in the radial velocity space . note that in the major - merger system abell 2256 @xcite revealed that gas and galaxies move in pairs within radial velocity and spatial spaces . based on asca observations @xcite observed an extended cool region @xmath144 east of the cluster center , alongside a ring - like region surrounding the east cool region and the core . they suggested a past collision of a poor cluster in a direction nearly parallel to the line of sight . our observation covers a substantial fraction of the east cool region as shown in fig . [ obs : rosat2 ] . to limit the current velocity of the east sub system we must assume the x - ray emission fraction of the moving system within the extracted spectra . we estimate this at 0.30.5 according to the relative deviation map of surface brightness in @xcite . combining this with our measurement in fig . [ ana - large - z : z1 ] ( @xmath145 km s@xmath1 ) , we constrain the radial velocity of the sub system @xmath146 km s@xmath1 . in future we would cover more galaxies by conducting deeper surveys and deep x - ray spectra covering larger volumes in a number of systems . gravitational lensing would also measure dark matter distributions . direct comparisons between these three components in dynamical and spatial distributions should reveal a number of local events and measure the dynamical cluster age , thus giving a systematic picture of structure formation of gas and galaxies but invisibly controlled by dark matter . considering that the virial radius ( @xmath147 ) of the cluster is 2.2 mpc ( or @xmath20 100 ) , we measured the gas motion typically around @xmath148 . the obtained upper limits for the bulk motions are typically @xmath149 taking account of the systematic errors . the turbulent motions are more loosely constrained ( @xmath150 ) . cosmological cluster simulation results suggest that the kinematic energy of the gas is expected to be left only in the level of @xmath151 % of the thermal energy in the corresponding radius range ( e.g. * ? ? ? * ; * ? ? ? furthermore , this value tends to be even lower for a sub sample of relaxed clusters . therefore , our results of the absence of both the bulk and turbulent motions except for the some hints at west of the core are consistent with these simulations . ) from the 2mass redshift catalog . there are some overlaps between the two samples . the velocity of ngc 1275 is shown by a solid line . star marks indicate the gas velocity taken from fig . [ ana - large - z : z1 ] . the central velocity of the gas components is shown by a dashed line . ] in addition to the gas bulk motion , we searched for the turbulent motion using the line width . we obtained the limits generally larger than the sound speed as given in table [ dis - com : summary ] . ota07 also reported a limit of the turbulence using the xis data of the centaurus integrated over a large region of @xmath152 ( table [ dis - com : summary ] ) . these two limits are similar to each other , because that both limits depend largely on the energy resolution of the xis and its calibration . our result is the first attempt to constrain the turbulent motion based on the spatially - resolved line x - ray spectra . the gas turbulent ( random ) motion could introduce an error on the h.e . mass estimate . the additional mass or kinetic energy due to the turbulence can be approximated by the factor given in subsection [ dis - core ] , @xmath153 with the turbulent velocity , @xmath109 replaced as @xmath99 . we obtained the limit on @xmath109 a few times larger than that on the bulk velocity ( table [ dis - com : summary ] ) . therefore the uncertainty from the turbulent motion dominates that from the ordered one . the _ xmm - newton _ rgs was used to constrain the gas turbulent motions in hot gas in clusters and galaxies ( e.g. * ? ? ? @xcite analyzed rgs spectra of dozens of sources and found evidence for @xmath154 km s@xmath1 velocity line broadening and limits down to 300 km s@xmath1 for some systems . the rgs spectra can be used only for centrally - peaked x - ray line emission below 1.5 kev ( dominated by fe - l lines ) and results are coupled with the brightness distribution of the line emission which is not straightforward to model . as described in subsection [ ana - center - lw ] , the limit on @xmath109 is mostly limited by the instrumental energy resolution . therefore to improve the limit significantly high energy resolution instruments such as x - ray calorimeters are necessary . for example , direct and robust study will be obtained by sxs onboard astro - h , providing 57 ev ( fwhm ) energy resolution , corresponding to 300 km s@xmath1 at the fe - k line . a 100 ksec sxs observation of the perseus center provides more than 10,000 counts in the he - like triplet emission and measurements of the turbulent along with those of bulk velocity structure . we thank the referee for useful comments and suggestions . _ chandra _ image of the perseus cluster was provided kindly by j.sanders . we thank all the _ suzaku _ team member for their supports . we acknowledge the support by a grant - in - aid for scientific research from the mext , no.24540243 ( tt ) , no.25400231 ( no ) , and no.a2411900 ( su ) . arnaud , k. a. 1996 , astronomical data analysis software and systems v , 101 , 17 ascasibar , y. , & markevitch , m. 2006 , , 650 , 102 bhringer , h. , voges , w. , fabian , a. c. , edge , a. c. , & neumann , d. m. 1993 , , 264 , l25 clowe , d. , brada , m. , gonzalez , a. h. , markevitch , m. , randall , s. w. , jones , c. , & zaritsky , d. 2006 , , 648 , l109 churazov , e. , forman , w. , jones , c. , bhringer , h. 2003 , , 590 , 225 churazov , e. , forman , w. , jones , c. , sunyaev , r. , bhringer , h. 2004 , , 347 , 29 dupke , r. a. , & bregman , j. n. 2001a , , 547 , 705 evrard , a. e. , metzler , c. a. , & navarro , j. f. 1996 , , 469 , 494 fabian , a. c. , sanders , j. s. , allen , s. w. , et al . 2003 , , 344 , l43 fabian , a. c. , johnstone , r. m. , sanders , j. s. , et al . 2008 , , 454 , 968 fabian , a. c. , sanders , j. s. , allen , s. w. , et al . 2011 , , 418 , 2154 fujita , y. , et al . 2008 , , 60 , 1133 furusho , t. , yamasaki , n. y. , ohashi , t. , shibata , r. , & ezawa , h. 2001 , , 561 , l165 hayato , a. , yamaguchi , h. , tamagawa , t. , et al . 2010 , , 725 , 894 heinz , s. , brggen , m. , & morsony , b. 2010 , , 708 , 462 huchra , j. p. , geller , m. j. , clemens , c. m. , tokarz , . p. , michel , a. 1995 , the cfa redshift catalogue , version june 1995 , harvard - smithsonian center for astrophysics [ http://vizier.cfa.harvard.edu/viz-bin/cat?vii/193 ] huchra , j. p. , macri , l. m. , masters , k. l. , et al . 2012 , , 199 , 26 hirayama , y. , & ikeuchi , s. 1978 , progress of theoretical physics , 60 , 1337 kent , s. m. , & sargent , w. l. w. 1983 , , 88 , 697 koyama , k , et al . 2007 , , 59 , s23 lau , e. t. , kravtsov , a. v. , & nagai , d. 2009 , , 705 , 1129 markevitch , m. , & vikhlinin , a. 2007 , , 443 , 1 matsushita , k. , sakuma , e. , sasaki , t. , sato , k. , & simionescu , a. 2013 , , 764 , 147 mitsuda , k , et al . 2007 , , 59 , s1 mitsuda , k. , et al . 2010 , , 7732 , nagai , d. , vikhlinin , a. , & kravtsov , a. v. 2007 , , 655 , 98 werner , n. , urban , o. , simionescu , a. , & allen , s. w. 2013 , , 502 , 656 nishino , s. , fukazawa , y. , hayashi , k. , nakazawa , k. , & tanaka , t. 2010 , , 62 , 9 nishino , s. , fukazawa , y. , hayashi , k. 2012 , , 64 , 16 ota , n. , et al . 2007 , , 59 , 351 ( ota07 ) ozawa , m. , et al . 2009 , , 61 , 1 sanders , j. s. , & fabian , a. c. 2013 , , 429 , 2727 sato , k. , matsushita , k. , ishisaki , y. , et al . 2008 , , 60 , 333 sato , t. , matsushita , k. , ota , n. , et al . 2011 , , 63 , 991 sawada , m. , nakashima , s. , nobukawa , m. , uchiyama , h. , & xis team 2012 , american institute of physics conference series , 1427 , 245 serlemitsos , p. et al . 2007 , , 59 , s9 simionescu , a. , allen , s. w. , mantz , a. , et al . 2011 , science , 331 , 1576 simionescu , a. , werner , n. , urban , o. , et al . 2012 , , 757 , 182 sugawara , c. , takizawa , m. , & nakazawa , k. 2009 , , 61 , 1293 suto , d. , kawahara , h. , kitayama , t. , et al . 2013 , , 767 , 79 takahashi , t. , et al . 2010 , , 7732 , takizawa , m. , nagino , r. , & matsushita , k. 2010 , , 62 , 951 tamura , t. , et al . 2009 , , 705 , l62 ( t09 ) tamura , t. , hayashida , k. , ueda , s. , & nagai , m. 2011 , , 63 , 1009 uchiyama , h. et al . 2009 , , 61 , s9 vazza , f. , brunetti , g. , kritsuk , a. , et al . 2009 , , 504 , 33 xu , h. , kahn , s. m. , peterson , j. r. , et al . 2002 , , 579 , 600 . , title="fig : " ] . , title="fig : " ]

i investigate the use of pearson s chi - square statistic , the maximum likelihood ratio statistic for poisson distributions , and the chi - square - gamma statistic ( mighell 1999 , apj , 518 , 380 ) for the determination of the goodness - of - fit between theoretical models and low - count poisson - distributed data . i demonstrate that these statistics should not be used to determine the goodness - of - fit with data values of 10 or less . i modify the chi - square - gamma statistic for the purpose of improving its goodness - of - fit performance . i demonstrate that the modified chi - square - gamma statistic performs ( nearly ) like an ideal statistic for the determination of goodness - of - fit with low - count data . on average , for correct ( true ) models , the mean value of modified chi - square - gamma statistic is equal to the number of degrees of freedom and its variance is like the distribution for degrees of freedom . probabilities for modified chi - square - gamma goodness - of - fit values can be calculated with the incomplete gamma function . i give a practical demonstration showing how the modified chi - square - gamma statistic can be used in experimental astrophysics by analyzing simulated x - ray observations of a weak point source ( s / n ; 40 photons spread over 317 pixels ) on a noisy background ( 0.06 photons per pixel ) . accurate estimates ( 95% confidence intervals / limits ) of the location and intensity of the x - ray point source are determined . # 1#1 # 1 # 1

astro-ph0007328

the goodness - of - fit between an observation of @xmath5 data values , @xmath6 , with errors , @xmath7 , and a model , @xmath8 , can be determined by using the standard chi - square statistic : @xmath9 ^ 2\ , . \label{eq : chisq_std}\ ] ] the standard chi - square statistic is the appropriate statistic to determine the goodness - of - fit whenever the errors can be described by a normal ( a.k.a . gaussian ) distribution . let us consider a more complicated situation where all the data values come from a pure counting experiment where each measurement , @xmath10 , is a random integer deviate drawn from a poisson ( @xcite , p. 205 et seq . ) distribution , @xmath11 with a mean value of @xmath12 . the use of equation ( [ eq : chisq_std ] ) in analyzing poisson - distributed data is technically never correct . while the poisson distribution approaches the normal distribution as the poisson mean approaches infinity , a poisson distribution never actually becomes a normal distribution even at very large poisson mean values . the normal distribution is always symmetric ; the coefficient of skewness for the normal distribution is zero . poisson distributions are always asymmetric ; the coefficient of skewness for a poisson distribution of mean @xmath12 is @xmath13 . while the poisson distribution is almost symmetric about the mean for large mean values , its shape becomes progressively more asymmetric as the mean approaches zero . thus the standard assumption that a poisson distribution is approximately normally distributed is a good approximation only when the coefficient of skewness is negligible ( e.g. , @xmath14 ) . how does one then determine the goodness - of - fit with poisson - distributed data ? historically , many @xmath0 statistics have been proposed for the analysis of poisson - distributed data . this paper will investigate the following four : + pearson s @xmath0 : @xmath15 where the expectation value of the mean of the parent poisson distribution of the @xmath16th data value is assumed to be equal to the poisson deviate [ @xmath17 and the square of the measurement error is assumed to be equal to the mean of the model poisson distribution [ @xmath18 ; + the modified neyman s @xmath0 : @xmath19 where the expectation value of the mean of the parent poisson distribution of the @xmath16th data value is assumed to be equal to the poisson deviate [ @xmath17 and the square of the measurement error is assumed to be equal to poisson deviate or one whichever is greater [ @xmath20 ; + the maximum likelihood ratio statistic for poisson distributions : @xmath21 \label{eq : x2l}\ ] ] ( see , e.g. , baker & cousins @xcite and references therein ) ; + and the chi - square - gamma statistic ( mighell @xcite ; hereafter ): @xmath22 ^ 2 } { n_i+1 } \label{eq : x2 g } \ , \ ] ] where the expectation value of the mean of the parent poisson distribution of the @xmath16th data value is assumed to be equal to the poisson deviate plus a small correction factor of zero for zero deviates and one in all other cases [ @xmath23 and the square of the measurement error is assumed to be equal to the poisson deviate plus one [ @xmath24 . in , i demonstrated that the application of the standard weighted mean formula , @xmath25 / \left[\sum_i { \sigma^{-2}_i}\right ] $ ] , to determine the weighted mean of data , @xmath10 , drawn from a poisson distribution , will , on average , underestimate the true mean by @xmath26@xmath27 for all poisson mean values larger than @xmath26@xmath28 when the common assumption is made that the error of the @xmath16th observation is @xmath29 . this small , but statistically significant offset , explains the long - known observation that chi - square minimization techniques which use the modified neyman s @xmath0 statistic [ eq . ( [ eq : x2n ] ) ] to compare poisson - distributed data with model values , @xmath8 , will typically predict a total number of counts that underestimates the actual total by about @xmath27 count per bin ( see , e.g. , bevington @xcite , wheaton @xcite ) . based on my finding that the weighted mean of data drawn from a poisson distribution can be determined using the formula @xmath30\left(n_i+1\right)^{-1 } \right ] / \left [ \sum_i \left(n_i+1\right)^{-1 } \right ] $ ] , i proposed that the chi - square - gamma statistic , @xmath31 [ eq . ( [ eq : x2 g ] ) ] , should always be used to analyze poisson - distributed data in preference to the modified neyman s @xmath0 statistic . following my own advice , i will not discuss the modified neyman s @xmath0 statistic in the remainder of this article . the chi - square distribution for @xmath3 degrees of freedom approaches a gaussian distribution with a mean equal to @xmath3 ( i.e. , @xmath32 ) and a variance equal to @xmath2 ( i.e. , @xmath33 ) as the number of degrees of freedom approaches infinity . ideally , a @xmath0 statistic for poisson distributions for @xmath3 ( independent ) degrees of freedom would exhibit the same behavior as the number of degrees of freedom approaches infinity _ for all poisson mean values _ ( i.e. , @xmath34 ) . do the @xmath35 , @xmath36 , and the @xmath31 statistics perform as expected for large poisson mean values ? these three @xmath0 statistics are applied to the same data set in fig . [ fig : x2p_x2l_x2g_mu100 ] ( _ top to bottom _ , respectively ) . for this example , an ideal @xmath0 statistic for poisson - distributed data would have a cumulative distribution similar to that of the chi - square distribution for @xmath37 degrees of freedom which well approximated as the cumulative distribution function of a gaussian distribution with a mean of @xmath37 and a variance of @xmath38 . the results of the top and bottom panels are well matched to the expected cumulative distribution ; the differences between the expected and measured mean and rms values are not statistically significant . the @xmath35 and the @xmath31 statistics perform as expected with a poisson mean value of 100 . the cumulative distribution of the middle panel , however , clearly deviates from the expected cumulative distribution ; the difference between the expected and measured mean and rms values , while small , is statistically significant . the @xmath36 does not perform like an ideal @xmath0 statistic for poisson distributions with a mean value of 100 a level that is generally considered to be well above the low - count regime ( @xmath39 ) . let us continue the investigation of the performance of the maximum likelihood ratio statistic for poisson distributions with 1000 samples of @xmath37 poisson deviates with poisson mean values of 100 , 10 , 1 , 0.1 , and 0.001 . figure [ fig : x2l_5mu ] confirms that the @xmath36 statistic does not perform like an ideal @xmath0 statistic in the low - count regime . the average contribution by the @xmath16th deviate to an ideal @xmath0 statistic for the analysis of poisson - distributed data would be exactly one and the average contribution to its variance would be exactly two . figure [ fig : x2lr_var ] expands the previous analysis of of the @xmath36 statistic over a wide range of poisson mean values from 0.001 to 1000 . the dashed lines of fig . [ fig : x2lr_var ] show the results for an ideal @xmath0 statistic ; one can clearly see that the average contribution to @xmath36 is not equal to one and the average contribution to its variance is not equal to two for poisson mean values @xmath4010 . the poor performance of the @xmath36 statistic with low - count data may come as a surprise to many readers since it has historically been advocated as being one of the best @xmath0 statistics for the analysis of poisson - distributed data . chi - square statistics can serve ( at least ) two distinct purposes : ( 1 ) their functional forms can be utilized as the core of parameter estimation algorithms , and ( 2 ) their values can serve as a measure of the goodness - of - fit between a model and a data set.while the functional form of the @xmath36 statistic can be successfully utilized for the purpose of parameter estimation with poisson - distributed data in the low - count regime ( see , e.g. , ) , _ the maximum likelihood ratio statistic for poisson distributions [ eq . ( [ eq : x2l ] ) ] should not be used to determine the goodness - of - fit with low - count data where the poisson mean is @xmath4010 . _ in this work , i investigate the use of pearson s @xmath0 statistic and the chi - square - gamma statistic for the determination of the goodness - of - fit between theoretical models and data derived from counting experiments . i develop a methodology in 2 which modifies pearsons s chi - square statistic for the purpose of improving its goodness - of - fit performance . this methodology is then be applied to modify the chi - square - gamma statistic ( 3 ) . the modified chi - square - gamma statistic is shown to perform ( nearly ) like an ideal @xmath0 statistic for the determination of goodness - of - fit with low - count data . simulated x - ray images are analyzed in 4 as a practical demonstration of the possible use of the modified chi - square - gamma statistic in experimental astrophysics . the summary of the paper is presented in 5 . let us continue the investigation of the performance of pearson s @xmath0 statistic with 1000 samples of @xmath37 poisson deviates with poisson mean values of 100 , 10 , 1 , 0.1 , and 0.001 . figure [ fig : x2p_5mu ] shows that the @xmath35 statistic does not perform like an ideal @xmath0 statistic for poisson mean values @xmath41 . the average contribution by the @xmath16th deviate to an ideal @xmath0 statistic for the analysis of poisson - distributed data would be exactly one and the average contribution to its variance would be exactly two . figure [ fig : x2pr_var ] expands the previous analysis of of pearson s @xmath0 statistic over a wide range of poisson mean values from 0.001 to 1000 . the dashed lines of fig . [ fig : x2pr_var ] show the results for an ideal @xmath0 statistic ; one can see that while the average contribution to @xmath35 is one , the average contribution to its variance is not equal to two for poisson mean values @xmath41 . the variance of pearson s @xmath0 statistic is , by definition , @xmath42 ^ 2\,,\ ] ] where @xmath43 is the number of independent degrees of freedom , n is the number of data values , and m is the number of free parameters . the variance of the _ reduced _ chi - square of a @xmath0 statistic for a large number of observations should ideally be two . in the limit of a large number of observations of a _ single poisson distribution with a mean value of @xmath12 _ , the variance of the reduced chi - square of the pearson s @xmath0 statistic is @xmath44 \nonumber \\ & \equiv & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - m_i)^2 } { m_i } - \left [ \frac{1}{\nu}~{\sum_{j=1}^{n } } \frac { ( n_j - m_j)^2 } { m_j } \right ] \right\}^2 \right ] \nonumber \\ & \equiv & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - m_i)^2 } { m_i } - \left [ \frac{\chi^2_{\rm{p}}}{\nu } \right ] \right\}^2 \right ] \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - m_i)^2 } { m_i } - \lim_{n \rightarrow \infty } \left [ \frac{\chi^2_{\rm{p}}}{\nu } \right ] \right\}^2 \right ] \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n - m}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - m_i)^2 } { m_i } - 1 \right\}^2 \right ] \hspace*{22truemm } \mbox{[see eq.\ ( 25 ) of \paperi ] } \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - \mu_{\rm{p}})^2 } { \mu_{\rm{p } } } - 1 \right\}^2 \right ] \hspace*{25truemm } \mbox{[see eq.\ ( 5 ) of \paperi ] } \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { \left(n_i - { \displaystyle\lim_{n \rightarrow \infty } } [ \mu_{\rm{p}}]\right)^2 } { { \displaystyle\lim_{n \rightarrow \infty } } [ \mu_{\rm{p } } ] } - 1 \right\}^2 \right ] \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { ( n_i - \mu)^2 } { \mu } - 1 \right\}^2 \right ] \hspace*{27truemm } \mbox{[see eq.\ ( 7 ) of \paperi ] } \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{k=0}^{\infty } } \left\ { n p(k;\mu ) \right\ } \left\ { \frac { ( k - \mu)^2 } { \mu } - 1 \right\}^2 \right ] \nonumber \\ & = & \frac{1}{\mu^2}~{\sum_{k=0}^{\infty } } p(k;\mu ) \left [ ( k-\mu)^2 - \mu \right]^2 \nonumber \\ & = & 2 + \frac{1}{\mu } \ . \label{eq : x2pr_var}\end{aligned}\ ] ] if we assume that pearson s @xmath0 applied to a large number of observations of a single poisson distribution with a mean value of @xmath12 always produces a normal distribution with a mean equal to the number of degrees - of - freedom ( @xmath3 ) [ see eq . ( 25 ) of ] and a variance of @xmath45 [ see eq . ( [ eq : x2pr_var ] ) ] , we can then attempt to create an ideal @xmath0 statistic for the analysis of poisson - distributed data by modifying pearson s @xmath0 as follows : @xmath46^{1/2 } \!\ ! + 1 \ , \right ] \label{eq : x2pm}\ ] ] where @xmath47 is the contribution of the @xmath16th data value to pearson s @xmath0 , @xmath48 is the expectation value of @xmath49 [ see eq . ( 25 ) of ] , @xmath50 is the variance of @xmath51 [ see eq . ( [ eq : x2pr_var ] ) ] . translating the mathematical notation to english , we have ( 1 ) shifted the mean of the standard @xmath35 distribution from @xmath3 times equation ( [ eq : expectation_value_x2p ] ) to zero , ( 2 ) forced the variance of the shifted distribution to be exactly @xmath2 , and then ( 3 ) shifted the mean of the variance - corrected distribution from zero back to @xmath3 . thus , _ by definition _ , the modified pearson s chi - square statistic ( @xmath52 ) will have a mean value of @xmath3 and a variance of @xmath2 in the limit of a large number of observations . let us now investigate the performance of the modified pearson s @xmath0 statistic with 1000 samples of @xmath37 poisson deviates with poisson mean values of 100 , 10 , 1 , 0.1 , and 0.001 . figure [ fig : x2pmt_5mu ] shows that @xmath52 results are significantly better than @xmath35 results [ fig . [ fig : x2p_5mu ] ] especially for poisson mean values less than 10 . figure [ fig : x2pmtr_var ] investigates the performance of the modified pearson s @xmath0 statistic over a wide range of poisson mean values from 0.001 to 1000 . the dashed lines of fig . [ fig : x2pmtr_var ] show the results for an ideal @xmath0 statistic ; one can see that while the average contribution to @xmath52 is 1 , as expected , and the average contribution to its variance is equal to 2 , as expected , the performance is not uniform for all poisson mean values the spread seen in the variance plot ( _ bottom panel _ ) increases as the poisson mean approaches zero . figures [ fig : x2pmt_5mu ] and [ fig : x2pmtr_var ] indicate the the modified pearson s @xmath0 statistic works well in the perfect case where one has _ a priori _ knowledge of the true poisson mean . in a real experiment , the true mean of the parent poisson distribution is rarely ( if ever ) known and model parameters must be estimated from the observations . how well does the modified pearson s @xmath0 statistic work with reasonable parameter estimates ? comparing fig . [ fig : x2pms_5mu ] with fig . [ fig : x2pmt_5mu ] and fig . [ fig : x2pmsr_var ] with fig . [ fig : x2pmtr_var ] , we see that the _ variances are significantly smaller _ when a _ realistic _ model ( i.e. , the sample mean ) is used instead of a _ perfect _ model ( i.e. , the true mean ) . a statistic that fails with reasonable parameter estimates is not a very useful statistic for the analysis of real observations with low - count data . let us continue the investigation of the performance of the chi - square - gamma statistic with 1000 samples of @xmath37 poisson deviates with poisson mean values of 100 , 10 , 1 , 0.1 , and 0.001 . figure [ fig : x2gt_5mu ] confirms that the @xmath31 statistic does not perform like an ideal @xmath0 statistic in the low - count regime . figure [ fig : x2gt_var ] expands the previous analysis of of the @xmath31 statistic over a wide range of poisson mean values from 0.001 to 1000 . the @xmath31 statistic clearly does not perform like an ideal @xmath0 statistic for poisson mean values @xmath41 . the variance of the chi - square - gamma statistic is , by definition , @xmath53 ^ 2 } { n_i + 1 } - \left ( \frac{1}{\nu } { \sum_{j=1}^{n } } \frac { [ n_j + { \rm{min}}(n_j,1 ) - m_j]^2 } { n_j + 1 } \right ) \right]^2\,,\ ] ] where @xmath43 is the number of independent degrees of freedom , n is the number of data values , and m is the number of free parameters . the variance of the _ reduced _ chi - square of a @xmath0 statistic for a large number of observations should ideally be two . in the limit of a large number of observations of a _ single poisson distribution with a mean value of @xmath12 _ , the variance of the reduced chi - square of the chi - square - gamma statistic is @xmath54 \nonumber \\ & \equiv & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - m_i \right]^2 } { n_i + 1 } - \left [ \frac{1}{\nu}~{\sum_{j=1}^{n } } \frac { \left [ n_j + { \rm{min}}(n_j,1 ) - m_j \right]^2 } { n_j + 1 } \right ] \right\}^2 \right ] \nonumber \\ & \equiv & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - m_i \right]^2 } { n_i + 1 } - \left [ \frac{\chi^2_{\gamma}}{\nu } \right ] \right\}^2 \right ] \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{\nu}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - m_i \right]^2 } { n_i + 1 } - \lim_{n \rightarrow \infty } \left [ \frac{\chi^2_{\gamma}}{\nu } \right ] \right\}^2 \right ] \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n - m}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - m_i \right]^2 } { n_i + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \right ] \hspace*{1truemm } \mbox{[see eq.\ ( 29 ) of \paperi ] } \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - \mu_{\gamma } \right]^2 } { n_i + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \right ] \hspace*{4truemm } \mbox{[see eq.\ ( 18 ) of \paperi ] } \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - { \displaystyle\lim_{n \rightarrow \infty } } [ \mu_{\gamma } ] \right]^2 } { n_i + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \right ] \nonumber \\ & = & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{i=1}^{n } } \left\ { \frac { \left [ n_i + { \rm{min}}(n_i,1 ) - \mu \right]^2 } { n_i + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \right ] \hspace*{6truemm } \mbox{[see eq.\ ( 19 ) of \paperi ] } \nonumber \\ & \approx & \lim_{n \rightarrow \infty } \left [ \frac{1}{n-1}~{\sum_{k=0}^{\infty } } \left\ { n p(k;\mu ) \right\ } \left\ { \frac { \left [ k + { \rm{min}}(k,1 ) - \mu \right]^2 } { k + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \right ] \nonumber \\ & = & { \sum_{k=0}^{\infty } } p(k;\mu ) \left\ { \frac { \left [ k + { \rm{min}}(k,1 ) - \mu \right]^2 } { k + 1 } - \left [ 1 + e^{-\mu}\left(\mu-1\right ) \right ] \right\}^2 \nonumber \\ & = & \mu^3 e^{-\mu}\left [ { \rm{ei}}(\mu ) - \gamma_{\rm{em } } - { \rm{ln}}(\mu ) + 4 \right ] -\mu^2 -\mu + e^{-\mu}\!\left [ -2\mu^2 + 2\mu + 1 \right ] + e^{-2\mu}\!\left [ -\mu^2 + 2\mu - 1 \right]\ ! , \label{eq : x2gr_var}\end{aligned}\ ] ] where ei@xmath55 is the exponential integral of @xmath56 @xmath57 $ ] and @xmath58 \approx 0.5772156649 $ ] is the euler - mascheroni constant . equation ( [ eq : x2gr_var ] ) approaches the expected value of 2 for large poisson mean values [ see the solid curve in the bottom panel of fig . [ fig : x2gt_var ] ] . 2gmx^2_i 2g^2 _ 2 g m_i^3 e^-m_i - m_i^2 -m_i + e^-m_i+e^-2m_i if we assume that chi - square - gamma statistic is applied to a large number of observations of a single poisson distribution with a mean value of @xmath12 always produces a normal distribution with a mean equal to the number of degrees of freedom ( @xmath3 ) times see equation ( 29 ) of and a variance of @xmath3 times equation ( [ eq : x2gr_var ] ) , we can then attempt to create an ideal @xmath0 statistic for the analysis of poisson - distributed data by modifying the chi - square - gamma statistic as follows : @xmath59^{1/2 } \!\ ! + 1 \ , \right ] \label{eq : x2gm}\ ] ] where @xmath60 ^ 2 } { n_i+1 } \label{eq : ith_contribution_x2g}\ ] ] is the contribution of the @xmath16th data value to the chi - square gamma statistic , @xmath61 is the expectation value of @xmath62 [ see equation ( 29 ) of ] , and @xmath63 -m_i^2 -m_i + \nonumber \\ & & e^{-m_i}\left [ -2m_i^2 + 2m_i + 1 \right ] + e^{-2m_i}\left [ -m_i^2 + 2m_i - 1 \right]\end{aligned}\ ] ] is the variance of @xmath64 [ see equation ( [ eq : x2gr_var ] ) ] . translating the mathematical notation to english , we have ( 1 ) shifted the mean of the standard @xmath35 distribution from @xmath3 times equation ( [ eq : expectation_value_x2 g ] ) to zero , ( 2 ) forced the variance of the shifted distribution to be exactly @xmath2 , and then ( 3 ) shifted the mean of the variance - corrected distribution from zero back to @xmath3 . thus , _ by definition _ , the modified chi - square statistic statistic ( @xmath65 ) will have a mean value of @xmath3 and a variance of @xmath2 in the limit of a large number of observations . let us now investigate the performance of the modified chi - square - gamma statistic with 1000 samples of @xmath37 poisson deviates with poisson mean values of 100 , 10 , 1 , 0.1 , and 0.001 . figure [ fig : x2gmt_5mu ] shows that @xmath65 results are significantly better than the @xmath31 results [ fig . [ fig : x2gt_5mu ] ] especially for poisson mean values less than 10 . figure [ fig : x2gmt_var ] investigates the performance of the @xmath65 statistic over a wide range of poisson mean values from 0.001 to 1000 . the dashed lines of fig . [ fig : x2gmt_var ] show the results for an ideal @xmath0 statistic ; one can see that while the average contribution to @xmath65 is 1 , as expected , and the average contribution to its variance is equal to 2 , as expected , the performance is not uniform for all poisson mean values . the bump seen in the bottom panel of fig . [ fig : x2gmt_var ] near the poisson mean value of 10 is an artifact caused by the @xmath66 offset in the numerator of the definition of the chi - square - gamma statistic [ eq . ( [ eq : x2 g ] ) ] . figures [ fig : x2gmt_5mu ] and [ fig : x2gmt_var ] indicate the the modified chi - square - gamma statistic works well in the perfect case where one has _ a priori _ knowledge of the true poisson how well does the modified chi - square - gamma statistic work with reasonable parameter estimates ? comparing fig . [ fig : x2gms_5mu ] with fig . [ fig : x2gmt_5mu ] and fig . [ fig : x2gms_var ] with fig . [ fig : x2gmt_var ] , we see that the results for the modified @xmath31 statistic with a realistic model ( i.e. , the sample mean ) are nearly identical are about 1 lower than the comparable value given in fig . [ fig : x2gmt_5mu ] . this is the result of losing one degree - of - freedom due to the determination of the sample mean from the data ( i.e. , @xmath3 drops from 10000 to 9999 ) . the scrambling caused by the modification of the @xmath31 statistic appears to have caused this expected loss of one degree - of - freedom to vanish in the very - low - count data regime ( @xmath67 ) . ] to those obtained with a perfect model ( i.e. , the true mean ) . i now demonstrate the new modified chi - square - gamma statistic by using it to study simulated x - ray images . cash ( @xcite ) applied his @xmath68 statistic to the problem of determining the position of a weak source in a x - ray image . let us use cash s point spread function ( psf ) but with a resolution of 100 pixels per unit area : @xmath69 where @xmath70 . this psf has the volume integral of @xmath71 & \mbox{for $ r\leq1 $ , } \\ 1 & \mbox{for $ r>1$. } \end{array } \right.\ ] ] figure [ fig : observation ] shows a simulated observation of a point source with an intensity of 40 x - ray photons on a background flux of 0.06 x - ray photons per pixel . this observation contains 2786 pixels with 0 photons , 204 pixels with 1 photons , and 10 pixels with 2 photons . there are a total of 56 photons found in the 317 pixels within a radius of 10 pixels of the center of the x - ray point source which is located at the @xmath72 position of @xmath73 . this is clearly a marginal detection of a weak x - ray point source on a noisy background ; the peak signal - to - noise ratio ( @xmath265.2 ) occurs at a radius of @xmath268 pixels . we will now use the modified chi - square - gamma statistic to answer the following questions about this x - ray is precluded by the fact that this particular observation contains only low - count data we must be content with realistic estimates for the location and intensity based on a detailed statistical analysis of the data . our first objective is to determine if there is an x - ray point source in the observation . one way this can be done is to investigate the region(s ) containing non - point - source pixels ( data values ) . this approach requires knowledge of the background flux level which we will henceforth assume is constant throughout the entire image . we begin by making a rough first estimate of the background flux level by dividing the total number of photon in the image by the total number of pixels : @xmath74 @xmath75 photons per pixel . the background flux level estimate may be significantly improved with a bit more work . there are 18 photons in the 317 pixels within 10 pixels of the position @xmath76 of fig . [ fig : observation ] . is the detection of 18 photons consistent with the expected value of 23.6799 @xmath77 photons ? the upper and lower 99.9% single - sided confidence limits for 18 photons are 35.35 and 7.662 , respectively [ see tables 1 and 2 of gehrels @xcite ] . i conclude that the pixel at @xmath76 is a background pixel ( shown as such with a gray box in fig . [ fig : first_sky ] ) because the _ expected _ number of background photons lies within the range of the upper and lower 99.9% single - sided confidence limits of the _ observed _ number of photons ( i.e. , @xmath78 ) . there are 56 photons in the 317 pixels within 10 pixels of the position @xmath73 of fig . [ fig : observation ] . the upper and lower 99.9% single - sided confidence limits for 56 photons are 83.1784 and 35.6834 , respectively [ see eqs . ( 10 ) and ( 14 ) of gehrels @xcite ] . i conclude that the pixel at @xmath73 is _ not _ a background pixel because the expected number of background photons is less than the lower 99.9% single - sided confidence limit of the observed number of photons ( i.e. , @xmath79 ) . this conclusion was expected since @xmath73 is the center of the x - ray source . the 2676 gray pixels in fig . [ fig : first_sky ] have a total of 162 photons . we can now make a second estimate of the background flux : @xmath80 @xmath81 photons per pixel . repeating this process once more yields the final estimate of the background flux : @xmath82 @xmath83 photons per pixel . the measurement error for this estimate is approximately 0.0049 @xmath84 . the final estimate of the x - ray background flux , @xmath85 , is in excellent agreement with the true value of 0.06 [ see fig . [ fig : final_sky ] ] . i conclude that fig . [ fig : observation ] has at least one x - ray point source because the entire data set is not consistent with a x - ray background flux of 0.0601 photons per pixel _ for every pixel_. assuming that there is only one x - ray source , we can make the first rough estimate of its location by stating that it probably is located at a non - gray pixel location in fig . [ fig : final_sky ] . there are 453 non - background pixels in fig . [ fig : final_sky ] with a total of 71 photons . this fact allows us to restrict the uncertainty of the location of the x - ray point source to about 15% @xmath86 of the total image . assuming a background flux of 0.06 photons per pixel , we expect that 27 @xmath87 photons of the total 71 photons found in the non - background pixels would be due to the background and not the point source . we can now make the first rough estimate of the intensity of the x - ray source : 44 @xmath88 photons . there are 48 photons in the 317 pixels within 10 pixels of the position @xmath89 of fig . [ fig : observation ] . is this photon sum consistent with a model of a 40 photon point source _ centered at that location _ on a background of 0.06 photons per pixel ? the upper and lower 95% single - sided confidence limits for 48 photons is 61.05 and 37.20 photons , respectively . the model predicts that we should find 58.9802 $ ] photons . the maximum number we would expect the model to predict is 59.0200 @xmath90 $ ] photons . the model prediction lies within these extremes : @xmath91 . ] photons within a radius of 10 pixels . the 48 photons found within 10 pixels of @xmath89 is consistent with the model ( shown as such with a dark - gray circle in fig . [ fig : final_sky ] ) , because the _ expected _ number of photons lies within the range of the upper and lower 95% single - sided confidence limits of the _ observed _ number of photons ( i.e. , @xmath92 ) . there are 217 circled pixels in fig . [ fig : final_sky ] . this fact allows us to further restrict the uncertainty of the location of the x - ray point source to @xmath267.2% @xmath93 of the total image . since the peak signal - to - noise ratio occurs near a radius of 8 pixels , we now investigate if we can improve our estimate of the location of the x - ray point source by considering photon sums within a smaller aperture with a radius of 8 instead of 10 pixels . there are 34 photons in the 197 pixels within a radius of 8 pixels of the position @xmath94 of fig . [ fig : observation ] . is this photon sum consistent with a model of a 40 photon point source centered at that location with a background of 0.06 photons per pixel ? the upper and lower 95% single - sided confidence limits for 34 photons is 45.27 and 25.01 photons , respectively . the model predicts that we should find 47.2664 photons within a radius of 8 pixels . the 34 photons found within 8 pixels of @xmath94 is _ not _ consistent with the model since the expected number of photons is greater than the upper 95% single - sided confidence limit of the observed number of photons ( i.e. , @xmath95 ) . however , the 39 photons found within 8 pixels of @xmath96 _ is _ consistent with a 40 photon point source centered at @xmath96 on a background of 0.06 photons per pixel ( i.e. , @xmath97 ) . there are a total of 111 circled pixels in fig . [ fig : green_radius8 ] . this fact allows us to further restrict the uncertainty of the location of the x - ray point source to @xmath263.7% @xmath98 of the total image . one way to boost the data out of the low - count regime is to compare the cumulative radial distribution of the model with the cumulative radial distribution of the data . the modified chi - square - gamma statistic was used to compare the cumulative radial distribution of the model ( 10 1-pixel - wide bins @xmath99 10 degrees - of - freedom ) with the cumulative radial distribution of the data ( similarly formatted ) . at the position of @xmath96 the value of @xmath100 for the cumulative radial distributions was computed to be 13.7338 ( @xmath101 ) for a model of a 40 photon point source centered at @xmath96 on a background of 0.06 photons per pixel . the 95th percentage point for the chi - square distribution with 10 degrees of freedom may be found in several standard references : 18.3 ( _ crc handbook of chemistry and physics _ , lide & frederikse @xcite , p. a-106 ) , 18.31 ( bevington @xcite , p. 315 ) and 18.3070 ( abramowitz & stegun @xcite , p. 985 ) . i conclude that the image location @xmath96 is within the 95% confidence interval because the value of the modified chi - square - gamma statistic is less than the 95th percentage point for the chi - square distribution with 10 degrees of freedom ( i.e. @xmath102 ) . the probability that the observed chi - square value for a correct model should be less than a value of @xmath0 for @xmath3 degrees of freedom is @xmath103 where the latter function is the incomplete gamma function [ @xmath104 ; see , e.g. , the routine in _ numerical recipes _ ( press @xcite ) ] . if we assume that @xmath100 is distributed like the @xmath0 distribution , then we can assign a probability for the modified chi - square - gamma value for 10 degrees of freedom : @xmath105 . there is thus a @xmath2681.5% chance that the observed modified chi - square - gamma statistic will be less than 13.7338 for 10 degrees of freedom . the contour in fig . [ fig : green_radius8 ] shows the 95% confidence interval of the x - ray point source based on the @xmath100 analysis of the cumulative radial distribution of the data . the value of @xmath100 for the cumulative radial distribution at @xmath94 was computed to be 30.4707 giving a probability of @xmath2699.9% ; this location in fig . [ fig : green_radius8 ] lies outside the 95% confidence interval . we can further use the modified chi - square - gamma statistic with the cumulative radial distribution to determine the 95% confidence limits of the intensity of the x - ray source in the image ( see fig . [ fig : crd_probability ] ) . the upper and lower single - sided 95% confidence limits for the intensity of an x - ray point source at @xmath73 in fig . [ fig : observation ] is 54.5 and 28.0 , respectively . the true intensity of the x - ray source is 40 photons . given a background flux uncertainty of @xmath106 photons per pixel ( see above ) , we can approximate the theoretical rms measurement error for a 40 photon point source spread over 317 pixels @xmath107 px@xmath108 as @xmath109 photons . the difference between the upper and lower 95% single - sided confidence limits is approximately 3.3 standard deviations of the normal probability function . this fact can be used to approximate an rms measurement error for our intensity estimate of @xmath110 [ @xmath111 photons . the @xmath100 analysis using the cumulative radial distribution has yielded an excellent intensity estimate . the analysis presented in figures [ fig : green_radius8 ] and [ fig : crd_probability ] is predicated on the assumption that the modified chi - square - gamma statistic is distributed like @xmath0 . but is this assumption valid ? figure [ fig:10000_cumulative_fraction ] shows that the analysis of @xmath37 simulated x - ray observations like fig . [ fig : observation ] yields modified chi - square - gamma values that are distributed like the chi - square distribution for @xmath112 degrees of freedom : a gaussian distribution with a mean of @xmath3 and a variance of @xmath2 . the above analysis has assumed that the probability that the observed modified chi - square - gamma value for a correct model should be less than a value of @xmath0 for @xmath3 degrees of freedom can be given as @xmath113 . assuming that the predicted probability from @xmath113 is an accurate prediction of the true probability , then the _ predicted _ probability of the 9500th simulated observation of a total of 10000 ( sorted by @xmath100 value ) should be very close to 95% ; fig . [ fig:10000_sorted_probability ] indicates that this is indeed the case ( i.e. , the probability for the @xmath100 value of the 9500th simulated observation is 94.8666% ) . since @xmath100 statistic is distributed ( nearly ) like the @xmath0 distribution , the usage of the incomplete gamma function to predict probabilities for modified chi - square - gamma values appears to be justified in practical analysis problems . i investigated the use of pearson s chi - square statistic [ eq . ( [ eq : x2p ] ) ] , the maximum likelihood ratio statistic for poisson distributions [ eq . ( [ eq : x2l ] ) ] , and the chi - square - gamma statistic [ eq . ( [ eq : x2 g ] ) ] for the determination of the goodness - of - fit between theoretical models and low - count poisson - distributed data . i concluded that none of these statistics should be used to determine the goodness - of - fit with data values of 10 or less . i modified pearson s chi - square statistic for the purpose of improving its goodness - of - fit performance . i demonstrated that modified pearson s @xmath0 statistic [ eq . ( [ eq : x2pm ] ) ] works well in the perfect case where one has _ a priori _ knowledge of the correct ( true ) model . in a real experiment , however , the true mean of the parent poisson distribution is rarely ( if ever ) known and model parameters must be estimated from the observations . i demonstrated that the modified pearson s @xmath0 statistic has a variance that is significantly smaller than that of the @xmath0 distribution when _ realistic _ models , defined as having parameters estimated from the observational data , are compared with poisson - distributed data . any statistic that fails with models based on reasonable parameter estimates is not a very practical statistic for the analysis of astrophysical observations . i concluded that the modified pearson s @xmath0 statistic should not be used to determine the goodness - of - fit with low - count data values of 10 or less . i modified the chi - square - gamma statistic for the purpose of improving its goodness - of - fit performance . i demonstrated that the modified chi - square - gamma statistic [ eq . ( [ eq : x2gm ] ) ] performs ( nearly ) like an ideal @xmath0 statistic for the determination of goodness - of - fit with low - count data . on average , for correct ( true ) models , the mean value of the modified chi - square - gamma statistic is equal to the number of degrees of freedom @xmath1 and its variance is @xmath2 like the @xmath0 distribution for @xmath3 degrees of freedom . an ideal @xmath0 statistic for the determination of goodness - of - fit with low - count data should _ fail _ in a predictable manner . hypothesis testing of low - count poisson - distributed data with the modified pearson s @xmath0 statistic will produce the peculiar and undesirable result that correct models are more likely to be rejected than realistic models [ cf . [ fig : x2pmt_5mu ] with fig . [ fig : x2pms_5mu ] ] . the modified chi - square - gamma statistic is a _ practical _ statistic to use for hypothesis testing of astrophysical data from counting experiments because it performs ( nearly ) like an ideal @xmath0 statistic for realistic _ and _ correct models in the low - count _ and _ the high - count data regimes ; accurate and believable probabilities for @xmath100 goodness - of - fit values can be calculated with the incomplete gamma function [ figs . [ fig:10000_cumulative_fraction ] and [ fig:10000_sorted_probability ] ] . a lot of nothing can tell you something as long as there are _ some _ observations with signal in them . vincent eke sent me an e - mail asking if i had an expression for the variance of the @xmath31 statistic which described the mysterious second hump of the solid curve of fig . 3 of . after a rapid exchange of email with me over the period of a week , he was the first to derive an analytical formula for @xmath114 [ eq . ( [ eq : x2gr_var ] ) ] . the knowledge that the variance of @xmath31 could in fact be expressed explicitly as an analytical expression turned out to be the breakthrough that i had needed in order to complete the development of the modified @xmath31 statistic . it is a pleasure to acknowledge his contribution to this research . i was supported by a grant from the national aeronautics and space administration ( nasa ) , order no . s-67046-f , which was awarded by the long - term space astrophysics program ( nra 95-oss-16 ) . this research has made use of nasa s astrophysics data system abstract service which is operated by the jet propulsion laboratory at the california institute of technology , under contract with nasa . abramowitz , m. , & stegun , i. 1964 , `` handbook of mathematical functions with formulas , graphs , and mathematical tables '' , applied mathematics series 55 , ed . m. abramowitz & i. stegun ( washington , d.c . : nbs ) 01cap [ fig : x2p_x2l_x2g_mu100 ] a simulated data set of 1000 samples ( `` observations '' ) of @xmath37 poisson deviates ( `` measurements '' ) per sample was created assuming a mean value @xmath12@xmath115@xmath116 for each poisson deviate . each sample in this data set was then analyzed using pearson s @xmath0 statistic [ _ top ; _ definition : eq . ( [ eq : x2p ] ) ] , the maximum likelihood ratio statistic for poisson distributions [ _ middle ; _ definition : eq . ( [ eq : x2l ] ) ] , and the chi - square - gamma statistic [ _ bottom ; _ definition : eq . ( [ eq : x2 g ] ) ] . the model of the @xmath16th deviate in each sample was set to the true mean value of parent poisson distribution ( i.e. , @xmath117 ) and the number of independent degrees - of - freedom was therefore equal to the number of deviates per sample ( i.e. @xmath118 ) . compare the cumulative distribution for each statistic with the cumulative distribution function of a gaussian distribution with a mean of @xmath37 and a variance of @xmath38 [ _ thick curve _ in each panel ] . the number and error shown on the right side of each panel is the mean and rms value of the 1000 samples shown in that panel ; ideally these values should be about @xmath119 . 02cap [ fig : x2l_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the maximum likelihood ratio statistic for poisson distributions [ definition : eq . ( [ eq : x2l ] ) ] . in all cases , @xmath118 and @xmath8 was set to the true mean value of the data set . other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 03cap [ fig : x2lr_var ] reduced chi - square as a function of the true poisson mean ( @xmath121 with 10 mean values per decade ) for the maximum likelihood ratio statistic for poisson distributions with the model of the @xmath16th deviate set to the mean value of the parent poisson distribution . open squares _ show the results of the analysis of one sample composed of @xmath122 poisson deviates ( @xmath123 ) at each given poisson mean value . the _ filled squares _ show the results of the analysis of 1000 subsamples of the @xmath122 poisson deviates ( @xmath124 ) previously analyzed as one large sample . the scatter of the filled squares with respect to the open squares is real and is due to random fluctuations of the parent poisson distributions . the _ dashed line _ shows the ideal value of one . _ bottom panel : _ the variance of the reduced chi - square values shown in the top panel . the _ dashed line _ shows the ideal value of two . 04cap [ fig : x2p_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using pearson s @xmath0 statistic [ definition : eq . ( [ eq : x2p ] ) ] ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the true mean value of the data set . other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 05cap [ fig : x2pr_var ] reduced chi - square as a function of the true poisson mean for pearson s @xmath0 statistic with the model of the @xmath16th deviate set to the true mean value of the parent poisson distribution ( same input data set as for fig . [ fig : x2lr_var ] ) . the _ solid line _ connecting the open squares in the _ bottom panel _ is the formula @xmath125 [ see eq . ( [ eq : x2pr_var ] ) ] . other details as in fig . [ fig : x2lr_var ] . 06cap [ fig : x2pmt_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the modified pearson s @xmath0 statistic [ definition : eq . ( [ eq : x2pm ] ) ] ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the true mean value of the data set . other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 07cap [ fig : x2pmtr_var ] reduced chi - square as a function of the true poisson mean for the modified pearson s @xmath0 statistic with the model of the @xmath16th deviate set to the true mean value of the parent poisson distribution ( same input data set as for fig . [ fig : x2lr_var ] ) . other details as in fig . [ fig : x2lr_var ] . 08cap [ fig : x2pms_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the modified pearson s @xmath0 statistic ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the _ sample mean_. other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 09cap [ fig : x2pmsr_var ] reduced chi - square as a function of the true poisson mean for the modified pearson s @xmath0 statistic with the model of the @xmath16th deviate set to the _ sample mean _ ( same input data set as for fig . [ fig : x2lr_var ] ) . other details as in fig . [ fig : x2lr_var ] . 10cap [ fig : x2gt_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the @xmath126 statistic [ definition : eq . ( [ eq : x2 g ] ) ] ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the true mean value of the data set . other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 11cap [ fig : x2gt_var ] reduced chi - square as a function of the true poisson mean for the @xmath126 statistic statistic with the model of the @xmath16th deviate set to the true mean value of the parent poisson distribution ( same input data set as for fig . [ fig : x2lr_var ] ) . the _ solid line _ connecting the open squares in the _ top panel _ is the formula @xmath127 [ eq . 29 of ] . the _ solid line _ connecting the open squares in the _ bottom panel _ is equation ( [ eq : x2gr_var ] ) . other details as in fig . [ fig : x2lr_var ] . 12cap [ fig : x2gmt_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the modified @xmath126 statistic [ definition : eq . ( [ eq : x2gm ] ) ] ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the true mean value of the data set . other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 13cap [ fig : x2gmt_var ] reduced chi - square as a function of the true poisson mean for the modified @xmath126 statistic with the model of the @xmath16th deviate set to the true mean value of the parent poisson distribution ( same input data set as for fig . [ fig : x2lr_var ] ) . other details as in fig . [ fig : x2lr_var ] . 14cap [ fig : x2gms_5mu ] the cumulative distribution functions for 1000 samples of @xmath37 poisson deviates ( _ top to bottom _ : @xmath120 , and 0.01 ) analyzed using the modified @xmath126 statistic ( same input data set as for fig . [ fig : x2l_5mu ] ) . in all cases , @xmath118 and @xmath8 was set to the _ sample mean_. other details as in fig . [ fig : x2p_x2l_x2g_mu100 ] . 15cap [ fig : x2gms_var ] reduced chi - square as a function of the true poisson mean for the modified @xmath126 statistic with the model of the @xmath16th deviate set to the _ sample mean _ ( same input data set as for fig . [ fig : x2lr_var ] ) . other details as in fig . [ fig : x2lr_var ] . 16cap [ fig : observation ] the simulated x - ray observation . a 40 photon x - ray point source is located at the @xmath72 position of @xmath73 on a background of 0.06 photons per pixel . the point spread function is @xmath128 $ ] where @xmath70 . this is the same psf used by cash ( @xcite ) but with a resolution of 100 pixels per unit area . 17cap [ fig : first_sky ] indicate pixels with a total number of photons within a radius of 10 pixels that are consistent ( within the 99.9% upper and lower single - sided confidence limits of the observed photon total ) with the estimated background flux level of 0.0747 photons per pixel . the _ x _ marks the center of the x - ray point source and the _ dotted circle _ has a radius of 10 pixels which is the maximum size of the psf . other details as in fig . [ fig : observation ] . 18cap [ fig : final_sky ] indicate the pixels with a total number of photons within a radius of 10 pixels that are consistent ( within the 99.9% upper and lower single - sided confidence limits of the observed photon total ) with the true background flux level of photons per pixel . _ dark - gray circles _ indicate the pixels with a total number of photons within a radius of 10 pixels that are consistent ( within the upper and lower single - sided confidence limits of the observed photon total ) with the model of a 40 photon point source _ at that pixel location _ on a background of 0.06 photons per pixel . other details as in fig . [ fig : first_sky ] . 19cap [ fig : green_radius8 ] indicate the pixels with a total number of photons within a radius of that are consistent ( within the 95% upper and lower single - sided confidence limits of the observed photon total ) with the model of a 40 photon point source at that pixel location on a background of 0.06 photons per pixel . all pixels within the _ solid black contour _ are within the as determined by the @xmath100 analysis of the cumulative radial distribution of the data within 10 pixels is compared with the cumulative radial distribution of a model of a 40 photon point source at that pixel location on a background of 0.06 photons per pixel . note how well the 95% confidence interval of the @xmath100 analysis of the cumulative radial distribution matches the region described by the circled pixels . other details as in fig . [ fig : final_sky ] . 20cap [ fig : crd_probability ] the photon distribution of the 317 pixels within a radius of 10 pixels of the location @xmath73 of fig . [ fig : observation ] was transformed to a cumulative radial distribution ( 10 1-pixel - wide bins @xmath99 10 degrees - of - freedom ) and then compared , using the modified chi - square - gamma statistic , with 80 models of the observation : a 1 to 80 photon ( in steps of 1 photon ) x - ray point source at @xmath73 on a background of 0.06 photons per pixel ( i.e. , the true background ) . the 95th percentage point for the chi - square distribution with 10 degrees of freedom is 18.31 [ i.e. @xmath129 . assuming that @xmath100 is distributed like @xmath0 , we see that the upper and lower single - sided 95% confidence limits for the intensity of an x - ray point source at @xmath73 in fig . [ fig : observation ] is 54.5 and 28.0 , respectively . the true intensity of the x - ray source is 40 photons . 21cap [ fig:10000_cumulative_fraction ] a data set of @xmath37 realizations of the the same model used to make fig . [ fig : observation ] was created . each sample in this data set was then analyzed using the modified chi - square - gamma statistic at the location @xmath73 the true location of the simulated x - ray point source of 40 photons on a background of 0.06 photons per pixel . all 317 pixels within a radius of 10 pixels ( the size of the psf ) were compared to the true model value at that location and the number of independent degrees - of - freedom was therefore equal to the number of pixels analyzed ( i.e. @xmath130 ) . compare the cumulative distribution with the cumulative distribution function of a gaussian distribution with a mean of @xmath131 and a variance of @xmath132 [ _ thick curve _ ] . the numbers with errors shown on the right side give the mean and rms value for the @xmath100 ( _ top _ ) and ideal @xmath0 ( _ bottom _ ) statistics . 22cap [ fig:10000_sorted_probability ] the @xmath37 simulations of fig . [ fig:10000_cumulative_fraction ] were sorted by the value of the modified chi - square - gamma statistic . the _ dark _ plot shows the probablity @xmath133 as a function of the sorted @xmath100 values . the _ gray _ plot on the bottom shows the residuals from the ideal one - to - one correspondance . the _ predicted _ probabilities for the 9000th , 9500th , and 9900th sorted simulations were 90.2069% , 94.8666% , and 98.9497% , which agrees very well with the _ expected _ probabilities of 90% , 95% , and 99% , respectively . for the entire simulation , the mean and rms value of the resiuals is @xmath134 percentage points note that the residuals never exceeds 1 percent . figures [ fig:10000_cumulative_fraction ] and [ fig:10000_sorted_probability ] indicate that the assumption that the @xmath100 statistic is distributed like the @xmath0 distribution was valid .

voids form a prominent aspect of the megaparsec distribution of galaxies and matter . not only do they represent a key constituent of the cosmic web , they also are one of the cleanest probes and measures of global cosmological parameters . the shape and evolution of voids are highly sensitive to the nature of dark energy , while their substructure and galaxy population provides a direct key to the nature of dark matter . also , the pristine environment of void interiors is an important testing ground for our understanding of environmental influences on galaxy formation and evolution . in this paper , we review the key aspects of the structure and dynamics of voids , with a particular focus on the hierarchical evolution of the void population . we demonstrate how the rich structural pattern of the cosmic web is related to the complex evolution and buildup of voids .

1611.01222

voids form a prominent aspect of the megaparsec distribution of galaxies and matter @xcite . they are enormous regions with sizes in the range of @xmath0 mpc that are practically devoid of any galaxy , usually roundish in shape and occupying the major share of space in the universe ( see fig.[fig : sdssvoids ] and @xcite for a recent review ) . forming an essential and prominent aspect of the _ cosmic web _ @xcite , they are instrumental in the spatial organization of the cosmic web @xcite . surrounded by elongated filaments , sheetlike walls and dense compact clusters , they weave the salient weblike pattern of galaxies and matter pervading the observable universe . several recent studies came to the realization that voids not only represent a key constituent of the cosmic mass distribution , but that they are also one of the cleanest probes and measures of the global cosmology . particularly interesting is the realization that their structure , morphology and dynamics reflects the nature of dark energy , dark matter and that of the possibly non - gaussian nature of the primordial perturbation field . another major aspect of voids is that their pristine environment represents an ideal and pure setting for the study of galaxy formation and the influence of cosmic environment on the evolution of galaxies . in addition , voids play a prominent role in the reionization process of the universe , forming the principal regions along which the ionizing radiation produced by the first stars in the universe propages . in a void - based description of the evolution of the cosmic matter distribution , voids mark the transition scale at which density perturbations have decoupled from the hubble flow and contracted into recognizable structural features . at any cosmic epoch the voids that dominate the spatial matter distribution are a manifestation of the cosmic structure formation process reaching a non - linear stage of evolution . on the basis of theoretical models of void formation one might infer that voids may act as the key organizing element for arranging matter concentrations into an all - pervasive cosmic network @xcite . as voids expand , matter is squeezed in between them , and sheets and filaments form the void boundaries . this view is supported by numerical studies and computer simulations of the gravitational evolution of voids in more complex and realistic configurations @xcite . a marked example of the evolution of a typical large and deep void in a @xmath1cdm scenarios is given by the time sequence of six frames in fig . [ fig : lcdmvoid ] . the relatively simple structure and dynamical evolution of voids remains strongly influenced by the evolving large scale environment , through dominant tidal influences and direct contact between neighbouring voids . it is this aspect that has been recognized in a surge of recent studies for its considerable potential for measuring the value of cosmological parameters . the evolution of their structure and shape appear to be a direct manifestation of the nature of dark energy , while their dynamics reflects the nature of dark matter or that of the possibly nonstandard nature of gravity . 0.0truecm [ fig : sdssvoids ] voids are an outstanding aspect of the weblike cosmic mass distribution @xcite . they have been known as a feature of galaxy surveys since the first surveys were compiled @xcite . following the discovery by @xcite and @xcite of the most dramatic specimen , the botes void , a hint of their central position within a weblike arrangement came with the first cfa redshift slice @xcite . this view has been dramatically endorsed and expanded by the redshift maps of the 2dfgrs and sdss surveys @xcite . they have established voids as an integral component of the cosmic web . the 2dfgrs maps and sdss maps ( see e.g. fig . [ fig : sdssvoids ] ) are telling illustrations of the ubiquity and prominence of voids in the cosmic galaxy distribution . recent studies have also revealed the prominence of voids in the distant and early u niverse . a beautiful example is that of the deep probe of the vipers galaxy redshift survey , which meticulously outlined the cosmic web up to redshifts @xmath2 @xcite . perhaps even more impressive is the identification of the considerably more intricate void population in the dark matter distribution , as revealed by its reconstruction in the 2mrs survey volume @xcite and the sdss volume @xcite . 0.0truecm [ fig : voidinventory ] a relatively crude analysis of voids in computer simulations provides an impression of the status of voids in the megaparsec universe . they clearly occupy a major share of the volume of the universe . this has been confirmed in a recent systematic analysis by cautun et al . ( 2014 ) , which included an inventory of morphological components of the cosmic web with respect in the dark matter distribution in the lcdm millennium simulation ( see fig . [ fig : voidinventory ] ) . around @xmath3 of the cosmic volume should be identified as a void region . nonetheless , it represents less than @xmath4 of the mass content of the universe . this implies that the average density in voids is @xmath5 of the average cosmic density , which is indeed reasonably close to what is expected on simple theoretical grounds ( see below ) . with voids being recognized as prominent aspects of the megaparsec galaxy and matter distribution , we should expect them to be a rich source of information on a range of cosmological questions . we may identify at least four : 1 . they are a prominent aspect of the megaparsec universe , instrumental in the spatial organization of the cosmic web @xcite . their effective repulsive influence over their surroundings has been recognized in surveys of the local universe @xcite . voids contain a considerable amount of information on the underlying cosmological scenario and on global cosmological parameters . they are one of the cleanest cosmological probes and measures of dark energy , dark matter and tests with respect to possible modifications of gravity as described by general relativity . this realization is based on the fact that it is relatively straightforward to relate their dynamics to the underlying cosmology , because they represent a relatively modest density perturbation . their structure and shape , as well as mutual alignment , are direct reflections of dark energy @xcite . notable cosmological imprints are also found in the outflow velocities and accompanying redshift distortions @xcite . in particular interesting is the realization that the dynamics and infrastructure of voids are manifestations of the nature of dark matter or of modified gravity @xcite . the cosmological ramifications of the reality of a supersized voids akin to the identified ones by @xcite , @xcite and @xcite would obviously be far - reaching . the pristine low - density environment of voids represents an ideal and pure setting for the study of galaxy formation and the influence of cosmic environment on the formation of galaxies ( e.g. * ? ? ? * ; * ? ? ? voids are in particular interesing following the observation by peebles that the dearth of low luminosity objects in voids is hard to understand within the @xmath1cdm cosmology @xcite . voids are prominent in the key reionization transition in the early universe , key targets of lofar and ska ( e.g. * ? ? ? * ; * ? ? ? 0.15 cm [ fig : localvoids ] the most detailed view of the structure and galaxy population of voids is offered by voids in the local universe . a particularly important source of information is the karachentsev lv catalog of galaxies in the local volume , a volume - limited sample of galaxies within a radius of 11 mpc around the milky way @xcite . it provides a meticulously detailed view of the local void that appears to dominate a major fraction of space in our immediate cosmic neighbourhood . the desolate emptiness of this vast volume is most strikingly borne out by the adhesion reconstruction of the local void by @xcite and @xcite . -0.25truecm [ fig : localvoids ] moving to a slightly larger volume of the surrounding universe , we obtain a more representative impression of the prominence and role of voids in the overall large scale mass distribution . the 2mrs survey @xcite provides a uniquely complete census of the galaxy distribution out to distances of 100 - 150 mpc . it entails the entire cosmic environment out to that distance , and has enabled remarkably precise and detailed reconstructions of the underlying matter distribution in our cosmic neighbourhood . figure [ fig : localvoids ] provides a remarkably detailed reconstruction of the cosmic web in the 2mrs volume . it shows the ( surface ) density of the weblike structures in the local universe . these are the result of adhesion simulations by @xcite , based on the the constrained bayesian kigen reconstruction by @xcite of the initial conditions in the local volume traced by the 2mrs redshift survey ( see * ? ? ? for a given gaussian primordial field , the adhesion formalism allows the accurate reconstruction of the rich pattern of weblike features that emerge in the same region as a result of gravitational evolution . the adhesion formalism was applied to 25 constrained realizations of the 2mrs based primordial density field @xcite . the mean of these realizations gives a reasonably accurate representation of the significant filamentary and wall - like features in the local universe . most outstanding is the clear outline of the void population in the local universe . the reconstruction also includes the velocity flow in the same cosmic region . it reveals the prominent nature of the outflow from the underdense voids , clearly forming a key aspect of the dynamics of the megaparsec scale universe . the local universe structure in figure [ fig : localvoids ] presents a telling image of a void dominated large scale universe . many of the voids in the adhesion reconstruction can be identified with the void nomenclature proposed by fairall @xcite , who mainly identified these voids by eye from the 6dfgrs survey . it is interesting to see that the socalled tully void appears to be a richly structured underdense region , containing at least the microscopium void , the local void and the `` trans tully void '' . [ fig : lcdmvoid ] voids emerge out of the density troughs in the primordial gaussian field of density fluctuations . early theoretical models of void formation concentrated on the evolution of isolated voids @xcite . nearly without exception they were limited to spherically symmetric configurations ( but see * ? ? ? neither of these simplifications appears to be close to what happens in reality . nonetheless , the spherical model of isolated void evolution appears to provide us with valuable insights in major physical void characteristics and , as important , with sometimes surprisingly accurate quantitative reference benchmarks ( see below ) . in the next section ( sect.[sec : ellipsoids ] ) , we will see that a major and dominant aspect of voids is the fact that they are the opposite of isolated objects . to understand the dynamical evolution of voids we need to appreciate two major aspects in which the dynamics of voids differs fundamentally from that off dark halos . because voids expand and increase in size , they will naturally run up against their expanding peers . their spatial distribution and organization will be substantially influenced by the way in which the mutually competing voids distribute their share of space . equally important is the fact that voids will always be limited to a rather modest density deficit : they can not become more empty than empty . an immediate consequence of this is that the dynamical influence of the external mass inhomogeneities retains its dominant role in the evolution of a void . voids will never and can not decouple from their surroundings ! the essence of void evolution stems from the fact that they are underdensities in the mass distribution . as a result , they represent regions of weaker gravity . it translates in an effective repulsive peculiar gravitational influence . most of the principal characteristics of void evolution can be recognized in the illustration of a typical evolving void in a @xmath1cdm universe in fig . [ fig : lcdmvoid ] ( also see * ? ? ? the sequence of 6 timestepe of the simulation reveals at least three of the principal aspects of void evolution . because of the effective repulsive and thus outward peculiar gravitational accleration , initially underdense regions expand faster than the hubble flow : they expand with respect to the background universe . also we see that mass is streaming out of the underdensity . as a result of this evacuation , the density within voids continuously decrease . a clear census of the continuously decreasing mass content in voids can be found in the study of cosmic web evolution by @xcite . isolated voids would asympotically evolve towards an underdensity @xmath6 , pure emptiness . also observable in figure [ fig : lcdmvoid ] is how the mass distribution inside the void appears to become increasingly uniform . at the same time we see an accumulation of mass around its boundary . this is a direct consequence of the differential outward peculiar gravitational acceleration in voids . because the density within underdense regions gradually increases outward , we see a decrease of the corresponding peculiar ( outward ) gravitational acceleration . it means that matter at the centre of voids moves outward faster than matter at the boundary regions . as a result , matter tends to accumulate in - filamentary or planar - ridges surrounding the void , while the interior evolves into a uniform low - density region resembling a low - density homogeneous frw universe @xcite . another key feature of void evolution is the diminishing prominence of substructure in its interior . figure [ fig : lcdmvoid ] clearly shows how the internal structure gradually disappears as mass moves out of the void . this is a direct manifestation of the complex hierarchical evolution of voids @xcite . a direct implication of this is that massive features and objects can not form in voids , which e.g. manifests itself in a strong shift of the mass spectrum of dark halos - and hence galaxies - towards small masses @xcite . @xcite showed how this can be rather accurately modelled by means of a modification of the spectrum of density fluctuations towards one more appropriate for a low - density frw universe . note that while by definition voids correspond to density perturbations of at most unity , @xmath7 , mature voids in the nonlinear matter distribution do represent _ highly nonlinear _ features . this may be best understood within the context of lagrangian perturbation theory @xcite . overdense fluctuations may be described as a converging series of higher order perturbations . the equivalent perturbation series is less well behaved for voids : successive higher order terms of both density deficit and corresponding velocity divergence alternate between negative and positive . while in reality voids will never be isolated , nor spherical , we may obtain a basic understanding of the quantitative aspects of void evolution from the simple model of an evolving isolated spherical void . figure [ fig : sphervoid ] illustrates the evolution of spherical isolated voids . the void in the lefthand is a typical pure and uncompensated bucket " void . we notice the principal characteristics of void formation that we discussed in the previous section . the expansion of the void is evident as we see the boundary edge , and the ridge at its boundary , move outward . meanwhile , matter moves out of the void , leading to an increasingly empty void . the spherical void model also clarifies an additional major aspect , the formation of a ridge around the expanding void . this is the result of the differential outward expansion of the mass in and around the void . while the uniform underdensity profile implies a uniform expansion within the void , near the boundary the inner layers move outward faster than the more outward layers this is the result of the latter feeling a more moderate interior underdensity . in all , this ultimately leads to the interior mass shells taking over the initially exterior shells . this leads to a _ fundamental _ evolutionary timescale for void evolution , that of _ shellcrossing_. @xcite demonstrated that once voids have passed the stage of shellcrossing they enter a phase of self - similar expansion . subsequently , their expansion will slow down with respect to the earlier linear expansion . this impelled @xcite to identify voids in the present - day galaxy distribution with voids that have just reached the stage of shell - crossing . it happens when a primordial density depression attains a linearly extrapolated underdensity @xmath8 ( strictly speaking for an eds universe ) . that happens when a perfectly spherical `` bucket '' void will have expanded by a factor of 1.72 at shellcrossing , and therefore have evolved into an underdensity of @xmath9 of the global cosmological density , ie . @xmath10 . in other words , the voids that we see nowadays in the galaxy distribution do probably correspond to regions whose density is @xmath9 of the mean cosmic density ( note that it may be different for underdensity in the galaxy distribution ) . -5,5truecm -7.2truecm interestingly , for a wide range of initial radial profiles , voids will evolve into a bucket shaped void profile . the righthand frame of fig.[fig : sphervoid ] shows the evolution of a spherical void that develops out of an initial underensity that is an angular averaged density profile of an underdensity in a cdm gaussian random field ( see * ? ? ? * ; * ? ? ? * ) . due to the differential expansion of the interior mass shells , we get an acccumulation of mass near the exterior and boundary of the void , meanwhile evening out the density distribution in the interior . we also recognized this behaviour in the more complex circumstances of the lcdm void in fig . [ fig : lcdmvoid ] . to a considerable extent this is determined by the steepness of the density profile of the protovoid depression @xcite . in nearly all conceivable situations the void therefore appears to assume a _ bucket shape _ , with a uniform interior density depression and a steep outer boundary ( fig . [ fig : sphervoid ] , righthand frame ) . the development of a _ bucket _ shaped density profile may therefore be considered a generic property of cosmic voids . recently , there have been a range of studies on the issue of void density profiles , and the question whether they display universal behaviour ( see e.g. * ? ? ? * ; * ? ? ? * ) . in a range of studies , hamaus et al . ( e.g * ? ? ? * ) concluded that spherically averaged density profiles of voids do indeed imply a universal density profile , that could be parameterized by 2 factors . interestingly , these density profiles have a less prominent bucket shaped interior profile than those seen for the spherical voids in fig . [ fig : sphervoid ] . this may be understood from the fact that voids in general are not spherical ( see sec . [ sec : voidshapes ] ) , so that spherical averaging will lead to the mixing of different layers in the void s interior . the recent study by @xcite confirms this : when taking into account the shape of voids , a remarkably strong bucket void density profile appears to surface . the corresponding void expansion velocity profile confirms the above . a uniform density distribution within a void s interior directly translates into a superhubble velocity outflow , ie . an outflow in which the expansion velocity scales linear with the distance to the void s center . in the next section we will discuss this within the context of the dynamics of voids . soon after their discovery , various studies pointed out their essential role in the organization of the cosmic matter distribution @xcite , based on the realization that they have a substantial dynamical influence on their surroundings . voids have an effective repulsive influence over their surroundings ( see below ) . even though they represent relatively limited density deficits - voids will never exceed @xmath11 - the fact that they occupy a major share of cosmic volume translates into a dominant factor in the dynamical interplay of forces on megaparsec scales . the flow in and around the void is dominated by the outflow of matter from the void , culminating into the void s own expansion near the outer edge . this has indeed been recognized in various galaxy surveys and surveys of galaxy peculiar motions in our local universe . a telling illustration is the map of velocity flows implied by the pscz galaxy redshift survey ( see fig . [ fig : psczsculptor ] ) . the gravitational impact of the sculptor void on our immediate cosmic vicinity is directly visible . the influence of voids on the peculiar velocities of galaxies have been even recognized at an individual level @xcite . -0.0truecm to fully to understand the dynamical influences on the evolution of voids , and as well to appreciate the dynamical impact of voids on their environment , we may obtain substantial insight from the idealized configurations of a spherically isolated void , and that of the force field on a homogeneous ellipsoidal void . the spherical model allows us to understand the principal aspects of the flow field in the interior of voids . the ellipsoidal model allows us to evaluate the importance of external forces on the dynamical evolution of voids . in terms of their dynamics , the key observation is that external mass inhomogeneities retain a dominant role in the evolution of void . key to this realization is the fact that voids are always be limited to a rather modest density deficit of not more than @xmath6 . this will strongly limit their own local gravitational action . it also means that the implicit tidal influences induces by the large scale environment of voids represent a major factor in their dynamical evolution . their shape , mutual alignment and overall expansion are aspects that are strongly dependent on these external tidal influences . it also means that global cosmological influences retain a strong impact on the void s evolution . while highly nonlinear overdensities internally hardly notice the presence of the repulsive force induced by the presence of dark energy , in the desolate interiors of voids dark energy plays an even considerably more prominent role than on global cosmological scales @xcite . along the same line , it is in the diluted density field of a void that the imprint of possible modifications of the force of gravity will be most noticeable @xcite . while this direct dynamical influence of dark energy , and gravity modifications , in voids may be one factor of importance , there is an equally and possibly even more sizeable secondary influence . the dominant role of external tidal influences on the void s dynamics depends on the amplitude of the inhomogeneities in the large scale environment of voids . the growth rate of mass fluctuations is a sensitive function of factors such as the nature of dark energy , and the nature of gravity on large scales , as these influence the dynamical timescales involved in the formation of structure . to fully appreciate the forces at work in voids , we first looks at their own principal imprint , that of the void s expansion and the corresponding outflow velocities . subsequently , we will investigate the external influences on the basis of the ellipsoidal model . because voids are emptier than the rest of the universe , they have an outward directed velocity flow : mass is flowing out of the void . evidently , the resulting void velocity field profile is intimately coupled to that of its interior density field profile . in the situation of a mature , evolved void , the velocity field of a void resembles that of a hubble flow , in which the outflow velocity increases linearly with distance to the void center . in other words , voids are _ super - hubble bubbles _ @xcite ! the linear velocity increase is a manifestation of its general expansion . the near constant velocity divergence within the void conforms to the _ super - hubble flow _ expected for the near uniform interior density distribution that voids attain at more advanced stages ( see fig.[fig : sphervoid ] ) . it is straightforward to appeciate this from the _ continuity equation_. for a uniform density field , it tells us that the velocity divergence in the void will be uniform , corresponding to a hubble - like outflow . because voids are emptier than the rest of the universe they will expand faster than the rest of the universe , with a net velocity divergence equal to @xmath12 where @xmath13 is defined to be the ratio of the super - hubble expansion rate of the void and the hubble expansion of the universe . inspection of the velocity flow in the sculptor void ( see fig.[fig : psczsculptor ] ) shows that this is indeed a good description of the observed reality . @xcite confirmed that the velocity outflow field in viable cosmological scenarios does indeed resemble that of a superhubble expanding bubble ( see fig . [ fig : superhubble]b ) . they also established that the superhubble expansion rate is directly proportional to the nonlinear void density @xmath14 ( see fig . [ fig : superhubble]a ) , @xmath15 this relation , known within the context of a linearly evolving spherical density perturbation , in the case of fully evolved voids appears to be valid on the basis of the _ nonlinear _ void density deficit . as @xcite illustrated , voids should therefore be considered as distinctly nonlinear objects . several recent studies ( e.g. * ? ? ? * ) have confirmed this finding for voids in a range of high resolution cosmological simulations . of decisive importance for understanding the dynamics and evolution of voids is the realization that voids will never and can not decouple from their surroundings . one aspect of this is that voids expand and increase in size , they will naturally run up against their expanding peers . their spatial distribution and organization will be substantially influenced by the way in which the mutually competing voids distribute their share of space . the second strong environmental influence on the evolution of voids is that of the tidal influences induced by the large scale environment of voids . their shape , mutual alignment and overall expansion are aspects that are strongly dependent on these external tidal influences . arguably one of the most direct and transparent means of developing our insight into the tidally affected dynamics of voids is to look at the evolution of homogeneous underdense ellipsoids . this idea was forwarded by @xcite . in particular the dominant role of external tidal forces on the evolution of voids can be understood by assessing the evolution of homogeneous underdense ellipsoids , within the context of _ homogeneous ellipsoidal model _ @xcite . it is interesting to realize that the ellipsoidal model , even while idealized , is a rather good approximation for the main aspects of a void s evolution . in many respects the homogeneous model is far more suitable as an approximation for underdense regions than it is for overdense ones : voids expand and evolve towards an increasingly uniform interior density field over a vast range of their volume ( see fig . [ fig : sphervoid ] ) : while they expand their interior gets drained of matter and develops a flat `` bucket - shaped '' density profile . overdense regions contract into more compact and hence steeper density peaks , so that the area in which the ellipsoidal model represents a reasonable approximation will continuously shrink . evidently , the approximation is restricted to the interior and fails at the void s outer fringes . the latter is a result of the accumulation of material near the void s edge , and the encounter with neighouring features of the cosmic web and surrounding voids . the homogeneous ellipsoidal model assumes an object to be a region with a triaxially symmetric ellipsoidal geometry and a homogeneous interior density , embedded within a uniform background density @xmath16 . consider the simple situation of the external tidal shear directed along the principal axes of the ellipsoid . the gravitational acceleration along the principal axes of an ellipsoid with over / underdensity @xmath17 can be evaluated from the expression for the corresponding scale factors @xmath18 , @xmath19\,{\cal r}_{\rm m}\,-\,\tau_{\rm m}\ , { \cal r}_{\rm m}\,+\,\lambda r_m\ , , \label{eq : ellipseqnmot}\end{aligned}\ ] ] where we have also taken into account the influence of the cosmological constant @xmath1 . the factors @xmath20 are the ellipsoidal coefficients specified by the integral equation , @xmath21 the influence of the external ( large - scale ) tidal shear tensor @xmath22 enters via the eigenvalue @xmath23 . from eqn . [ eq : ellipseqnmot ] , it is straightforward to appreciate that as @xmath17 grows strongly nonlinear , the relative influence of the large - scale ( near-)linear tidal field will decline . however , the density deficit @xmath17 of voids will never exceed unity , @xmath24 , so that the importance of the factor @xmath23 remains relatively large . -8.0truecm -0.25truecm the impact of the external tidal forces can be so strong they not only effect an anisotropic expansion of the void , but even may manage to make an initially spherically expanding void to collapse . the latter can be seen in fig . [ fig : voidcoll ] , where we compare the evolution of the two initially spherical and equivalent void regions . the isolated one assumes the regular evolution of a spherical isolated void , while its peer develops into an increasingly anisotropic configuration . at some point the external tides go as far to squeeze the void into contraction and ultimately towards collapse . this is the situation that we find in the hierarchical buildup of voids , to be described in the next section [ sec : voidsocio ] . as illustrated in the accompanying illustration of the hierarchically evolving void population , small voids near the boundary of large void bubbles get squeezed into collapse as a result of the tidal impact of their overdense surroundings . @xcite pointed out that any ( isolated ) aspherical underdensity will become more spherical as it expands . the effective gravitational acceleration is stronger along the short axis than along the longer axes . for overdensities this results in a stronger inward acceleration and infall , producing increasingly flattened and elongated features . by contrast , for voids this translates into a larger _ outward _ acceleration along the shortest axis so that asphericities will tend to diminish . for the interior of voids this tendency has been confirmed by n - body simulations @xcite . in reality , voids will never reach sphericity . even though voids tend to be less flattened or elongated than the halos in the dark matter distribution , they are nevertheless quite nonspherical : @xcite find that they are slightly prolate with axis ratios of the order of @xmath25 . this agrees with the statement by @xcite and @xcite that in realistic cosmological circumstances voids will be nonspherical . this is also quite apparent in images of simulations , such as those in figure [ fig : voidhieradh ] . the flattening is a result of large scale dynamical and environmental factors , amongst which we can identify at least two important factors @xcite . even while their internal dynamics pushes them to a more spherical shape they will never be able to reach perfect sphericity before encountering surrounding structures such as overdense filaments or planar walls . even more important is the fact that , for voids , the external tidal influences remain important ( see discussion above ) . these external tidal forces are responsible for a significant anisotropic effect in the development of the voids . in extreme cases they may even induce fullscale collapse and demolition of the void . figure [ fig : devoids ] illustrates a typical example of the shape distribution of voids in a @xmath1cdm cosmology . the identification of voids by means of the watershed transform , which delineates the region of a void independent of its scale and shape , a direct objective measurement of the volume , shape and orientation of the void population is obtained . this may be directly inferred from the comparison between the bottom panels with the dark matter distribution in the top panel of fig.[fig : devoids ] . introduced and proposed by @xcite ( also see contribution by jones & van de weygaert in this volume ) , the watershed void finder ( wvf ) identifies voids via a watershed transform applied to the dtfe ( delaunay tessellation field estimator ) density field reconstruction @xcite . the latter exploits the scale and shape sensitivity of voronoi and delaunay tessellations to retain the intricate multiscale and weblike nature of the mass distribution probed by a discrete particle or galaxy distribution . the idea of the use of the watershed transform for the objective identification of voids , given the close relation to the topology of the weblike megaparsec matter distribution ( see * ? ? ? * ) , has in the meantime been recognized as a true watershed with respect to setting a standard definition for what should be considered as a void @xcite . large scale tidal influences not only manifest themselves in the shaping of individual voids . they are also responsible for a distinct alignment of substructures along a preferred direction , while they are also instrumental in their mutual arrangement and organization . locally , the orientation of a void turns out to be strongly aligned with the tidal force field generated by structures on scales up to at least @xmath26 . this goes along with a similar mutual alignment amongst voids themselves . they have strongly correlated orientations over distances @xmath27 @xcite , a scale considerably exceeding the typical void size . it forms a strong confirmation of the large scale tidal force field as the dominant agent for the evolution and spatial organization of the megaparsec universe , as emphasized by the cosmic web theory @xcite . -0.0truecm -0.25truecm [ fig : devoids ] a third interesting influence on the shape of vois is the possible impact of dark energy on the dynamics and evolution of voids . this influence is a result of its direct repulsive effect on the force field of the void , as well as indirectly via the external tidal force field induced by the surrounding large scale inhomogeneities . following the earlier suggestions by @xcite , studies by wandelt and collaborators @xcite showed that void shapes may be used as precision probes of dark energy . @xcite even quoted the possibility of improving the figure of the dark energy task force figure of merit by an order of hundred for future experiments like euclid . an elaborate study by @xcite of void shape evolution in simulations within a range of different dark energy cosmologies confirmed the high level of sensitivity of the shapes of dark matter voids to the underlying dark energy ( see fig . [ fig : devoids ] ) . however , it is less straightforward to reach similar conclusions of the shapes of voids in the observed galaxy distribution @xcite . @xcite forwarded the suggestion of using the alcock - paczysnki test on the stacking of voids in a sufficiently large cosmic volume . assuming that the orientation of voids is random , the shapes of properly scaled voids would average out to purely spherical . this comes along with the significant advantage of substantially increasing the signal - to - noise of the resulting void stack . given the observation of the voids in a galaxy redshift survey , as a result of the cosmic expansion the resulting void stack in redshift space will be stretched differently in the radial direction than in the transverse direction . the ratio between the transverse and radial stretching of an intrinsically spherical feature yields direct information on the angular diameter distance and hubble parameter at a redshift z. since the suggestion by @xcite , several groups realized the large cosmological potential of the application of the alcock - paczysnki test to voids in large surveys . it has also resulted in a range of interesting studies ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , and is generating substantial interest for exploiting voids in upcoming large galaxy surveys ( e.g. * ? ? ? * ) . of key importance for a succesful application of the alcock - paczysnki test to samples of voids is to correct for some systematic effects . one concerns the correction for the peculiar velocity outflow of the void itself ( see * ? ? ? * for a clear discussion ) . another effect that should be accounted for is the alignment of voids on large scales , as discussed in the previous subsection [ sec : voidalign ] ( also see fig.[fig : devoids ] ) . clearly , if the region from which the void sample is extracted is not much larger than the scale on which tidally induced alignment are expected , the stacked voids may not define an intrinsically spherical object and hence influence the outcome of the determined cosmological parameters . various studies have found strong indications for the imprint of voids in the peculiar velocity flows of galaxies in the local universe . @xcite made the first claim of seeing pushing influence of voids when assessing the stronger velocity flows of galaxies along a filament in the first cfa slice . stronger evidence came from the extensive and systematic potent analysis of mark iii peculiar galaxy velocities @xcite in the local universe @xcite . potent found that for a fully selfconsistent reconstruction of the dynamics in the local universe , it was inescapable to include the dynamical influence of voids @xcite . with the arrival of new and considerably improved data samples the dynamical influence of voids in the local universe has been investigated and understood in greater detail . the reconstruction of the density and velocity field in our local cosmos on the basis of the 2mass redshift survey has indeed resulted in a very interesting and complete view of the dynamics on megaparsec scales . this conclusion agree with that reached on the basis of an analysis of the peculiar velocity of the local group by @xcite . their claim is that the local void is responsible for a considerable repulsive influence , accounting for @xmath28 km s@xmath29 of the @xmath30 km s@xmath29 local group motion with respect to the cmb ( also see fig.[fig : localvoids ] ) . the substantial dynamical role of voids in the large scale universe has been most convincingly demonstrated in the recent advances in the study of cosmic flows enabled by the completion of the cosmicflows2 and cosmicflows3 surveys @xcite . the implied velocity - based reconstructions of the local universe , culminating in the suggestion of the hypercluster laniakea as local dynamical entity @xcite , clearly reveal the substantial repulsive influence of voids in the local universe . computer simulations of the gravitational evolution of voids in realistic cosmological environments do show a considerably more complex situation than that described by idealized spherical or ellipsoidal models @xcite . in recent years the huge increase in computational resources has enabled n - body simulations to resolve in detail the intricate substructure of voids within the context of hierarchical cosmological structure formation scenarios @xcite . they confirm the theoretical expectation of voids having a rich substructure as a result of their hierarchical buildup ( see e.g. fig . [ fig : voidhieradh ] ) . @xcite treated the emergence and evolution of voids within the context of _ hierarchical _ gravitational scenarios . it leads to a considerably richer view of the evolution of voids . the role of substructure within their interior and the interaction with their surroundings turn out to be essential aspects of the _ hierarchical _ evolution of the void population in the universe . an important guideline are the heuristic void model simulations by @xcite , and the theoretical void study by @xcite within the context of a lagrangian adhesion model description . in some sense voids have a considerably more complex evolutionary path than overdense halos . their evolution is dictate by two processes : their _ merging _ into ever larger voids as well as the _ collapse _ and disappearance of small ones embedded in overdense regions ( see fig . [ fig : voidhier ] ) . as argued by @xcite , the implied hierarchical development of voids , akin to the evolution of overdense halos , may be described by an _ excursion set _ formulation @xcite . to take account of the more complex evolutionary history , the evolution of voids needs to be described by a _ two - barrier _ excursion set formalism . the resulting hierarchical buildup of the void population resembles that of a gradually diluting soapsud . at early time , emerging from a primordial gaussian random field , the first bubbles to appear are small matured voids . the ones coexisting within the realm of a larger underdense region , will gradually merge into a larger void . this is the typical fate of a void in a _ void - in - void _ configuration . meanwhile , small voids that find themselves in overdense regions or near the boundary of larger underdensities get squeezed out of existence , the fate for a void in a _ void - in - cloud _ configuration . what remains is a sud of larger void bubbles . this sequence proceeds continuously . given the almost volume - filling nature of voids , this void hierarchical process conjures up the impression of a weblike pattern evolving as the result of continuously merging bubbles , with filaments , walls and cluster nodes at the interstices of the network ( see * ? ? ? * ; * ? ? ? * for a geometric model along these lines ) . first , consider a small region which was less dense than the critical void density value . it may be that this region is embedded in a significantly larger underdense region which is also less dense than the critical density . many small primordial density troughs may exist within the larger void region . once small density depressions located within a larger embedding underdensity have emerged as true voids at some earlier epoch , their expansion tends to slow down . when the adjacent subvoids meet up , the matter in between is squeezed in thin walls and filaments . the peculiar velocities perpendicular to the void walls are mostly suppressed so that the flow of matter is mostly confined to tangential motions . the subsequent merging of the subvoids is marked by the gradual fading of these structures while matter evacuates along the walls and filaments towards the enclosing boundary of the emerging void @xcite ( also see top row fig . [ fig : voidhier ] ) . the timescale on which the internal substructure of the encompassing void is erased is approximately the same as that on which it reaches maturity . the final result is the merging and absorption of the subvoids in the larger void emerging from the embedding underdensity . hence , as far as the void population is concerned only the large void counts , while the smaller subvoids should be discarded as such . only a faint and gradually fading imprint of their original outline remains as a reminder of the initial internal substructure . a _ second _ void process is responsible for the radical dissimilarity between void and halo populations . if a small scale minimum is embedded in a sufficiently high large scale density maximum , then the collapse of the larger surrounding region will eventually squeeze the underdense region it surrounds : the small - scale void will vanish when the region around it has collapsed completely . alternatively , though usually coupled , they may collapse as a result of the tidal force field in which they find themselves . if the void within the contracting overdensity has been squeezed to vanishingly small size it should no longer be counted as a void ( see fig . [ fig : voidhier ] , bottom row ) . when inspecting the evolving matter distribution in and around voids , such as in the high resolution configuration of fig . [ fig : voidhieradh ] , we find that most of the collapsing or squeezed voids are small voids to be found in and near the boundary regions of large underdense void regions . the small voids in these regions are strained by the high density boundary regions , mostly filaments and clusters , and in many situations squeezed out of existence . on the other hand , subvoids in the interior of a larger void tend to merge with surrounding peers . a detailed assessment of the evolution of these regions reveals that the void collapse process is an important aspect of the evolution and buildup of the cosmic web . interestingly , recent observational studies have indeed identified the existence of collapsing void regions by studying the redshift structure around voids in the sdss galaxy redshift survey @xcite . the collapse of small voids is an important aspect of the symmetry breaking between underdensities and overdensities . in the primorial universe , gaussian primordial conditions involve a perfect symmetry between under- and overdense . any inspection of a galaxy redshift map or an n - body simulation shows that there is a marked difference between matter clumps and voids . while the number density of halos is dominated by small objects , void collapse is responsible for the lack of small voids . marked by the two processes of merging and collapse , the complex hierarchical buildup of the void population may be modelled by a two - barrier excursion set formalism @xcite . the barriers refer to the critical ( linear ) density thresholds involved with the _ merging _ and _ collapse _ of voids . whenever the linearly extrapolated ( primordial ) @xmath31 on a scale @xmath32 , @xmath33 exceeds a density threshold @xmath34 it will collapse . for an einstein - de sitter @xmath35 universe the critical value has the well - known value @xmath36 @xcite . a void will form when an underdensity reaches the critical density threshold of _ shell - crossing _ , corresponding to a value of @xmath37 for spherical voids in an einstein - de sitter universe . the linear density growth factor @xmath38 , normalized to unity at the present epoch , follows from the integral @xcite , @xmath39 emerging from a primordial gaussian random field , many small voids may coexist within one larger void . small voids from an early epoch will merge with one another to form a larger void at a later epoch . the excursion set formalism takes account of this _ void - in - void _ configuration by discarding these small voids from the void count . to account for the impact of voids disappearing when embedded in collapsing regions , the two - barrier formalism also deals with the _ void - in - cloud _ problem . by contrast , the evolution of overdensities is governed only by the _ cloud - in - cloud _ process : the _ cloud - in - void _ process is much less important , because clouds which condense in a large scale void are not torn apart as their parent void expands around them . this asymmetry between how the surrounding environment affects halo and void formation is incorporated into the _ excursion set approach _ by using one barrier to model halo formation and a second barrier to model void formation . only the first barrier matters for halo formation , but both barriers play a role in determining the expected abundance of voids . the analytical evaluation of the two - barrier random walk problem in the extended press - schechter approach leads directly to a prediction of the distribution function @xmath40 for voids on a mass scale @xmath41 ( or corresponding void size @xmath32 is done by assuming the simplest approximation , that of the spherical tophat model . according to this model a void has expanded by a factor of 1.7 by the time it has mature , so that @xmath42 . ] ) . the resulting void spectrum is nearly equal to the corresponding void mass deficit . ] . is peaked , with a sharp cutoff at both small and large values of the peak mass @xmath43 ( fig . [ fig : voidhiersvdw ] , bottom lefthand frame ) , @xmath44 where @xmath45 is the fractional relative underdensity , @xmath46 in which with the dependence on the mass scale @xmath41 entering via the rms density fluctuation on that scale , @xmath47 . the quantity @xmath48 is the `` _ void - and - cloud parameter _ '' , it parameterizes the impact of halo evolution on the evolving population of voids : the likelihood of smaller voids being crushed through the _ void - in - cloud _ process decreases as the relative value of the collapse barrier @xmath50 with respect to the void barrier @xmath51 becomes larger . assessment of the relation above shows that the population of large voids is insensitive to the _ void - in - cloud _ process . the large mass cutoff of the void spectrum is similar to the ones for clusters and reflects the gaussian nature of the fluctuation field from which the objects have condensed . characteristic void size _ increases with time : the gradual merging of voids into ever larger ones is embodied in a self - similar shift of the peak of the void spectrum . when evaluating the corresponding fraction of contained in voids on mass scale m , @xmath52 , we find that this also peaks at the characteristic void scale . it implies a mass fraction in voids of approximately thirty percent of the mass in the universe , with most of the void mass to be found in voids of this characteristic mass . while the two - barrier excursion set formalism offers an attractive theoretical explanation for the distinct asymmetry between clumps and voids and for the peaked void size distribution , realistic cosmological simulations are needed to identify where the disappearing small - scale voids are to be found in a genuine evolving cosmic matter distribution . an important aspect of the implied void population is that it is approximately _ space - filling_. it underlines the adagio that the large scale distribution of matter may be compared to a _ soapsud of expanding bubbles_. this follows from evaluation of the cumulative integral @xmath53 where the factor 1.7 is an estimate of the excess expansion of the void based upon the spherical model for void evolution ( see footnote ) . the top righthand panel of fig . [ fig : voidhiersvdw ] shows the resulting ( current ) cumulative void volume distribution : for a finite value of void radius @xmath32 the whole of space indeed appears to be occupied by voids , while there is the gradual shift of the cumulative volume distribution towards larger voids . in other words , the correct image appears to be that of a gradually unfolding bubbly universe in which the average size of the voids grows as small voids merge into ever larger ones . while the svdwmodel has proven to provide a good model for the evolving void population , it should be considered as a basic framework . substantial finetuning to realistic circumstances , devoid of spherical symmetry and other implicit simplifications , will likely improve the model . one major issue concerns the value of the typical void formation threshold or density barrier . the simple choice for the shellcrossing of a hypothetical spherical configuration certainly deserves scrutiny , and considerable improvements to this have been forwarded @xcite . also , to connect the svdw formalism to the voids in the observed galaxy distribution involves a range of factors . galaxies are diluted and biased tracers of the underlying dark matter distribution . while the dark matter voids define a rich and complex structural pattern ( see e.g. * ? ? ? * ; * ? ? ? * ) , voids in the galaxy distribution will only represent a limited aspect of the void populations . @xcite elaborated on the svdw formalism to describe what it would imply for voids in the galaxy distribution . the advantage of having an analytical formalism for describing the void population has recently been recognized in the context of casting constraints on the nature of dark energy . by evaluating the implied void population in a range of dark energy scenarios , @xcite demonstrated that in upcoming surveys such as euclid it will be possible to infer competitive estimates of the equation of state of dark energy , as well as on its time derivative . the use of the analytical description of the two - barrier excursion set model provides highly versatile and flexible framework for readily identifying the most sensitive factors and situations . the same philosophy was followed in the elaboration by @xcite , who applied the excursion set formalism towards a general description of the evolving void population in modified gravity scenarios . an important issue within the hierarchically proceeding evolution of voids and the cosmic web is the fate of its substructure . in voids the diluted and diminished infrastructure remains visible , at ever decreasing density contrast , as cinders of the earlier phases of the _ void hierarchy _ in which the substructure stood out more prominently . n - body simulations show that voids do retain a rich infrastructure . examples such as the fig . [ fig : voidhieradh ] , and images of large cosmological simulations such as the millennium and the illustris simulation @xcite show that while void substructure does fade , it does not disappear . we may find structures ranging from filamentary and sheetlike structures to a population of low mass dark matter halos and galaxies ( see e.g. * ? ? ? although challenging , these may also be seen in the observational reality . the galaxies that populate the voids do currently attract quite some attention ( see next section ) . also , the sdss galaxy survey has uncovered a substantial level of substructure within voids , such as a the vgs31 filament by @xcite . the gradual fading of the initially rich substructure as a void expands , and its matter rcontent evacuates its interior , is clearly visible in the development of the void region in fig . [ fig : lcdmvoid ] . this may be seen as the slowing of structure growth in the low - density environment of voids . the resulting observation is one in which the density distribution in a void region undergoes a manifest and continuous transformation towards an ever larger dominating scale , while it mostly retains the topological features of the initial density field . in a sense , it reflects of the buildup of the cosmic web itself , its basic pattern already recognizable in the primordial matter distribution @xcite . i wish to thank the many collaborators and students that over many years have accompanied me in my explorations of the desolate but wildly interesting void regions in our universe . in particular i wish to thank ravi sheth , johan hidding , marius cautun , patrick bos , bernard jones , erwin platen , mathijs dries , francisco kitaura and vincent icke for major contributions to the work described in this paper . rvdw is particularly grateful to j. hidding and f. kitaura for the use of the images in figures 3 and 4 in advance of publication and to j. hidding for figure 11 in advance of publication . aragon - calvo m.a . , platen e. , van de weygaert r. & szalay a.s . 2010 , _ apj _ 723 , 364 aragon - calvo m.a . , van de weygaert r. , araya - melo p.a . , platen e. & szalay a.s . 2010 , _ mnras _ 404 , 89 aragon - calvo m.a . , szalay a.s . 2013 , _ mnras _ 28 , 3409 aragon - calvo m.a . 2014 , arxiv:14409.8661 bardeen j.m . , , kaiser n. & szalay a.s . ( bbks ) 1986 , _ apj _ , 304 , 15 e. 1985 , _ apj _ 295 , 1 e. , dekel a , faber s.m . , dressler a. & burstein d. 1990 , _ apj _ 364 , 370 beygu b. , kreckel k. , van de weygaert r. , van der hulst j.m . & van gorkom j.h . 2013 , _ aj _ 145 , 120 biswas r. alizadeh e. & wandelt b. 2010 , _ phrvd _ 82 , 3002 g.r . , da costa l. , goldwirth d.s . , lecar m. & piran t. , 1992 , _ apj _ 388 , 324 j.r . , cole s. , efstathiou g. & kaiser n. 1991 , _ apj _ 379 , 4440 j.r . & myers s.t . 1996 _ apjs _ , 103 , 1 j.r . , kofman l. & pogosyan d.yu . 1996 _ nature _ 380 , 603 bos e.g.p . , van de weygaert r. , dolag k. & pettorino v. 2012 , _ mnras _ 426 , 440 g.d . , geller m.j . , kurtz m.j . , huchra j.p . & schild r.e . 1992 _ apj _ 395 , 347 cai y - c .. padilla n. & li b. 2015 , _ mnras _ 451 , 1036 cautun m. , van de weygaert r. , jones b.j.t . , & frenk c.s . 2014 , _ mnras _ , 441 , 2923 cautun m. , cai y - h . & frenk c.s . 2016 , _ mnras _ 457 , 2540 l. , padilla n.d . , valotto c. & lambas d.g 2006 _ mnras _ 373 , 1440 g. & rood h.j . 1975 _ nature _ 257 , 294 clampitt j. , cai y - c . , li b. 2013 , _ mnras _ 431 , 749 j.m . , sheth r.k . , diaferio a. , gao l. & yoshida n. 2005 , _ mnras _ 360 , 216 j.m . et al . 2008 , _ mnras _ 387 , 933 m. , et al . 2003 , arxiv:0306581 courtois h. , hoffman y. , tully b. & gottlber s. 2012 , _ apj _ 744 , 43 a. , bertschinger e. & faber s.m . 1990 , _ apj _ 364 , 349 a. 1994 , _ araa _ 32 , 371 a. & rees m.j . 1994 , _ apjl _ 443 , l1 v. 2008 , _ mnras _ 388 , 638 j. , da costa l.n . , goldwirth d.s . , lecar m. & piran t. 1993 , _ apj _ 410 , 458 j. , joeveer m. & saar e. 1980 , _ nature _ 283 , 47 einasto j. , et al . 2011 , _ a&a _ 534 , 128 d.j . & loeb a. 1995 , _ apj _ 439 , 520 a.p . 1998 , _ large - scale structures in the universe _ ( wiley ) s.r . & piran t. 2006 , _ mnras _ 366 , 467 furlanetto s. , peng oh.s . & briggs f. 2006 , _ phys . rep . _ 433 , 181 d.m . & vogeley m.s . 2004 , _ apj _ 605 , 1 s. , okas e.l . , klypin a. & hoffman y. 2003 , _ mnras _ 344 , 715 b.r . , neyrinck m.c . & szapudi i. 2008 , _ apj _ 701 , 414 s.a . & thompson l.a . 1978 , _ apj _ 222 , 784 j.e . & gott j.r . 1972 , _ apj _ 176 , 1 guzzo l. & teh vipers team 2013 , _ the messsenger _ , 151 , 41 guzzo l. , et al . 2014 , _ aa _ , 566 , 108 n. , sutter p.m. & wandelt b.d . 2014 , _ phrvl _ 112 , 1302 a.j.s . 2001 , _ mnras _ 322 , 419 d.j . 1977 , _ mnras _ 179 , 351 hess s. , kitaura f .- s . & gottlber s. 2013 , _ mnras _ , 435 , 2065 hidding j. , van de weygaert r. , vegter g. , jones b.j.t . & teillaud m. 2012 , video publication socg12 , symposium on computational geometry , arxiv1205.1669 hidding j. , van de weygaert r. , kitaura f .- s . & hess s. 2016 _ mnras _ , to be subm . hidding j. 2016 , phd thesis , univ . groningen y. & shaham j. 1982 , _ apjl _ 262 , l23 j.p . et al . 2012 , _ apjs _ , 199 , 26 v. 1973 , _ a&a _ , 27 , 1 v. 1984 , _ mnras _ 206 , 1p v. & van de weygaert r. 1987 , _ a&a _ 18 , 16 e. , li y. & hu w. 2013 , _ mnras _ 434 , 2167 i.d . , karachentseva v.e . , huchtmeier w.k . & makarov d.i . 2004 , _ aj _ 127 , 2031 r.p . , oemler a. , schechter p.l . & shectman s.a . 1981 , _ apjl _ 2448 , l57 r.p . , oemler a. , schechter p.l . & shectman s.a . 1987 , _ apj _ 314 , 493 kitaura f .- s . 2012 , _ mnras _ , 429l , 84 kreckel k. , platen e. , aragon - calvo m.a . , van gorkom j.h . , van de weygaert r. , van der hulst j.m . , kovac k. , yip c .- w . & peebles p.j.e . 2011 , _ aj _ 141 , 4 kreckel k. , platen e. , aragon - calvo m.a . , van gorkom j.h . , van de weygaert r. , van der hulst & j.m . beygu b. 2012 , _ aj _ 144 , 16 o. & suto y. 2004 , _ living reviews in relativity _ 7 , 82pp lavaux g. & wandelt b. 2010 , _ mnras _ 403 , 1391 lavaux g. & wandelt b. 2012 , _ apj _ 754 , 109 v. , geller m.j . & huchra j.p . 1986 , _ apjl _ 302 , l1 leclercq f. , jasche j. , sutter p.m. , hamaus n. & wandelt b. 2015 , _ jcap _ , 03 , 047 leclercq f. 2015 , _ bayesian large - scale structure inference and cosmic web analysis _ , phd thesis , univ . pierre et marie curie , institut dastrophysique de paris j. & park d. 2009 , _ apj _ , 696 , 10 li b. 2011 , _ mnras _ 411 , 2615 li b. & efstathiou g. 2012 , _ mnras _ 421 , 1431 h. & wasserman i. 1990 , _ apj _ 348 , 1 h. & white s.d.m . 2002 , _ mnras _ 337 , 1193 morales m.f . & wyithe j.s.b . 2010 , _ araa _ 48 , 127 nadathur s. & hotchkiss s. 2015 , _ mnras _ 454 , 2228 nadathur s .. 2016 , _ mnras _ in press , arxiv:1602.04752 neyrinck m. 2008 , _ mnras _ 386 , 2101 n.d . , ceccarelli l. & lambas d.g . 2005 , _ mnras _ 363 , 977 p.l & voglis n. 1983 , _ mnras _ 205 , 543 pan d. , vogeley m.s . , hoyle f. , choi y .- y . & park c. 2012 _ mnras _ 421 , 926 a. , lam t.y . & sheth r.k . 2012 , _ mnras _ 420 , 1648 d. & lee j. 2007 , _ phys . _ 98 , 081301 d. , lares m. , ceccarelli l. , padilla n. & lambas d.g . 2013 , _ mnras _ 436 , 3480 p. j. e. 1980 , _ the large - scale structure of the universe _ ( princeton university press ) p.j.e . 2001 , _ apj _ 557 , 495 peebles p.j.e . , nusser a. 2010 , _ nature _ 465 , 565 pisani a. , lavaux g. , sutter p.m. & wandelt b.d . 2014 , _ mnras _ 443 , 3238 pisani a. 2014 , _ cosmology with cosmic voids _ , phd thesis , univ . pierre et marie curie , institut dastrophysique de paris pisani a. , sutter p.m. , hamaus n. , alizadeh e. , biswas r. , wandelt b.d . & hirata c.m . 2015 , _ phrvd _ 92 , 3531 e. , van de weygaert r. & jones b.j.t . 2007 , _ mnras _ 380 , 551 e. , van de weygaert r. & jones b.j.t . 2008 , _ mnras _ 387 , 128 pontzen a. , slosar a. , roth n. & peiris h.v . 2016 , _ phrvd _ 93 , 3519 w.h . & schechter p. 1974 , _ apj _ 187 , 425 e. & geller m.j . 1991 , _ apj _ 373 , 14 rieder s. , van de weygaert r. , cautun m. , beygu b. , portegies - zwart s. 2013 _ mnras _ 435 , 222 l. , brown s. & williams l.r . 2007 , _ apj _ 671 , 40 b.s . & melott a.l . 1996 , _ apj _ 470 , 160 v. , sathyaprakash b.s . & shandarin s.f . 1994 , _ apj _ 431 , 20 v. & shandarin s.f . 1996 , _ mnras _ 282 , 641 w. e. & van de weygaert r. 2000 , _ a&a _ 363 , l29 s. , feldman h.a . , heitmann k. & habib s. 2006 _ mnras _ 376 , 1629 r.k . 1998 , _ mnras _ 300 , 1057 r. k. & van de weygaert r. 2004 , _ mnras _ 350 , 517 v. et al . 2005 , _ nature _ 435 , 629 sutter p. , lavaux g. , wandelt b.d & weinberg d.h . 2012 , _ apj _ 761 , 44 sutter p. , lavaux g. , wandelt b.d & weinberg d.h . 2012b , _ apj _ 761 , 187 sutter p. , pisani a. , wandelt b.d & weinberg d.h . 2014 , _ 443 _ , 2983 sutter p.m. , elahi p. , falck b. , onions j. , hamaus n. , knebe a. , srisawat c. , schneider a. 2014b , _ mnras _ 445 , 1235 sutter p.m. , et al . 2015 _ a & c _ 9 , 1 szapudi i. , et al . 2015 , _ mnras _ 450 , 288 m. & sdss collaboration 2004 , _ apj _ 606 , 702 r.b . , shaya e.j . , karachentsev i.d . , courtois h. , kocevski d.d . , rizzi l. & peel a. 2008 , _ apj _ 676 , 184 tully r.b . , courtois h. , hoffman y. & pomarede d. 2014 , _ nature _ 513 , 71 r. 1991 , _ voids and the geometry of large scale structure _ , phd thesis , leiden university r. & van kampen e. 1993 , _ mnras _ 263 , 481 van de weygaert r. & bertschinger e. 1996 , _ mnras _ , 281 , 84 van de weygaert r. & bond j.r . 2008 , _ clusters and the theory of the cosmic web _ , in m. plionis , o. lpez - cruz & d. hughes eds . , _ a pan - chromatic view of clusters of galaxies and the large - scale structure _ , lnp 740 ( springer ) , p. 335 r. & platen e. 2011 , _ ijmps _ 1 , 41 , arxiv:0912.2997 r. & schaap w.e 2009 , in v. martnez et al . _ data analysis in cosmology _ , lnp 665 , 291 s.d.m . & silk j. 1979 , _ apj _ 231 , 1 j.a . , courteau s. , faber s.m . , burstein d. , dekel a. & strauss m.a . 1997 , _ apjs _ 109 , 333 wojtak r. , powell d. & abel t. 2016 , _ mnras _ 458 , 4431 zeldovich ya . b. , einasto j. & shandarin s.f . 1982 , _ nature _ 300 , 407

we have carried outco j=21 line observations of the supernova remnant ( snr ) hb 21 in order to search for evidence of interaction with molecular clouds . we mapped the eastern half ( ) of the snr almost completely . molecular gas appears to be distributed mainly along the boundary of the snr , but the overall distribution has little correlation either with the distortion of the snr boundary or with the distribution of radio brightness . along the eastern boundary , where the snr was considered to be interacting with molecular clouds in previous studies , we have not found any strong evidence for the interaction . instead we detected broad ( 2040 ) co emission lines in the northern and southern parts of the snr . in the northern area , the broad - line emitting cloud is composed of a small ( or 0.5 pc ) , very bright , u - shaped part and several clumps scattered around it . there is a significant enhancement of radio emission with flat ( ) spectral index possibly associated with this cloud . in the southern area , the broad - line emitting cloud is filamentary and appears to form an elongated loop of in extent . small ( or 0.3 pc ) , bright clumps are seen along the filamentary structure . we have obtained sensitive j=10 and j=21 spectra ofco andco molecules toward several peak positions . the intensity ofco j=21 emission is low ( k ) and the ratio ofco j=21 to j=10 integrated intensities is high ( 1.62.3 ) , which suggests that the emission is from warm , dense , and clumpy gas . we have applied an lvg analysis to derive their physical parameters . the detected broad co lines are believed to be emitted from the fast - moving molecular gas swept - up by the snr shock . the small ( ) shock velocity suggests that the shock is a non - dissociating c - shock . we discuss the correlation of the shocked molecular gas with the previously detected , shocked atomic gas and the associated infrared emission .

astro-ph0012341

the number of supernova remnants ( snrs ) with convincing evidence for interaction with ambient molecular clouds has increased considerably in recent years . the evidence ranges from a simple morphological relation to the detection of broad and/or shock - excited emission lines from various molecules . although circumstantial evidence could be very suggestive , it is the molecular lines from the shocked gas that are essential for understanding the physical and chemical processes associated with the molecular shock . in this regard , there are still only a few snrs adequate for the study of molecular cloud - shock interaction ; perhaps w28 ( arikawa et al . 1999 ) , w44 ( seta et al . 1998 ) , w51c ( koo & moon 1997 ) , 3c391 ( reach & rho 1996 , 1999 ) , and the classical source ic 443 ( denoyer 1979 ; tauber et al . 1994 and references therein ) . in this paper , we report the discovery of broad emission lines from the shocked co gas in hb 21 . hb 21 ( g89.0 + 4.7 ) is one of those snrs with mixed morphology , e.g. , shell - like in radio and center - filled in the x - ray ( rho & petre 1998 ) , where the center - filled , thermal x - ray emission is suggested to be due to interaction with molecular clouds . it has a nearly complete radio - continuum shell with an angular extent of @xmath9 ( hill 1974 ; tatematsu et al . 1990 , hereafter t90 ) . the shell is elongated along the northwest - southeast direction . the brightness distribution of the shell is not uniform , but enhanced in scattered areas . particularly noticeable features are the v - shaped northern boundary , a @xmath4-sized loop structure in the south central area , and the one in the central region of the eastern boundary ( see fig . 1 ) . optical nebulosity associated with hb 21 has not been detected in h@xmath10 or [ sii ] plates ( van den bergh 1978 ) . x - ray emission from hb 21 was detected by leahy ( 1987 ) and studied in detail by leahy & aschenbach ( 1996 ) . the distance to the snr is uncertain . we adopt 0.8 kpc following t90 , which is the distance to the cyg ob 7 complex ( humphreys 1978 ) . hb 21 has been a suspect for interaction with molecular clouds based on its radio appearance and the ambient molecular clouds ( erkes & dickel 1969 ; huang & thaddeus 1986 ; t90 ) . erkes & dickel ( 1969 ) suggested that the distorted boundaries with enhanced radio brightness might be the places where the snr is interacting with dense ambient gas . huang & thaddeus ( 1986 ) found that the giant molecular cloud associated with cyg ob 7 appears to be partially surrounding hb 21 ( see also dobashi et al . t90 obtained a higher - resolution ( @xmath11 ) co map of the eastern part of the snr and found that the eastern boundary of the snr appeared to be in contact with molecular and atomic clouds . they also made a coarsely - sampled map of the regions with enhanced radio emission and detected molecular clumps . but no direct evidence for the shocked _ molecular _ gas has been detected for hb 21 . the search for oh masers , which are known to be an indicator for the interaction between a snr and molecular cloud , gave negative results too ( frail et al . on the other hand , koo & heiles ( 1991 ) detected shocked h i gas moving at 40 to 120 in the southern part of the snr , although the limited angular resolution ( @xmath12 ) hindered any detailed study of the shocked gas . we detected broad co emission lines in the northern and southern parts of hb 21 , but not in the eastern part . we have found no evidence suggesting that the molecular clouds in the eastern part are interacting with the snr , although they appear to be located along the boundary of the snr on the sky . we summarize the observations and the results of molecular line observations in 2 and 3 , respectively . in 4 . we discuss the implications of our results on the interaction between the snr and molecular clouds . 5 summarizes the main results of our paper . @xmath0co j=21 line observations were carried out using the 12 m telescope of the national radio astronomy observatory at kitt peak in 1999 june and 2000 january . the fwhm of the telescope at 230 ghz was 27@xmath13 . we mapped the eastern half ( @xmath1 ) of the snr almost completely using the otf ( on - the - fly ) observing technique . we used two 256 channel filter banks ; one with 500 khz , and the other with 1 mhz resolution . we split each filter bank into two sections and observed two linear polarizations simultaneously . the velocity resolution and coverage of the 500 khz filter bank were 0.65 and 83 , while those of the 1 mhz filter bank were two times greater . typical system temperatures were 350450 k. during the observing run in january 2000 , we had some trouble because of telluric co j=21 emission , which appeared at @xmath145 . the emission was not cancelled out completely by usual position - switching observation and produced a hill - and - valley feature in spectra , the strength of which depends on elevation . we were able to avoid the contamination from telluric co emission by averaging out the contaminated velocity channels because the telluric co emission line is narrow ( 0.8 ) and the absolute strength of the ` hill ' and ` valley ' features are equal . we also obtained sensitive spectra of @xmath6co j=21 , @xmath0co j=10 and j=21 lines toward several peak positions . for co j=10 observations , we used 1 and 2 mhz filter banks , so that they have the same velocity resolution and coverage with those of co j=21 line observations . we also used the millimeter autocorrelator ( mac ) for the @xmath0co observations , which provided high resolution ( 0.25 after smoothing ) spectra . we have converted the observed temperatures ( @xmath15 ) to the main - beam brightness temperatures ( @xmath16 ) using the corrected main - beam efficiency provided by the nrao . additional observations of @xmath0co and @xmath6co j=10 lines were performed using the taeduk radio astronomy observatory ( trao ) 13.7 m telescope ( hpbw=49@xmath13 at 115 ghz ) in 2000 january and march . an sis receiver equipped with a quasi - optical sideband filter was used along with a 250 khz , 256-channel filter bank . the main beam efficiency was 0.41 at 115 ghz ( roh & jung 1999 ) , and the pointing accuracy was better than 10@xmath13 . typical system temperatures were about 750 k at 115 ghz and 450 k at 110 ghz . figure 1 shows the distribution of the integrated intensity of co j=21 emission . the velocity range is between @xmath17 and @xmath18 , which covers most of the emission . the overlaied contour map shows the 1420 mhz brightness distribution of hb 21 ( t90 ) . the overall distribution of co gas in figure 1 is not very different from the low - resolution map of t90 . but figure 1 shows much detailed structure because of its high resolution , high sensitivity , and complete sampling . figure 1 immediately shows that molecular gas is distributed mainly along the boundary of the snr . ( we observed the central area in figure 1 , which had not been covered in our co j=21 observations , in co j=10 line emission , and have detected only several small [ @xmath19 clumps other than some faint extension associated with the clouds in southern and eastern parts of the remnant . ) but the overall distribution has little correlation either with the distortion of snr boundaries or with the distribution of radio brightness ( cf . 3.2 ) . for the purpose of discussion , we divide the remnant into three areas ( fig . 2 ) : ( 1 ) eastern area ( ra@xmath20 20@xmath21 47@xmath22 ) where three relatively large ( @xmath23 ) clouds and several filamentary clouds are present , ( 2 ) northern area centered at ( 20@xmath21 46@xmath22 , 51@xmath24 00@xmath25 ) , where a small ( @xmath2 ) , very bright u - shaped cloud is noticeable , and ( 3 ) southern area centered at ( 20@xmath21 44@xmath22 , 49@xmath24 50@xmath25 ) , where clumpy and filamentary clouds with complicated structures are present . we detected broad ( 2040 ) emission lines from the clouds in the northern and southern areas , which will be discussed in detail in the next sections . in the following , we summarize the results on the eastern area . in the eastern area , there are three clouds centered at declinations @xmath26 49@xmath25 , @xmath27 15@xmath25 , and @xmath28 @xmath29 . we call these three clouds by clouds a , b , and c following t90 ( see fig . 2 for the location of these clouds ) . figure 3 shows the channel maps of the eastern area . the velocity ranges of the channel maps were chosen to show the essential features clearly . cloud a appears at @xmath30 to @xmath31 and is composed of two velocity components centered at + 6 and @xmath32 , respectively . the former component ( 6 ) , which is seen in figure 3a , is extended and the emission peaks at the southern part ( @xmath33 45@xmath25 ) of the cloud , while the latter component ( @xmath32 ) , which is seen in figures 3b and 3c , is spatially confined and comprises the northern part of the cloud . their maximum brightnesses are @xmath34 and 7 k , respectively . cloud b appears at @xmath35 to @xmath36 and is seen in figures 3c and 3d . the south central part of the cloud , e.g. , the region between @xmath3717@xmath25 , is bright and appears to be connected to cloud c. cloud c has two components at very different velocities , e.g. , one at + 2 to @xmath38 , which is seen in figures 3c and 3d , and the other at + 17 to + 10 , which is not shown in figure 3 but has a distribution similar to the other velocity component . according to the result of t90 , cloud c extends to @xmath39 . an interesting feature in figure 3 is the semi - circular loop that appears above cloud b in figure 3b . the ratio of the minor , which is along the ns direction , to the major axis is 0.8 . if it is at 0.8 kpc , the linear size of the semimajor axis would be @xmath40 pc . the velocity increases systematically from both ends to the northern top of the loop , which is consistent with an expanding loop . the top portion is redshifted with respect to the both ends by @xmath41 . if we assume that the ellipticity is due to projection , then the expansion velocity would be @xmath42 , so that the dynamical age of the ring is probably shorter than @xmath43 yrs . this is much greater than the age ( @xmath44 yrs , koo & heiles 1991 , scaled to 0.8 kpc adopted in this paper ) of hb 21 and , therefore , the loop might not be associated with hb 21 . we suspect that the loop is originated from some energetic phenomena in cloud b. a faint v - shaped structure that connects cloud b and the ends of the loop in figure 3b seems to indicate the association of the two . the cloud in the northern area is composed of a small ( @xmath2 ) , very bright , u - shaped cloud and several clumps scattered around it ( fig . 1 ; see also fig . 9 for enlarged view ) . figure 4 shows its velocity structure . there are several points to be made from figure 4 : first the u - shaped cloud is composed of several clumps , whose central velocities shift systematically from + 3 to @xmath31 as we move from ne to nw along the structure . the integrated intensity attains a maximum at ( 20@xmath21 46@xmath22 [email protected] , 51@xmath24 00@xmath25 00@xmath13 ) , which we call hb21:bml - n1 ( broad molecular line northern position 1 ) , or simply n1 . second , there are several other clumps in the field . these clumps , except the one near the southeastern corner at @xmath46 , appear over a wide ( @xmath47 ) velocity range . among them , the one at ( 20@xmath21 45@xmath22 55.@xmath450 , 51@xmath24 03@xmath25 30@xmath13 ) , which we call hb21:bml - n2 ( or n2 ) , appears over the widest ( @xmath48 ) velocity interval . third , there is a diffuse emission at @xmath49 to the ne of the cloud . its line is narrow ( 23 ) and it is part of a large ( @xmath50 ) cloud that appears to be connected to cloud a. we consider that the clumps aligned along the ne - sw direction in figure 4 are associated and call them cloud n , i.e. , cloud n does not include the diffuse emission in the northeastern area and the clump in the southeastern corner . t90 detected in this area only the diffuse molecular gas at @xmath51 to 6 and called it cloud d. ) as can be expected from figure 4 , most clumps in cloud n have broad emission lines . as an example , we show the spectra of n1 and n2 in the top frames in figure 5 , where we see that the spectrum of n1 is box - shaped and its full width ( at zero intensity ) is 30 km / s , while that of n2 is asymmetric and extends from @xmath52 to + 11 km / s . for comparison , the spectrum of the diffuse , extended structure in the northeastern part of this area has narrow ( 23 ) emission lines centered at @xmath49 , a sample of which is shown in the right bottom frame in figure 5 . we have obtained sensitive j=10 and j=21 spectra of @xmath0co and @xmath6co molecules at the two peak positions , n1 and n2 , and figure 5 shows the spectra . the molecule and transition are marked in each spectrum . the second spectrum from the top is @xmath0co j=21 emission convolved to the j=10 beam size ( @xmath53 ) to be compared with the j=10 spectra . the difference between the top and convolved spectra indicates that some velocity components , e.g. , the narrow component centered at + 4 of n1 and the broad component at @xmath54 of n2 , are confined to small areas . by comparing the j=10 and the convolved j=21 spectra , we notice that the ratio of j=21 to j=10 intensities is high and that it varies over the profile : for n1 , the ratio is between 1.2 and 2.0 in the central parts of the spectrum , while it increases at the wings , e.g. , @xmath55 at + 5 and @xmath5613 . for n2 , the ratio varies between 0.8 and 2.8 , and it is higher between @xmath38 and 0 . the ratios of the @xmath0co j=21 and j=10 integrated intensities @xmath57 are 1.6 and 1.7 for n1 and n2 , respectively . the @xmath6co j=21 line is clearly detected toward n1 , while it is marginally detected toward n2 . for n1 , the line has double peaks centered at + 1 and @xmath36 , while the @xmath0co j=21 line profile toward n1 is composed of several narrow peaks . the narrow peaks might indicate that the emission is from several , unresolved subclumps . presumably , the @xmath6co emission might be from these subclumps too , which is not apparent in the profile in figure 5 because of low signal - to - noise ratio and low velocity resolution . ( note that the velocity resolutions of the @xmath0co j=2 - 1 and @xmath6co j=2 - 1 lines are 0.25 and 0.68 , respectively . ) the ratios of @xmath0co j=21 to @xmath6co j=21 integrated line intensities @xmath58 are @xmath59 and @xmath60 for n1 and n2 , respectively . ( the errors are statistical errors . ) table 1 summarizes the line parameters of the peak positions , i.e. , their coordinate , velocity range ( @xmath61 , @xmath62 ) , co j=21 peak brightness temperature @xmath63 , @xmath57 , @xmath58 , and the ratio of @xmath0co j=10 to @xmath6co j=10 integrated line intensities @xmath64 . the detailed line diagnostics based on the observed line parameters in table 1 is discussed in 3.4 . in the southern part of the snr , the emission is detected at @xmath65 to @xmath66 . the velocity structure is shown in figure 6 . at positive velocities , we see several clouds with narrow lines come and go , e.g. , a diffuse cloud that extends @xmath67 along the ns direction centered at ( 20@xmath21 [email protected] , 50@xmath24 00@xmath25 ) between @xmath68 and + 5 . the co distribution at negative velocities is fairly complicated : the distribution is filamentary , and small ( @xmath5 or 0.3 pc ) , bright clumps are seen along the filamentary structure . the filamentary structure , which we call cloud s , appears to form a loop of @xmath69 in extent , elongated along the ns direction . the eastern part of the loop is particularly clumpy and has a semicircular shape ( see the channel map centered at @xmath70 ) . the clumps generally have broad ( @xmath71 ) lines . among them , three clumps marked by crosses in figure 6 have broadest ( 3040 ) lines and we show the @xmath0co j=21 and j=10 spectra at their peak positions , which we call s1 , s2 , and s3 from east to west ( see fig . 2 ) , in figure 7 . again we show the convolved j=21 spectra together , although the line shapes do not change significantly by convolution toward these peak positions . we have obtained some sensitive j=10 and j=21 spectra of @xmath0co and @xmath6co molecules at these peak positions , and the line parameters are listed in table 1 . note that @xmath57(=1.72.3 ) and @xmath72 toward s1 ) are similar to those of northern positions , while @xmath64 s , although they have large uncertainties , appear to be much greater than that of n1 . the broad co lines are presumably emitted from the shocked gas , where physical parameters vary greatly over a short distance scale . but still it would be worthwhile to estimate their excitation parameters based on elementary considerations . first , the observed @xmath7340 are significantly less than either the average ratio ( @xmath74 ; langer 1997 ) of @xmath0c/@xmath6c in the solar neighborhood or the terrestrial value ( 89 ) , implying that the @xmath0co j=21 lines are _ not _ optically thin . if we adopt @xmath0c/@xmath6c=67 and assume that the emission is thermalized , then the optical depth for the @xmath0co j=21 line @xmath753.4 . on the other hand , the @xmath64(@xmath76 ) close to or greater than the terrestrial value imply that @xmath0co j=10 lines are optically thin , except at n1 where j=10 and j=21 lines appear to have comparable optical depths . this , however , is not conclusive because of large uncertainties associated with @xmath64 . second , the large values of @xmath772.3 imply that the broad - line emitting region is warm and dense . for typical molecular clouds , where the j=2 level is subthermally excited , the ratio is usually less than unity . for example , molecular clouds in the local arm exhibit ratios ranging from 0.53 ( taurus ) to 0.75 ( orion a ) ( sakamoto et al . 1994 , 1997 ) . our spectra of ambient gas also show this , e.g. , see the spectra toward s2 in fig . 7 where the narrow component at + 2 has @xmath78 . but the large ratio is common for the shocked molecular gas in snrs ( see 4.1 ) . third , the low ( @xmath79 k ) brightness temperature of j=21 lines , regardless of their moderate optical depths , imply that the emitting region must be clumpy , i.e. , composed of subclumps , and the emission is beam - diluted . from the above considerations , we may conclude that the broad emission lines are from warm , dense clumps with significant column densities so that the 21 lines are optically thick . we have applied the large - velocity - gradient ( lvg ) model ( scoville & solomon 1974 ; goldreich & kwan 1974 ) to our broad co lines in order to derive their excitation parameters . the model assumes an uniform , spherical cloud with a constant velocity gradient ( @xmath80 ) . if the lines are emitted from the shocked region where temperature and density vary greatly , the resulting parameters may be considered as ` average ' values . since we have found that the emission is beam - diluted , we have used the line ratios , @xmath57 and @xmath58 , instead of brightness temperatures to determine the excitation parameters . according to our lvg analysis , the observed ratios are possible for @xmath81 k. figure 8a shows the result of our model computations when @xmath82 k , where curves of constant @xmath57 and constant @xmath58 are drawn in ( @xmath83 , @xmath84 ) plane . @xmath83 is the fractional abundance of @xmath0co relative to h@xmath85 ( @xmath86 ) per unit velocity gradient interval . the asterisks ( * ) mark the observed ratios toward the peak positions where both ratios are obtained , i.e. , n1 , n2 , and s1 . according to figure 8a , @xmath87@xmath88 @xmath89 and @xmath90@xmath91 pc ( ) @xmath92 at the three peak positions . there are multiple choices for n1 , e.g. , the same ratios are obtainable when @xmath93 @xmath89 and @xmath94 pc ( ) @xmath92 . we adopt the lower density because it is comparable to the densities in the other peak positions and because the density of @xmath95 @xmath89 appears to be too high for the co emission to explore . if the temperature becomes higher , both @xmath84 and @xmath83 need to be greater . in figure 8b , we plot the _ expected _ co j=21 radiation temperature @xmath96 $ ] which is just the brightness temperature when @xmath97 and the expected @xmath64 from the same lvg model . note that the expected radiation temperatures are much greater than the observed main - beam brightness temperature . we have estimated beam filling factors of ( 7.78.8)@xmath98 from the ratio of these two brightnesses . we have estimated the co column densities @xmath99@xmath100 @xmath101 at these peak positions by @xmath102 \delta v$ ] , where @xmath10318 is the velocity width . the excitation parameters derived from the lvg analysis are listed in table 2 . note that the @xmath64 expected from the lvg model differ from the observed ones : at n1 , the observed value is small by a factor of 2 , while , at s1 , it is large by a factor of @xmath104 . considering the weakness of @xmath6co j=10 lines and various uncertainties associated with different telescopes , however , it is not obvious if this difference is critical . we have made a crude estimate of the mass of the broad - line clouds as follows . if the co j=10 line emission is optically thin , then @xmath105 can be obtained from the co j=10 luminosity @xmath106 by @xmath107 $ ] where @xmath108 is the fraction of co molecules at @xmath109 level and the other coefficients have their usual meanings . in our case , co j=10 emission has less optical depth than the j=21 emission , but is not very optically thin , so that the above formula might yield an underestimate . what we have is the luminosity of co j=21 emission @xmath110 , which has moderate optical depth . but , since @xmath772.3 at the peak positions , we may obtain @xmath106 by assuming that @xmath111 where @xmath112 and @xmath113 are co @xmath11410 and @xmath11421 line frequencies respectively . finally we assume @xmath115 , which is a mean value of those ( 0.150.26 ) at the three peak positions obtained from the lvg analysis . we have found that the h@xmath85 masses of clouds n and s are @xmath116 and @xmath117 , respectively . the mass of the central u - shaped part of cloud n is @xmath118 while the masses of the small clumps in cloud s are @xmath1191 . broad co emission lines with large @xmath57 in clouds n and s strongly suggest that they are being shocked . the observed velocity width is as large as @xmath120 . note that , toward this direction ( @xmath121 ) , the lsr velocity permitted by the galactic rotation is @xmath122 , so that it is not impossible for broad lines to be produced by molecular clouds accidentaly aligned along the line of sight . but it is highly improbable that such alignment ( over a few kpcs ) occurs in very small ( 12@xmath25 ) areas on the sky . we also searched for protostellar candidates around the broad line emitting regions using the _ infrared astronomical satellite _ ( _ iras _ ) point source catalog , because broad lines can be emitted from the high - velocity gas associated with protostellar object too . we did not find any suspicious sources except one , iras 20444 + 4954 , which is located close to the s2 clump , i.e. , at ( @xmath123 , @xmath124 ) from the peak position in table 1 . the source has been detected in two iras wavebands , i.e. , 60 and 100 @xmath125 , with flux densities of @xmath126 jy and @xmath127 jy . we have found that the source is located within a small ( @xmath128 ) , bright ( @xmath129 k ) co j=21 core at @xmath130 , so that it might be a young stellar object associated with the @xmath67-sized , diffuse cloud in the northern part of the + 2.9 map in figure 6 , not with the s2 clump . also the velocity of the s2 clump is similar to those of the other fast - moving clumps in this area , which suggests that they have the common origin . therefore , the broad lines that we detected are almost certainly from the fast - moving molecular gas swept - up by the snr shock in hb 21 . another indication that the broad co lines are from the shocked gas is their high @xmath772.3 . as we have shown in 3.4 , the high ratio implies that the emitting gas is warm and dense , which might be manifestation of shock . indeed high @xmath57 is a common property of the broad lines from the shocked molecular gas in snrs : all six snrs that are known to have broad molecular emission lines , i.e. , w28 ( arikawa et al . 1999 ) , 3c391 ( reach & rho 1999 ) , w44 ( seta et al . 1998 ) , w51c ( koo & moon 1997 ) , hb 21 ( this paper ) , and ic 443 ( e.g. , van dishoeck et al . 1993 ) , have ratios greater than 1 , which implies that the broad co lines in these snrs are all emitted from warm and dense , shocked gas . meanwhile , the maximum brightness temperature ( @xmath131 k ) of co j=21 line in hb 21 is significantly less than those of other snrs even if it was obtained with a higher spatial resolution ( 0.1 pc ) , e.g. , it is 33 k for w28 ( reach & rho 2000b ) and ic 443 ( van dishoeck et al . 1993 ) when observed with a resolution of 0.2 pc . the much smaller co j=21 brightness temperature with comparable @xmath57 imply that either the shocked gas in hb 21 is composed of much smaller clumps or is less dense . on a large scale , hb 21 appears to be in contact with a giant molecular cloud ( gmc ) along its eastern boundary ( huang & thaddeus 1986 ; tatematsu et al . the gmc is 130 pc@xmath13270 pc in extent ( dobashi et al . 1994 ) , and the structure that we call clouds a , b , and c defines the western boundary of the gmc . t90 inspected the correlation between these clouds and the snr in detail , and concluded that cloud a might be interacting with the snr because it is located where the radio continuum boundary of the snr is distorted . on the other hand , they concluded that clouds b and c might not , because there is no indication of the interaction in the distribution of radio brightness . according to our high - resolution observations , however , there is little relationship between the boundaries of cloud a and the snr . instead , since the velocity of the ambient molecular gas around hb 21 might be negative ( see 4.3 ) , we consider that clouds b and c have a better chance of interaction . but we have detected broad co lines toward none of these eastern clouds . even if a snr is interacting with a molecular cloud , the broad lines may be absent , however . 3c 391 ( g31.9 + 0.0 ) , which appears to be located at the edge of a large molecular cloud , for example , has no broad co lines along the interface ( wilner , reynolds , & moffett 1998 ; reach & rho 1999 ) . but strong [ oi ] 63 @xmath133 m emission has been detected near the interface , which indicates that the snr is interacting with the molecular cloud ( reach & rho 1996 ) . this would happen if the shock is dissociative and molecules have not reformed . figure 9 shows an enlarged view of the northern area overlaied with a 325 mhz radio continuum map of the snr . note that there is an enhanced radio continuum emission elongated along the ne - sw direction in the central area . its peak position falls exactly inside the u - shaped part of cloud n. the positional coincidence suggests that the enhancement is possibly associated with cloud n. we have derived a spectral index ( flux@xmath134 ) of @xmath135 for the radio emission associated with the u - shaped central part using the 325 mhz and 1420 mhz maps . the derived index has a large uncertainty because of the confusing background " level , but it appears to be flatter than the mean spectral index @xmath136 of hb 21 ( willis 1973 ) . for cloud s , there is no obvious correlation between the co emission and radio continuum brightness , although the radio continuum appears to be bright around the cloud in general . it is not obvious , observationally or theoretically , what determines whether or not radio synchrotron emission becomes enhanced when sn shock hits a dense cloud . observationally , we see very limited correlation between radio continuum brightness and shocked molecular gas in some snrs , i.e. , the shocked molecular gas is not usually associated with radio continuum enhancement and vice versa ( see also chevalier 1999 ) . in ic 443 , for example , shocked molecular gas is distributed in a fragmentary , flattened , ring , which partly overlaps with the radio continuum shell ( e.g. , see dickman et al . 1992 for the shocked molecular gas and green 1986 for the radio continuum ) . but the radio continuum is brightest in the northeastern part of the shell where there is no shocked molecular gas , and the radio continuum is not particularly bright toward the shocked molecular gas , perhaps except around the southern part of the molecular ring ( see next ) . in 3c 391 , there is a shocked molecular clump in the southern part of the snr , but there is only a faint , local radio continuum peak at @xmath137 pc apart from the shocked clump , whose association can not be confirmed ( reach & rho 1999 ) . in w28 , on the other hand , there is a ridge of radio continuum emission associated with the shocked molecular gas ( arikawa et al . theoretically , it could be either the dense cloud or the surrounding intercloud medium where synchrotron emission becomes enhanced . if the shock propagating through the dense cloud is radiative , there will be a large compression of cosmic rays and magnetic field , which would increase the synchrotron emissivity ( van der lann 1962 ; blandford & cowie 1982 ) . but this mechanism may not work because the molecular shocks in old snrs are not ionizing shocks and , therefore , high energy particles may escape from the shocked region ( draine & mckee 1993 ; chevalier 1999 ) . on the other hand , the shocked intercloud medium surrounding the cloud , particularly the medium behind the cloud , could have enhanced synchrotron emissivity because of the increased magnetic field strength there ( e.g. , jones & kang 1993 ; mac low et al . 1994 ) . in hb 21 , the peak of the enhanced radio emission is located behind the shocked cloud ( fig . it is also noteworthy that the cloud has a u - shape , which is similar to what we would expect when a small cloud is swept up by a strong shock ( e.g. , klein , mckee , & colella 1994 ; mac low et al . these morphological characteristics seem to suggest that the enhanced emission is not physically associated with the shocked cloud but with the shocked surrounding intercloud medium . in some respects , the region around cloud n in hb 21 is similar to the flat ( @xmath138 ) spectral region in ic 443 , which is also located behind a shocked molecular cloud ( green 1986 ; keohane et al . 1997 ) . in ic 443 , keohane ( 1997 ) found that the flat spectral region is particularly bright in hard x - rays , which are most likely due to synchrotron radiation . they concluded that the enhanced hard x - ray emission and the flat spectral index is due to shock acceleration of cosmic rays behind dense clouds . more observational study is certainly needed for hb 21 in order to reveal the relationship between the northern u - shaped cloud and the enhanced radio emission . cold molecular gas around hb 21 in general has central velocities between @xmath139 and + 6 , which must be the velocity range of the preshock gas . in the areas where broad lines have been detected , there is diffuse molecular gas at positive ( 0 to + 5 ) velocities ( see figures 4 and 6 ) . if this gas represents the preshock gas , then , since the broad lines are centered and spread out mostly at negative velocities , the gas should have been shocked and accelerated toward us _ systematically_. if we take + 3 as the velocity of the preshock gas and if we take either the central or peak positions of the broad lines as the systematic velocities of the shocked gas , then _ the line - of - sight _ velocities of the shock would be @xmath140 and @xmath8 for clouds n and s , respectively . and , if we take @xmath141 and @xmath142 ( @xmath143=the radius of the snr ) as their projected distances from the center of the snr , then their deprojected velocities are @xmath144 and @xmath145 , respectively . such coasting clumps are indeed theoretically expected for the old snrs such as hb 21 : molecular clumps swept up by a snr blast wave are accelerated by the shock propagating into the clumps and also by the ram pressure of interclump gas ( e.g. mckee 1988 ) . the characteristic timescale for acceleration is @xmath146 where @xmath147 is the radius of the clump , which becomes @xmath148 yrs ( @xmath149 yrs ) for the clumps in hb 21 using @xmath150 pc and @xmath15125 . the wings of the broad lines may be attributed to the gas accelerated by shocks propagating from sides . but one difficulty in this scenario would be that there is no clear correlation in the distributions of the broad lines between the preshock and postshock gases . in the southern area , for example , the preshock gas ( 23 component ) is distributed in a filamentary cloud extended along the ns direction , while the broad - line emitting clumps are distributed over a much wider area to the sw of this cloud . also it appears rather awkward that there are no broad lines centered near the velocity of the preshock gas . alternatively , it could be the molecular gas at negative velocities that is associated with the snr , and , in the northern and southern areas , most , if not all , of the ambient gas may have been shocked . in this picture , the shocked gas may be coasting too , but not necessarily at high speeds because the velocity of the preshock gas is somewhere within the velocity range of the broad lines , e.g. , near the velocity of the peak . the shock velocity determined from the line width is @xmath8 . between the two possible interpretations , we prefer the latter because of the difficulty with the former mentioned above . but , since the difference in the shock velocities between the two interpretations is @xmath140 , the following discussion remains valid basically even if the velocity of the preshock gas is 0 to + 5 . ( the derived properties of the shocked gas in 3.4 should remain valid too in either case because the emission from the preshock gas might be in narrow lines and its contribution to the integrated intensity of broad lines is expected to be small . ) the _ observed _ shock velocity is @xmath8 . this is less that the critical velocity for the dissociation of molecules , which is 2550 depending on preshock density and magnetic field strength ( hollenbach & mckee 1980 ; draine , roberge , & dalgarno 1983 ) . hence , the shock might be a _ non - dissociating c - shock_. the observed integrated intensity of the co j=21 emission is ( 39)@xmath152 ergs @xmath101 s@xmath92 sr@xmath92 at the peak positions . if we consider the beam dilution ( 3.4 ) , the actual surface brightness may be greater by an order of magnitude , e.g. , ( 410)@xmath153 ergs @xmath101 s@xmath92 sr@xmath92 . this is much larger than the _ angle - averaged _ surface brightness predicted from shock model computations . draine & roberge ( 1984 ) , for example , computed surface brightnesses expected for steady - state c - shocks propagating through molecular gas with different preshock conditions . according to their result , the angle - averaged surface brightness of co j=21 emission varies from @xmath154 to @xmath155 ergs @xmath101 s@xmath92 sr@xmath92 for a 1020 shock propagating through a molecular cloud with @xmath156 to @xmath157 @xmath89 . larger shock velocity does not raise the surface brightness while higher preshock density may yield @xmath158 ergs @xmath101 s@xmath92 sr@xmath92 . the much higher surface brightness toward the peak positions would be possible if these are directions where we are observing the shock tangentially . we want to briefly discuss the non - detection of oh 1720 mhz masers in hb 21 , because such masers are known to indicate interaction of snrs with molecular clouds ( frail et al . firstly , the oh masers may require a very specific set of physical conditions that might not be realized in hb 21 . according to lockett , gauthier , & elitzer ( 1999 ) , the 1720 mhz masers arise only in c - shocks when @xmath159 - 125 k , @xmath160 , and oh column density of @xmath161@xmath162 @xmath101 . according to our result in 3.4 , the density of the shocked molecular gas in hb 21 appears to be much lower than required . secondly , it is not impossible that the oh maser emission , even if present , had been missed in the survey by the frail et al . ( 1996 ) who mapped the snrs in rectangular grids with full - beam ( or @xmath163 full - beam ) grid spacing . it would be worthwhile to search for oh masers toward the shocked co gas in hb 21 . koo & heiles ( 1991 ) detected shocked h i gas associated with hb 21 . the shocked h i gas moves at @xmath164123 and is confined to the southern part ( @xmath165 0@xmath2550@xmath24 30@xmath25 ) of the snr . the highest velocity component coincides with cloud s , although the angular resolution ( @xmath12 ) of the hi observation is too large for a detailed comparison . koo & heiles ( 1991 ) assumed that the shocked h i gas represents a cap portion of a large expanding h i shell and derived a mean _ ambient _ h i density of 3.7 @xmath89 ( when scaled to 0.8 kpc adopted in this paper ) . if the molecular shock has been driven by this h i shell , then we can roughly estimate the density of the shell as follows : we assume that the shocked molecular gas is confined to a thin slab and that the radiative h i shell has an uniform density of @xmath166 . then it is straightforward to show that @xmath166 is related to the density of the molecular cloud @xmath167 by @xmath168 ( for @xmath169 ) where @xmath170 and @xmath171 are the velocities of radiative shell and shocked molecular slab , respectively ( e.g. , see chevalier 1999 ) . for hb 21 , @xmath172 and @xmath173 . and if we take the h@xmath85 density of the cloud @xmath174 @xmath89 , the density of the h i shell would be @xmath175 @xmath89 . by comparing with the mean ambient density ( 3.7 @xmath89 ) , this implies a compression factor @xmath176 for the h i shell . such moderate compression would be obtained if the ambient magnetic field strength tangential to the shell is @xmath177 @xmath133 g where @xmath178 is the density of the ambient medium . ( the equation is obtained by assuming that ambient magnetic field is uniform and magnetic pressure dominates the pressure in the shell . see chevalier 1974 for a discussion . ) alternatively , the fast - moving h i gas could be the gas originally associated with the molecular clouds , i.e. , the atomic and molecular shocks may be produced when the snr shock hits a large molecular cloud . in this case the fast - moving h i gas represents the swept - up interclump medium or h i envelope of molecular cloud . high - resolution h i observation is needed to reveal the relation between the shocked atomic and molecular gases . we used archival data from _ iras _ to search for infrared emission associated with the remnant . in his catalog of infrared emission from supernova remnants , arendt ( 1989 ) called hb 21 a `` probable '' infrared source at 12 and 60 @xmath133 m , with total fluxes of @xmath179 and @xmath180 jy , respectively . the main source of uncertainty is confusion with unrelated emission in the galactic plane , which can not be easily separated in infrared images . using the _ iras _ _ sky survey atlas _ ( _ issa _ ; wheelock et al . 1993 ) , we created an image covering the region around hb 21 at 60 and 100 @xmath133 m . there is extensive emission to the south , east , and west of the remnant , but with no clear correlation with the radio or co image . the region toward the center of the remnant is relatively fainter than these edges , but the northern part of the remnant is also faint . it is not possible to tell whether the emission is related to a partial shell around the remnant or just fluctuations in the background emission . to search for infrared emission associated with the remnant in more detail , we obtained a dedicated _ iras _ hires ( aumann , fowler , & melnyk 1990 ) image for a @xmath181 field centered on hb 21 . hb 21 is visible in the _ iras _ hires images at all four wavelengths ( 12 , 25 , 60 , and 100 @xmath133 m ) . the 60 @xmath133 m image , where hb 21 is most prominent , is shown in figure 10 . comparing the _ iras _ and co images , it is evident that the southern filamentary cloud , cloud s , is detected as a long arc , with very similar location , shape and width . the 60 @xmath133 m surface brightness of the filament is typically 7 mjy sr@xmath92 , and its structure is clumpy , like that of the co emission . but the peaks of co and infrared emission do not match in detail , suggesting that the 60 @xmath133 m emission does not arise from the exact same regions as the co. in the northern area , there is also a good correspondence between the infrared and co emission . the general correspondence in the south and north , and partial overlap in the east , show that many of the infrared features around the edge of hb 21 are related to the remnant , although the infrared emitting regions differ in detail from the co emitting regions . the nature of the infrared emission from hb 21 could be either dust grains surviving the shock , or from spectral lines from shock - excited gas , or both . we have estimated the mean surface brightness and color of the co clouds by using several faint regions in the field as background . the results are summarized in table 3 . the far - infrared color ratio , @xmath182 , for clouds a , b , and c is almost identical to that of diffuse cirrus clouds in the solar neighborhood ( boulanger & prault 1988 ) . this suggests that the infrared emission from clouds a , b , and c is most likely due to dust heated at the surface of the molecular clouds by the interstellar radiation field ; specifically , it suggests that the infrared emission is not related to shock fronts into the clouds . on the other hand , clouds n and s have a significantly higher color ratio @xmath183 . this enhanced color , and the morphological correspondence with the broad molecular line emitting regions , suggests that the infrared emission from clouds n and s is due to shocks in propagating into the clouds . conversely , the normal infrared color of clouds a , b , and c is consistent with their being due to ambient molecular clouds . if the infrared emission from clouds n and s is due to dust , then the relatively higher @xmath184 could be due to smaller or warmer dust grains . for dust heated by the average interstellar radiation field in the solar neighborhood , about half of the emission at 60 @xmath133 m is thought to be due to small , transiently heated grains ( draine & anderson 1985 ; dsert , boulanger , & puget 1990 ) . thus , if the enhanced 60 @xmath133 m emission is due to dust grains , then clouds n and s may contain a larger fraction of small grains . an enhanced abundance of small grains would be expected if a significant fraction of larger grains were shattered behind the shock front . observations of local cirrus clouds with significant velocities revealed that @xmath185 is typical for clouds with @xmath186 km s@xmath92 , suggesting that grain shattering was significant in the shocks that accelerated local clouds to intermediate velocities ( heiles , reach , & koo 1988 ) . theoretically , significant shattering is not predicted for slow shocks such as inferred from the widths of the co lines , but faster shocks through somewhat lower - density interclump gas , with @xmath187 km s@xmath92 , could produce the enhanced 60 @xmath133 m emission ( jones , tielens , & hollenbach 1996 ) . spectral lines could contribute significantly to the infrared emission from clouds n and s. the most important lines in the _ iras _ passbands , based on infrared spectra of similar supernova remnants ( oliva et al . 1999 , cesarsky et al . 1999 , reach & rho 2000a ) , are [ ] 88 @xmath133 m in _ iras _ band 4 , [ ] 63 @xmath133 m in band 3 , [ ] 26 @xmath133 m in band 2 , and [ ] 12.8 @xmath133 m and h@xmath85 lines in band 1 . if we were to interpret all of the _ iras _ emission from cloud s as due to the ionic lines listed above , then , using the system response and bandwidths ( beichman et al . 1988 ) , we find that the brightest line would be [ ] 88 @xmath133 m , with intensity @xmath188 erg s@xmath92 @xmath101 sr@xmath92 . relative to this line , the other bright lines would have ratios @xmath18963/88=0.27 , @xmath18926/88=0.13 , and @xmath18912.8/88=0.26 . the implied brightness of the [ ] 63 @xmath133 m line can be easily produced by shocks with velocities @xmath190 km s@xmath92 into moderate - density ( @xmath191 @xmath89 ) gas ( hollenbach & mckee 1989 ) ; however , such shocks do not produce as much [ ] emission as observed , because the column density of highly - ionized gas is insufficient . slower shocks into denser gas can also produce [ ] 63 @xmath133 m lines this bright ( draine , roberge , & dalgarno 1983 ) . however , slower ( c - type ) shocks produce essentially no ionic line emission especially not an ion such as . nor would the slow shocks destroy grains adequately to produce significant gas - phase . therefore , if the infrared emission is from slow shocks , the _ iras _ 100 @xmath133 m band emission is from dust , and the _ iras _ 60 @xmath133 m band emission is from a mix of dust and the [ ] 63 @xmath133 m line . the nature of the _ iras _ 12 and 25 @xmath133 m band emission is more difficult to constrain . if the infrared emission is from slower shocks , such as inferred from the co observations , then there is likely a contribution from h@xmath85 lines . rotational lines are the dominant coolant for a range of molecular shocks ( rho et al . 2000 , reach & rho 2000a ) . for now , it is not possible to clearly tell what fraction of the infrared emission is from gas or dust . the nature of the infrared emission from hb 21 ( and other supernova remnants ) can be determined in the future using spectroscopy or narrow - band imaging . we have mapped the eastern half ( @xmath192 ) of the snr hb 21 in @xmath0co j=21 line emission almost completely . our map , which has been completely sampled with @xmath193 resolution , shows the detailed structure of molecular clouds in this area . we have detected broad co lines with large @xmath57 in the northern and southern parts of hb 21 , which is direct evidence for the interaction between molecular clouds and the snr . in the following , we summarize the main results of this paper : \(1 ) we detected shocked molecular clouds , clouds n and s , with broad ( 2040 ) co lines in the northern and southern parts of the snr . cloud n is composed of a small ( @xmath2 or 0.5 pc ) , very bright , u - shaped , clumpy part and several clumps scattered around it . cloud s is filamentary and appears to form an elongated loop of @xmath4 in extent . small ( @xmath5 or 0.3 pc ) , bright clumps are seen along the filamentary structure . the h@xmath85 masses of clouds n and s are @xmath116 and @xmath117 , respectively.(2 ) we have obtained sensitive j=10 and j=21 spectra of @xmath0co and @xmath6co molecules toward several peak positions of clouds n and s. they have @xmath772.3 and @xmath7340 with co j=21 main - beam brightness temperature less than 7 k. according to our lvg analysis , @xmath81 k , and , for @xmath82 k , @xmath194(37)@xmath195 @xmath89 and @xmath196(2.411 ) @xmath197 @xmath101 . the emitting region appears to fill a small ( 0.0770.088 ) fraction of the beam . ( 3 ) there is an enhanced radio emission which attains a maximum exactly inside the central u - shaped part of cloud n. the emission has a spectral index ( @xmath198 ) flatter than that of the whole remnant . the association of this emission with cloud n needs to be explored.(4 ) clouds n and s are visible in the _ iras _ hires images at all four wavelengths ( 12 , 25 , 60 , and 100 @xmath133 m ) . they have the far - infrared color ratio @xmath183 , which is significantly greater than that ( 0.20 ) of the other clouds in this area . this enhanced color , and the morphological correspondence with the broad molecular line emitting regions , suggests that the infrared emission from clouds n and s is due to shocks in propagating into the clouds.(5 ) along the eastern boundary of the snr , three relatively large ( @xmath23 or 3.5 pc ) clouds and several filamentary clouds are present . no broad co emission or enhanced 60/100 @xmath133 m color were detected in any of these clouds , and there is little relationship between the boundaries of the clouds and the snr . therefore , there is no strong evidence for the interaction of the snr with molecular clouds along the eastern boundary . arendt , r. g. 1989 , apjs , 70 , 181 arikawa , y. , tatematsu , k. , sekimoto , y. , & takahashi , t. 1999 , pasj , 51 , l7 aumann , h. h. , fowler , j. w. , & melnyk , m. 1990 , aj , 99 , 1674 beichman , c. a. et al . 1988 , _ infrared astronomical satellite ( iras ) catalogs and atlases : volume 1 . explanatory supplement _ , nasa rp-1190 ( nasa : washington , dc ) blandford , r. d. , & cowie , l. l. 1982 , apj , 260 , 625 boulanger , f. , & prault , m. 1988 , apj , 330 , 964 cesarsky , d. , cox , p. , pineau des forts , g. , van dishoeck , e. f. , boulanger , f. , & wright , c. m. 1999 , a & a , 348 , 945 chevalier , r. a. 1974 , apj , 188 , 501 1999 , apj , 511 , 798 denoyer , l. k. 1979 , apjl , 232 , l165 dsert , f. x. , boulanger , f. , & puget , j .- l . 1990 , a&a , 237 , 215 dickman , r. l. , snell , r. l. , ziurys , l. m. , & huang , y .- l . 1992 , apj , 400 , 203 dobashi , k. , bernard , j .- yonekura , y. , & fukui , y. 1994 , apjs , 95 , 419 draine , b. t. , & anderson , n. 1985 , apj , 292 , 494 draine , b. t. , & mckee , c. f. 1993 , araa , 31 , 373 draine , b. t. , roberge , w. g. , & dalgarno , a. 1983 , apj , 264 , 485 draine , b. t. , & roberge , w. g. 1984 , apj , 282 , 491 erkes , j. w. , & dickel , j. r. 1969 , aj , 74 , 840 frail , d. a. , goss , w. m. , reynoso , e. m. , giacani , e. b. , green , a. j. , & otrupcek , r. 1996 , aj , 111 , 1651 goldreich , p. , & kwan , j. 1974 , apj , 189 , 441 green , d. a. 1986 , mnras , 221 , 473 heiles , c. , reach , w. t. , & koo , b .- c . 1988 , apj , 332 , 313 hill , i. e. 1974 , mnras , 169 , 59 hollenbach , d. , & mckee , c. f. 1980 , apjl , 241 , l47 1989 , apj , 342 , 306 huang , y .- l . , & thaddeus , p. 1986 , apj , 309 , 804 humphreys , r. m. 1978 , apjs , 38 , 309 jones , t. w. , & kang , h. 1993 , apj , 402 , 560 jones , a. p. , tielens , a. g. g. m. , & hollenbach , d. j. 1996 , apj , 469 , 740 keohane , j. w. , petre , r. , gotthelf , e. v. , ozaki , m. , & koyama , k. 1997 , apj , 484 , 350 klein , r. l. , mckee , c. f. , colella , p. 1994 , apj , 420 , 213 koo , b .- c . , & heiles , c. 1991 , apj , 382 , 204 koo , b .- c . & moon , d .- s . 1997 , apj , 485 , 263 langer , w. d. 1997 , in co : twenty - five years of millimeter - wave spectroscopy , iau symp . latter , w. b. et al . ( dordrecht : kluwer ) , 98 leahy , d. a. 1987 , mnras , 228 , 907 leahy , d. a. , & aschenbach , b. 1996 , a&a , 315 , 260 lockett , p. , gautheir , e. , elitzur , m. 1999 , apj , 511 , 235 mac low , m .- mckee , c. f. , klein , r. i. , stone , j. m. , & norman , m. l. 1994 , apj , 433 , 757 mckee , c. f. 1988 , in supernova remnants and the interstellar medium , proceedings of iau colloquium no . roger and t.l . landecker ( cambridge : cambridge ) , 473 oliva , e. , moorwood , a. f. m. , drapatz , s. , lutz , d. , & sturm , e. 1999 , a & a , 343 , 943 reach , w. t. , & rho , j. 1996 , a&a , 315 , l277 . 1999 , apj , 511 , 836 . 2000a , aph0007148 . 2000b , in preparation rengelink , r. b. , tang , y. , de bruyn , a. g. , miley , g. k. , bremer , m. n. , rttgering , h. j. a. , & bremer , m. a. r. 1997 , a&as , 124 , 259 rho , j. , & petre , r. 1998 , apjl , 503 , l167 rho , j. , jarrett , t. , cutri , r. m. , & reach , w. t. 2000 , apj , submitted roh , d .- jung , j. h. 1999 , publications of the korean astronomical society , 14 , 123 sakamoto , s. , hasegawa , t. , handa , t. , hayashi , m. , & oka , t. 1997 , apj , 486 , 276 sakamoto , s. , hayashi , m. , hasegawa , t. , handa , t. , & oka , t. 1994 , apj , 425 , 641 scoville , n. z. , & solomon , p. m. 1974 , apjl , 187 , l67 seta , m. , hasegawa , t. , dame , t. m. , sakamoto , s. , oka , t. , handa , t. , hayashi , m. , morino , j. , sorai , k. , & usuda , k. s. 1998 , apj , 505 , 286 tatematsu , k. , fukui , y. , landecker , t. l. , & roger , r. s. 1990 , a&a , 237 , 189 ( t90 ) tauber , j. a. , snell , r. a. , dickman , r. l. , & ziurys , l. m. 1994 , apj , 421 , 570 van den bergh , s. 1978 , apjs , 38 , 119 van der laan , h. 1962 , mnras , 124 , 179 van dishoeck , e. f. , jansen , d. j. , & phillips , t. g. 1993 , a&a , 279 , 541 wheelock s. l. , gautier , t. n. , chillemi , j. , kester , d. , mccallon , h. , oken , c. , white , j. , gregorich , d. , boulanger , f. , and j. good 1994 , _ iras sky survey atlas : explanatory supplement_. ( jpl / caltech : pasadena ) willis , a. g. , 1973 , a&a , 26 , 237 wilner , d. j. , reynolds , s. p. , & moffett , d. a. 1998 , aj , 115 , 247 hb21:bml - n1 & ( 20 46 03.2 , 51 00 00 ) & @xmath199 & 3.6 & 1.6&@xmath59&@xmath200hb21:bml - n2 & ( 20 45 55.0 , 51 03 30 ) & @xmath201 & 3.1 & 1.7&@xmath202& ... hb21:bml - s1 & ( 20 44 37.2 , 49 47 10 ) & @xmath203 & 2.7 & 1.9&@xmath204&@xmath205hb21:bml - s2 & ( 20 44 31.0 , 49 55 20 ) & @xmath206 & 2.9 & 1.7& ... &@xmath207hb21:bml - s3 & ( 20 42 45.2 , 49 56 50 ) & @xmath208 & 6.9 & 2.3& ... &@xmath209 a & 1.1(0.2 ) & 1.1(0.1 ) & 4.2(0.2 ) & 20.8(2.3 ) & 0.20(0.03)b & 1.7(0.1 ) & 1.8(0.1 ) & 5.9(1.1 ) & 29.5(2.4 ) & 0.20(0.04)c & 1.7(0.1 ) & 1.7(0.1 ) & 8.0(1.1 ) & 38.0(2.4 ) & 0.21(0.03)n & 1.0(0.2 ) & 0.8(0.1 ) & 4.8(0.2 ) & 17.0(2.3 ) & 0.28(0.04)s & 1.2(0.1 ) & 1.2(0.1 ) & 6.8(1.1 ) & 26.3(2.4 ) & 0.26(0.05 )

this is an investigation of the period - luminosity relation of classical cepheids in samples of different metallicity . it is based on 481 cepheids in the large and small magellanic clouds from the blue and red filter ccd observations ( most similar to & ) of the french eros microlensing project . the data - set is complete and provides an excellent basis for a differential analysis between lmc and smc . in comparison to previous studies of effects on the pl - relation , the eros data - set offers extremely well - sampled light curves and well - filled instability strips . this allows reliable separation of cepheids pulsating in the fundamental and the first overtone mode and derivation of differential reddening . our main result concerns the determination of distances to galaxies which are inferred by using the lmc as a base and using two color photometry to establish the amount of reddening . we find a zero - point offset between smc and lmc which amounts to a difference between inferred and true distance modulus of 0.14 mag in the system . the offset is exactly the same in both sets of pl - relations of the fundamental and of the first overtone mode cepheids . no effect is seen on the slopes of the pl - relations , although the fundamental and the first overtone mode cepheids have different pl slopes . we attribute the color and the zero - point offset to the difference in metallicity between the smc and lmc cepheids . a metallicity effect of that small magnitude still has important consequencies for the inferred cepheid distances and the determination of . when applied to recent estimates based on cepheid observations , our metallicity dependence makes the low- values ( sandage _ et al . _ 1994 ) and the high- values ( freedman _ et al . _ 1994b ) , thus bringing those discrepant estimates into agreement near km.s .

astro-ph9612216

the cepheid period - luminosity ( pl ) relation is widely accepted as being one of the most accurate primary distance indicators to nearby galaxies . it has now been applied as far as the virgo cluster of galaxies ( freedman _ et al . _ 1994b , pierce _ et al . _ 1994 , sandage _ et al . _ 1994 , tanvir _ et al . _ 1995 ) and holds the greatest promise to settle the debate over the value of the hubble constant by providing a @xmath12% accurate determination ( the goal of the @xmath6 key project ) . the cepheid variable stars offer a simple way of measuring distances : their pulsation periods ( easy to obtain ) are strongly correlated with their luminosities ; then distances are determined from their apparent brightness via the pl relation . the reliability of these distances stems from the good theoretical understanding of the cepheids and their pl relation ( iben & tuggle 1975 ) . theory predicts a small , but not negligible , abundance effect on the pl zero point ( see stothers 1988 for review ) . however the ambiguous results of previous empirical tests for this effect have led all recent @xmath6 studies quoted above to assume that the cepheid pl relation is insensitive to metallicity . three sources contribute to an abundance dependence of the pl relation : ( 1 ) theory of stellar pulsation , through the dependence of period on mass and radius ; ( 2 ) theory of stellar evolution , through the mass - luminosity relation ; and ( 3 ) theory of stellar atmospheres , through line blanketing and backwarming , @xmath13 the relations between effective temperature , absolute magnitudes in bandpasses , and bolometric correction . as a result , a metal - poor cepheid is always fainter than a metal - rich cepheid ( at a fixed period and temperature ) ; metal - poor cepheids are also hotter ( bluer ) on the average . given the sensitivity of each of the above three sources to metallicity , the overall weakness of the effect on the bolometric pl relation is remarkable ( stothers 1988 , chiosi _ et al . theory predicts that the slope of the pl relation is nearly independent of metallicity , only the zero point is affected . the observed pl relations are not bolometric , hence the effect would depend on the bandpass . most of the above predictions are valid for a limited range of cepheid pulsation periods @xmath4 cepheids with p@xmath1450 days are expected ( and observed ) to deviate from a linear pl relation . theory in itself predicts a period - luminosity - color relation ( sandage 1958 ) . however we study the pl relation , because this is the current one of choice in the determination of @xmath5 . theory predicts abundance effects on the pl relation due to helium ( y ) and heavy elements ( z ) . however we only know the heavy elements abundances ( z ) in magellanic clouds cepheids , hence we speak of metallicity effects in this paper . previous studies have looked for metallicity effects on the cepheid pl relation since the beginning of the 70s , but the issue remains unsettled observationally . partly to blame is the near degeneracy between three important observed properties of cepheids : the lines of constant period , reddening , and metallicity , in optical and near - infrared bandpasses . the lines of constant period represent the range of temperatures over which a star can sustain a stable cepheid pulsation in a given mode and period ; this introduces a natural width ( in luminosity ) to the pl relation . the amount of obscuration ( reddening ) is a more serious problem , as it is common practice to derive it from the photometry of the cepheids themselves @xmath4 thus the color difference due to metallicity will affect the reddening estimate and the inferred distance ( stothers 1988 , freedman & madore 1990 ) . the first large scale comparison between smc , lmc , and galaxy cepheids is due to payne - gaposchkin & gaposchkin ( 1973 ) . payne - gaposchkin ( 1974 ) concluded that there was no evidence for composition differences . however , gascoigne ( 1974 ) reinvestigated the apparent color differences between lmc and smc cepheids as a metallicity effect and found that the smc cepheids would be fainter than lmc cepheids by 0.1 mag ( in @xmath15 ) . the color shift between lmc and smc cepheids was confirmed by martin , warren , & feast ( 1979 ) and clearly distinguished from differential extinction . subsequently , iben & tuggle ( 1975 ) and iben & renzini ( 1984 ) argued for a much smaller effect , mostly from theoretical considerations . stothers ( 1988 ) offered a critical review of all these attempts and pointed out the effect of the reddening correction . in a different approach to the problem , caldwell & coulson ( 1985 , 1986 ) merged theoretical and empirical relations and individual reddenings from color - color diagrams to derive pl relations @xmath16 for abundance differences . these fit well their data - set of about 130 cepheids in lmc and smc . caldwell & coulson s results confirm the existence of a metallicity effect . their very different approach provides no base for a detailed comparison with the current extragalactic use ( stothers 1988 , madore & freedman 1991 ) . in addition , caldwell & coulson s sample of about 130 cepheids appears to be too small for quantifying the tiny effect . this can be seen in a recent comparison of primary distance indicators to 15 galactic and extragalactic objects in which caldwell & coulson s pl relations ( @xmath16 for metallicity ) were used ( huterer , sasselov , & schechter 1995 ) @xmath4 the uncertainties within and between the four primary indicators are larger than the weak metallicity dependence . the main reason extragalactic distance measurements have assumed that cepheid luminosity does not depend on metallicity is the work by freedman & madore ( 1990 ) . they found that any distance differences in three m 31 fields with varying metallicity are consistent with statistical noise , and much smaller than stothers(1988 ) prediction . the conclusion is based on a sample of 38 cepheids and 152 @xmath17 observations ; the method used is the same used in all recent distance determinations . the situation was reviewed by feast ( 1991 ) , who urged for the need to test the metallicity corrections empirically . this was partially accomplished by laney & stobie ( 1994 ) , who concluded that their @xmath18 data on 21 galactic and 115 mc cepheids do not support the implications of freedman & madore s result that intrinsic cepheid color is independent of metallicity . at the same time gould ( 1994 ) challenged freedman & madore s conclusion by pointing out the high degree of correlation among the @xmath17 measurements ( treated by freedman & madore as independent ) . gould obtained a pl zero - point shift similar to that of stothers ( 1988 ) by reanalyzing the same @xmath17 observations of 36 cepheids in m 31 . however , he showed that the data - set suffers from some systematic uncertainty , which affects the derived size of the effect . stift ( 1995 ) confirmed independently gould s conclusions using current theoretical evolutionary and atmosphere models . thus the issue remained unsettled . as a result all recent hst measurements of cepheid distances assume that no metallicity effects are present at v and i ( freedman _ et al . _ 1994b , sandage _ et al . _ 1994,tanvir _ et al . _ 1995 ) . in this paper we use a new data - set of two - color photometry ( @xmath19 3 million observations ) of about 500 cepheids in the lmc and smc to derive the dependence of the optical pl relations on metallicity . we find that the apparent distance modulus depends on metallicity roughly as : @xmath20 + 0.3)$ ] . the dependence is weaker than some previous claims , though by no means negligible . it is derived from a differential lmc - smc analysis which incorporates all correlations between cepheid measurements , independently derived metal abundances for cepheids and supergiants , and known limits to the extinction laws . the photometric database is a byproduct of the eros microlensing survey ( aubourg _ et al . _ 1993a , _ 1995 , beaulieu _ et al . _ 1996 ) . the eros ( exprience de recherche dobjets sombres ) french collaboration ( aubourg _ et al . _ 1993a , 1993b , 1995 ) , and the macho project ( alcock _ et al . _ 1993 ) are both searching for baryonic dark matter in the galactic halo through microlensing effects on stars of the magellanic clouds . the eros ccd equipment has been described by arnaud _ et al . _ ( 1994 ) . the observation , reduction , and calibration procedures have been described by grison _ et al . _ all details on the cepheids discovered and analyzed in the lmc and smc are given in beaulieu _ et al . _ ( 1995 ) and beaulieu _ et al . _ ( 1996 ) , respectively . in brief , all observations were obtained at eso , la silla using a 0.4-m , f/10 reflector and a 2@xmath218 mosaic of 16 ccds . two broad - band filters were used @xmath4 @xmath22 and @xmath23 , with central wavelengths which fall roughly between johnson @xmath24 and @xmath15 and @xmath25 and @xmath26 , respectively . the observations were obtained during the 1991 - 92 and 1993 - 95 seasons . our differential analysis relies on no external zero - point , and is thus completed within the @xmath22,@xmath23 system . however , in applying our metallicity effect to the @xmath27 work with the hst , we have investigated our transformation between @xmath22,@xmath23 and @xmath28 . the net transmission of the @xmath22 band is affected by the blue cutoff of the ccds sensitivity , and is thus much closer to johnson @xmath15 than to @xmath24 . the @xmath23 bandpass remains intermediate between cousins @xmath25 and @xmath26 . the @xmath22 filter is broader than johnson @xmath15 in a similar way as the @xmath6 f555w filter is broader than @xmath15 . the @xmath29 and @xmath30 colors transform well between each other as a result : @xmath31 . the lmc data from the 1991 - 92 season was obtain with somewhat different set of blue - red filters . therefore we treated the lmc cepheid data as two different sets and compared each of them separately to the smc data . the 1991 - 92 season lmc data is useful because of the large number of observations . with thousands of stars available in the cepheids range of magnitudes and colors on the eros ccd frames , constructing an accurate and reliable photometric transformation we transformed all old @xmath32 photometry to the new @xmath33 system by : @xmath34 @xmath35 the two sets of lmc data are very similar and produced the same result in the final differential analysis . the identification of cepheids in both clouds is based on the fourier components of their light curves and mean magnitudes in the expected luminosity range . several cepheids were excluded from our analysis on the basis of strong blending . these could be either actual physical binaries , or unresolved stars along the line of sight . they are easy to identify by the form of the pulsation loop on the color - magnitude diagram ( figure 1 ) and by the accompanying deviation of their mean magnitude and color from the rest of the cepheid sample . examining the pulsation loops shows that background blending for the lmc and smc cepheids is not a problem ( affects less than 10% of them ) , as should be expected for luminous stars like them . we investigated several possible sources of difference between the lmc and smc photometry , like airmass and the pinatubo eruption , but found their effect to be insignificant . the concept is simple @xmath4 we have two complete cepheid samples with known difference in metallicity @xmath36_{lmc - smc}=0.35 $ ] ( spite & spite 1991 ; luck & lambert 1992 ) . this is an average difference between spectroscopic abundances of longer - period cepheids . we compare the two samples in the period - magnitude - color ( plc ) manifold to derive 2 independent sources of difference @xmath4 distance and extinction , and to search for a @xmath37 source @xmath4 metallicity . for each cepheid we have 3 observed quantities : two coordinates and a pulsation period ; and 2 observed data - sets : complete light curves in @xmath22 and @xmath23 . the 2 data - sets ( light curves ) give two observed quantities : the time - averaged , intensity - weighted mean magnitudes ; thus each cepheid has 5 observed quantities in total . the light curves also provide independently the means to separate cepheids of different pulsation modes and to treat background blending . we construct the pl relations in each band , @xmath22,@xmath23 , of lmc and smc cepheids , @xmath13 the observed mean magnitudes , @xmath38 , for each cepheid of period @xmath39 and each band,@xmath40 , are fit to a linear function : @xmath41 where @xmath42 are the zero points and @xmath43 are the slopes ; within our completeness @xmath4 @xmath44days , the pl relations are found to be linear . we compare the two cepheid sets in the pl plane . they are offset in luminosity ( @xmath45 ) due to two sources : difference in distance between lmc and smc , and difference in extinction towards lmc and smc . we search for a @xmath37 source @xmath4 a term due to the difference in metallicity . two pl relations ( pl-@xmath22 and pl-@xmath23 ) in each of the clouds can give us the distance difference and the reddening difference , but the metallicity effect remains degenerate . we search to lift this degeneracy by going to the period - magnitude - color ( plc ) space , and imposing constraints on the interstellar extinction . we fix the mean reddening in lmc as e(b - v)=0.10 , the value adopted by all @xmath6 teams , and we adopt independent estimates for the foreground reddenings to lmc and smc . now , given a large observed sample ( in a statistical sense ) , we can compare differentially the plc distributions of lmc and smc cepheids . such a comparison will constitute a fit to a multi - parameter model . two of the model parameters are metallicity terms . we chose as our goal to evaluate the metallicity effects on the lmc - based technique used by all @xmath6 teams ( @xmath46 madore & freedman 1991 ; hereafter @xmath4 the modern technique ) for two reasons . first , the current cepheid - based @xmath5 is exclusively derived by it . second , the eros cepheid sample provides an excellent opportunity to accomplish that goal , being obtained with two filters ( @xmath22,@xmath23 ) similar to the @xmath6 f555w , f814w filters , and being lmc - based too . we emphasize the need to account for the correlations between magnitude measurements , which has not been done in its applications to - date . this rigorous approach was pioneered by gould ( 1994 ) ; our analysis draws heavily on it and expands it . due to the strong correlations and near - degeneracies present in the problem , we see no alternative to the rigorous approach . hereafter the notation @xmath47 will refer to the metallicity dependence of the quantity @xmath48 in the sense : @xmath49 , where @xmath50 . our method applies the same basic techniques for modeling of data ( press _ et al . _ 1994 , 15 ) introduced earlier by gould ( 1994 ) , but deviates substantially from gould s analysis . the difference is that we solve for a metallicity effect which is a function of bandpass / wavelength . therefore we model our data in the three - dimensional plc manifold , as opposed to just the pl plane . thus our analysis follows the current state of theoretical understanding ( stothers 1988 ; chiosi _ et al . _ 1993 ; stift 1995 ) , and takes advantage of the fact that our smc data - set fills densely the plc manifold . here we present a simultaneous fit of the lmc and smc data . the two ( @xmath22 , @xmath23 ) magnitude measurements of each cepheid are treated as correlated with each other , but not with the measurements of any other cepheid . then we can write the @xmath51 merit function as @xmath52 where @xmath53 correspond to the lmc data ( with @xmath54 cepheids ) on the period - magnitude and color - magnitude planes , respectively ; @xmath55 correspond similarly to the smc data ( with @xmath54 cepheids ) ; @xmath56 are the two bandpasses ( @xmath22,@xmath23 ) , and where the residuals are defined as follows : @xmath57 @xmath58,\ ] ] @xmath59 @xmath60 @xmath61,\ ] ] with @xmath62 being the number of cepheids in lmc and smc , respectively . the covariance matrices of the data are then @xmath63 for @xmath53 , and @xmath64 for @xmath55 , where @xmath65 in equation ( 2 ) , @xmath66 is the inverse covariance matrix of the data , @xmath67 . the above set of equations illustrates our differential analysis of lmc and smc cepheids as _ ensembles _ of stars . each cepheid with a period @xmath39 , and mean magnitudes @xmath38 is fitted ( for each band ) to a linear function of the period @xmath4 the pl relation ( @xmath68 ) , and to the instability strip ( @xmath69 ) . thus , for the lmc cepheids we have the residuals in equations ( 3 ) and ( 4 ) . the residuals of the smc cepheids will contain four additional linear terms : ( 1 ) one due to the distance difference ( @xmath70 ) ; ( 2 ) one due to extinction @xmath4 the relative reddening @xmath71 , and @xmath72 being the adopted reddening vector ( see 6 below ) ; and ( 3 ) two ( one for each band ) due to metallicity difference ( @xmath73 ) . all of these terms simply add on to the zero point of the pl relation ( for each band ) , and that is what equation ( 5 ) is all about . the smc residuals from equation ( 6 ) have a slight complication , which will become clear here . let @xmath74 ( @xmath56 ) be the absolute magnitudes in @xmath22 and @xmath23 , respectively . they define the pl relations for each band : @xmath75 . then following the metallicity term , @xmath76 , introduced above , the resulting changes in @xmath74 are simply : @xmath77 and @xmath78 . color is defined as : @xmath79 , where the subscript 0 " means corrected for extinction ; hence the color change at a fixed period due to the metallicity difference between lmc and smc will be : @xmath80 this term , @xmath81 , will appear in the period - color plane together with the reddening @xmath82-term alone . the equation of the linear period - color relation is not added to equations ( 3)-(6 ) to minimize redundancy ; its slope will be simply @xmath83 . now we can describe the meaning of equation ( 6 ) and the residuals in the color - magnitude plane . in the color - magnitude plane we have the more complex case , where both the color and the magnitude are affected by both the reddening @xmath84 the metallicity . while the terms for the magnitude remain the same as in equation ( 5 ) , the terms for the color will come from equation ( 10 ) for the metallicity , and from the definition of extinction color correction . the latter definition is : @xmath85 , where the @xmath86 are the total extinctions in each band , @xmath87 ; hence @xmath88 . thus equation ( 6 ) describes the following aspect of our model : more extinction makes cepheids redder and fainter . as far as the metallicity terms are concerned , they are not constrained ( although theory predicts that more metals make cepheids redder and brighter ) . we constrain the reddening term by not allowing unphysical negative values , and further by considering the foreground extinction towards the smc as its lower limit ( see 5 and 6 ) . to obtain the true distance modulus @xmath89 , the observed mean magnitude must be first corrected for extinction : @xmath90 . if @xmath86 are derived from the observations of the cepheids themselves , the metallicity dependence of the true distance modulus will be a combination of the dependencies of the absolute magnitudes and the color ( stothers 1988 ) . using the standard definitions for the ratio of total to selective extinction , we have : @xmath91 where @xmath92 . keeping @xmath72 fixed , we obtain for the change of distance modulus with metallicity : @xmath93 @xmath94 which applies to the use of a pl relation and is in units of stellar magnitude per @xmath95z of metals by mass . stothers ( 1988 ) used this case ( when @xmath86 are derived from the cepheids themselves ) to derive the prototype of equation ( 11 ) from theoretical considerations . we follow the same case ( referred to as the modern technique ) , but derive the correction , @xmath96 , empirically from the differential analysis of the magnitudes , colors , and periods of lmc and smc cepheids . therefore the application of our method ( see next section ) is restricted to this specific case . we derive the metallicity correction as @xmath97 . we want to emphasize again that our fit is in the plc space for the sole purpose of finding the residual color difference between the lmc and smc cepheids ; we are not interested in deriving or using a plc relation . it is our way of lifting the degeneracy between the effects of reddening and metallicity in the pl planes in our differential analysis of lmc and smc . freedman & madore ( 1990 ) used a different approach to break this degeneracy @xmath4 by setting @xmath98 , @xmath13 by observing three fields at the same distance but different metallicity . they can not use our method , because their data - set ( 36 cepheids ) is too small to fill densely the plc space . in other words , in the language of @xmath99 statistics , their number of data points ( 36 ) will be comparable to the number of model parameters . as described by equations ( 3)-(6 ) of the previous section , our model of the lmc - smc comparison has 12 model parameters . to solve for all 12 parameters we adopt the following external constraints : ( 1 ) mean lmc extinction e(b - v)=0.10 as used by the hst teams ; ( 2 ) foreground extinctions of 0.06 for lmc and 0.05 for smc ( bessell 1991 ) ; ( 3 ) no depth dispersion in the eros lmc sample ; and ( 4 ) the line of nodes for smc from caldwell & coulson ( 1986 ) to derive the eros smc sample depth dispersion . these constraints are discussed in the next two sections . obviously , there is one more projection @xmath4 on the period - color plane , which is not included in equations ( 3)-(6 ) , but is trivially derived from them . one estimates the model parameters by minimizing in a statistical sense the residuals , given independent constraints on some of the parameters . to find the best fit one differentiates @xmath100 , where @xmath101 is the 12 element vector of model parameters : @xmath102 the @xmath99 merit function of equation ( 2 ) can be rewritten in a more explicit form for our multidimensional fit : @xmath103\ ] ] @xmath104\ ] ] the differentiation yields a matrix equation which can be solved for @xmath101 : @xmath105 where @xmath106 is the covariance matrix of the model parameters vector , @xmath107 with @xmath108 being the basis functions from equation ( 13 ) , given explicitly in equations ( 3)-(6 ) ; in a similar fashion , the vector @xmath109 relates the basis functions to the mean magnitudes @xmath110 . the procedure requires an iteration ( see gould 1994 ) . first , we construct period - magnitude and color - magnitude relations for the cepheids in the lmc and derive the eight parameters @xmath111 from linear fits to the data . these are then used to find the covariance matrices of the lmc data ( equation 7 ) . next , we compute the covariance matrices of the smc data and find the best - fit values . we found that the slopes of the pl relations do not differ ( within the uncertainties ) between lmc and smc ; we adopt two unique slopes ( one for fundamental , one for first overtone cepheids ) for both lmc and smc . at this point we derive @xmath112 , by setting @xmath113 . then we minimize the residuals along a given reddening vector r@xmath114 and additionally constraining @xmath115 so that smc cepheids have no unphysical extinctions ( negative ) . this constraint is derived independently of the fit ( see next section ) . we use the same form of the galactic extinction law ( cardelli , clayton , & mathis 1989 ) for all samples . we iterate this procedure until the best - fit parameters agree with the trial parameters used to estimate the covariances . note that the entire procedure is separately done for the fundamental mode and first overtone mode cepheids , and thus we have a strong constraint on the final derived parameters , which are invariant to the type of cepheids used to derive them . for illustration , the covariance matrix of the lmc fundamental cepheids ( equation 7 ) is : @xmath116 and for the smc fundamental cepheids ( equation 8) : @xmath117 the correlation coefficients are obviously very high , as also seen in figs.5 & 6 . for cepheids all at the same distance and selected only by magnitude the pl relation is subjected to bias , which can be avoided by using the inverse regression ( schechter 1980 , feast 1995 ) . we do not expect strong magnitude bias in our samples , but used both direct and inverse regressions in our fits . the method above contains as a subset the procedures which comprise the modern technique for determining cepheid distances ( madore & freedman 1991 ) . namely , one begins by constructing the pl relations in the lmc , then fitting the mean magnitudes of the smc cepheids to the lmc in each band , and deriving apparent distance moduli for each band . these distance moduli are then fitted to an extinction law with an adopted parameter ( @xmath118 ) to derive the amount of reddening in the smc data . finally the reddening corrected , true distance modulus is determined . as brief as it is in this description , the technique overlooks several sources of physical scatter and systematics which could be present in the data . we study those below , since they are also part of our method to derive the metallicity dependence . they provide the independent constraints , which make possible the derivation of both @xmath119 and @xmath120 . in our differential analysis between lmc and smc the effect of interstellar reddening ( and obscuration ) is represented by the term @xmath121 , where @xmath122 and the reddening vector @xmath123 we use the @xmath118-dependent galactic extinction law of cardelli , clayton , & mathis ( 1989 ) and the above values correspond to a convolution of the eros filters transmission curves with the @xmath124 extinction law ( fig.2 ) ; here @xmath125 . an extinction law with @xmath124 is widely accepted for the extinction in lmc ( @xmath46 freedman _ et al . _ 1994a ) , while for our galaxy @xmath126 is more common on the average . there are a few lines - of - sight with @xmath118 as high as 5.3 , but that s very rare . physically , high values of @xmath118 appear to be related to systematically larger dust particles in dense regions . however , an empirical relation between the spectral bump " at 2200 and @xmath118 seems to lack physical grounds ( cardelli , clayton , & mathis 1989 ) . when used , this relation predicts a high @xmath118 value ( as high as 5 ) for the smc , where the 2200 feature is known to be very weak ( bessell 1991 ) . this is a good enough reason for us to investigate how our results are affected by the value of the extinction law parameter . therefore we have convolved the transmission curves of the eros filters with @xmath126 and @xmath127 extinction laws as well ( fig . 2 ) . after deriving the apparent distance moduli between smc and lmc for each band @xmath84 each mode of pulsation , we plot them against inverse wavelength at the effective centers of the photometric bands ( fig.3 ) . we fit through the points galactic extinction laws with different amount of extinction , as measured by e(b - v ) , and with different parameters , @xmath118 . concerning the reddening of the magellanic clouds , it is important for our study to know the existing observational limits for the average values . first of all , we have the limit of foreground reddening . towards the smc the foreground ( due mostly to our galaxy ) is smooth and e(b - v ) lies between 0.04 and 0.06 mag ( bessell 1991 ) . within the smc the reddening could vary between 0.06 and 0.3 mag . on the average the reddening of the lmc is very similar : foreground of 0.04 to 0.09 , and same within ( bessell 1991 ) . these values are derived independent of cepheid observations and will prove very useful in our derivation of a metallicity dependence . also , when we refer to cepheids in lmc and smc , one should bear in mind that the cepheid samples are located in the bar of lmc and near the central region of smc . this is particularly important in terms of their mean reddenings ( as samples ) ; the central region of smc is known to be dusty . the extinction diagram in fig.3 provides an opportunity to derive individual reddenings for each cepheid on our list , by fitting an extinction law with a fixed reddening parameter . the individual cepheid reddenings are excellent for the study of the structure of the pl relation ( see next section ) . in deriving individual reddenings we required that no cepheid be assigned a reddening smaller than the known foreground reddening to the mc given above . as it will become clear from the next section , the so derived individual reddenings offer no additional advantage in the differential analysis . it is well known that interstellar dust absorbs or scatters light in the optical range , hence extinction is by definition a non - negative quantity . in our estimate of the extinction , @xmath86 , we need to bear this cutoff in mind , because @xmath86 is small towards the lmc and smc , and for a large subsample of our cepheids we get negative values . given the crucial use of the reddening cutoff in our analysis and final fit to the model , it is important to truncate the measurements of @xmath86 properly , and preserve a useful estimate of the measurment uncertainty , @xmath128 in the process of doing so . the solution to finding the value of @xmath86 and its error @xmath129 with prior knowledge that @xmath86 can not be less than zero is given by bayes s theorem . for an estimate of @xmath86 with value @xmath48 and normal error @xmath130 , and our knowledge of the distribution of the observed @xmath86 , @xmath131 , the probability distribution for the true @xmath86 will be given by the bayesian filter : @xmath132 a practical , and entirely satisfactory assumption would be that @xmath131 is a one - sided gaussian which has a maximum at @xmath133 and declines for large @xmath86 ( see press 1996 ) . the conclusion of our analysis is that the reddening law in the bar of smc is not very different from that of the lmc with a reddening parameter much closer to 3.3 than 5 . concerning the absolute value for the lmc reddening ( currently used : e(b - v)=0.10 and @xmath118=3.3 ) , our mean value , from the individual reddenings , is e(b - v)[email protected] . with the derived parameters in the next section we could constrain the mean reddening of the smc cepheids to e(b - v)@xmath140.10 . now we are ready to apply our method . we start by constructing the pl relations ( in each band ) for the lmc and smc cepheids , separately for fundamental and first overtone pulsators ( fig.4 ) . all pl relations are essentially linear in the range of periods in our data . the slopes , @xmath135 ( eq.1 ) , which are insensitive to moderate amounts of reddening , are the same in lmc and smc : + @[email protected]@xmath1340.16 and @[email protected]@xmath1340.14 for fundamental mode cepheids in lmc , and + @[email protected]@xmath1340.07 and @[email protected]@xmath1340.06 for smc . similarly , for the first overtone cepheids we have : + @[email protected]@xmath1340.22 and @[email protected]@xmath1340.20 for lmc , and @[email protected]@xmath1340.14 and @[email protected]@xmath1340.13 for smc . this lack of metallicity dependence of the slopes , as well as the difference between @xmath136 and @xmath137 are in very good agreement with theory ( stothers 1988 ) . the scatter about the pl relations results from physical reasons , as well as statistical noise . on the basis of our very well sampled light curves in two bands , we have already excluded cepheids which show evidence of blending , as well as stars which are not cepheids , and can analyse the other physical sources of scatter . individually , each cepheid in the lmc can be offset from the pl relation of the lmc cepheids due to 3 @xmath138 sources : ( 1 ) position in the instability strip ; ( 2 ) differential reddening within the lmc ; ( 3 ) differential distance ( depth ) within the lmc . these three additional independent sources of scatter contribute to a natural width ( in l ) of the pl relation . they are manifested in the structure of the lmc covariance matrix ( equation 7 ) , where none of the sources lies in the plane defined by the other two sources ; hence the covariance matrix would be three - dimensional . we illustrate this by plotting against each other the magnitude residuals defined in equation ( 3 ) . in figure 5 we have the diagram of the residuals for the lmc fundamental cepheids . the high level of correlation is evident . the diagram of the residuals shows that reddeing ( solid line ) within the lmc is not the only source of scatter . in figure 6 we have the same diagram for the smc cepheids ( equation 8) ; all of the above applies to them as well . the diagrams of the residuals provide a nice illustration to the structure of the pl relations . for example , if the pl relation had no dispersion , all points would lie in the center ( 0,0 ) of the diagram ( fig . 5 ) . alternatively , if there were groupings of cepheids separated in distance along the line of sight , they will appear clearly as clumps on the diagram . if we deredden each cepheid as described in the previous section 6 , we are mostly left with the dispersion due to depth ( fig . 7 ) . the reason is that the individual dereddening corrects also for the width of the instability strip along the lines of constant period , at least partially . the correction is partial for two reasons : ( 1 ) the reddening and the @xmath139 lines are not completely degenerate ( fig . 8) ; and ( 2 ) the individual reddening is limited ( from below ) by the amount of foreground reddening ( independently known ) . the near - degeneracy of reddening and @xmath139 lines in the color - magnitude plane is worth investigating . we find that the slopes of the @xmath139 lines in lmc and smc are practically the same : @xmath140 ; while the slopes of reddening ( @xmath22 ) are 2.86 ( for @xmath118=3.3 ) and 3.53 ( for @xmath118=5.0 ) . a limited illustration to this is given in fig.8 ; the cepheids shown have been selected to have exactly the same periods ( to the second digit ) @xmath4 in practice we use the whole data - set to derive the slopes . now we turn to accounting for the depth of lmc and smc . in this we employ for the first time the last two observed cepheid quantities : the coordinates . we plot the magnitude residuals versus angle on the sky across the smc ( 1@xmath141=0.0389 mag ) using the line of nodes for smc from caldwell & coulson ( 1986 ) . they derived them over areas which are about 25 times larger than the eros ccd - array areas . the eros areas cover only the lmc bar ( which is seen face on ) and the smc bar ( seen edge - on ) and have dimensions of 25@xmath14271@xmath143 on the sky . that is one reason we did not attempt our own derivation of the line of nodes of the clouds . in deriving the depth dispersion in the smc we used cepheids along the line of nodes and orthogonal to it . with the width of the instability strip derived from the color - magnitude plane to be @xmath144 mags , we derive the depth dispersion in our smc sample to be @xmath145 mags . our uncertainty in the value of @xmath118 limits our ability to derive the internal amount of reddening within smc to the ranges already known ( 6 ) . however , it provides a very valuable constraint from below on the overall reddening towards smc as e(b - v)@xmath140.10@xmath146 ; otherwise half of the cepheids will have negative extinctions , which is unphysical . we use this value to constrain @xmath115 in our multidimensional minimization ( equations 12 - 15 ) . finally , we would like to point out that the eros sample of smc cepheids happen to be mostly from the far arm and the main bar ( see figs.7&9 of caldwell & coulson 1986 ) . hence the sample is not likely to define the centroid of smc well , and would result in a larger distance modulus difference with the lmc ( by about 0.15 @xmath147 ) . this is of no consequence to our differential analysis here . the diagrams of the residuals are independent of differential reddening or metallicity between the mcs , however understanding the structure of each pl relation and quantifying the sources of scatter improve the reliability of the differential analysis . when we @xmath148 the lmc and smc pl relations , the above three sources of scatter are not involved ( as long as we have a statistically large sample of data points , @xmath13 cepheids ) @xmath4 we use @xmath149 quantities . then it is only consistent to use the ensemble " reddening for each mc , the individual reddenings will introduce a bias . other sources of scatter have been proposed in a critique of the plc relation @xmath4 mass loss and different strip crossings ( stift 1982 , 1995 ) . we do not identify another source in our data and in the remaining scatter ; such effects must be subtle for the pl relation alone . we apply the procedure in 5 to the data - sets of fundamental mode and first overtone mode cepheids separately and obtain very similar results . there are enough first overtone cepheids , 27 and 141 respectively , to make this result statistically significant . thus , the two sets of pl relations give the same distance modulus difference ( smc - lmc ) of [email protected] mag ( before corrections for reddening and metallicity ) . this illustrates for the first time the usefulness of the pl relation of first overtone cepheids for determining distances . the model parameters derived from the fits of the different datasets are as follows : @xmath150= 17.61@xmath151,-2.72@xmath152 ; 17.74@xmath153,-2.95@xmath154 ; @xmath155= 14.49@xmath156 , -11.78@xmath157 ; 14.36@xmath158 , -12.77@xmath159 ; @xmath160= 0.62@xmath161 , @xmath115= 0.010@xmath146 , @xmath119= 0.06@xmath146 , and @xmath120= -0.01@xmath146 . this is the fit with the lmc 1993 - 94 data ; the zero points @xmath162 , @xmath163 refer to the smc linear fits . the correction due to the metallicity dependence of the inferred distance modulus by the described technique is then @xmath164mag . if the lmc 1991 - 92 dataset is used , the color shift is smaller by 0.02 @xmath147 to @xmath81= 0.05@xmath146 , but a corresponding change in the pl relations leads to @xmath160= 0.63@xmath161 , @xmath115= 0.025@xmath165 , @xmath119=0.05@xmath146 , and @xmath120= -0.01@xmath146 . therefore , @xmath166mag . in the error budget , in order of importance , the sources are : ( 1 ) the photometric transformation between the 1991 - 92 and present data ( 0.06 mag ) ; ( 2 ) the scatter in the pl relations ( 0.04 mag ) ; ( 3 ) the uncertainty in the reddening ( and depth ) estimate in smc ( 0.03 mag ) . the first error source is not applicable to the lmc 1993 - 94 dataset . the above result is for a fixed reddening vector , @xmath72 ( equation 18 ) . if we change the parameter to @xmath118=5.0 , @xmath13 @xmath72=(5.51 , 3.95 ) , the above result for the metallicity effect is no longer unambiguous . the reason for this is that now it is possible to satisfy the constraint from the foreground smc reddening and correcting for the offset of the smc cepheids in the color - magnitude plane . in other words , if in smc @xmath118=5.0 and e(b - v)=0.04 , one need not invoke a metallicity effect in the differential comparison to lmc . the fact that the smc cepheids appear bluer than the lmc cepheids is then explained with the small amount of extinction and steep extinction law . while a value of @xmath1675.0 is not excluded ( but unlikely ) by our data , an extinction of e(b - v)=0.04 is very difficult to accommodate in our analysis of the magnitude residuals ( see 7 ) . our mean value is [email protected] , and that is in very good agreement with a number of independent estimates , as reviewed by bessell ( 1991 ) . therefore we consider this alternative explanation ( very high @xmath118 and very low reddening ) as very unlikely . we propose instead that the observed effect is due to the known difference in metallicity between the cepheid samples of the lmc and smc : 0.14 @xmath147 for a factor of 2 difference in heavy elements . the sign of the effect would lead to the distance of a metal - poor sample to be overestimated . in applying the above result to other galaxies , we need to address two issues : ( 1 ) about the form of the metallicity dependence of inferred cepheid distances , and ( 2 ) about the role of the helium abundance ( @xmath168 ) . the form of the dependence can be inferred from current theoretical models . photometric colors and bolometric corrections ( from model atmospheres ) have linear dependencies on @xmath169(@xmath170 ) , where @xmath171 is the abundance of metals ( by mass ) and z=0.016 in the galactic cepheids . mass - luminosity relations , hence @xmath4 @xmath74 , seem to depend linearly on @xmath171 ( chiosi , wood , & capitanio 1993 ; buchler _ et al . in addition , these models show a significant linear dependence on the abundance of helium ( @xmath168 ) . we have currently no data on @xmath95y between the lmc and smc cepheids , therefore in applying our metallicity dependence on distance to other galaxies , we have to assume that @xmath172 . that assumption is strongly supported by the current level of understanding and measurement in stellar populations relevant to our study ( e.g. olive , skillman , & steigman 1996 ) , and will not affect any conclusions with regards to distances . we also want to point out that the above needs to be born in mind in comparisons to results from theoretical modelling , e.g. chiosi _ et al . _ ( 1993 ) use y=0.27 for both lmc ( z=0.008 ) and smc ( z=0.004 ) in table 15 , which may not be a good assumption . in view of the above and our model , we derive the following metallicity dependence of inferred cepheid distances : @xmath173 where @xmath171 is the abundance of metals ( by mass ) in the studied cepheids and @xmath174=0.0085 . the metallicity dependence is valid in the spectral region covered by the eros filters . due to these filters transformation properties into the standard @xmath17 system , and the application at hand , we have derived it for v , and v - i use in particular , where @xmath175 . the above metallicity dependence applies to distances inferred by differencing against the lmc cepheids and cepheid derived reddenings ( what we refer to as the _ modern technique _ ) , and @xmath176=3.3 . the metallicity dependence we find is two times less steep than that found by gould ( 1994 ) from freedman & madore s m31 data ( [email protected] [ fe / h ] ) . it is also smaller than stothers prediction ( 28.7 @xmath177 , for @xmath178 ) . however our result is in perfect agreement with the metallicity effect seen in galactic cepheids as a function of galactocentric distance . the observed abundance gradient for cepheids is @xmath179}/\delta{r_{gc}}=-0.07 \pm 0.02 kpc^{-1}$ ] ( giridhar 1986 ) . two independent cepheid samples show the same effect of progressively bluer color with increasing galactocentric distance ( caldwell & coulson 1987 , and gieren _ _ 1993 ) , which converts to the following values for @xmath81 ( of equation 10 ) in units of [ fe / h ] : @xmath180 for ( b - v ) , and @xmath181 for ( v - i ) . for ( v - i ) of our system we obtained @xmath182 . exactly the same value is obtained ( and a nice illustration of the effect ) from the compilation of lmc and smc cepheids in the study by di benedetto ( 1995 ) @xmath4 see his fig.1 & 2 , and also 8 . our comparison to the work of caldwell & coulson ( 1986 ) in @xmath183 and of laney & stobie ( 1994 ) in @xmath18 is limited for at least two reasons . first , we can not reproduce their reddening determinations with only two bands . second , we are unable to comment on the theoretical models which they used as constraints . nevertheless , we find a similar intrinsic difference in @xmath29 colors between lmc and smc as caldwell & coulson ; our value is two times larger . we also agree with laney & stobie s main conclusions , despite the uncertainty of a @xmath184 to @xmath26 transformation needed for such a comparison . the metallicity dependence is not strong enough to upset the overall agreement of the four primary distance indicators in fifteen local distances ( huterer , sasselov , & schechter 1995 ) . similarly , the discrepant distance moduli involving ic 1613 noted by gould ( 1994 ) are less discrepant with our weaker metallicity dependence ; the rr lyrae distances may be partly responsible for the remaining difference . despite its relative weakness , this metallicity dependence has a significant effect on extragalactic distance measurements . the effect of metallicity on cepheid luminosity , when not accounted for , can lead to two types of errors in distance measurements ( gould 1994 ) : incorrect @xmath5 value , or a dependence of @xmath5 on distance appearing as proper motions or malmquist bias . here we illustrate the effect of our metallicity dependence on the value of @xmath5 , as based on cepheids . recent efforts to determine @xmath5 from cepheid distances have focused on : ( 1 ) the virgo and fornax cluster galaxies ( freedman _ et al . _ 1994b , 1996 ) , ( 2 ) the leo i group , containing ellipticals ( tanvir _ et al . _ 1995 ) ; and ( 3 ) the parent galaxies of supernovae of type ia ( sandage _ et al . all these studies use the same modern technique with the lmc as a base and all the same initial assumptions ; we share them in our analysis . yet they result in three different values of @xmath5 , with hardly overlaping error bars . the abundances of metals in the young populations of the host galaxies for these studies differ and , apparently , in a systematic way . therefore we propose that the metallicity dependence may be responsible for most of this discrepancy . to illustrate our idea , below we apply equation ( 19 ) ; please note that the uncertainties in the metallicities of the extragalactic cepheids are not yet well known , hence we do not try to provide rigorous error estimates to the corrected values . with the first approach and hst @xmath27 photometry of cepheids in m100 , freedman _ et al . _ ( 1994b ) derive @xmath185 km.s@xmath10@xmath11 . for the metallicity of the cepheids we adopt [ fe / h]=+0.1 ( z=0.021 ) , derived from the abundances of h ii regions ( zaritsky _ et al . _ 1994 ) , assuming a solar [ o / fe ] ratio . with the uncertainty range for the slope of our metallicity effect , this leads to @xmath186 km.s@xmath10@xmath11 . with some of the color difference due to metallicity effects , the estimate of interstellar reddening will also change ( the metal - rich m100 cepheids are intrinsically redder " than the lmc ones ) @xmath4 @xmath46 from e(v - i)=0.13@xmath187 ( after ferrarese _ et al . _ 1996 ) we obtain 0.05 @xmath147 . recent interim results reported on half - dozen more galaxies ( freedman , madore , & kennicutt 1996 ) include mostly spirals which are not as metal rich as m100 , _ e.g. _ m101 has lmc abundances in its outer field . all this brings their interim result down to 73@xmath12 from their initial m100-based value , as we would expect from our metallicity effect . a complete analysis should be done when all data becomes available . with the second approach and hst @xmath27 photometry of cepheids in m96 , tanvir _ et al . _ ( 1995 ) derive @xmath188 km.s@xmath10@xmath11 . the abundance estimate for m96 comes form oey & kennicutt ( 1993 ) , slightly below solar at [ fe / h][email protected] . this implies a small change in the cepheid distance to m96 and a decrease in the value of @xmath5 by 3@xmath45 km.s@xmath10@xmath11 . the reddening changes from e(v - i)=0.09@xmath189 to 0.03 @xmath147 . with the third approach and hst @xmath27 photometry of cepheids in ic4182 and ngc5253 , sandage _ et al . _ ( 1994 ) derive @xmath190 km.s@xmath10@xmath11 . abundances in ngc5253 have been measured and discussed by pagel _ et al . _ ( 1992 ) ; discussion of abundances in ic4182 is given by saha _ et al . _ ( 1994 ) @xmath4 we adopt [ fe / h][email protected] and caution on the large uncertainties . with this metallicity the above estimate of @xmath5 should be increased by 6 to 13 km.s@xmath10@xmath11 . the reddening estimates change from e(v - i)[email protected]@xmath191 and 0.03@xmath192 , to e(v - i)=0.09@xmath191 and 0.23@xmath192 , respectively ( these metal - poor cepheids are bluer " than the lmc ones ) . here we should point out that the interpretation of the light curves of sn ia themselves has been improved considerably by riess , press , & kirshner ( 1995 ) , leading to an additional systematic shift upwards by 6 km.s@xmath10@xmath11 . this brings about the estimate to @xmath193 , which is also in good agreement with the theoretical calibration of sn ia and the value @xmath194 km.s@xmath10@xmath11 by hflich & khokhlov ( 1996 ) . the new value of @xmath195 km.s@xmath10@xmath11 by sandage et al . ( 1996 ) includes two sne ia in spiral galaxies , in which cepheid metallicity is most likely similar to that of the lmc . note , however , that the two sne @xmath4 1981b ( ngc4536 ) and 1990n ( ngc4639 ) are by [email protected] @xmath147 dimmer than the three sns in ic4182 and ngc5253 . therefore , the result we obtained above by accounting the effect of metallicity on the cepheid distances for each galaxy , remains virtually unchanged , and so does the consistency with the results of riess , press , & kirshner ( 1995 ) . the metallicity dependence we found from the lmc / smc analysis brings all the derivations of @xmath5 to good agreement . it also leads to reasonable amounts of reddening , especially regarding the negative values obtained otherwise in metal - poor samples . secondary distance indicators , like the tully - fisher ( tf ) relation , the surface brightness fluctuations ( sbf ) method , and the planetary nebulae luminosity function ( pnlf ) method , have their zero points calibrated with cepheids . the tf relation is calibrated primarily in m81 ( freedman _ et al . _ 1994a ) with 25 cepheids . the metallicity of these cepheids is high , [ fe / h]@xmath1960.05 ( garnett & shields 1987 ; zaritsky _ et al . the sbf and pnlf zero points come from m31 , and in a similar fashion @xmath4 from metal - rich cepheids ( freedman & madore 1990 ) . given our metallicity effect , we are not surprised to note that all these methods give a high value for @xmath5 . we use the new eros microlensing survey data - set of 3 million two - color observations of about 500 cepheids in the lmc and smc to search for and derive the dependence of the optical pl relations on metallicity . we find that : + ( 1 ) the pl relations for both types of cepheids have the same zero - point offset ( but no slope difference ) , which is attributed to the effect of metallicity under a reasonable assumption about the extinction law . as expected from theory , this effect of metal content is manifested in a color shift of the instability strip . it amounts to about @xmath197 of the strip width , and we could detect it unambiguously thanks to the large number of cepheids and extremely well sampled light curves . + ( 2 ) with the known ensemble difference in metal content between lmc and smc cepheids , we derive a linear relation between the distance modulus correction and metallicity : @xmath173 it applies to distances which are inferred by using lmc as a base _ and _ using two color @xmath27 photometry of the cepheids to establish the reddening . the linearity of metallicity dependence is a good assumption , but needs to be confirmed empirically outside the range of application ( a factor of few lower than smc and higher than the galaxy ) . + ( 3 ) the first overtone cepheids have pl relations which provide distances fully consistent with the pl relations of fundamental mode cepheids . our result can be applied to the long - standing discrepancy between the low-@xmath5 scale and the high-@xmath5 scale . the host galaxies on which each of these scales relies appear to have systematically different metallicities . a simple application of our correction to several recent derivations makes the low-@xmath5 values ( sandage _ et al . _ 1994 ) @xmath7 and the high-@xmath5 values ( freedman _ et al . _ 1994b ) @xmath8 , thus bringing those discrepant estimates into agreement near @xmath9 km.s@xmath10@xmath11 . alcock c. , akerlof c.w . , allsman r.a . , axelrod t.s . , bennett d.p . , chan s. , cook k.h . , freeman k.c . , griest k. , marshall s.l . , park h - s . , perlmutter s. , peterson b.a . , pratt m.r . , quinn p.j . , rodgers a.w . , stubbs c.w , sutherland w. nature , 1993 , * 365 * , 621 aubourg e. , bareyre p. , brehin s. , gros m. , lachieze - rey m. , laurent b. , lesquoy e. , magneville c. , milsztajn a. , moscoso l. , queinnec f. , rich j. , spiro m. , vigroux l. , zylberajch s. , ansari r. , cavalier f. , moniez m. , beaulieu j.p . , ferlet r. , grison ph . , vidal - madjar a. , guibert j. , moreau o. , tajahmady f. , maurice e. , prevot l. , gry c. , the messenger , 1993a , * 72 * , 20 aubourg e. , bareyre p. , brehin s. , gros m. , lachieze - rey m. , laurent , b. , lesquoy e. , magneville c. , milsztajn a. , moscoso l. , queinnec f. , rich j. , spiro m. , vigroux l. , zylberajch s. , ansari r. , cavalier f. , moniez m. , beaulieu j.p . , ferlet r. , grison ph . , vidal - madjar a. , guibert j. , moreau o. , tajahmady f. , maurice e. , prevot l. , gry c. , nature , 1993b , * 365 * , 623 beaulieu , j.p . , _ et al . _ 1996 , in preparation . bessell , m.s . 1991 , a&a , 242 , l17 buchler , j.r . , kollath , z. , beaulieu , j.p . , & goupil , m.j . , 1996 , apjl , 462 , l83 . caldwell , j.a.r . & coulson , i.m . , 1985 , mnras , 212 , 879 caldwell , j.a.r . & coulson , i.m . , 1985 , mnras , 218 , 223 caldwell , j.a.r . & coulson , i.m . , 1987 , aj , 93 , 1090 chiosi , c. , wood , p.r . , & capitanio , n. , 1993 , apjs , 86 , 541 . di benedetto , g.p . , 1995 , apj , 452 , 195 . feast , m.w . 1991 , in _ observational tests of cosmological inflation _ , eds . shanks et al . , p.147 . feast , m.w . 1995 , in _ astrophysical applications of stellar pulsation _ , eds . stobie & whitelock , asp ( san francisco ) , p.209 . ferrarese , l. , freedman , w.l . , hill , r.j . , saha , a. , _ et al . _ 1996 , apj , 464 , 568 freedman , w.l . , & madore b.f . , 1990 , apj , 365 , 186 freedman , w.l . , hughes , s. m. , madore , b.f . , _ et al . _ 1994a , apj , 427 , 628 freedman , w.l . , madore , b.f . , mould , j. r. , _ et al . _ 1994b , nature , 371 , 757 freedman , w.l . , madore , b.f . , & kennicutt , r.c . 1996 , in stsci symposium on `` the extragalactic distance scale '' , ed . m.livio , in press . gieren , w.p . , barnes , t.g . , & moffett , t.j . 1993 , apj , 418 , 135 giridhar , s. 1986 , j.astrophys.astron . , 7 , 83 gould , a. 1994 , apj , 426 , 542 hflich , p. & khokhlov , a. 1996 , apj , 458 , 500 huterer , d. , sasselov , d.d . , & schechter , p.l . 1995 , aj , 110 , 2705 laney , c.d . , & stobie , r.s . 1994 , mnras , 266 , 441 . madore , b.f . , & freedman , w.l . 1991 , pasp , 103 , 933 martin , w.l . , warren , p.r . , & feast , m.w . 1979 , mnras , 188 , 139 . oey , m.s . , & kennicutt , r.c . 1993 , apj , 411 , 137 olive , k.a . , skillman , e. , & steigman , g. , 1996 , preprint ( umn - th-1514/96 ) , astro - ph/9611166 . pagel , b.e.j . , simonson , e.a . , terlevich , r.j . , & edmunds , m.g . 1992 , mnras , 255 , 325 pierce , m.j . , welch , d.l . , mcclure , r.d . , van den bergh , s. , racine , r. , & stetson , p.b . 1994 , nature , 371 , 385 press , w.h . , teukolsky , s.a . , vetterling , w.t . , & flannery , b.p . 1992 , _ numerical recipes _ , 2nd ed . , ( cambridge university press , cambridge ) press , w.h . 1996 , preprint , astro - ph/9604126 . riess , a.g . , press , w.h . , & kirshner , r.p . 1995 , apj , 438 , l17 saha , a. , labhardt , l. , schwengeler , h. , macchetto , f.d . , panagia , n. , sandage , a. , & tammann , g.a . 1994 , apj , 425 , 14 sandage , a. , saha , a. , tammann , g.a . , labhardt , l. , schwengeler , h. , panagia , n. , & macchetto , f.d . 1994 , apj * 423 * , l13 sandage , a. , saha , a. , tammann , g.a . , labhardt , l. , panagia , n. , & macchetto , f.d . 1996 , apjl , in press . schechter , p.l . 1980 , aj , 85 , 801 stift , m.j . 1982 , a&a , 112 , 149 stift , m.j . 1995 , a&a , 301 , 776 stothers , r. b. 1988 , apj , 329 , 712 tanvir , n.r . , shanks , t. , freguson , h.c . , & robinson , d.r.t . 1995 , nature , 377 , 27 zaritsky , d. , kennicutt , r.c . , & huchra , j. p. 1994 , apj , 420 , 87

we study partially and totally associative ternary algebras of first and second kind . assuming the vector space underlying a ternary algebra to be a topological space and a triple product to be continuous mapping we consider the trivial vector bundle over a ternary algebra and show that a triple product induces a structure of binary algebra in each fiber of this vector bundle . we find the sufficient and necessary condition for a ternary multiplication to induce a structure of associative binary algebra in each fiber of this vector bundle . given two modules over the algebras with involutions we construct a ternary algebra which is used as a building block for a lie algebra . we construct ternary algebras of cubic matrices and find four different totally associative ternary multiplications of second kind of cubic matrices . it is proved that these are the only totally associative ternary multiplications of second kind in the case of cubic matrices . we describe a ternary analog of lie algebra of cubic matrices of second order which is based on a notion of-commutator and find all commutation relations of generators of this algebra . 17a40 , 20n10 . [ firstpage ]

0901.2506

a ternary algebra or triple system is a vector space @xmath1 endowed with a ternary law of composition @xmath2 which is a linear mapping with respect to each its argument , and we will call this mapping a ternary multiplication or triple product of a ternary algebra @xmath1 . hence a ternary algebra is an algebra which closes under a suitable triple product . obviously any binary algebra which closes under double product can be considered as a ternary algebra if one defines the ternary multiplication as twice successively applied binary one , and in this case the ternary multiplication is generated by a binary one . however there are ternary multiplications which can not be obtained as twice successively applied binary multiplication . for instance , pure imaginary numbers or elements of grading one of a superalgebra closes under triple product . a well known example of a ternary matrix algebra is the vector space @xmath3 of @xmath4 matrices endowed with the ternary multiplication @xmath5 , where @xmath6 and @xmath7 is transpose of the matrix @xmath8 . since lie algebras play a fundamental role in physics , particular attention was given to ternary algebras when they were shown to be building blocks of ordinary lie algebras . given ternary algebra one can construct a lie algebra by using the method proposed by kantor in @xcite . this method was extended to super lie algebras in @xcite and later was applied by the same authors in @xcite to construct a gauge field theory by introducing fundamental fields associated with the elements of a ternary algebra . a skew - symmetric bilinear form is an important component in the large class of algebraic structures such as lie algebras , grassmann algebras and clifford algebras . for example , the lie brackets @xmath9:{\frak l}\times{\frak l}\to { \frak l}$ ] of a lie algebra @xmath10 is the skew - symmetric bilinear form , and the multiplication @xmath11 of a grassmann algebra @xmath12 restricted to the subspace of odd elements is the skew - symmetric bilinear form . a skew - symmetry of a bilinear form can be interpreted by means of the faithful representation of the symmetric group @xmath13 , where @xmath14 is the identity permutation , as follows : a bilinear form @xmath15 is skew - symmetric if @xmath16 . making use of this interpretation we can construct a ternary analog of a skew - symmetric bilinear form replacing @xmath17 by @xmath18 with its faithful representation by cubic roots of unity @xmath19 , i.e. @xmath20 , where @xmath14 is the identity permutation and @xmath21 are the cyclic permutations , as follows : a trilinear form @xmath22 is called @xmath0-skew - symmetric if for any elements @xmath23 of a vector space @xmath1 it satisfies @xmath24 the notion of a @xmath0-skew - symmetric form can be assumed as a basis for a ternary analog of grassmann , clifford and lie algebras . these ternary structures were developed in @xcite and applied to construct a ternary analog of supersymmetry algebra in @xcite . in this paper we study algebras with ternary law of composition . in section 2 we consider partially and totally associative ternary algebras of first and second kind . we show that a triple product of a ternary algebra induces three binary multiplications and find the sufficient and necessary condition a triple product of a ternary algebra must satisfy in order to induce the associative binary algebra . assuming the vector space underlying a ternary algebra to be a topological space and a triple product to be continuous mapping we consider the trivial vector bundle over a ternary algebra and show that a triple product induces a structure of binary algebra in each fiber of this vector bundle . the sufficient and necessary condition a ternary multiplication must satisfy in order to induce a structure of associative binary algebra in each fiber is given in terms of the vector bundle over a ternary algebra . the relations for different kinds of partial and total associativity of a ternary algebra and induced by it binary algebras are found in the terms of the structure constants of a ternary algebra . it should be pointed out that the cohomologies of a ternary algebra of associative type are studied in @xcite . in section 3 we consider an algebraic structure consisting of two bimodules over unital associative algebras with involution and construct a ternary algebra by means of this algebraic structure . choosing different modules , unital associative algebras and homomorphisms we show that this structure allows to construct a large class of ternary algebras including a ternary algebra of rectangular matrices and ternary algebras of sections of a vector bundle over a smooth finite dimensional manifold . we end the section 3 by constructing the binary lie algebra of matrices whose entries are the elements of bimodules and unital associative algebras . it should be mentioned that there are @xmath25-ary generalizations of lie algebra which include the concepts such as @xmath25-ary algebra of lie type enclosing @xmath25-ary nambu algebra , @xmath25-ary nambu - lie algebra . the concept of @xmath25-ary hom - algebra structure generalizing previously mentioned @xmath25-ary generalizations of lie algebra is introduced and studied in @xcite . a good and detailed survey on the theory of ternary algebras can be found in @xcite . it is well known that a large class of associative algebras can be constructed by means of square matrices and their multiplication . though the rectangular matrices can be successfully used to construct a ternary algebra we think that probably more appropriate objects to construct ternary algebras are the cubic matrices . our aim in section 4 is to construct ternary algebras of cubic matrices and to study their structures . we find four different totally associative ternary multiplications of second kind of cubic matrices and prove that these are the only totally associative ternary multiplications of second kind in the case of cubic matrices . it is worth mentioning that our search for associative ternary multiplications of cubic matrices has shown that there is no totally associative ternary multiplication of first kind in the case of cubic matrices . i section 5 we describe the ternary analog of lie algebra of cubic matrices of second order by finding all commutation relations of generators of this algebra with respect to @xmath0-commutator . in this section we remind a notion of a ternary algebra and its partial or total associativity of first or second kind . holding fixed one argument of a ternary multiplication we get the binary multiplications and study the relation between the associativity of a ternary multiplication and the associativity of induced binary multiplication . we propose to use a vector bundle approach to describe the family of binary algebras induced by a ternary algebra . let @xmath26 be complex vector spaces , and @xmath27 be a @xmath28-valued trilinear form . we will call @xmath22 a ternary law of composition or ternary multiplication on @xmath1 if @xmath22 is a @xmath1-valued trilinear form . the pair @xmath29 is said to be a ternary algebra or triple system if @xmath1 is a complex vector space , and @xmath30 is a ternary law of composition on @xmath1 . it is obvious that relation analogous to binary associativity in the case of ternary law of composition should contain at least five elements of @xmath1 . there are three different ways to apply twice a ternary multiplication @xmath22 to ordered sequence of five elements @xmath31 which lead us to the following relations defining a notion of partial associativity for ternary multiplication : @xmath32 hence we have three different kinds of partially associative ternary algebra @xmath33 which will be called -partially associative ternary algebra ( [ lr - kind ] ) , -partially associative ternary algebra of first kind ( [ lc - kind ] ) and -partially associative ternary algebra of first kind ( [ cr - kind ] ) . a ternary algebra @xmath33 is said to be totally associative ternary algebra of first kind if its ternary multiplication @xmath22 satisfies any two of the relations ( [ lr - kind])([cr - kind ] ) . it is obvious that in the case of totally associative ternary algebra of first kind a ternary multiplication @xmath22 satisfies the relations @xmath34 where @xmath31 . the notion of totally associative ternary algebra of first kind can be viewed as a direct ternary generalization of classical associativity @xmath35 , where @xmath36 are the elements of an algebra @xmath37 with a binary law of composition @xmath38 , when one applies twice algebra multiplication ( binary or ternary ) to ordered sequence of elements of algebra successively shifting the first ( interior ) multiplication from left to right and setting equal obtained products . in this sense the notion of -partial associativity can be considered as most similar to classical associativity whereas the notion of -partial or -partial associativity can be defined for the first time only in the case of ternary multiplication because in the case of binary multiplication there is no central group of two elements in the middle of a sequence @xmath39 . since the notion of -partial or -partial associativity appears for the first time in the case of ternary multiplication there is no reason to keep the requirement of fixed order of a sequence @xmath31 looking for a possible analogs of associativity in the case of ternary algebras . it turns out that we get a useful notion of ternary associativity giving up the requirement of fixed order of elements in a sequence @xmath31 . this means that unlike the case of ternary associativity of first kind we not only successively shift the first ( interior ) multiplication inside a sequence of elements @xmath31 from left to right but at the same time permute the elements @xmath40 in the middle of sequence . obviously we should use non - cyclic permutation in order to get the initial order of a sequence @xmath41 on the second step . this reasoning leads us to the following relations : @xmath42 a ternary algebra @xmath29 is said to be -partially associative ternary algebra of second kind if ternary multiplication @xmath22 satisfies ( [ lc - secondkind ] ) and -partially associative ternary algebra of second kind if @xmath22 satisfies ( [ cr - secondkind ] ) . a ternary algebra @xmath29 is said to be totally associative ternary algebra of second kind if it is -partially associative and either -partially associative of second kind or -partially associative of second kind . hence in the case of totally associative ternary algebra of second kind we have @xmath43 a ternary multiplication @xmath22 of ternary algebra @xmath33 has three arguments @xmath44 , where @xmath45 , and if we fix one of them then @xmath22 induces the binary multiplication on @xmath46 . it is obvious that this allows us to split the ternary multiplication @xmath22 into three binary ones . we can study the structure of ternary multiplication @xmath22 from this point of view by making use of known concepts and methods of the theory of binary algebras . given an element @xmath47 a ternary multiplication @xmath22 induces three binary multiplications @xmath48 on @xmath1 defined as follows : @xmath49 where @xmath50 . the binary multiplications @xmath51 are not independent because of the relations @xmath52 a vector space @xmath1 equipped with the binary multiplication @xmath53 becomes the binary algebra which will be denoted by @xmath54 . considering an element @xmath55 in @xmath54 as a parameter ranging within a vector space @xmath1 we have three families of binary algebras @xmath56 induced by a ternary multiplication @xmath22 . the family of binary algebras @xmath54 is said to be an associative family of binary algebras induced by a ternary algebra @xmath33 if for any @xmath31 it holds @xmath57 taking @xmath58 in the previous relation we see that each associative family of binary algebras @xmath54 is the family of associative binary algebras . it is useful to describe the above mentioned families of binary algebras in terms of vector bundle . for this purpose we will assume that @xmath46 is a topological vector space , and a ternary multiplication @xmath2 is a continuous mapping . let us consider the direct product @xmath59 as the trivial vector bundle over the base space @xmath46 with the fiber @xmath1 and the projection @xmath60 defined as usual @xmath61 , where @xmath62 . any fiber @xmath63 of @xmath64 is isomorphic to @xmath46 , and we will denote this isomorphism at a point @xmath55 of the base space @xmath1 by @xmath65 , i.e. @xmath66 and @xmath67 , where @xmath68 . let @xmath69 be two points of the base space of a vector bundle @xmath64 . then @xmath70 is the isomorphism between two fibers . in order to apply the constructed vector bundle @xmath71 to describe the families of binary algebras induced by a ternary algebra @xmath33 within the framework of a single structure we assume that the base space @xmath1 of this bundle is equipped with a ternary multiplication @xmath22 . for any point @xmath47 of the base space a fiber @xmath72 at this point is endowed with one of the binary multiplications @xmath48 which we carry over from the family of binary algebras to fibers of @xmath71 by requiring @xmath65 to be an isomorphism of algebras , i.e. @xmath73 where @xmath74 . if each fiber of @xmath71 is endowed with a binary multiplication @xmath75 then in order to emphasize this algebraic structure of fibers we will denote the corresponding vector bundle by @xmath76 . thus @xmath77 , where the base space ( the first factor in the direct product ) is a ternary algebra @xmath33 , and a fiber @xmath63 is the binary algebra @xmath54 . we will call @xmath76 , where @xmath78 , the vector bundle of binary algebras over a ternary algebra @xmath33 . a section @xmath79 of the vector bundle @xmath76 is a continuous mapping @xmath80 satisfying @xmath81 , and the vector space of continuous sections will be denoted by @xmath82 . evidently this vector space equipped with the binary multiplication @xmath83 where @xmath84 , is the binary algebra . the notion of an associative family of binary algebras defined by ( [ associativefamily ] ) can be described in the terms of vector bundle @xmath76 . let @xmath85 be two points of a vector bundle @xmath76 and @xmath47 . a vector bundle of binary algebras @xmath76 is said to be an associative vector bundle of binary algebras over a ternary algebra @xmath33 if for any @xmath85 and @xmath47 it holds @xmath86 particularly ( [ associative_family ] ) implies the associativity of a fiber @xmath63 for any @xmath47 if we take @xmath87 in ( [ associative_family ] ) , i.e. any associative vector bundle of binary algebras @xmath76 is a vector bundle of associative binary algebras whereas the converse is generally not true . now it is natural to pose a question concerning the associativity of induced binary algebras @xmath54 provided a ternary algebra @xmath33 is partially or totally associative of first or second kind . [ lr - associativity ] a ternary algebra @xmath33 is -partially associative ternary algebra if and only if @xmath88 is the associative vector bundle of binary algebras over a ternary algebra @xmath33 . particularly if a base space @xmath33 is -partially associative ternary algebra then each fiber of the vector bundle @xmath88 is an associative binary algebra and the binary algebra of sections of this bundle @xmath89 is associative algebra . indeed the left side of ( [ associative_family ] ) can be transformed as follows : @xmath90 analogously for the right side of ( [ associative_family ] ) we have @xmath91 and this proves the -partial associativity of a ternary algebra @xmath33 . it is well known that any associative binary algebra is the lie algebra with respect to the commutator defined with the help of a binary multiplication of this algebra , and the associativity of a binary multiplication implies the jacoby identity for the commutator . it follows from the proposition [ lr - associativity ] that if a ternary algebra @xmath92 is -partially associative ternary algebra then each fiber @xmath63 , where @xmath93 , of the vector bundle @xmath88 is the lie algebra with respect to the commutator @xmath9_a$ ] defined by @xmath94_a=\binc(p , q)-\binc(q , p)\ ] ] where @xmath74 . clearly the associative binary algebra of sections @xmath89 is the lie algebra under the commutator @xmath95(a)=[\xi(a),\eta(a)]_a$ , \quad where\quad $ \xi,\eta\in \gamma({\frak e}_2)\ ] ] a ternary algebra @xmath33 is said to be a ternary algebra of lie type of first kind ( of second kind ) if for any @xmath96 it holds @xmath97 where @xmath98 is the symmetry group of third order and @xmath99 is its cyclic subgroup . clearly any ternary algebra of lie type of second kind is a ternary algebra of lie type of first kind whereas the converse is generally not true . from ( [ grassmann_type ] ) it follows that any element @xmath55 of a ternary algebra of lie type ( of first or second order ) satisfies @xmath100 . it is pointed out in the introduction that we can construct a ternary analog of the notion of skew - symmetry by means of a faithful representation of @xmath99 by cubic roots of unity . let @xmath101 be the primitive cubic root of unity . a ternary multiplication @xmath22 of a ternary algebra @xmath33 is said to be @xmath0-skew - symmetric if @xmath102 where @xmath45 . if a ternary multiplication @xmath22 of @xmath33 is @xmath0-skew - symmetric then @xmath33 is a ternary algebra of lie type of second order . indeed in this case we have @xmath103 we see that the notion of @xmath0-skew - symmetric ternary multiplication is based on the faithful representation of the cyclic group @xmath99 by cubic roots of unity . given a ternary algebra @xmath33 we can make it the ternary algebra of lie type of second order by endowing it with the ternary @xmath0-brackets or the ternary @xmath0-commutator which is defined by @xmath104=\tau(a , b , c)+j\,\tau(b , c , a)+j^2\,\tau(c , a , b)\ ] ] if @xmath1 has an involution @xmath105 then @xmath22 will be called hermitian if it satisfies @xmath106 . hence a hermitian @xmath0-skew - symmetric ternary multiplication @xmath22 satisfies ( [ cyclic j - symmetric ] ) and @xmath107 let us suppose that @xmath108 is a -partially nonassociative ternary algebra , i.e. in general we have @xmath109 , where @xmath110 . a ternary algebra @xmath108 is said to be a ternary algebra of jordan type if its ternary multiplication @xmath111 satisfies the following identities : @xmath112 where @xmath45 . it is easy to see that if @xmath108 is a ternary algebra of jordan type then for any @xmath47 the binary algebra @xmath113 is the jordan algebra . indeed in this case the identities ( [ first_identity],[second_identity ] ) take on the form @xmath114 if @xmath33 is -partially associative ternary algebra then the ternary algebra @xmath108 , where @xmath115 is the ternary algebra of jordan type . we see that having fixed one variable in a triple product @xmath44 of a ternary algebra @xmath33 we can study the structure of a ternary multiplication @xmath22 by splitting it into three binary ones . what kind of structures induces a ternary multiplication of @xmath33 if one fixes two variables in @xmath44 ? obviously fixing two variables we get the linear operator acting on @xmath46 , and this is the second way for studying the structure of a ternary multiplication . let @xmath116 be the algebra of linear operators of the vector space @xmath1 . given a pair @xmath117 we define the linear operators @xmath118 , where @xmath78 , as follows : @xmath119 actually these operators are not independent because for any @xmath120 we have the relations @xmath121 it is easy to see that for every @xmath122 the linear operator @xmath123 is bilinear with respect to its variables @xmath124 , and therefore the family of linear operators @xmath125 determines the bilinear mapping @xmath126 . on the other hand if there is a vector space @xmath1 equipped with a bilinear mapping @xmath127 then one can construct the ternary algebra @xmath33 by letting @xmath128 . now we can introduce an analog of identity element for a ternary algebra @xmath33 by means of the bilinear mappings @xmath129 , @xmath78 . indeed given an element @xmath47 we define @xmath130 by @xmath131 where @xmath132 is the identity operator . a pair @xmath133 is said to be an identity @xmath122-pair for an element @xmath55 if @xmath134 . let @xmath135 and @xmath136 . if @xmath137 then we will call an element @xmath138 an identity @xmath122-pair , and similarly if @xmath139 then we will call an element @xmath140 an identity pair of a ternary algebra @xmath33 . let us now assume that the vector space @xmath1 of a ternary algebra @xmath141 is a finite dimensional vector space , i.e. @xmath1 is an @xmath142-dimensional vector space and @xmath143 is a basis for @xmath144 then for any element @xmath145 we have @xmath146 , and the triple product of elements @xmath120 can be expressed as follows : @xmath147 where @xmath148 are the structure constants of a ternary algebra @xmath33 defined by @xmath149 if @xmath150 is another basis for a vector space @xmath151 and @xmath152 where @xmath153 is the transition matrix then @xmath154 where @xmath155 are the structure constants of a ternary algebra @xmath141 with respect to a basis @xmath156 , i.e. @xmath157 , and @xmath158 is the inverse matrix of @xmath159 . if we require a ternary algebra @xmath33 to be a partially or totally associative ternary algebra either of first or second kind then this requirement leads to the relations the structure constants of @xmath33 have to satisfy . these relations for different kinds of associativity of first kind have the following form : @xmath160 it follows from the above relations that if @xmath141 is a totally associative ternary algebra of first kind then the structure constants satisfy @xmath161 in the case of ternary associativity of second kind we have the following relations : @xmath162 the structure constants of a totally associative ternary algebra of second kind satisfy @xmath163 for any @xmath47 a ternary algebra @xmath33 induces three binary algebras @xmath164 , with the binary multiplications defined by relations @xmath165 . the structure constants @xmath166 of binary algebra @xmath167 defined by @xmath168 can be expressed in terms of the structure constants of a ternary algebra @xmath33 as follows : @xmath169 where @xmath170 . in this section we propose few different methods for constructing ternary algebras and apply these methods to construct a ternary algebra of vector fields on a smooth finite dimensional manifold and a ternary algebra of rectangular matrices . particularly the curvature of an affine connection determines the structure of a ternary algebra on the module of vector fields on a smooth manifold . given two modules over the algebras with involutions we construct a ternary algebra which is used to construct a lie algebra . our approach generalizes the approach proposed in @xcite , where the authors use the rectangular complex matrices . let @xmath1 be a vector space over the complex numbers @xmath171 and @xmath172 be the dual space . given a @xmath171-multilinear mapping @xmath173 we construct the ternary algebra @xmath33 by defining the ternary multiplication @xmath22 as follows : @xmath174 where @xmath175 . particularly given a @xmath171-bilinear mapping @xmath127 , where @xmath116 is the algebra of linear operators of a vector space @xmath1 , we define @xmath176 and applying ( [ dual_t ] ) we get the ternary algebra @xmath33 whose ternary multiplication @xmath22 can be described implicitly by the formula @xmath177 applying this construction to a module over an associative unital algebra we can construct a ternary algebra by means of ( [ dual_t ] ) or ( [ operator_way ] ) . indeed if @xmath1 is a left @xmath178-module , where @xmath178 is a binary unital associative complex algebra , @xmath172 is the dual module and @xmath179 is an @xmath178-multilinear mapping then @xmath33 is the ternary algebra with the ternary multiplication defined by ( [ dual_t ] ) . similarly given an @xmath178-module @xmath1 and a @xmath178-bilinear mapping @xmath127 , where @xmath116 is the algebra of @xmath178-linear operators of a module @xmath1 , then @xmath46 is the ternary algebra @xmath33 with the ternary multiplication @xmath22 defined by ( [ operator_way ] ) . we can use ( [ dual_t],[operator_way ] ) to construct the ternary algebras by means of well known structures of differential geometry on a smooth manifold . let @xmath180 be a vector bundle over a smooth finite dimensional manifold @xmath181 , @xmath182 be the algebra of smooth functions on a smooth manifold @xmath181 , and @xmath183 be the module of smooth sections of @xmath184 . given a @xmath182-multilinear mapping @xmath185 , where @xmath186 is the dual bundle , we obtain the ternary algebra @xmath187 of smooth sections of a vector bundle @xmath184 with the ternary multiplication @xmath188 , where @xmath189 are sections of @xmath184 , and @xmath190 . particularly if @xmath181 is a smooth manifold , @xmath191 is the tangent bundle , @xmath192 is the cotangent bundle , @xmath193 is the module of vector fields , @xmath194 is the module of 1-forms , @xmath195 is an affine connection on @xmath181 , and @xmath196 } , \quad x , y\in { \frak d}(m)\ ] ] is the curvature of @xmath195 then we have the @xmath182-multilinear mapping @xmath197 induced by the curvature @xmath198 and this mapping induces the structure of the ternary algebra @xmath199 on the module of vector fields with the ternary multiplication @xmath200 let @xmath201 be ( binary ) unital associative algebras over @xmath171 with involutions respectively @xmath202 and @xmath203 , where @xmath204 . let @xmath205 be a @xmath206-bimodule . we suppose that @xmath207 is an abelian group which is isomorphic to the abelian group @xmath205 , where @xmath208 is the corresponding isomorphism . then @xmath207 can be endowed with the structure of @xmath209-bimodule if we define the right and left multiplication by elements of algebras @xmath210 , @xmath211 as follows @xmath212 let @xmath213 , @xmath214 be the tensor products of modules where the first tensor product has the structure of @xmath178-bimodule , and the second has the structure of @xmath215-bimodule . it is clear that the algebras @xmath210 , @xmath215 can be viewed respectively as @xmath178-bimodule and @xmath215-bimodule . we also assume that there are two homomorphisms @xmath216 , @xmath217 respectively of @xmath178-bimodules and @xmath215-bimodules which satisfy @xmath218 where @xmath219 . evidently @xmath205 has the structure of vector space over @xmath171 . we define the ternary law of composition @xmath22 on @xmath220 by the formula @xmath221 @xmath222 is -partially associative ternary algebra . [ lr - associativity - of - m ] indeed for any quintuple of elements @xmath223 of @xmath205 we have @xmath224 on the other hand @xmath225 and this ends the proof . from this proposition and the proposition [ lr - associativity ] it follows that for any @xmath226 the binary algebra @xmath227 , where @xmath228 , is an associative algebra . let us define the ternary multiplication @xmath111 on @xmath220 by @xmath229 , where @xmath219 . the ternary algebra @xmath230 is the ternary algebra of jordan type , and the ternary multiplication @xmath111 of this algebra satisfies the identity @xmath231 where @xmath232 . we use the ternary algebra @xmath233 to construct a lie algebra which will be constructed by means of defining commutation relations . we denote the set of generators of this lie algebra by @xmath234 , where @xmath235 , i.e. we assign to each element @xmath236 the generator @xmath237 , to each element @xmath238 the generator @xmath239 , and to each pair @xmath240 the generator @xmath241 . let @xmath242 be the linear spans induced by the corresponding generators and @xmath243 . we define @xmath244 = s_{mn},\quad\;\;\;\;\;\;\;\ ; [ s_{pq},u_{m } ] = u_{\sigma(p , q , m ) } \label{brackets-1}\\ [ s_{pq},u_{\bar n } ] = -u_{\overline{\sigma(q , p , n)}},\;\ ; [ s_{mn},s_{pq } ] = s_{\sigma(m , n , p)\,q}-s_{p\,\sigma(n , m , q ) } \label{brackets-2}\end{gathered}\ ] ] the vector space @xmath245 endowed with the brackets defined by ( [ brackets-1 ] ) , ( [ brackets-2 ] ) is the lie algebra , and the identity ( [ identity ] ) leads to the jacoby identity for the brackets ( [ brackets-1 ] ) , ( [ brackets-2 ] ) . we can construct a matrix representation for the lie algebra @xmath245 if we consider the set @xmath246 of all @xmath247 square matrices of the type @xmath248 given two such matrices @xmath249 we define their product as follows @xmath250 the vector space of matrices @xmath246 endowed with the multiplication ( [ product of matrices ] ) is a unital associative ( binary ) algebra with the unity element @xmath251 where @xmath14 is the unity element of @xmath178 and @xmath252 is the unity element of @xmath215 . now we construct the matrix representation for the lie algebra @xmath245 as follows @xmath253 and the matrix representation of @xmath241 can be found by explicit calculation . having constructed the lie algebra @xmath245 and its matrix representation we can go further and construct a gauge field theory based on ternary algebra @xmath230 . for this purpose we take a vector bundle @xmath184 over a smooth finite - dimensional manifold @xmath181 with the fiber @xmath230 . a section of this bundle is a gauge field of our theory may be called ternon . now in each fiber we construct the lie algebra @xmath245 and this leads us to the vector bundle of lie algebras . next we construct the lie group corresponding to @xmath245 by means of exponential mapping and campbell - hausdorf series . we get the principal fiber bundle and the we proceed in constructing the gauge field theory as usual . in this section we consider a vector space of cubic matrices , where by cubic matrix we mean a quantity @xmath254 with three subscripts @xmath255 each running some set of integers . we use this vector space to construct a ternary algebra by means of triple product of cubic matrices . a triple product or ternary multiplication of cubic matrices is constructed in analogy with the classical product of two rectangular matrices by means of summation which is taken over certain system of subscripts of three cubic matrices . our aim in this section is to find all totally associative ternary multiplications of first or second kind , and we prove that there are four ternary multiplications of cubic matrices each yielding the associative ternary algebra of second kind . let @xmath256 , where @xmath257 and @xmath258 are integers satisfying @xmath259 , @xmath260 , @xmath261 . we will call @xmath159 a complex @xmath262space matrix provided that its entries @xmath263 are arranged in the vertices of a 3dimensional lattice and this structure is shown in particular case of a cubic matrix on the figure below . let us denote the set of all such matrices by @xmath264 , i.e. @xmath265 the set of @xmath262space matrices is the vector space if we define the addition of space matrices and multiplication by complex numbers as usual @xmath266 our main concern in this paper is a special case of space matrices when @xmath267 . in this case we will call @xmath254 , where @xmath268 , a complex @xmath269-cubic matrix and denote the vector space of such matrices by @xmath270 . particularly , if @xmath271 is a cubic matrix of third order then we will place its entries into the vertices of @xmath272dimensional lattice as follows @xmath273\ar@{.}'[d]'[dd][ddd ] & & & a_{123 } \ar@{.}[rrr]\ar@{.}'[d]'[dd][ddd ] & & & a_{133 } \ar@{.}[ddd ] \\ & a_{112 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}'[d][ddd ] & & & a_{122 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}'[d][ddd ] & & & a_{132 } \ar@{-}[ur ] \ar@{-}[ddd ] \\ a_{111 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}[ddd ] & & & a_{121 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}[ddd ] & & & a_{131 } \ar@{-}[ur ] \ar@{-}[ddd ] \ar@{-}[ddd ] \\ & & a_{213 } \ar@{.}'[r]'[rr][rrr ] \ar@{.}'[d]'[dd][ddd ] & & & a_{223 } \ar@{.}'[r]'[rr][rrr ] \ar@{.}'[d]'[dd][ddd ] & & & a_{233 } \ar@{.}[ddd ] \\ & a_{212 } \ar@{-}[ur ] \ar@{-}'[rr][rrr ] \ar@{-}'[d][ddd ] & & & a_{222 } \ar@{-}[ur ] \ar@{-}'[rr][rrr ] \ar@{-}'[d][ddd ] & & & a_{232 } \ar@{-}[ur ] \ar@{-}[ddd ] \\ a_{211 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}[ddd ] & & & a_{221 } \ar@{-}[ur ] \ar@{-}[rrr ] \ar@{-}[ddd ] & & & a_{231 } \ar@{-}[ur ] \ar@{-}[ddd ] \\ & & a_{313 } \ar@{.}'[r]'[rr][rrr ] & & & a_{323 } \ar@{.}'[r]'[rr][rrr ] & & & a_{333 } \\ & a_{312 } \ar@{-}[ur ] \ar@{-}'[rr][rrr ] & & & a_{322 } \ar@{-}[ur ] \ar@{-}'[rr][rrr ] & & & a_{332 } \ar@{-}[ur ] \\ a_{311 } \ar@{-}[rrr]\ar@{-}[ur ] & & & a_{321 } \ar@{-}[rrr]\ar@{-}[ur ] & & & a_{331 } \ar@{-}[ur ] } \nonumber\end{aligned}\ ] ] the above @xmath272dimensional lattice clearly shows that if one fixes a value of subscript @xmath274 in @xmath275 with @xmath276 ranging from 1 to 3 then the corresponding entries of cubic matrix @xmath159 form the square matrix of order @xmath272 . hence we get three square matrices of order @xmath272 which all together give us a cubic matrix of third order @xmath159 . therefore any cubic matrix @xmath159 of third order can be represented as the set of square matrices of order @xmath272 as follows @xmath277 \nonumber\end{aligned}\ ] ] now our aim is to construct a multiplication of space matrices . we will do this for cubic matrices of order @xmath269 because constructed multiplication can be extended to a vector space of @xmath262space matrices in an obvious way . if @xmath278 is a cubic matrix of order @xmath269 then it induces two mappings as follows : 1 . @xmath279 2 . @xmath280 where @xmath281 is the vector space of linear operators acting on @xmath282 . these mappings determined by a cubic matrix of order @xmath269 show that if we wish to construct a multiplication of cubic matrices of order @xmath269 which is based on composition of linear mappings as in the calculus of rectangular matrices then we should take three cubic matrices in order to close a corresponding multiplication in the sense that the product of three cubic matrices of order @xmath269 will be the cubic matrix of order @xmath269 . given two cubic matrices @xmath283 we can form two products of corresponding mappings @xmath284 and @xmath285 which are not close with respect to composition because the first product is the linear mapping @xmath286 , and the second is the linear mapping @xmath287 . it is obvious than we can close a procedure of taking compositions of this sort of mappings by adding one more cubic matrix @xmath288 . in this case the triple product @xmath289 closes the operation of taking compositions giving the mapping @xmath290 . this kind of reasoning suggests a possible way of constructing the cubic matrix of order @xmath269 from three given ones by means of summation which is taken over the certain pair of subscripts . given three cubic matrices of order @xmath269 we have nine subscripts , and because the number of possible combinations of subscripts is finite we can use the methods of computer algebra to find all associative ternary multiplications either of first or second kind . our analysis shows that there is no total associative ternary multiplication of first kind , and all total associative ternary multiplications of second kind are described by the following theorem . there are only four different triple products of complex cubic matrices of order @xmath269 which obey the total ternary associativity of second kind . these are 1 . @xmath291 2 . @xmath292 3 . @xmath293 4 . @xmath294 [ theorem ] now our aim is to construct a ternary analog of lie algebra with the help of ternary multiplication of cubic matrices and ternary analog of lie bracket . a notion of ternary analog of lie bracket is based on a faithful representation of cyclic group @xmath295 by cubic roots of unity . we construct a ternary analog of lie algebra which may be considered as an analog of the lie algebra generated by pauli matrices . in our construction we use the cubic matrices of second order with certain symmetries with respect to the subscripts as generators of our algebra and find all commutation relations . in this section we use the ternary multiplication of cubic matrices which has the property that any cyclic permutation of the matrices in the product is equivalent to the same permutation on the subscripts and this multiplication is studied in @xcite . let @xmath296 @xmath297 @xmath298 be cubic matrices of order @xmath269 . we define the triple product @xmath299 by the following formula @xmath300 it is easy to see that any cyclic permutation of the matrices in the product is equivalent to the same permutation on the subscripts , i.e. @xmath301 it should be mentioned that the ternary multiplication ( [ multiplication ] ) is neither partially nor totally associative . the ternary algebra of cubic matrices of order @xmath269 with respect to the multiplication ( [ multiplication ] ) can be decomposed into direct sum of subspaces of cubic matrices with certain symmetries according to the irreducible representation of the symmetry group @xmath302 it should be mentioned that we have the similar decomposition in the case of square matrices of order @xmath303 where the algebra of square matrices can be decomposed into the direct sum of subspaces of symmetric and skew - symmetric matrices according to representation of the permutation group @xmath304 the symmetry group @xmath98 possesses a full and faithful representation on the complex plane , which can be generated by two elements representing a cyclic and an odd permutation . this representation can be constructed by assigning the operator of multiplication by the cubic root of unity @xmath305 to the cyclic permutation @xmath306 and assigning the operator of complex conjugation to the odd permutation @xmath307 . in order to construct the subspace of cubic matrices with certain symmetries we use the cyclic part of this representation where the cyclic group @xmath295 is represented by cubic roots of unity . a cubic matrix @xmath308 of order @xmath269 is said to be a @xmath0-skew - symmetric if it satisfies @xmath309 similarly a cubic matrix @xmath310 of order @xmath269 is said to be a @xmath311-skew - symmetric if it satisfies @xmath312 it can be shown that the space of cubic matrices of order @xmath269 can be decomposed into the direct sum of the subspace of @xmath0-skew - symmetric matrices , the subspace of @xmath311-skew - symmetric matrices and the subspace of symmetric matrices , where under symmetric matrix we mean a cubic matrix @xmath313 of order @xmath269 which satisfies @xmath314 . each of subspaces of @xmath0-skew - symmetric matrices and of @xmath311-skew - symmetric matrices has the dimension @xmath315 the subspace of symmetric matrices can be decomposed into the direct sum of the subspace of diagonal cubic matrices ( all entries excepting diagonal ones are equal to zero ) and the subspace of symmetric matrices with zeros on diagonal . obviously the dimension of the subspace of diagonal cubic matrices is equal to @xmath269 and the subspace of symmetric matrices with zeros on diagonal is equal to @xmath315 let @xmath316 be a basis for the subspace of @xmath0-skew - symmetric matrices , @xmath317 be a basis for the subspace of @xmath311-skew - symmetric matrices , and @xmath318 be a basis for the subspace of symmetric matrices with zeros on diagonal , where @xmath319 a basis for the subspace of diagonal cubic matrices will be denoted by @xmath320 where @xmath321 thus we have @xmath322 now our aim is to construct a ternary analog of lie algebra by means of the ternary @xmath0-commutator which we define as follows @xmath323=a\circledcirc b\circledcirc c+jb\circledcirc c\circledcirc a+j^2c\circledcirc a\circledcirc b,\quad a , b , c\in \mbox{cmat}_{n}({\mathbb c } ) \label{commutator}\ ] ] where the ternary multiplication @xmath324 is defined by ( [ multiplication ] ) . it is easy to see that the ternary @xmath0-commutator ( [ commutator ] ) may be viewed as an analog of the binary lie commutator @xmath325=xy - yx,$ ] where @xmath326 @xmath327 are the elements of an associative binary algebra , because we replace the symmetric group @xmath328 and its representation @xmath329 respectively by the cyclic group @xmath295 and its representation @xmath330 indeed in the case of the binary lie commutator we have the property @xmath325=-[y , x].$ ] for the ternary @xmath0-commutator we have the similar property @xmath331=j[b , c , a]=j^2[c , a , b].$ ] in analogy with the binary lie commutator for any @xmath332 we have @xmath333=0.$ ] let us consider the space of cubic matrices of order @xmath334 i.e. each subscript runs from 1 to 2 . in this case all subspaces mentioned above are 2-dimensional , which means that basis for each subspace consists of two matrices , i.e. @xmath335 we construct the basis for the subspace of diagonal cubic matrices @xmath336 by choosing @xmath337 we choose the basis for the subspace of symmetric cubic matrices with zeros on diagonal by fixing @xmath338 @xmath339 with other entries are equal to zero . we get the basis for the subspace of @xmath0-skew - symmetric cubic matrices by taking @xmath340 @xmath341 with other entries equal to zero and the basis for the subspace of @xmath311-skew - symmetric cubic matrices if we put @xmath342 @xmath343 with other entries equal to zero . the space of cubic matrices of order 2 equipped with the ternary @xmath0-commutator can be considered as a ternary analog of matrix lie algebra and all commutation relations of this algebra in the basis @xmath344 are given in the following table : @xmath345 & = -\orhob , & [ \rhob , \rhoa , \rhob ] & = -\orhoa & [ \orhoa , \orhob , \orhoa ] & = 2\orhob , & [ \orhob , \orhoa , \orhob ] & = 2\orhoa \\ [ \omegaa , \omegab , \omegaa ] & = -\orhob , & [ \omegab , \omegaa , \omegab ] & = -\orhoa & [ \etaa , \etab , \etaa ] & = 0 , & [ \etab , \etaa , \etab ] & = 0 \\ \\ [ \rhoa , \orhoa , \rhoa ] & = 3\orhoa , & [ \rhob , \orhob , \rhob ] & = 3\orhob , & [ \orhoa , \rhoa , \orhoa ] & = 0 , & [ \orhob , \rhob , \orhob ] & = 0 \\ [ \rhoa , \orhob , \rhoa ] & = -\orhob , & [ \rhob , \orhoa , \rhob ] & = -\orhoa , & [ \orhoa , \rhob , \orhoa ] & = -\orhob , & [ \orhob , \rhoa , \orhob ] & = -\orhoa \\ [ \rhoa , \orhoa , \rhob ] & = 2\orhob , & [ \rhoa , \orhob , \rhob ] & = 2\orhoa , & [ \rhoa , \orhoa , \orhob ] & = -\orhob , & [ \rhob , \orhob , \orhoa ] & = -\orhoa \\ \\ [ \rhoa , \omegaa , \rhoa ] & = 0 , & [ \omegaa , \rhoa , \omegaa ] & = 3\orhoa , & [ \rhob , \omegab , \rhob ] & = 0 , & [ \omegab , \rhob , \omegab ] & = 3\orhob \\ [ \rhoa , \omegab , \rhoa ] & = 2\orhob , & [ \omegab , \rhoa , \omegab ] & = 2\orhoa , & [ \rhob , \omegaa , \rhob ] & = 2\orhoa , & [ \omegaa , \rhob , \omegaa ] & = 2\orhob \\ [ \rhob , \rhoa , \omegaa ] & = -\orhob , & [ \omegab , \rhob , \rhoa ] & = -\orhoa , & [ \omegaa , \rhoa , \omegab ] & = -\orhob , & [ \omegaa , \rhob , \omegab ] & = -\orhoa \\ \\ [ \orhoa , \omegaa , \orhoa ] & = 3\orhoa , & [ \omegaa , \orhoa , \omegaa ] & = 0 , & [ \orhob , \omegab , \orhob ] & = 3\orhob , & [ \omegab , \orhob , \omegab ] & = 0 \\ [ \orhoa , \omegab , \orhoa ] & = -\orhob , & [ \omegab , \orhoa , \omegab ] & = -\orhoa , & [ \orhob , \omegaa , \orhob ] & = -\orhoa , & [ \omegaa , \orhob , \omegaa ] & = -\orhob \\ [ \orhoa , \omegaa , \orhob ] & = -\orhob , & [ \orhoa , \omegab , \orhob ] & = -\orhoa , & [ \orhoa , \omegaa , \omegab ] & = 2\orhob , & [ \omegaa , \omegab , \orhob ] & = 2\orhoa \\ \\ [ \rhoa , \etaa , \rhoa ] & = \orhoa , & [ \etaa , \rhoa , \etaa ] & = 0 , & [ \rhob , \etab , \rhob ] & = \orhob , & [ \etab , \rhob , \etab ] & = 0 \\ [ \rhoa , \etab , \rhoa ] & = \orhob , & [ \etaa , \rhob , \etaa ] & = \orhob , & [ \rhob , \etaa , \rhob ] & = \orhoa , & [ \etab , \rhoa , \etab ] & = \orhoa \\ [ \rhoa , \rhob , \etaa ] & = \orhob , & [ \etab , \rhoa , \etab ] & = \orhoa , & [ \etaa , \rhoa , \etab ] & = 0 , & [ \etaa , \rhob , \etab ] & = 0 \\ \\ [ \orhoa , \etaa , \orhoa ] & = \orhoa , & [ \etaa , \orhoa , \etaa ] & = 0 , & [ \orhob , \etab , \orhob ] & = \orhob , & [ \etab , \orhob , \etab ] & = 0 \\ [ \orhoa , \etab , \orhoa ] & = \orhob , & [ \etaa , \orhob , \etaa ] & = \orhob , & [ \orhob , \etaa , \orhob ] & = \orhoa , & [ \etab , \orhoa , \etab ] & = \orhoa \\ [ \orhoa , \etaa , \orhob ] & = \orhob , & [ \orhoa , \etab , \orhob ] & = \orhoa , & [ \etaa , \orhoa , \etab ] & = 0 , & [ \etaa , \orhob , \etab ] & = 0 \\ \\ [ \etaa , \omegaa , \etaa ] & = 0 , & [ \etaa , \omegab , \etaa ] & = \orhob , & [ \etab , \omegaa , \etab ] & = \orhoa , & [ \etab , \omegab , \etab ] & = \orhob \\ [ \omegaa , \etaa , \omegaa ] & = \orhoa , & [ \omegaa , \etab , \omegaa ] & = \orhob , & [ \omegab , \etaa , \omegab ] & = \orhoa , & [ \omegab , \etab , \omegab ] & = \orhob \\ [ \etaa , \omegaa , \etab ] & = 0 , & [ \etaa , \omegab , \etab ] & = 0 , & [ \etaa , \omegaa , \omegab ] & = \orhob , & [ \etab , \omegab , \omegaa ] & = \orhoa \\ \\ [ \rhoa , \orhoa , \omegaa ] & = 0 , & [ \omegaa , \rhoa , \orhob ] & = 2\orhob , & [ \rhoa , \orhoa , \omegab ] & = -\orhob , & [ \rhoa , \rhob , \omegab ] & = -\orhoa \\ [ \etaa , \rhoa , \orhoa ] & = \orhoa , & [ \etaa , \rhoa , \orhob ] & = \orhob , & [ \orhoa , \etab , \rhoa ] & = \orhob , & [ \rhoa , \orhob , \etab ] & = \orhoa \\ [ \orhoa , \omegaa , \rhob ] & = -\orhob , & [ \rhob , \orhob , \omegaa ] & = -\orhoa , & [ \omegab , \rhob , \orhoa ] & = 2\orhoa , & [ \rhob , \orhob , \omegab ] & = 0 \\ [ \etaa , \orhoa , \rhob ] & = \orhob , & [ \rhob , \etaa , \orhob ] & = \orhoa , & [ \etab , \rhob , \orhoa ] & = \orhoa , & [ \etab , \rhob , \orhob ] & = \orhob \\ [ \etaa , \orhoa , \omegaa ] & = \orhoa , & [ \etaa , \omegab , \orhoa ] & = \orhob , & [ \omegaa , \etab , \orhoa ] & = \orhob , & [ \etab , \orhoa , \omegab ] & = \orhoa \\ [ \etaa , \orhob , \omegaa ] & = \orhob , & [ \orhob , \etaa , \omegab ] & = \orhoa , & [ \orhob , \omegaa , \etab ] & = \orhoa , & [ \etab , \orhob , \omegab ] & = \orhob \\ [ \etaa , \rhoa , \omegaa ] & = \orhoa , & [ \rhoa , \etaa , \omegab ] & = \orhob , & [ \omegaa , \etab , \rhoa ] & = \orhob , & [ \rhoa , \etab , \omegab ] & = \orhoa \\ [ \omegaa , \etaa , \rhob ] & = \orhob , & [ \omegab , \etaa , \rhob ] & = \orhoa , & [ \omegaa , \etab , \rhob ] & = \orhoa , & [ \etab , \rhob , \omegab ] & = \orhob\end{aligned}\ ] ] .1 cm we would like to point out that the above table of commutation relations demonstrates a peculiar property of the algebra of cubic matrices of second order in the basis @xmath344 : the @xmath0-commutator of any three generators is proportional to one of the generators @xmath346 . moreover taking @xmath311-commutator @xmath323^{*}=a\circledcirc b\circledcirc c+j^2\,b\circledcirc c\circledcirc a+j\,c\circledcirc a\circledcirc b,\quad a , b , c\in \mbox{cmat}_{n}({\mathbb c } ) , \label{con - commutator}\ ] ] which can be considered as a conjugate commutator to @xmath0-commutator ( [ commutator ] ) , we get the similar table of commutation relations ( we do not demonstrate it here in order not to overburden the paper ) which clearly shows that the @xmath311-commutator of any three generators is proportional to one of the generators @xmath347 . finally considering a ternary analog of anti - commutator defined by @xmath348 and applying it to the cubic matrices of second order we obtain the table of commutation relations for the generators @xmath349 @xmath350 @xmath351 @xmath352 @xmath353 @xmath354 @xmath355 , @xmath356 which can be shortly described as follows : the ternary anti - commutator of any three generators is a linear combination of the generators @xmath357 . we conclude this section by pointing out an analogy between the generators of our ternary algebra of cubic matrices of second order and the pauli matrices @xmath358 indeed applying the ternary @xmath0-commutator to pauli matrices we get @xmath359=2i\sigma_2 , \quad[i\sigma_2,i\sigma_1,i\sigma_2]=2i\sigma_1 \nonumber\end{aligned}\ ] ] in the table of commutation relations of cubic matrices of order 2 we have @xmath360 = 2 \orhob,\quad\;\;\;\ ; [ \orhob , \orhoa , \orhob]=2\orhoa \nonumber\end{aligned}\ ] ] which clearly demonstrates a striking analogy between the @xmath311-skew - symmetric generators of ternary algebra of cubic matrices of order 2 and the skew - hermitian matrices @xmath361 @xmath362 @xmath363 where @xmath364 @xmath365 @xmath366 are pauli matrices . v. abramov , o. liivapuu and s. shitov gratefully acknowledge the financial support of their research by the estonian science foundation under the research grant etf 7427 . v. abramov , r. kerner and o. liivapuu would like to thank the french - estonian scientific program `` g. f. parrot '' for financial support under the travel grant p-2/2007 .

the narrow emission line spectra of active galactic nuclei are not accurately described by simple photoionization models of single clouds . recent hubble space telescope images of seyfert 2 galaxies show that these objects are rich with ionization cones , knots , filaments , and strands of ionized gas . here we extend to the narrow line region the `` locally optimally emitting cloud '' ( loc ) model , in which the observed spectra are predominantly determined by powerful selection effects . we present a large grid of photoionization models covering a wide range of physical conditions and show the optimal conditions for producing many of the strongest emission lines . we show that the integrated narrow line spectrum can be predicted by an integration of an ensemble of clouds , and we present these results in the form of diagnostic line ratio diagrams making comparisons with observations . we also predict key diagnostic line ratios as a function of distance from the ionizing source , and compare these to observations . the predicted radial dependence of the [ ] / [ ] ratio may be matched to the observed one in ngc 4151 , if the narrow line clouds see a more intense continuum than we see . the loc scenario when coupled with a simple keplerian gravitational velocity field will quite naturally predict the observed line width versus critical density relationship . the influence of dust within the ionized portion of the clouds is discussed and we show that the more neutral gas is likely to be dusty , although a high ionization dust - free region is most likely present too . this argues for a variety of nlr cloud origins .

astro-ph9705083

the failure of `` simple '' photoionization models to describe the narrow emission line spectrum of active galactic nuclei ( agn ) was shown dramatically by filippenko & halpern ( 1984 ) , who found that the line widths of the optical forbidden lines were correlated with the critical density and ionization potential of those lines . this correlation indicates stratification of the photoionized region , with the lines emitting near their critical densities . recent hubble space telescope images of seyfert 2 galaxies show that these objects are rich with ionization cones ( evans et al . 1991 ; tsvetanov & walsh 1992 ; wilson et al . 1993 ; macchetto et al . 1994 ; and arribas et al . 1996 ) , knots ( acosta - pulid et al . 1996 ) , and filaments , and strands ( falcke et al . 1996 ) of ionized gas . we should not expect that these complex environments can be described with simple , single - component photoionization models . similar images of seyfert 1 type objects show a less complex , symmetric structure that may indicate a viewing angle preference consistent with the unification model of seyfert galaxies ( schmitt & kinney 1996 ) . two - component photoionization models have been proposed by binette , wilson , & storchi - bergmann ( 1996 ) and have met with some success in describing the emission line spectrum of seyfert galaxies . this model assumes that the narrow line region ( nlr ) is composed of both matter - bounded and ionization - bounded clouds , and that the covering fraction of the two populations can account for the range of observed line emission . a multi - component photoionization model has been proposed by komossa & schulz ( 1997 ) , who assume that the nlr is effectively composed of a few clouds with a range of gas densities and distances from the central agn . both of these types of models do much better at reproducing the scatter in the line ratio observations than single component photoionization models . an alternative to photoionization by the central source is photoionization by fast moving shocks . dopita & sutherland ( 1995 ) describe in detail these `` photoionizing '' shocks and argue that all narrow emission lines can be excited by these processes . however , morse , raymond , & wilson ( 1996 ) and wilson ( 1997 ) discuss the viability of such shocks in describing the emission spectrum of active galaxies . in particular , wilson ( 1997 ) argues that the energetics of radiative shocks may not account for the observed line luminosities . it is still unclear exactly what role shocks play in the nlr environment ; that is , whether they provide the bulk of the energy input for the observed emission or merely compress gas so that it emits more strongly when photoionized by a central source ( pogge 1997 ) . we will assume that photoionization by the central source is the dominating mechanism responsible for nlr emission . we extend the locally optimally - emitting clouds ( loc ) hypothesis of baldwin et al . ( 1995 ) to nlr gas . this assumes that in nature , gas with a wide range of physical conditions is present and the emission line spectrum we observe is predominately a result of a simple , yet powerful , selection effect : we observe lines from clouds best able to emit them . to this end we present the results of a large grid of photoionization models and make predictions of the nature of the integrated spectrum . for simplicity we will assume that the source of narrow line emission observed in active galaxies is from photoionized clouds that are similar to illuminated molecular clouds in our own galaxy . the source of these clouds may be four - fold : ( 1 ) molecular cloud complexes in the disk of the galaxy ; ( 2 ) the dusty molecular torus that is believed to surround the central source and which will certainly be photoionized ; ( 3 ) ism entrained behind a radio jet so that photoionization by the central source becomes important ; and ( 4 ) material `` snow plowed '' in the form of bow shocks in front of a radio jet such that it too may be photoionized by the central source . we will ignore emission that may be due to shock excitation coming from photo - ionizing shocks , although how important this emission is to the overall spectrum is unclear ( wilson 1997 ) . we have used the spectral synthesis code cloudy ( version 90.03 ; ferland 1996 ) to calculate the emission from plane parallel , constant hydrogen density clouds ionized by a continuum similar to that expected from a typical seyfert galaxy with @xmath0 ergs s@xmath1 . the shape of the ionizing continuum was chosen to be a combination of a uv - bump of the form @xmath2 and an x - ray power law of the form @xmath3 spanning 13.6 ev to 100 kev . the uv - bump cutoff temperature , @xmath4 , was chosen such that the uv - bump peaked ( in @xmath5 ) at 48 ev . the uv and x - ray components were combined with a typical seyfert uv to x - ray spectral slope , @xmath6 . the ionization / thermal equilibrium and radiative transfer calculations for a single cloud proceeded until one of the following three conditions were met . first , the electron temperature dropped below 3000 k. gas with lower temperatures does not contribute significantly to the emission lines presented in this paper . second , we do not allow clouds to have thicknesses in excess of 10% of its distance from the central continuum source . third , we do not allow the total hydrogen column density to exceed 10@xmath7 @xmath8 ; this mainly affects the very high ionization clouds and prevents them from becoming thomson thick . in practice the third condition had little impact on the narrow emission lines . the largest cloud of the grid presented below is one with a thickness of @xmath916 pc , consistent with the size of large molecular clouds . we will comment further on the effects of these stopping criteria in a later section . finally , we assume that each cloud sees the full continuum with no obscuration . at large enough distances , intervening clouds or diffuse ism may attenuate the ionizing spectrum significantly . we have looked at the effects of such ism attenuation on the incident continuum : an ism with gas densities as low as 0.1 @xmath10 such as found in the local ism of the milky way ( wood & linsky 1997 ) would have no effect on the results presented below . a higher density ism ( n(h ) = 0.5 @xmath10 ) however , would have an effect at large distances from the ionizing source , due to the larger column density and lower ionization resulting in a higher gas opacity . for the higher density gas and the shape and luminosity of the incident continuum given above , the attenuated continuum has an optical depth of unity at 4 ryd at a distance of @xmath11 cm from the ionizing source , resulting in 5 20% changes in the line luminosities . at larger distances the attenuation by a dense ism would result in a decrease in the luminosity of the emitted spectrum of up to a factor of 10 . we neglect this complication for simplicity . we have computed , on the university of kentucky convex exemplar parallel supercomputer , a grid of 1,881 of the single cloud models described above in a plane of hydrogen gas density and distance from the central ionizing source . we chose a range of parameters in order to cover all the possible physical conditions of narrow line region gas . the hydrogen gas density , @xmath12 , ranged from 10@xmath13 10@xmath14 @xmath10 . the incident continuum flux was varied so that for an assumed ionizing continuum luminosity @xmath15 ergs s@xmath1 ( monochromatic luminosity @xmath161450 ) = 39.558 ergs s@xmath1 @xmath1 for the continuum shape used here ) the distance from the ionizing source of radiation , @xmath17 , varied from 10@xmath18 10@xmath19 cm . this distance range scales with the luminosity as @xmath20 . the conversion from incident flux to luminosity and radius assumes that the ionizing radiation is emitted isotropically . however , observations of ionizing `` cones '' ( evans et al . 1991 ; tsvetanov & walsh 1992 ; wilson et al . 1993 ; macchetto et al . 1994 ; and arribas et al . 1996 ) and uv - photon deficits ( binette , fosbury , & parker 1993 ; and morse , raymond & wilson 1996 ) imply that the gas may see a different ionizing continuum than we do . calculations were performed assuming two abundance sets : solar and a set including dust that varied with distance from the ionizing source . we will describe each set of abundances in turn and compare the two in terms of diagnostic line ratio diagrams in a later section . figure 1 shows the results of grid calculations , for solar abundance clouds , in the form of contour plots of logarithm equivalent width referred to the incident continuum at 4860 . the solar abundances are from grevesse & anders ( 1989 ) and grevesse & noels ( 1993 ) and have the following values : h : 1.00e+00 he:1.00e-01 li:2.04e-09 be:2.63e-11 b : 7.59e-10 c : 3.55e-04 n : 9.33e-05 o : 7.41e-04 f : 3.02e-08 ne:1.17e-04 na:2.06e-06 mg:3.80e-05 al:2.95e-06 si:3.55e-05 p : 3.73e-07 s : 1.62e-05 cl:1.88e-07 ar:3.98e-06 k : 1.35e-07 ca:2.29e-06 sc:1.58e-09 ti:1.10e-07 v : 1.05e-08 cr:4.84e-07 mn:3.42e-07 fe:3.24e-05 co:8.32e-08 ni:1.76e-06 cu:1.87e-08 zn:4.52e-08 furthermore , we assume that there are no dust present in the solar abundance computations . the ionization parameter provides a useful homology relationship with the emission lines ( davidson 1977 ) . @xmath21 is defined as the ratio of hydrogen ionizing photon density to hydrogen density , @xmath22 , where @xmath23 is the flux of hydrogen ionizing photons and @xmath24 is the speed of light . for reference a dashed line is placed on the upper left hand plot of the first panel representing @xmath25 . the ionization parameter increases from top right to bottom left in these diagrams ; lines of constant @xmath21 have a slope of @xmath26 . shown in figure 1 are 23 strong optical , uv and infrared recombination , resonance , and forbidden lines . the temperature of the illuminated face of the cloud is also shown in the upper right hand plot of the first panel . for many emission lines a ridge of near maximum equivalent width runs diagonally across the distance gas density plane , roughly parallel to lines of constant ionization parameter . most lines are emitted optimally for a narrow range of ionization parameter spanning this ridge . for larger ( smaller ) ionization parameters the gas is over ( under ) ionized , and the line is not efficiently emitted . this explains the sudden drops in the emission line equivalent widths on either side of their ridges . moving _ along _ the ridge to increasing gas densities , at near constant ionization parameter , the forbidden lines become collisionally deexcited and the line equivalent width falls off . moving along the ridge at constant ionization parameter to smaller gas densities , the equivalent widths of some lines also diminish as other lines of similar ionization , but lower critical density , become important coolants . this effect is especially dramatic for @xmath271549 and @xmath271035 . the stopping criteria assumed in @xmath282.1 do affect the resulting spectrum . the grid of models shown in figure 1 includes clouds that are both optically thin and optically thick to hydrogen ionizing radiation . optically thin clouds generally occupy the lower left portion of the distance density plane below a line with slope @xmath29 @xmath302/3 and intersecting the lower right hand corner . the temperature of this gas ( see fig . 1 ) ranges from 20,000 k to several million k. emission lines presented in this paper are not emitted by gas with t@xmath31 10@xmath32 k ( x - ray lines _ are _ ) . these hot clouds are generally truncated by the plane parallel condition discussed earlier . optically thick clouds are in the upper right portion of the figures , and the gas generally has nebular temperatures , except the extreme upper right corner where the electron temperatures can be as low as a few hundred kelvin and the gas is mostly molecular . generally the optically thick clouds were stopped because the electron temperature fell below 3000 k ( @xmath282.1 ) . the total hydrogen column density restriction mentioned in @xmath282.1 applies to a small number of high density optically thick clouds in a triangular region bounded by a distance of 1 pc on the the topside and a line roughly parallel to @xmath33 on the leftside . the clouds in this region do not significantly contribute to the narrow forbidden line spectrum . different types of lines can be seen in figure 1 to be emitted from clouds with a range in gas densities and distances . recombination lines , such as @xmath34 and @xmath35 , generally form broad planes of maximum equivalent width whereas the forbidden lines , such as [ ] , are either narrow ridges or peaked islands ( [ ] @xmath277325 ) . figure 1b shows the differences in the nebular and auroral lines of [ ] and [ ] . notice that the auroral features are formed optimally at higher densities , because of their higher critical density . figure 1c shows the contour plots of a few of the stronger uv - lines . figure 1d shows a few strong [ ] infrared fine - structure lines from a 16-level model atom . the model atom has the same configuration as that shown in thompson ( 1995 ) and uses the collision data of pradhan & zhang ( 1993 ) and einstein as from quinet , le dourneuf & zeippen ( 1996 ) . verner et al . ( 1997 ) compare the differences between the 16-level model atom and a much larger definitive 371-level calculation . preliminary comparisons show that predicted line intensities of the 16-level atom differ typically by a few percent and at maximum by 10% from the larger 371-level calculation . to illustrate the problems encountered when trying to use conventional diagnostic line intensity ratios for an ensemble of clouds with a wide range of physical conditions , we momentarily assume that each emission line in the total spectrum is produced with the maximum possible equivalent width found in the figures . we then use this spectrum to reproduce the standard density and temperature diagnostic line ratios . this is the simplest way to document the effects of a distribution of clouds . three density indicators are given by osterbrock ( 1989 ) . these include the [ ] @xmath276716/@xmath276731 and [ ] @xmath273729/@xmath273726 doublets and the [ ] @xmath271907/ ] @xmath271909 ratio . for [ ] , pradhan & zhang ( 1993 ) use the 1.534 @xmath36m/1.644 @xmath36 m ratio as a density indicator . assuming a temperature of 10,000 k , the [ ] ratio we predict gives an n@xmath31 of 300 @xmath10 , for the predicted [ ] doublet the density is 700 @xmath10 , for the [ ] ratio n@xmath31 is 20,000 @xmath10 , and the [ ] / ] ratio gives a density of nearly 300,000 @xmath10 . the [ ] ( @xmath274959 + @xmath275007)/@xmath274363 temperature density diagnostic using the peak emission of figure 1 and an electron temperature of 10,000 k gives an electron density on the order of a few million @xmath10 . the point of this exercise is clear : _ line diagnostics fail when the gas has a wide range of physical conditions_. as figure 1 shows , many of these lines form at different densities and temperatures . however , by using the diagnostic line ratios , one assumes that the lines are formed in gas of similar physical conditions , which is obviously not the case in figure 1 , and is likely not the case in nature . we will in a later section integrate the `` visibility functions '' of figure 1 to calculate an emission spectrum by assuming that the observed spectrum is the result of an ensemble of clouds with differing properties . in the previous section we have assumed that the narrow line region is a dust free environment . we now recompute the grid assuming that the clouds are dusty . grains will not survive in an environment with high photoionizing flux due to the various sublimation processes discussed by laor & draine ( 1993 ) . therefore , after the manner of netzer & laor ( 1993 ) we vary the abundances as a function of distance from the ionizing source , in order to mimic these destruction processes . at small distances from the central source we assume the abundances are solar , as given in the previous section . at a distance of approximately @xmath37 cm from the central source ( for the ionizing luminosity given above ) , orion type graphite grains ( baldwin et al . 1991 ) are just at their sublimation temperature ( @xmath91750 k ) at the cloud face . at distances larger than this we include graphite grains within the calculation of the single cloud and deplete carbon such that its abundance is typical of an region ( see below and baldwin et al . orion type silicate grains sublimate at @xmath91400 k , corresponding to a distance of @xmath38 cm . beyond this radius the abundances are set to values which approximate those found in the orion nebula : h : 1.00e+00 he:9.50e-02 c : 3.00e-04 n : 7.00e-05 o : 4.00e-04 ne:6.00e-05 na:3.00e-07 mg:3.00e-06 al:2.00e-07 si:4.00e-06 s : 1.00e-05 cl:1.00e-07 ar:3.00e-06 ca:2.00e-08 fe:3.00e-06 ni:1.00e-07 these abundances are based upon the results of several recent studies of the orion nebula ( baldwin et al . 1991 , rubin et al . 1991 , 1992 , and osterbrock , tran & veilleux 1992 ) . although this discontinuous `` turning on '' of grains is unphysical and our choices of grain composition and gas depletions are specific , it serves as an example to simulate the presence of dust in the narrow line region . the grain physics used here is described by baldwin et al . ( 1991 ) . figure 2 shows the same contour plots as figure 1 except that the individual models include the grains as specified above . the discontinuous grain sublimation distances are easily seen in the figure as sharp discontinuities in the contour plots at @xmath39 and @xmath40 cm . comparisons of figures 1 and 2 show four major effects of grains : ( 1 ) the depletion of refractory elements ; ( 2 ) the weakening of emission lines due to absorption of the incident continuum by dust at large @xmath21 ; ( 3 ) the photoelectric heating of the gas by grains ; and ( 4 ) in the case of resonance lines such as @xmath35 @xmath271216 and @xmath271549 , line destruction by grains . higher ionization lines , such as [ ] , [ ] , and [ ] , tend to be emitted less efficiently than in the dust - free models , with the equivalent width falling by as much as a factor of 2 . the weakening of these lines is caused by the absorption of the incident continuum by the dust . the lower ionization lines ( [ ] , [ ] , [ ] , [ ] etc . ) actually have their peak emission increase , though at lower ionization parameters , due to the photoelectric heating of the gas by the grains . resonance line destruction is readily apparent in figure 2c . the heavy depletion of iron is also apparent in figure 2d , wherein the peak equivalent width drops by nearly a factor of ten . this last point will be discussed more fully in a later section . the equivalent widths shown in figures 1 and 2 reflect the efficiency with which the clouds reprocess the incident continuum . it is clear that different lines are formed optimally in different places in the gas density distance plane . if the distribution function of clouds in this plane is known , then the total line luminosity emitted by a set of clouds will be given by @xmath41 where @xmath42 is the emission line flux of a single cloud at radius @xmath43 , gas density @xmath44 , and @xmath45 is the cloud distribution function , which is not necessarily analytical . the remaining question is the spatial distribution of the nlr clouds . many type 2 seyferts have resolved nlr s , often with complex structures that are associated with linear or double - lobed radio sources ( bower et al . 1995 ; pogge & de robertis 1995 ; capetti et al . 1996 ; cooke et al . 1997 ) , whereas the structure of the nlr in type 1 seyferts is generally more compact and axisymmetric , at least in projection ( schmitt & kinney 1996 ; nelson et al . since it is not clear how the emitting gas is distributed , in _ all _ its dimensions , we will assume that the distribution function @xmath45 can be approximated by @xmath46 , where @xmath47 and @xmath48 are the cloud covering fractions with distance and gas density , respectively ( see baldwin et al . 1995 ) . for simplicity we assume simple power laws : @xmath49 the power law indices , @xmath50 and @xmath51 , will be the only free parameters of our predicted integrated spectra . in the case where both weighting functions are proportional to the inverse of @xmath44 or @xmath43 , then the line equivalent widths of the clouds are equally weighted ( in log space ) across the gas density distance plane . in the integrations computed below we assumed certain integration limits in the gas density distance plane in order to exclude gas that may not be present in the nlr . we exclude gas with densities @xmath52 , the rough maximum critical density of the optical forbidden lines observed . by including gas with @xmath53 , the line ratios do not change significantly except for those integrations that are weighted to higher densities ( flatter @xmath48 ) . if present , the contribution of high density gas in the inner regions of the nlr can not be significant or else the narrow line spectrum would appear vastly different ( forbidden lines would be too weak relative to the hydrogen lines ) . we exclude gas that is closer than 0.1 pc to the ionizing source . this distance corresponds to the grain sublimation point described in @xmath282.3 . integration limits were also included in ionization parameter space in order to exclude gas that is far over ( under ) ionized and does not contribute to the observed emission . these limits effectively exclude gas from the broad emission line region and very high density gas hundreds of parsecs away from the central engine . integrations described above have been computed and plotted after the manner of baldwin , phillips & terlevich ( 1981 ; hereafter bpt ) . intensity ratios for the integrated model spectra are plotted in this manner in figure 3 . two types of lines are shown in the figure . solid lines represent dusty models with depleted refractory elements typical of region type abundances ( hereafter `` dusty '' ) . the dashed lines represent dust - free clouds with solar metallicities ( hereafter `` dust - free '' ) . different values of @xmath48 ( cf . [ 2 ] ) are represented by different lines , with the power of @xmath47 ranging from @xmath54 in 0.25 increments along each line with the direction of more negative @xmath47 shown by the arrows . for the dusty simulations ( solid lines ) three values of the @xmath48 index are shown ( @xmath51 = @xmath551.0 , @xmath551.6 , @xmath551.8 ) , and the trend of more negative @xmath51 are indicated by the appropriate arrows in the figures . for the dust - free ( dashed lines ) simulations the @xmath48 index are @xmath51 = @xmath551.0 , @xmath551.4 , @xmath551.8 , similarly the trends for the more negative index are indicated . the `` s '' and `` l '' indicate in the figures the average values of the line ratios for seyfert , taken from ferland & osterbrock ( 1986 ) , and liner type spectra taken from netzer ( 1990 ) , with the exceptions of @xmath274686 , [ ] @xmath274363 , from ho , filippenko & sargent ( 1995 ) for the liner galaxy m81 . the observed range in [ ] @xmath277325 and [ ] @xmath569069,9532 strengths were taken from dopita & sutherland ( 1995 ) . to make it possible to follow particular models from panel to panel , we have marked two fiducial models in figure 3 . the stars represent a dusty model ( solid lines ) with @xmath57 and @xmath58 , while the squares represent a dust - free model ( dashed lines ) with @xmath59 and @xmath60 . these particular models were chosen because they approximately reproduce the [ ] @xmath273727/ [ ] @xmath275007 and [ ] @xmath275007/h@xmath51 line ratios in the average seyfert 2 spectrum . interestingly , solar abundance , grainless calculations are not excluded by the line ratios presented in figure 3 . the dashed , dust - free , lines in figure 3 tend to converge in the area of seyfert observations . the dusty integrations tend to be shifted toward and into the liner region of the diagrams , that is with higher [ ] , [ ] , and [ ] relative to h@xmath61 and higher [ ] relative to [ ] . table 1 compares the fiducial models represented by the star and square in figure 3 with the observed average seyfert 2 spectra from ferland & osterbrock ( 1986 ) , dopita & sutherland ( 1995 ) , thompson ( 1995 ) , and villar - martn & binette ( 1997 ) . the uv / optical line ratios given by ferland & osterbrock , likely suffer from inaccurate absolute optical spectral flux calibrations coupled with substantially different aperture sizes between the iue uv spectra and ground based optical spectra coupled with the extendedness of the nlr , making many of these uv / optical narrow line ratios highly uncertain . the table also lists the observed range of equivalent width of h@xmath51 taken from netzer ( 1990 ) and binette , fosbury , & parker ( 1993 ) . in column 4 of table 1 we give the dereddened uv / optical hst / fos spectrum of the seyfert 2 galaxy ngc 3393 , from cooke et al . the cooke et al . observations were taken at two overlapping but significantly different aperture positions within a complicated two - dimensional line - emitting structure ; spectra covering 1100 2300 were taken at one position while the range 2300 6800 was covered at the other position . because of the unintended positional offset it is impossible to unambiguously join together the two spectral regions . we have normalized the two sections of the ngc 3393 spectrum so that the uv / optical ratios would agree with case b predictions . this does not guarantee consistency amongst the other emission lines , but it is not clear that the uncertainty due in this case to a 0.4 arcsec offset between two 1 arcsec diameter apertures is any worse than the errors incurred through the comparison of iue to ground based data , whereas the signal : noise in the ngc 3393 spectra is much higher than in the earlier data . observations by oliva et al . ( 1994 ) of the seyfert type 2 circinus galaxy are given in column 5 of table 1 . these observations include the ground - based optical and infrared spectra of many high ionization `` coronal lines '' ( ionization potentials greater than 100 ev ) . columns 6 and 7 of table 1 show the results of the predicted integrated line spectrum for the dust - free and the dusty fiducial points . recall that the integration parameters for the two fiducial models were chosen to simultaneously reproduce the [ ] / [ ] and [ ] /h@xmath51 emission line ratios of the average seyfert galaxy , and with this in mind , the integrated spectrum is representative of the observed values . compared to the mean seyfert 2 spectrum the uv lines , l@xmath61 , , and ] , are generally underpredicted by the solar , dust - free integration shown in column 5 of table 1 . in column 6 of the table , the dusty integration underpredicts l@xmath61 , and even worse than the dust - free case ( generally because of resonance line destruction by dust grains ) . the exceptions being that ] @xmath271909 line , is within the observed range , and is on the low end of the range . the predictions of the weaker uv emission lines observed in ngc 3393 , not included in the mean seyfert 2 spectrum of table 1 , are generally consistent with the observations . the execptions and will be elaborated upon in the next section . the integrations of the dust - free clouds predict l@xmath61/h@xmath51 @xmath928 , which is close to the high density case b limit ( @xmath934 ) , while the observations compiled by ferland & osterbrock ( 1986 ) suggest much larger ratios ( 30 70 ) . our calculations find a maximum l@xmath61/h@xmath51 @xmath9100 occuring at @xmath62= 9.75 @xmath10 . l@xmath61/h@xmath51 significantly larger than @xmath934 implies the existance of a substantial population of high density clouds ; more specifically 40 @xmath63 l@xmath61/h@xmath51 @xmath63 70 occurs for 5.5 @xmath64 8.5 . however , the spectrum resulting from integrations that heavily weight this dense gas would not be consistent with the strong optical forbidden line strengths relative to h@xmath51 . our predicted hydrogen line spectrum is in better agreement with the hst data of cooke et al . ( 1997 , see table 1 ) . the corrected observed l@xmath61/h@xmath51 @xmath919 is inconsistent with the compilations of ferland & osterbrock ( 1986 ) , with the caveat that this observed ratio is somewhat uncertain . l@xmath61/h@xmath51 ratios smaller than 23 are not possible , unless l@xmath61 is weak either due to very dense ( @xmath65 10@xmath66 @xmath10 ) or dusty gas . the ngc 3393 /h@xmath51 ratio is also consistent with the dusty simulations . hst observations of the narrow lines of several high luminosity agn ( wills et al . 1993 ) also yield ratios that are more consistent with the dusty simulations . clearly , uniform , long slit , high signal - to - noise , uv to infrared spectra are required to understand the narrow emission line spectra of agns . future observations using the stis on hst will accomplish a good part of this for the first time . the predicted narrow optical forbidden lines are generally within the observed ranges of the mean spectrum with the exceptions of [ ] and [ ] both being slightly underpredicted by @xmath950% and @xmath920% , respectively , indicating possible differences in abundances . however , these lines are predicted to be stronger relative to h@xmath51 than observed in ngc 3393 . the [ ] , [ ] lines are within the observed ranges of the mean seyfert 2 spectrum . both integrations of the predicted nitrogen line [ ] , are consistent with the observed mean seyfert 2 range , although the [ ] line is predicted to be lower that observed by nearly half . comparisons between both integrations and ngc 3393 of the nitrogen emission lines ( [ ] , [ ] , and ) relative to h@xmath51 indicate that nitrogen may be enhanced in that object relative to solar by a factor of @xmath92 . the strength of the @xmath274686 permitted line relative to h@xmath51 is often used to constrain the ionizing continuum . a stronger @xmath274686 line ( relative to h@xmath51 ) is indicative of more he@xmath67 ionizing photons . simple photoionization models often underpredict the line while doing well with the rest of the optical spectrum ( ferland & osterbrock 1986 ) . figure 3 and table 1 show that the integrated model can predict this line within the observed range . our fiducial integrations given in columns 5 and 6 of the table overpredict the entire spectrum relative to h@xmath51 when compared to ngc 3393 , indicative that ngc 3393 has a slightly softer continuum than the one assumed here ( @xmath682.1 ) . the predictions of the relative strengths of the optical iron coronal lines , [ ] , [ ] , [ ] , and [ ] to h@xmath51 are in accord with the observations of the circinus galaxy ( oliva et al . 1994 ) for the dust - free integrations . the dusty integrations underpredict these lines by factors of @xmath9 5 10 . however , oliva et al . ( 1994 ) , morisset & pquignot ( 1996 ) , and ferguson et al . ( 1997 ) discuss the apparent large uncertainties in the collision strengths of [ ] @xmath69 , [ ] @xmath70 , and [ ] @xmath71 . given these uncertainties we set the collision strengths for these transitions equal to 1 prior to running these simulations ( note : ferguson et al . used the collision strengths from the literature ) . these lines will not be useful probes of the high ionization narrow line region until this issue is resolved . the weakness of the optical doublet [ ] @xmath567291,7324 has been cited as evidence for the presence of dust grains with the nlr ( ferland 1993 ; and kingdon , ferland & feibelman , 1995 ) . table 1 indicates that the computations assuming solar abundances overpredict the [ ] @xmath277291 line by factors of 3 160 , with the dusty calculations on the very low end of the observations . this suggests that calcium is actually depleted by factors up to 160 compared to its solar abundance . the infrared forbidden [ ] lines from the integrations described above are compared with the observations of ngc 4151 made by thompson ( 1995 ) . because thompson s observations included p@xmath51 , we can reference the [ ] to h@xmath51 , as in table 1 , by assuming case b conditions for hydrogen . this is a valid assumption since p@xmath51/h@xmath51 predicted by our integrations is very nearly the case b ratio . table 1 shows that the predicted intensities of the `` dust - free '' integration ( column 3 ) are clearly too strong to match the observations , with some lines overpredicted by a factor of 10 . even for the `` dusty '' integration ( column 4 ) , where the iron abundance is depleted by a factor of ten , relative to solar , the strongest [ ] lines are too strong by a factor of 2 compared with h@xmath51 this evidence suggests that iron may be depleted by factors up to 20 in ngc 4151 . simpson et al . ( 1996 ) and veilleux et al . ( 1997 ) present ir spectroscopy of a number of seyfert 2 galaxies . simpson et al . found [ ] 1.257 @xmath36m / p@xmath51 @xmath72 1.21 ( @xmath73 0.81 ) , and veilleux et al . found [ ] 1.257 @xmath36m / p@xmath51 @xmath72 1.24 ( @xmath73 0.90 ) . as in ngc 4151 these observed values are better matched to the predicted ratio from the dusty integration in table 1 ( @xmath72 1.4 ) . the dust - free integration predicts a ratio of @xmath72 7 , much like the ratios arising in the shocked ( grain free ) environments of supernova remnants , plotted by both simpson et al . and veilleux et al . although the [ ] line ratios are too strong compared with the hydrogen lines , indicating an abundance difference , the relative [ ] spectrum is fairly consistent with the observations . thompson ( 1995 ) discusses at length the probable difficulties with the observation of weaker lines , such as 1.279 @xmath36 m and 1.677 @xmath36 m . table 1 also includes predictions of the relative strength of the far - ir [ ] 25.98@xmath36 m line , the lowest transition within the @xmath74d ground term . this line is predicted to be quite strong in the solar integration ( 4 times p@xmath51 ) . recall that the simulations presented here excluded gas with t@xmath75 3000 k , and emission from these low excitation lines might still be important at lower electron temperatures . in fact , we find that this line and others that arise within the ground term will have substantial contributions from agn - heated photodissociation regions , if present . none of the @xmath9 1 @xmath36 m [ ] lines are emissive in gas temperatures below 3000 k. our simulations of the infrared coronal lines shown in table 1 are generally in accord with the observations of the circinus galaxy for the dust - free integration . the [ ] , [ ] , and [ ] spectrum is well matched by the fiducial dust - free integrations . the dusty models do a poor job at matching the observations , particularly the [ ] line . our dust - free fiducial model predicts this line to within a factor of two , whereas the dusty integration underpredicts this line by over a factor of 30 . the predicted strength of the optical coronal line [ ] @xmath276087 is also consistent with the proposition that the high ionization lines are emitted in a dust - free environment . filippenko & halpern ( 1984 ) showed that the line width of optical forbidden lines is correlated with the critical density of those lines . other observers ( filippenko 1985 , appenzeller & stereicher 1988 , espey et al . 1994 and ho , filippenko , & sargent 1996 ) have extended this work . it was realized by filippenko & halpern that the probable explanation for the observed correlation is that the lines emit most efficiently near their critical densities and that the line emission is spatially stratified . our loc models are a quantitative formulation of this idea . by just glancing at figure 1 , one observes that the peak emission of the set of lines shown occurs over a range of gas density and distance . if the dominant velocity field is one which diminishes with distance from the nucleus , then we expect that those lines that are emitted optimally at smaller distances ( larger @xmath12 ) to have broader line widths than those that are optimally emitted farther away ( smaller @xmath12 ) and qualitatively agree with the observed correlation . consider the contour plots of [ ] @xmath274363 and [ ] @xmath276720 shown in figure 1 . it is clear by looking at their places of peak emission that [ ] will be much broader than [ ] , as is observed . figure 4 shows the results of a simple calculation of relative line widths assuming that the density distance phase space system of clouds obey the covering fraction and density indices of the fiducial square shown in figure 3 . in lieu of a more realistic velocity field which takes into account the mass distribution of the bulge in addition to the supermassive black hole , we assume virial velocities about a simple point mass of 10@xmath76 m@xmath77 . in the figure we show the full width at half maximum ( fwhm ) and the full width at 10% intensity ( fw10% ) for the ions indicated versus the logarithm of the critical density ( @xmath78 ) . the slopes of the fitted linear regression in the figure were computed to be 0.193 ( fw10% ) and 0.118 ( fwhm ) , similar to those observed by filippenko & halpern ( 1984 ) and espey at al . ( 1994 ) , which range from 0.095 to 0.200 . in a number of nearby seyfert galaxies the nlr is spatially resolved even on ground - based long - slit and imaging fabry perot spectra . these nlrs frequently have the form of either ionization cones ( cf . pogge 1988 ; tadhunter & tsvetanov 1989 ; wilson et al . 1993 ; robinson et al . 1994 ) or of apparent bow - shocks and cocoon shocks around linear radio sources ( bower et al . 1995 ; pogge & de robertis 1995 ; capetti et al . 1996 ; cooke et al . typical scales are 200 - 400 @xmath79 pc arcsec@xmath1 , where @xmath80h@xmath81 km s@xmath1 mpc@xmath1 . with the imminent commisioning of the hst stis spectrograph and ground - based adaptive optics systems , it should be possible to study spectroscopically a reasonable sample of nlr s with @xmath82 pc ( @xmath83 cm ) resolution . the exact scaling of such dimensions onto the y - axis of figures 1 and 2 is complicated by the fact that the ( probably beamed ) ionizing continuum luminosity seen by these nlrs is not directly measured . however , simple arguments based on counting recombination photons ( cf . binette , fosbury & parker 1993 ) imply that the gas sees an ionizing luminosity that is on average a factor of 10 larger than what we see . therefore , the range in incident flux used in the model grid presented here is applicable to these resolved nlrs , and it is of interest to consider the predicted radial dependence of various line intensity ratios when integrated over the cloud gas density distribution . figure 5 shows four different line ratios indicating the gas density , temperature , and ionization as functions of distance and gas density in the form of logarithmic contour plots . the figure also plots the ratios of the line intensities integrated over clouds of all densities , as functions of distance from the ionizing source ; shown are the results of two different integrations along the density axis . the density indicator [ ] @xmath276716/@xmath276731 is shown in the top two panels of figure 5a . the line ratio has the linear value of 1.36 in the extreme upper left corner of the plot at ( @xmath84 ) . moving down and to the right along a line of constant ionization parameter the solid lines are 1 dex increments and the dashed lines are 0.1 dex steps . the large flat plain of constant ratio corresponds to the high density limit or a linear value of @xmath9 0.44 . the line ratio as a function of distance shows that for the integrations presented the high density limit is predicted at small distances . the solid line ( integrated assuming that @xmath85 ) remains at that limit for larger distances , whereas the dashed line ( integrated assuming a steeper density distribution , i.e. @xmath86 ) turns upwards sooner . the dashed curve reaches a peak value nearly equal to the low density limit given by osterbrock ( 1989 ) . the gas density inferred from the dashed curve falls off as @xmath87 , while the density inferred from the solid curve is pinned at the high - density limit for larger values of r , then falls off more steeply than @xmath87 . the temperature indicator [ ] @xmath275007/@xmath274363 ( note that @xmath274959 is not included ) is shown in the bottom two plots of figure 5a . the contour plot has a peak value of 112 in the upper left hand corner ( @xmath88 ) . following a line of constant ionization parameter down and to the right , the solid contour line ( which has a backwards `` s '' shape ) has the logarithmic value of 2.0 , and the dashed lines are 0.2 dex increments . the outer contour in the lower right part of the plot has a logarithmic value of @xmath551.0 at ( @xmath89 ) . the contour plot shows that in the low density limit ( @xmath90 4.5 @xmath10 ) the ratio is a good temperature indicator in that moving perpendicular to the ionization parameter the ratio changes are roughly independent of density . at higher densities the ratio becomes dependent upon both temperature and density . the gas density integrated line ratio as a function of distance shows that for large distances from the ionizing source the ratio is independent of the weighting of the density axis ; i.e. , the ratio is in the low density limit . using the analytical expression from osterbrock ( 1989 ) the ratio can be converted into temperature in the low density limit . for the range 20.0 cm @xmath91 22.0 cm the inferred temperature has the range 14,000 k @xmath92 8400 k , falling off very gradually as approximately @xmath93 . at smaller radii the @xmath275007/@xmath274363 ratio further decreases , reflecting , mainly , the drop in @xmath275007 due to collisional quenching and the increase of @xmath274363 which continues to emit efficiently up to its critical density , rather than reflecting an increase in temperature . the [ ] @xmath275755/@xmath276584 line ratio is also a useful temperature indicator . wilson , binette & storchi - bergmann ( 1997 ) use this ratio in concert with the [ ] @xmath275007/@xmath274363 ratio as evidence for the presence of matter - bounded clouds in the nlr . defining @xmath94}$ ] as the [ ] line ratio and @xmath95}$ ] as the [ ] line ratio , our model integrations predict that @xmath95}$ ] versus @xmath94}$ ] depends little upon the density weighting factor @xmath48 . we find a linear - log relationship of the form @xmath96 } = 0.1593\log r_{[nii ] } + 0.1932.\ ] ] the fiducial dust - free integration from figure 3 ( @xmath683.1 ) predicts @xmath95 } = 0.0182 $ ] and @xmath94 } = 0.0215 $ ] , and the fiducial dusty simulation predicts @xmath95 } = 0.0206 $ ] and @xmath94 } = 0.0229 $ ] . these correspond to temperature differences ( @xmath97 } - t_{[nii]}$ ] ) of 3000 k and 3500 k , respectively . these predictions lie in the range of observed values ( wilson , binette & storchi - bergmann ) , and do not require the presence of a significant population of matter - bounded clouds , although we do not exclude this possibility . the ratio [ ] @xmath275007/(h@xmath61 + [ ] @xmath276584 ) has been used as an ionization indicator and this ratio is shown in the top left panel of figure 5b . the outer contour has the linear value of 1.0 and the dashed lines are @xmath980.2 dex steps . we have chosen to include both h@xmath61 and [ ] , since many filters include both lines . the contour plot of this ratio is relatively flat over a wide range in density , distance and ionization parameter . in fact , for a constant density of @xmath99 @xmath10 , the [ ] /(h@xmath61 + [ ] ) ratio would be constant from @xmath9 10 50 @xmath100 pc , even though the ionization parameter decreases by @xmath9 10@xmath101 over the same range . a ratio that behaves in similar manner is [ ] @xmath275007/h@xmath51 , and is shown in the bottom left panel of figure 5b . the outer contour has the linear value of 1.0 and the dashed lines are @xmath980.2 dex steps . the [ ] /h@xmath51 ratio is just as flat over a wider range in ionization parameter as is [ ] /(h@xmath61 + [ ] ) . thus , while these line ratios are useful in separating region galaxies from active galaxies ( c.f . bpt ) , they are not reliable ionization parameter indicators . in the top right panel of figure 5b the [ ] @xmath275007/(h@xmath61 + [ ] @xmath276584 ) as a function of distance is shown . the ratio has the feature of being constant from @xmath9 3 300 @xmath100 pc for the integration assuming a steep distribution ( @xmath86 ; dashed line ) of clouds along the density axis . the solid line ( @xmath102 ) is flat over a much smaller distance range , because in this integration emission from higher density gas is being included . the sharp falloff of the ratio at large distances ( @xmath103 pc ) occurs because the [ ] @xmath275007 emissivity is dropping very quickly with distance at the lowest gas density ( 10@xmath13 @xmath10 ) considered here , since the gas is underionized and can not produce the line efficiently , while the [ ] @xmath276584 line is simultaneously increasing in strength ( see figure 1 ) . the bottom right panel of figure 5b shows the integrated [ ] /h@xmath51 ratio as a function of distance ; this ratio behaves similarly to the [ ] /(h@xmath61 + [ ] ) ratio . both of these line ratios are observed to be relatively constant over factors of 3 4 in distance in several seyfert 2 galaxies . ( robinson et al . 1994 ; capetti et al . 1996 ) . figure 6a shows the [ ] @xmath275007/ [ ] @xmath273727 line ratio as a contour plot . in the figure , the outer contour has a linear value of 1.0 , and the dashed lines are @xmath980.2 dex steps . at low densities , the contours are closely spaced and parallel with ionization parameter , indicating that this line ratio is a good ionization indicator . at densities @xmath9 10@xmath104 @xmath10 , [ ] becomes collisionally quenched , while [ ] does not , thus the ratio becomes dependent upon density as well . in figure 6b we show the predicted [ ] / [ ] line ratio as a function of radius for 4 models compared with the observations of robinson et al . ( 1994 ) for ngc 4151 . the data have been placed on the distance axis by assuming that 10 @xmath72 100 pc , as used by robinson et al . the model integrations have been rescaled in distance by factors of 4 , corresponding to a 16-fold increase in the ionizing luminosity incident upon the clouds . the observed strengths of these and a number of other emission lines in ngc 4151 suggest that the ionizing luminosity incident upon the narrow - line clouds may be 10 20 times the observed one ( penston et al . 1990 ; robinson et al . ) , though the magnitude of the photon deficit is very uncertain ( see also schulz & komossa 1993 ) . since the ionizing luminosity employed here happens to match that of ngc 4151 emitted toward earth ( robinson et al . ) , our rescaling is consistent with the continuum beaming inferred from the observations . the predicted [ ] / [ ] ratio shown in figure 6b did not include emission from matter - bounded clouds ; including such clouds would tend to enhance the [ ] / [ ] ratio where it is small in figure 6b . robinson et al . ( 1994 ) include several other line ratios as functions of projected distance for ngc 4151 . these include [ ] / [ ] and [ ] / [ ] which behave similarly to [ ] / [ ] shown in figure 6b . also observed are [ ] /h@xmath61 , [ ] /h@xmath61 , [ ] /h@xmath61 , and [ ] /h@xmath61 . our simulations qualitatively match the general trends in these line ratios . the presence of dust in the nlr is a vital clue to the origin of the gas . for instance , if shocks are prevalent in the emitting region , then grains will not survive ( donahue & voit 1993 ) . both the dust - free and dusty simulations reproduce the average narrow emission line ratios shown in figure 3 . however , other lines do indicate substantial depletions relative to solar . the absence of an observed [ ] @xmath277291 line has led to speculation that the nlr is a dusty place ( ferland 1993 ; and kingdon , ferland & feibelman , 1995 ) . recent observations of the [ ] line ( villar - martn & binette 1997 ) show that the line is much weaker than the solar integrations predict ( table 1 ) , supporting villar - martn & binette conclusion that ca is sharply depleted . the simulations presented here indicate that ca is depleted relative to solar by factors of 3 160 based on the range of observed line strengths when compared to the dust - free simulations in table 1 . the results of the simulations for the infrared [ ] lines , as discussed in @xmath283.1.3 , indicate that fe is also underabundant relative to solar by factors of 10 20 in ngc 4151 . comparisons of the compilations of the [ ] 1.257@xmath36m / p@xmath51 line ratio by simpson et al . ( 1996 ) and veilleux et al . ( 1997 ) with the simulations provide further evidence that the nlr is a dusty environment . a similar conclusion was reached by simpson et al . two line ratios , l@xmath61/h@xmath51 and /h@xmath51 , observed in ngc 3393 and discussed in @xmath283.1.1 , also suggest that dust exists in the emitting gas . however , the uncertainty in the correction for the unintended positional error in the observations , makes the conclusion from l@xmath61/h@xmath51 uncertain . there is no positional error for the /h@xmath51 ratio , and this ratio makes a stronger statement supporting the presence of dust in ngc 3393 . evidence that a portion of the emitting gas is dust free is apparent in the observations of strong coronal lines in many agn . oliva et al . ( 1994 ) have observed a very strong [ ] 2.32 @xmath36 m infrared line and strong optical [ ] @xmath276087 , [ ] @xmath276374 , [ ] @xmath277892 , and [ ] @xmath275303 lines , indicating that the gas is dust - free . simulations of coronal line emitting gas by korista & ferland ( 1989 ) , oliva et al . ( 1994 ) ferguson , korista , & ferland ( 1997 ) , and the results of those shown in table 1 indicate that these elements are not depleted . taking the observations at face value , we are presented with a conundrum : the low ionization uv , optical [ ] , and infrared [ ] lines clearly indicate that dust is present , but the strong coronal lines suggest a dust - free environment . one possible reconciliation is to postulate that there is dust in the nlr , but only in the neutral or partially neutral portions of the clouds . in the simulations presented in this paper we have assumed that the entire cloud is either dust - free or dusty . it is not unreasonable to imagine that the highly ionized portion of nlr clouds to be dust free , perhaps because the grains have been exposed to a strong uv continuum for quite some time , and that in more neutral regions grains are free to exist . another possibility is that high ionization clouds have been processed through shocks , but those which emit more neutral ions have not . both of these possiblities argue for a very complex nlr environment , where the emitting gas has a variety of origins . we have shown that the agn narrow emission line spectra , from [ ] to [ ] , can largely be reproduced by assuming that the emitting region consists of clouds with a wide range of gas densities and distances from the ionizing source . here we chose simple power - law cloud distribution functions and the appropriate power law indices ( @xmath51 and @xmath50 ) whose resulting spectrum matched the observed [ ] / [ ] and [ ] /h@xmath51 ratios . while smooth power law distributions in cloud properties may be too simple , its success in predicting the observed spectrum points to the strength of the `` locally optimally emitting cloud '' scenario . the loc model integrations result in the following : \(2 ) the predicted h@xmath51 equivalent width , assuming full source coverage and an isotropic continuum , is more than sufficient to account for the observed emission in most seyfert galaxies ( table 1 ) . an anisotropic continuum will strengthen the predicted line equivalent widths and allow for smaller covering fractions . the predicted [ ] / [ ] ratio as a function of radius implies that the narrow line clouds see a continuum that is an order of magnitude more luminous than we observe in ngc 4151 . \(3 ) the loc integration simultaneously reproduces the low [ ] @xmath276716/@xmath276731 density and the high [ ] @xmath275007/@xmath274363 temperature ( cf . figure 3b ) , resolving a long standing problem . \(4 ) the observed line width critical density correlation for the optical forbidden lines can be reproduced . a natural consequence of an ensemble of clouds in the gas density distance plane is that lines with higher critical densities are formed at smaller distances ( cf . figure 1 ) . we have shown that if we assume that this ensemble of emitting clouds is placed in a gravitational potential , then the line width critical density correlation is very similar to the observed will naturally result . \(5 ) the presence of a large density range at each radius can change the values of the classical density and ionization - level indicators to at least more closely mimic the observed radial gradients in these intensity ratios . \(6 ) low ionization lines show clear evidence of depletion ( figure 3 and table 1 ) , but high ionization regions do not . it is possible that dust grains are selectively destroyed in high ionization regions by shocks or sublimation and that the neutral regions are undisturbed . this argues for a variety of nlr cloud origins . we thank andrew cooke for agreeing to our use of the ngc 3393 data . we acknowledge suport from nasa and nsf through nag-3223 and ast93 - 19034 , and stsci for go-06006.02 . we also thank the university of kentucky center for computational science and its director john connolly for access to the exemplar parallel supercomputer . 1. contours of constant logarithmic line equivalent widths as a function of @xmath105 and @xmath90 for the 23 ions indicated , referenced to the incident continuum at 4860 . the bold lines represent 1 dex increments and the dotted lines are 0.2 dex steps . the triangle is the peak of the equivalent width distribution and the contours decrease downward to the outer value of 1 . the reader will sometimes find it convenient to view the contour plots _ along _ the ridge at large inclination angle to the sheet of paper . the upper right hand plot on the first panel of the figure is the log t@xmath31 of the front face of the cloud . the temperature decreases from 10@xmath106 k in the lower left hand corner of the plot to 10@xmath107 k in the upper right hand corner . again , the bold lines represent 1 dex increments and the dotted lines are 0.2 dex steps . fig . 3. line diagnostic diagrams after the manner of baldwin , phillips & terlevich ( 1981 ) . different line types represent different abundance types , dashed are solar or dust - free , and solid lines are the dusty abundance set described in the text . different lines represent different families of the density weighting index , @xmath51 with different covering fraction indices , @xmath50 along each line with the arrows indicating the direction of the more negative index . the star represents the same model integration of dusty abundances with indices @xmath57 and @xmath108 . the square represents the same model integration with solar abundances with indices @xmath59 and @xmath109 . the large `` s '' and `` l '' symbols refer to the observations of the mean seyfert type 2 objects and liner type objects , respectively , as discussed in @xmath283.1 . fig . 4. line width versus critical density of the square point shown in fig . the solid circles are the full width at 10% intensity with a fitted slope of 0.193 . the open squares are fwhm with a fitted slope of 0.118 . the observed range in the slope is from 0.095 to 0.200 . 5. line ratio diagnostic diagrams of gas density , temperature , and ionization shown in two different forms . the left hand column of plots in the figure are logarithmic contour diagrams of the ratios indicated taken from the simulations shown in figure 1 . the right hand column of plots in figure 5 are graphs of the indicated line ratio versus distance from the central source for two different integrations as discussed in @xmath283 , but integrated in the density dimension only . the solid lines are integrations assuming that @xmath102 , and the dashed lines assume @xmath86 . each line ratio is shown on a different scale and we will discuss each in turn in the text in @xmath285 . 6. the predicted [ ] @xmath275007/ [ ] @xmath273727 ratio shown in two forms : ( a ) is a logarithmic contour diagram of the ratio taken from the simulations shown in figure 1 . the outer contour has a linear value of 1 , and the solid lines are 1.0 dex increments with the dashed lines being 0.2 dex steps . ( b ) the line ratio as function of distance from the ionizing source . the lower solid line is an integration in density assuming @xmath85 , and the upper solid line assumes @xmath86 . the dashed line is the fiducial integration with @xmath60 for the dust - free simulation . the dotted line is the fiducial integration with @xmath108 for the dusty simulation . all integrations have been rescaled by factors of 4 in distance corresponding to 16 in luminosity ( see @xmath285.3 ) . the filled circles are the observations of ngc 4151 by robinson et al . ( 1994 ) , assuming 10 @xmath72 100 pc . w@xmath110(h@xmath51)@xmath111 & 5 30 & n & & & 152 & 61 + w@xmath110(h@xmath51)@xmath111 & 15 100 & bfp & & & & + + l@xmath61 @xmath271216 & 55 @xmath112 20 & fo & 20.9 & & 28.7 & 9.62 + @xmath271240 & & & 1.42 & & 0.43 & 0.60 + @xmath271397 & & & @xmath1130.58 & & 0.09 & 0.03 + ] @xmath271402 & & & @xmath1130.84 & & 0.22 & 0.45 + @xmath271549 & 12 @xmath112 8 & fo & 4.75 & & 3.70 & 2.47 + @xmath271640 & & & 1.75 & & 2.18 & 2.82 + @xmath271909 & 5.5 @xmath112 3.7 & fo & 1.60 & & 1.25 & 3.63 + @xmath272326 & & & 0.25 & & 0.39 & 0.97 + @xmath114 @xmath272424 & & & 0.32 & & 0.19 & 0.31 + @xmath114 @xmath272471 & & & 0.11 & & 0.16 & 0.17 + @xmath272798 & 1.8 @xmath112 1.5 & fo & 0.43 & & 1.42 & 0.23 + @xmath273205 & & & 0.13 & & 0.12 & 0.15 + @xmath114 @xmath273426 & 1.2 @xmath112 0.2 & fo & 0.36 & & 0.62 & 0.67 + @xmath114 @xmath273727 & 3.2 @xmath112 1.7 & fo & 2.27 & & 2.32 & 3.20 + @xmath114 @xmath273869 & 1.4 @xmath112 0.4 & fo & 0.81 & & 1.13 & 1.00 + @xmath114 @xmath274074 & 0.3 @xmath112 0.2 & fo & 0.12 & & 0.26 & 0.28 + h@xmath115 @xmath274102 & & & 0.18 & & 0.27 & 0.27 + h@xmath50 @xmath274340 & & & 0.48 & & 0.48 & 0.47 ccccccc + + & & & & & & + & & & & & & + @xmath114 @xmath274363 & 0.2 @xmath112 0.1 & fo & 0.09 & & 0.21 & 0.22 + @xmath274686 & 0.3 @xmath112 0.1 & fo & 0.25 & 0.30 & 0.30 & 0.37 + h@xmath51 @xmath274861 & 1.00 & fo & 1.00 & 1.00 & 1.00 & 1.00 + @xmath114 @xmath275007 & 10.8 @xmath112 3.0 & fo & 10.2 & 11.8 & 11.7 & 10.8 + @xmath114 @xmath275199 & 0.15 @xmath112 0.09 & fo & 0.15 & 0.35 & 0.07 & 0.28 + @xmath114 @xmath275303 & & & & @xmath1130.10 & 0.033 & 0.004 + @xmath275876 & 0.13 @xmath112 0.06 & fo & @xmath1130.11 & & 0.13 & 0.11 + @xmath114 @xmath276087 & 0.10 @xmath112 0.05 & fo & & 0.055 & 0.049 & 0.009 + @xmath114 @xmath276300 & 0.57 @xmath112 0.20 & fo & 0.35 & 0.36 & 0.56 & 0.84 + @xmath114 @xmath276374 & 0.04 @xmath112 0.04 & fo & & 0.010 & 0.015 & 0.002 + h@xmath61 @xmath276563 & 3.1 @xmath112 0.1 & fo & 3.00 & 2.73 & 2.96 & 2.91 + @xmath114 @xmath276584 & 2.9 @xmath112 1.0 & fo & 3.64 & 3.19 & 1.18 & 1.69 + @xmath114 @xmath276720 & 1.5 @xmath112 0.5 & fo & & 1.64 & 1.20 & 1.87 + @xmath114 @xmath277135 & 0.24 @xmath112 0.07 & fo & & 0.16 & 0.22 & 0.27 + @xmath114 @xmath277291 & 0.04 @xmath112 0.04 & vb & & & 0.16 & 0.002 + @xmath114 @xmath277325 & 0.10 0.46@xmath116 & ds & & 0.073 & 0.20 & 0.21 + @xmath114 @xmath277892 & & & & 0.066 & 0.068 & 0.008 + @xmath114 @xmath27@xmath279069,9532 & 0.5 4.5@xmath116 & ds & & 0.94 & 1.40 & 1.61 + @xmath114 @xmath279915 & & & & 0.044 & 0.024 & 0.022 + @xmath114 1.252 @xmath36 m & & & & 0.042 & 0.11 & 0.086 + @xmath114 1.257 @xmath36 m & 0.12 @xmath112 0.01 & t & & 0.09 & 1.16 & 0.23 + @xmath114 1.279 @xmath36 m & 0.03 @xmath112 0.01 & t & & & 0.05 & 0.01 + p@xmath51 1.2818 @xmath36 m & 0.16 @xmath112 0.01@xmath117 & t & & 0.15 & 0.16 & 0.15 @xmath114 1.534 @xmath36 m & @xmath1130.02 @xmath112 0.01 & t & & & 0.10 & 0.01 + @xmath114 1.644 @xmath36 m & 0.09 @xmath112 0.01 & t & & 0.07 & 1.12 & 0.23 + @xmath114 1.677 @xmath36 m & 0.02 @xmath112 0.01 & t & & & 0.07 & 0.01 + @xmath114 1.96 @xmath36 m & & & & 0.088 & 0.088 & 0.015 + br@xmath50 2.166 @xmath36 m & & & & 0.027@xmath117 & 0.029 & 0.026 + @xmath114 2.32 @xmath36 m & & & & 0.065 & 0.15 & 0.002 + @xmath114 2.48 @xmath36 m & & & & 0.15 & 0.11 & 0.020 + @xmath114 3.94 @xmath36 m & & & & 0.19 & 0.17 & 0.027 + br@xmath61 4.052 @xmath36 m & & & & 0.079@xmath117 & 0.080 & 0.073 + @xmath114 25.98 @xmath36 m & & & & & 0.64 & 0.13 = 2.em @xmath118the range of observed values taken from the reference in column 3 . + = 2.em @xmath119reference of the observed values : ( n ) is netzer 1990 , ( bfp ) is binette , fosbury , & parker ( 1993 ) , ( fo ) is ferland & osterbrock ( 1986 ) , ( vb ) is from villar - martn & binette ( 1997 ) , ( ds ) denotes a range of values from dopita & sutherland ( 1995 ) , and ( t ) is from an observation of ngc 4151 by thompson ( 1995 ) . + = 2.em @xmath120observations of ngc 3393 from cooke et al . + = 2.em @xmath121observations of the circinus galaxy from oliva et al . + = 2.em @xmath122predictions from the integrated spectrum assuming solar abundances indicated by the square in fig . 3 . + = 2.em @xmath123predictions from the integrated spectrum assuming region abundances indicated by the star in fig . 3 . + = 2.em @xmath111equivalent width of h@xmath51 in . + = 2.em @xmath116these observations are given as a range . + = 2.em @xmath117assuming case b conditions .

it is shown that the electronic conduction in silicon - on - insulator ( soi ) layers exhibits a metallic regime which is very similar to that in high - mobility si - metal oxide semiconductor structures ( mos ) . the peak in the electron mobility versus density , the strong drop in resistivity and the critical concentration for the metal - insulator transition are all consistent . on the basis of our soi data for the temperature and in - plane magnetic field dependence of the resistivity , we discuss several models for the metallic state in two dimensions . we find that the observed behavior can be well described by the theory on the interaction corrections in the ballistic regime . for the investigated regime , the temperature dependent screening of scattering potentials gives also a good description of the data . 2

cond-mat0207170

the discovery of an apparent metallic state in si - metal oxide semiconductor ( mos ) structures @xcite has attracted much attention as it seemed to contradict the two - dimensional ( 2d ) localization behavior @xcite . it was found that above a critical carrier concentration @xmath0 , the resistivity strongly decreases with decreasing temperature @xmath1 ( metallic regime ) , whereas it increases for lower densities @xmath2 ( insulating regime ) . the scaling parameter @xmath3 shows a critical behavior around @xmath0 @xcite . at first the discovery of the metallic state in two dimensions was met with great scepticism . but the metal - to - insulator transition ( mit ) was confirmed first in si - mos structures from a different source @xcite and further in several other 2d material systems , which are n- and p- sige and algaas and n - alas structures ( for references see @xcite ) . after these experimental findings , the question was raised whether the metallic behavior manifests a new electronic ground state with dominating quantum behavior or if the strong resistivity drop towards low @xmath1 is based on familiar ( i.e. non coherent ) effects ( for a detailed discussion see @xcite and references therein ) . in this work , we report on the finding of a metallic state in silicon - on - insulator ( soi ) structures . in section [ sec : experiment ] , on the experiment , we describe the investigated samples and the setup . we further present the data on the temperature and the magnetic field dependence of the resistivity and we show the mobility versus density behavior . the experimental results of the soi system are compared in detail with the properties of si - mos structures and the differences are discussed . the discussion of the results follows in section [ sec : discussion ] . here we discuss and compare five important suggestions on the physical origin of the metallic state in two dimensions . it turns out that the coherent interaction corrections to the conductivity in the ballistic regime @xcite are able to describe the observed resistivity behavior quite well . for the investigated regime also the temperature dependent screening of scattering potentials @xcite gives a fairly good description of the data . after a short summary , we describe in appendix [ sec : appendixtrapmodel ] our numerical calculations on the charged trap state model in detail . our investigations were performed on two high - mobility soi structures which were prepared recently in finland . the soi hall bars have identical peak mobilities of @xmath4@xmath5/vs . they were fabricated on commercial unibond wafers produced by the smart - cut process @xcite . the thickness of the si - layer amounts 200 nm and that of the buried oxide layer 400 nm in these wafers . the background doping is n - type of about 10@xmath6@xmath7 which freezes out at low temperature . after a two - step thermal oxidation , the final thickness of the si - layer is 60 nm . the gate oxide thickness is about 90 nm with a dielectric strength of approximately 1v / nm . the electric contacts were phosphorus - implanted to an activated carrier concentration of @xmath8@xmath7 . the al - metallization is 300 nm thick and sintered at 400@xmath9c . a scheme of the final layer sequence is shown in figure [ fig : samplescheme ] . the etched hall bars have a size of 800@xmath10 m times 100@xmath10 m giving a length to width ratio of 8 . all resistivity and hall measurements were performed in a four terminal ac - technique at low frequencies of typically 13 or 17hz . the soi structures were investigated in a @xmath11he cryostat down to 1.4k . most investigations were performed on sample number l2 , sample l1 showed in several measurements practically the same behavior . the soi sample number l2 was intensively investigated in a density range from @xmath12 to @xmath13@xmath14 . at the low density side of this range , the contacts became high ohmic . due to capacitive effects , the measured voltages showed a delay time of more than 100ms before they came into a steady state . thus voltage measurements in dc - technique lead to other values than those with the 13hz ac - technique . we limited our measurements to the range were the ac- and dc - technique gave the same results and performed all investigations in the ac - technique as this gives a better signal - to - noise ratio . at the higher limit of the investigated density range , the gate started to leak and the current between gate and sample would have lead to errors in the resistivity calculations . the investigations were thus kept in the above given density range . from the resistivity and hall measurements , the carrier density @xmath15 and the mobility @xmath10 were calculated assuming the linear drude behavior to be valid . this is not clear a priori as quantum interference effects might give relevant contributions to the conductivity and would alter the necessary evaluation . the possible importance of quantum effects will be discussed later in this work . for now , @xmath15 and @xmath10 should be used as apparent drude parameters for the 2d electron gas . we have evaluated the dependence of mobility @xmath10 on the electron density @xmath15 by varying the gate voltage @xmath16 at a constant temperature of @xmath17k . in fig . [ fig : msisoi ] , the @xmath18-dependence for soi is compared with that for several si - mos samples @xcite with different mobilities . the soi structure has a peak mobility @xmath19 of @xmath20@xmath5/vs at a density of @xmath21@xmath14 . as can be seen from fig . [ fig : msisoi ] , the si - mos samples have lower or higher peak mobilities , depending on the sample quality , but the overall behavior of @xmath18 is very similar for the two sample types . at high densities , the electron wave function is squeezed by the strong electric field of the triangular potential towards the gate - sided si / sio@xmath22 interface and thus interface roughness scattering dominates . at low @xmath15 , the dominating scattering process is caused by impurities . at the transition region between the two scattering mechanisms , the mobility reaches its peak value @xmath19 @xcite . the most typical feature of the metallic state in 2d systems is expressed by the strong resistivity drop towards lower temperatures . this strong drop in @xmath23 is clearly visible in fig . [ fig : rsisoi]b for the soi sample in the range of @xmath24@xmath14 . the maximum decrease in @xmath23 amounts up to about a factor 3.5 for @xmath25@xmath14 . for the two lower density curves with @xmath26 and @xmath27@xmath14 , an insulating behavior is observed . the critical concentration @xmath0 , at which the behavior changes from insulating to metallic , lies at about @xmath28@xmath14 . for comparison , in fig . [ fig : rsisoi]a and [ fig : rsisoi]c the very high and medium mobility si - mos samples @xcite si-15 and si-4/32 with @xmath29 and @xmath30@xmath5/vs , respectively , are shown . in sample si-15 , the strong decrease in @xmath23 is shifted to lower @xmath1 and the maximum decrease amounts up to a factor 7 , in si-4/32 the decrease is at somewhat higher @xmath1 in comparison to the soi structure and the maximum decrease is about a factor 3 . this shows a clear trend of the properties of the metallic regime with the sample quality , i.e. the peak mobility @xmath19 and manifests the similarities between soi and si - mos structures . additionally , in fig . [ fig : rsisoi ] , dashed - dotted lines at @xmath31 , @xmath32 and @xmath33 are shown . 2 figure [ fig : rsisoisingle ] shows a direct comparison of the strong changes in @xmath23 for the four samples si-15 , si-43 , soi - l2 and si-4/32 for the same low-@xmath1 resistivity @xmath34 . the same behavior , as described above , is visible : the higher the peak mobility , the stronger is the change in @xmath23 and the lower is the temperature where the increase starts . for si - mos structures , it was shown that the critical conductance @xmath35 changes systematically with the peak momentum relaxation time @xmath36 , which is a measure of the sample quality @xcite . here @xmath37 denotes the effective conductivity mass which is the transverse mass of the two lower lying valleys of @xmath38@xmath39 and @xmath40 the elementary charge . for the soi structure , @xmath41 is about @xmath42 and is compared in fig . [ fig : gctp ] with the values for several si - mos samples @xcite . it can be seen , that the @xmath41-value for soi coincides quite well with the general behavior of the si - mos samples . as was seen so far , the soi structure behaves very similar to a high mobility si - mos sample . therefore , we ask the question : what are the similarities and differences between the two sample types ? due to the different effective masses @xmath37 perpendicular to the si / sio@xmath22 interfaces and due to strain effects , the six - fold conduction band ( cb ) degeneracy at the @xmath43-points of bulk si is lifted and only the two lower lying valleys with their longitudinal axis perpendicular to the 2d layer are occupied . the si layer in soi is only 60 nm thick , below there is the buried oxide layer , which constitutes also of a thermally grown oxide fabricated before the wafer bonding process . the active si - layer has an n - type background doping of about @xmath44@xmath7 which freezes out at low @xmath1 . in the si - mos samples , the electrons are confined in the surface inversion layer of a nominally p - type doped bulk si crystal . our high mobility si - mos samples were doped with an acceptor density of about @xmath45@xmath7 . due to the applied gate voltage @xmath16 , an approximately triangular potential well results for the electrons at the upper si / sio@xmath22 interface . nevertheless , the extension of the electronic wave function is different for the soi and the si - mos structure due to the different background doping . in the n - type soi structure , the fermi level @xmath46 is pinned underneath the 2d layer between the donor level and the cb edge . the band bending below the 2d layer is therefore quite small . in p - type si - mos , there is a wide depletion layer of about 0.85@xmath10 m below the 2d layer , as @xmath46 is pinned here between the acceptor level and the valence band ( vb ) edge in the undistorted p - type region . the different band bending below the 2d layer has an influence on the extension of the wave functions especially for small electron densities . we have calculated the wave function @xmath47 and the triangular potential @xmath48 by solving the schrdinger and poisson equations in a self consistent way ( with @xmath49 the coordinate perpendicular to the interface layer ) . in order to solve the schrdinger equation , the charge distribution latexmath:[$\rho(z ) \propto among the electrons were considered , thus giving a first order approximation of @xmath47 and @xmath48 . figure [ fig : psiwidth ] shows the width of the wave function versus electron density @xmath15 . the width of @xmath51 is defined as the length where the central 90 percent of the expectation probability @xmath52 is contained , with 5 percent at the left and at the right side ignored . for a high electron density of @xmath53@xmath14 the width is with 7.6 and 9.2 nm nearly the same for both structures . at lower densities , the difference in the width of @xmath51 increases . the depletion width of 0.85@xmath10 m at an acceptor density of @xmath54@xmath7 corresponds to a 2d charge density of @xmath55@xmath14 . thus , at that 2d electron density , the electric field at the si / sio@xmath22 interface is in si - mos twice as high as in the soi structure . as can be seen from fig . [ fig : psiwidth ] , the difference in the wave function extension is indeed large at that density . at still lower density , the width saturates in si - mos , as the field from the conducting 2d electrons is less important than that from the acceptors in the depletion layer . on the opposite , the width in the soi structure further increases , as there is no electric field confinement from other charges than the conducting electrons itself . figure [ fig : psipot ] shows for comparison the potential shape @xmath48 and the wave function @xmath47 at @xmath56@xmath14 . at that density , the confining potential is much steeper in si - mos and thus the energy @xmath57 of the lowest electronic subband is with 31mev much higher than in soi with 14mev . in fig . [ fig : psipot ] , the difference in the width of the wave functions is also clearly visible and is caused by the difference in the potential steepness . for the case that higher subbands @xmath58 are occupied , the differences in their energies and extensions should be even larger . in the considered density range , the width of the wave function is always smaller than the si - layer thickness of the soi structure and thus the electron do not feel the lower si / sio@xmath22 interface towards the buried oxide layer . in order that such an effect could become significant , the si layer would have to be as thin as 20 nm for @xmath59@xmath14 and even thinner for higher densities . we have also investigated the soi structures in perpendicular and parallel ( i.e. in - plane ) magnetic field @xmath60 . figure [ fig : rhobperp ] shows typical magnetoresistivity curves @xmath61 for @xmath60 perpendicular to the 2d layer for different densities at @xmath62k . for the lower densities , the weak localization peak around @xmath63 is clearly visible . at about 1 t , the shubnikov - de haas oscillations set in and gain strongly in amplitude towards higher magnetic fields . at around 4 t , the spin splitting is visible . the behavior in perpendicular magnetic field is again very similar to that of si - mos samples with comparable mobility . for in - plane magnetic field , the relative change @xmath64 of the magnetoresistivity of soi sample l2 is shown in fig . [ fig : rhobpar ] for different densities at @xmath65k . at high densities of about @xmath66@xmath14 , the changes are quite small , whereas for the lower densities there is a strong increase in @xmath67 by up to a factor 4 . at the lowest density of @xmath68@xmath14 , the observed change is again lower than for the previous densities as can be seen from the crossing of the upper curves in fig . [ fig : rhobpar ] . this indicates a saturation of the strong increase in @xmath69 at about @xmath70@xmath14 . an even stronger @xmath71-dependence was observed in high - mobility si - mos structures in the metallic regime @xcite . again the changes were stronger for the lower densities @xmath15 . the increase with @xmath72 was first attributed to the spin polarization of the conducting electrons @xcite and later to the increase in disorder caused by the magnetic field @xcite . a strong change of @xmath71 was also observed in p - type gaas / algaas structures . above , we have presented new results on the 2d metallic state in silicon - on - insulator structures and compared its properties with that of high - mobility si - mos samples . as the physical origin of the 2d metallic state is still an open question , we like to discuss the different possibilities in detail . there appear to remain five important suggestions for the 2d metallic state at low temperatures . these are ( i ) a new quantum mechanical ground state for electrons in a solid , ( ii ) scattering of electrons at charged hole traps in the oxide layer , ( iii ) interband scattering between different bands , ( iv ) temperature dependent screening of impurity scattering , and ( v ) coherent interaction corrections in the ballistic regime . if the first suggestion proves to be correct , the 2d mit would indeed be the manifestation of some new physics , whereas the next three suggestions are based on settled physical phenomena , which may be described by terms like conventional , semiclassical or non - coherent ( on the first sight ) . the fifth suggestion was proposed quite recently and it has to be discussed whether it can be classified in the terminology used above or not . the first suggestion ( i ) of a new quantum mechanical ground state is based on the scalability of the resistance in the neighborhood of the apparent mit in high - mobility si - mos structures @xcite . it was found that the @xmath73-curves for @xmath74 and for @xmath75 can be brought to coincidence on two distinct curves for the insulating and the metallic regime , respectively , by scaling the temperature with @xmath76 . the scaling parameter exhibits a power law @xmath77 , with @xmath78 and @xmath79 . such a critical scaling behavior indicates the metal - insulator transition to be a quantum phase transition @xcite . the seminal paper on the absence of quantum diffusion in two dimensions @xcite was based on the assumption of non - interacting electrons . in this case the weak localization contribution decreases the conductivity and leads to continuing higher resistivities with decreasing temperatures . in the recent quantum - phase - transition models for the unexpected mit in 2d , the electron - electron ( e - e ) interaction plays a dominant role @xcite . the quantum coherent interaction corrections consist of positive and negative contributions to the conductivity and may lead to a decrease of the resistivity under certain conditions@xcite . in any case , quantum coherence is a prerequisite that the mit can be a quantum phase transition . a metallic behavior , i.e. a strong drop in the resistivity , was not only observed in the near vicinity of the critical concentration @xmath0 but also far away , for densities up to @xmath80@xmath14 ( i.e. more than 10 times @xmath0 ) . as the decrease in @xmath23 looks quite similar near @xmath0 and far away , it is quite instructive to look for quantum effects over an extended density range . in a recent work @xcite , the temperature dependence of the weak localization was investigated in the metallic state of high - mobility si - mos structures in order to find the borders of quantum coherence . in that work , the temperature threshold @xmath81 for the single - electron quantum coherence was deduced from the crossing point of the temperature dependence of the phase coherence time @xmath82 and the momentum relaxation time @xmath83 . when @xmath84 , the coherence time is long enough that an electron can return in the coherent state to its origin whereas for @xmath85 this is not possible . it was found for the si - mos system @xcite that for @xmath86@xmath14 , the strong decrease in @xmath23 takes place above @xmath81 and thus should not be related to a single - electron coherent effect . also the second border @xmath87 lies below the strong resistivity drop at still higher densities . therefore also the e - e interaction induced quantum corrections to conductivity should not be responsible for the resistivity drop in this regime . this should at least be deduced in the picture where quantum coherent effects and screening are attributed to two different physical phenomena . but recently an interesting paper by zala et al . has appeared @xcite , which may indicate that the two before mentioned effects are caused by the same physics . this will be discussed in more detail later in this work . in any case , the work on the borders for quantum coherence @xcite is strictly applicable only in the higher density regime and does not give a direct answer on the importance of quantum coherence in the close vicinity of the critical density @xmath0 . in a recent work , punnoose and finkelstein show @xcite by a renormalization group analysis that especially for high - mobility si - mos structures , the presence of two degenerate conduction band valleys , stabilizes and enhances coherent e - e interaction effects in the vicinity of @xmath0 . they are able to describe the temperature dependent changes in conductance in an @xmath15-range between 0.83 and @xmath88@xmath14 without any adjustable parameter . so it is quite surprising that at low @xmath15 the whole drop in @xmath89 seems to come from quantum coherent interaction effects while at high @xmath15 the drop takes place without quantum coherence and nevertheless the strong @xmath1-dependence looks so similar in the different regions . it is not yet clear whether the metallic and insulating behavior belong to a quantum phase transition or not . this work focuses on the metallic regime and will not be able to judge on the metal - to - insulator transition itself . the second suggestion ( ii ) on the scattering of electrons at charged hole traps in the oxide layer by altshuler and maslov ( am ) @xcite is prepared especially for si - mos structures with the si / sio@xmath22 interface on the gated side of the 2d layer . the same model should be applicable also for our soi structures , as an equivalent si / sio@xmath22 interface is present for the 2d electron confinement . it is well known that in the sio@xmath22 interface layer there exist several types of defects states @xcite . in the am model , it is assumed that a relative large number of hole trap states exists at a certain trap energy @xmath90 , which is spatially homogeneously distributed in the oxide layer . it is further assumed that these trap state lie at a well defined energy @xmath91 if no external electric field is applied . the potential gradient due to an applied gate voltage @xmath16 causes a linear increase of the trap energy position . in order to describe the total energy of the trap states , also the mirror charge of the traps inside the 2d electron gas has to be included . this leads to a down bending or the energetic position towards the interface and causes a maximum in the total trap energy @xmath92 , with @xmath49 the position inside the oxide layer ( see fig . [ fig : trap_energy ] in appendix [ sec : appendixtrapmodel ] ) . when the fermi energy lies below certain trap states at energies @xmath92 ( for low @xmath15 ) these states are positively charged and cause a large scattering probability for the 2d electrons . at high densities all trap states lie below @xmath46 and only due to the boltzmann occupation tail , an exponentially small number of traps contributes to the electron scattering . the charged trap states form dipoles together with their mirror charge in the 2d layer which has to be taken into account in the calculation of the scattering cross section . in order to calculate the resistivity , one needs to know how the fermi energy @xmath46 varies with the temperature @xmath1 ( in the work of am , the chemical potential at @xmath93 is called the fermi energy @xmath46 whereas the chemical potential at @xmath94 is denoted by @xmath95 ; in this work we use @xmath96 instead ) . am suggest two scenarios for the @xmath1 dependence of the fermi energy of the 2d electron gas : ( a ) it coincides with the si substrate or ( b ) it is decoupled from the substrate . for case ( a ) , the calculation of @xmath23 by am show a behavior which looks quite similar to the observed temperature dependence near the critical density @xmath0 for high mobility si - mos structures . the strong increase in @xmath89 with increasing @xmath1 occurs around @xmath97 as then the occupied trap states are partially emptied ( and charged ) due to the broadening of the fermi occupation function and the scattering rate is increased accordingly . for case ( a ) , the insulating and metallic regions are distributed quite symmetrically around the separatrix which occurs for @xmath98 . for case ( b ) , the resistivity is decreasing in both regimes ( @xmath99 ) according to the calculations of am @xcite and seems to be unable to explain the experimentally observed behavior . as mentioned by am , the more realistic case is ( b ) as the depletion layer isolates the 2d electron gas from the substrate and both regions should have different quasi fermi energies . this is also verified experimentally @xcite . this situation causes some difficulty in order to apply the charge trap model to the high - mobility si - mos structures as well as to the soi samples considered in this work . we have performed numerical calculations of the am - trap model for the soi data . these calculations are described in appendix a in detail . if one applies an increasing gate voltage @xmath16 in order to increase the electron density @xmath15 in the 2d layer , the spatial maximum in the trap energy shifts quite fast to lower energies . this leads to a strong occupation of the trap states as soon as the maximum shifts below the fermi energy @xmath46 and the scattering efficiency decreases drastically . in fig.[fig : rho_trap_soi ] , we have calculated the @xmath23 dependence at an intermediate temperature of @xmath100k , where we have a set of measurement data . the temperature @xmath101 was chosen so that @xmath102 is for a large density interval neither in the low-@xmath1 saturation nor in the `` normal '' high-@xmath1 regime . the resistivity behavior should be dominated by thermal excitation of carriers according to the am - trap model . the low-@xmath1 saturation of @xmath23 in the metallic regime can not be described with the am - trap model alone , as all traps get filled according to the exponential boltzmann - tail and the resistivity tends to zero . only by assuming additional residual scattering centers , the resistivity saturates . we have evaluated the measurement data at @xmath101 only for densities where the low-@xmath1 saturation should not have a large influence yet . figure [ fig : rho_trap_soi ] shows that the decrease in resistivity vs. density is too strong by many orders of magnitude in the am - trap model . at @xmath103@xmath14 , @xmath89 should be below @xmath104 . this low value results from a very small number of positively charged trap states according to the fermi - dirac occupation function . temperature dependent screening effects were not considered in this calculation as they give only a change in @xmath89 by one or two orders of magnitude , but the calculated resistivity is by about 30 orders of magnitude too small compared with the measured values . in the same density interval , the measured resistivity decreases just by about one order of magnitude , which appears as a nearly flat line on the scale of figure [ fig : rho_trap_soi ] . we conclude on the am - trap model that it can not explain the experimental behavior of the soi ( and similarly of the si - mos ) samples at intermediate temperatures as the carrier freeze out on the trap states occurs by far too fast if the trap maximum is below @xmath46 . further improvements of the am - model may be possible , but one has to note that the addition of trap states at other energies or the energetical broadening of the trap states around @xmath90 destroys the critical behavior around @xmath0 and the modified model would probably not be able to describe the abrupt metal to insulator transition . moreover , case ( a ) of the trap model by am seems to be able to explain the @xmath23 dependence only in a very narrow density range around @xmath0 but can not explain the metallic behavior for the wide @xmath15 range as observed in si - mos and soi samples . thus it seems to be possible that the hole trap model in the present form may explain some important contributions to @xmath23 in si - mos and soi but is unable to describe the whole temperature dependence . as mentioned above , the am model is especially prepared for samples with an sio@xmath22 oxide layer as on border of the 2d electron gas . the model will thus probably not be able to explain the metal - insulator transition in other material systems with strong changes in @xmath23 like p - type algaas or si / sige . on the other hand , it is not clear anyway whether the apparent metal - insulator phase transition has a universal origin or if it is caused by several different effects in the different material systems . the third suggestion ( iii ) is attributed to interband scattering between different conduction or valence bands . the correlation between the existence of two conduction bands and the metallic behavior was proposed by pudalov for si - mos.@xcite he assumed that the spin - orbit coupling due to the asymmetry of the inversion layer potential is strong enough in order to create the splitting . but the calculation of the spin - orbit coupling is not straight forward and have to be performed with great care.@xcite a later estimate by pudalov et al . @xcite gives a clearly smaller value for the spin - orbit coupling constant of @xmath105kcm , so that the corresponding splitting should not be the origin of the metallic behavior in si - mos structures . for p - type algaas papadakis et al . have argued @xcite on the spin - orbit coupling due to the inversion asymmetry of the electron layer potential , but it turned out that the effect is due to an anomalous behavior of the magneto - oscillations @xcite . nevertheless , the apparent metallic behavior in p - type algaas layers strongly points towards a two - band transport effect . @xcite there , the heavy hole band is split and inelastic interband coulomb scattering between the two bands with different dispersion takes place . this was confirmed by magnetotransport measurements , where the positive magnetoresistance is attributed to the classical two - band scattering effect . but for n - type si - based heterostructures , the two - band transport should not be of great importance , as the conduction band splitting seems to be rather small . the forth suggestion ( iv ) on temperature dependent screening evolves from calculations of the mobility versus electron density dependence for si - mos @xcite , si / sige @xcite and iii - v @xcite structures in the 80th . in these early works , the resulting @xmath23-dependence was assumed to be relative small . the temperature dependence of the resistivity is dominated by the behavior of the screening function @xmath106 . for @xmath93 , this function has a distinct kink at @xmath107 with @xmath108 the fermi wave vector . for higher @xmath1 , this kink is smeared out @xcite and changes the scattering efficiency especially for @xmath109 . thus for large angle scattering , the temperature dependence of the screening function strongly influences the scattering efficiency . gold and dolgopolov give analytic results for the screening @xcite that consist of linear - in-@xmath1 and @xmath110 terms @xmath111^{3/2 } , \label{eq : dsigmagold}\ ] ] with @xmath112 , @xmath113 ^ 2 $ ] and @xmath114 a constant depending whether the scattering is dominated by impurities ( @xmath115 ) or by surface - roughness ( @xmath116 ) . we have fitted the @xmath23 behavior with the eq . [ eq : dsigmagold ] , assuming @xmath117 and @xmath118 . the prefactor @xmath119 was used as a fitting parameter . as the analytic results are only valid for @xmath120 , the fit was limited to that range . figure [ fig : rhotfitscreening ] shows that the fit gives a very good agreement with the observed @xmath1 dependence . the fitted prefactor @xmath119 increases from @xmath121 at @xmath122@xmath14 to @xmath123 at @xmath124@xmath14 . for lower densities , the fit does not give reasonable results as the slope of @xmath23 changes sign at the critical density @xmath0 which can not be described by the screening approximation alone . in fig . [ fig : cn ] , the range of the increase in @xmath119 towards lower densities is compared with the calculated behavior according to gold and dolgopolov @xcite . as can be seen , there is good agreement with the curve of the local field correction parameter of @xmath125 . but if one inserts @xmath126 for @xmath127 in @xmath128 as given for hubbard s approximation in ref . @xcite , one obtains @xmath129 ( with @xmath130 the valley degeneracy ) . thus the above screening description is able to describe the temperature dependence but has to use a larger value for the local field correction parameter @xmath131 as follows from the hubbard approximation . this discrepancy possibly results from the fact that in local field correction parameter only exchange effects are taken into account and no correlation effects . but the latter might become important for large electron - electron interaction parameters @xmath132 . the parameter @xmath132 describes the ratio of coulomb to kinetic energy and can be written as @xmath133 , with @xmath134 the bohr radius . recently , das sarma and hwang calculated @xcite that due to temperature dependent screening , changes in @xmath23 by even up to one order in magnitude can be explained by numerical calculations . another puzzling phenomenon in high - mobility si - mos structures , but also in the soi samples investigated in this work and in other material systems , is the strong positive magnetoresistance in parallel magnetic field @xmath72 . in si - mos structures a more than ten - fold and in soi a nearly four - fold increase in @xmath71 was observed . dolgopolov and gold calculated @xcite that an increase of up to a factor 4 in @xmath71 can take place as the electron system becomes completely spin polarized . this strong increase is caused by changes in the density of states at @xmath46 , in the fermi wave vector @xmath108 and in the screening wave number @xmath135 @xcite . we have fitted our data on soi according to this theory , the results are shown in fig.[fig : rhobparfit ] . as can be seen , at high electron densities ( i.e. low resistivity ) the fit in fig . [ fig : rhobparfit ] is in good agreement with the data , whereas at lower densities ( i.e. high resistivity and strong changes in @xmath71 ) , the screening behavior leads to curves which are rather linear and do not show the curvature of the measured values at low @xmath72 fields . for low density , at high @xmath72 fields , the model calculation shows an abrupt saturation of the resistivity which corresponds to the complete spin polarization of the electron system . both discrepancies between calculation and measurement may be caused by the fact that the model @xcite was formulated for @xmath93k . in that case , the fermi - dirac distribution is a step function and leads to the kink at the @xmath72 saturation field . the approximately linear behavior of @xmath71 for small @xmath72 is probably caused by the distinct kink @xcite in the screening function at zero @xmath1 . for @xmath63 only one kink exists at @xmath107 which splits into two kinks for @xmath136.@xcite at @xmath137 , the kinks in the screening function are rounded and thus a small splitting of the two screening functions ( by @xmath72 ) for the spin up and spin down electron systems will not be so effective . in order to fit the screening model to the data , an effective @xmath138 factor was used as the only fitting parameter instead of a constant value for @xmath139 . the resulting ratio @xmath140 is shown by cross symbols in the inset of fig.[fig : rhobparfit ] in comparison to the ratio @xmath141 from the analysis of shubnikov - de haas measurements on si - mos samples @xcite ( solid line ) . over all , there is a good agreement between @xmath141 from the screening fit and the si - mos values . just at the lowest @xmath15 , the deviation is non - monotonic , but this may be caused by the large deviations in the calculated and measured @xmath71 values as the calculation is performed for @xmath93 . on the other hand , the above screening model @xcite is also limited by the hubbard approach , which does not take into account electronic correlation effects . a further work by gold indicates @xcite that even larger ratios of @xmath142 than four can be explained beyond the random phase approximation by taking into account exchange / correlation ( i.e. many body effects ) and multiple scattering effects . in this framework possibly the data for high - mobility si - mos samples can be explained as well . the fifth and last suggestion ( v ) on the origin of the metallic state in 2d concerns coherent interaction corrections ( cic ) to the conductivity in the ballistic regime . the theory for it was recently introduced by zala , narozhny , and aleiner ( zna ) @xcite . previously , altshuler and aronov have calculated quantum corrections to conductivity ( which include especially the interaction corrections ) in the diffusive regime ( i.e. at low temperatures , where @xmath143 ) and found a typical @xmath144 behavior @xcite . for the intermediate @xmath1 range ( i.e. the ballistic regime , where @xmath145 ) , zna found a linear in @xmath1 contribution . @xcite in addition , they calculated the coherent interaction corrections for the transition regime , where the behavior changes from diffusive to ballistic . in the ballistic regime , the dominant processes can be understood as coherent backscattering of electrons at the friedel oscillations of the other electrons around some perturbation centers.@xcite in contrast , the coherent scattering in the diffusive regime occurs at two or more centers into a wider range of angles . in the approach of zna , where the interaction corrections are calculated in terms of the fermi liquid parameter @xmath146 , correlation ( hartree ) and exchange ( fock ) terms are included . as the fock terms have opposite sign to the hartree terms and the relative strength of the two contributions varies with temperature and electron density , the resulting interaction corrections may be either positive or negative . the relation of the ballistic interaction correction to the former screening approach will be discussed later . we have fitted the temperature dependence of the soi - resistivity by the theory of zna , but adopting the situation for two - fold valley degeneracy in n - type si inversion layers . as was shown by punnoose and finkelstein @xcite , the valley degeneracy leads to 15 spin triplet channels instead of 3 for the one - valley situation . in both situations there is just on charge channel participating . this modification of the zna theory was also taken into account in the evaluation of the resistivity in si - mos by vitkalov et al . @xcite and pudalov et al . @xcite . in order to fit the experimental data @xmath23 , the functions @xmath147 and @xmath148 for the charge and triplet channels , respectively , were calculated according to ref . @xcite and tabulated . the fermi liquid parameter @xmath146 and the conductivity @xmath149 are used as fitting parameters in a true multi - parameter least - square - fit procedure . as in the screening case ( iv ) , the fit is performed for @xmath150 . it can be seen from fig . [ fig : rhotfitzna ] that the fitted curve ( solid line ) describes the experimental values ( circle symbols ) quite well . the inset of fig . [ fig : rhotfitzna ] shows the @xmath146-values as obtained from the fitting procedure . in the density range of 2 to @xmath151@xmath14 the @xmath146 values lie between @xmath152 and @xmath153 . for the lowest density of @xmath154@xmath14 , we find a significantly higher value of @xmath155 . as can be seen from fig . [ fig : rhotfitzna ] , the fit was performed only in a short temperature interval and the fit is not so significant there . on the other hand , one sees also that slop of the @xmath1-dependence is indeed drastically changing and one eventually comes into a different regime of the sample behavior . for even lower electron densities the fit did not give any reasonable results as the sample finally becomes insulting below @xmath156@xmath14 . in our fitting procedure , we use @xmath157 as a fitting parameter . in the work zna @xcite and of vitkalov et al . @xcite , it is suggested to perform a linear extrapolation of the resistivity @xmath23 towards zero @xmath1 . but this has the disadvantage that the other contributions ( e.g.logarithmic ones ) may lead to an offset and the value obtained be the extrapolation towards zero is not the optimal one . as we have included the diffusive ( i.e. logarithmic ) terms in our fit , we find a discrepancy between the fitted value @xmath157 and the one from the linear extrapolation . with our procedure , we get a better consistency for the fit . if we perform our fit on @xmath23 under the assumption of no valley degeneracy , then the resulting @xmath146 values have to be further negative and are typically in the range of @xmath158 to @xmath159 . this comes from the fact that the single charge channel is always localizing whereas the triplet channels contribute the delocalizing parts to the conductivity and the strength of the triplet channels increases towards negative values of @xmath146 . if there is just one valley present , the @xmath146 parameter has to be larger in order to lead to the experimentally observed delocalizing behavior towards low @xmath1 whereas in the case of two valleys with 15 triplet channels , the @xmath146 value can be lower and still leads to the delocalizing behavior . for one valley the border between localizing / delocalizing behavior is given by @xmath160 , for two valleys it is about @xmath161 . from the quality of the fit , it has to mentioned that the screening fit of ( ii ) is nearly identical , the differences are very small . if one calculates the prefactor of the linear term in the zna theory and compares it with @xmath119 from the screening fit , one gets very similar values . this means that that in our case the fock contribution seems to be not very important . the interaction corrections at intermediate temperatures were also calculated for an in - plane magnetic field @xmath72 by zna.@xcite due to the bare zeeman splitting @xmath162 , the triplet states with different total spin component @xmath163 are split and the interference effects ( i.e. the triplet channels ) are partly suppressed . @xcite note that one must not take the renormalized zeeman splitting @xmath164 , as homogeneous collective modes are not renormalized by electron - electron interaction.@xcite we assumed again a two - fold valley degeneracy . in addition to the effective zeeman splitting , a possible valley splitting also contributes to the conductivity corrections . the expression for the different corrections is given by @xcite @xmath165 where the terms correspond in the order of appearance to the charge and triplet channels , to zeeman and valley splitting and to combinations of the latter two . the behavior of the triplet states in parallel field and/or valley splitting is described by the functions @xmath166 and @xmath167 for the ballistic and diffusive case , respectively . @xcite here @xmath168 and @xmath169 is any of the energies given as an argument in @xmath170 of eq . [ eq : dsezdv ] . we have again calculated and tabulated these functions in order to perform a true multi - parameter least - square fit procedure . first we describe a fit for @xmath171 ; the case of @xmath172 is discussed later . for the case of @xmath171 , the conductivity corrections reduce to @xmath173 each @xmath174 is able to freeze two triplet channels @xmath175 ( see eq . 11 in zna @xcite ) , so that in the high magnetic field limit @xmath176 only 7 triplet channels remain . figure [ fig : rhobparfitzna ] presents the result of the fit with @xmath146 and @xmath177 the fitting parameters and @xmath178 . as can be seen from the figure there is a very good agreement for high densities over the entire @xmath72-field rang and also for the lower densities for the low field part . at high parallel magnetic field and low density there is a discrepancy between calculations and experiment . this can be understood , as a limit of applicability of eq . [ eq : dsezdv ] is that @xmath179 . for the lower electron densities in our experiment , we reach the total spin polarization regime and the latter relation is not fulfilled any more . at first we fitted the whole @xmath72-range , but after recognizing the deviations at high fields , we limit the fit range to @xmath180 t as is shown by the full lines in fig . [ fig : rhobparfitzna ] . the extrapolated @xmath71 behavior is indicated by the dashed part of the lines . the inset of fig . [ fig : rhobparfitzna ] depicts the obtained values for the fitting parameter @xmath146 , which are in the range of @xmath181 to @xmath182 for the higher electron densities . for the lower densities , we find a sudden upturn of @xmath183 . it occurs for the same densities , for which there is a large and systematic deviation at high magnetic fields . possibly , we reach a limit of the applicability of the theory . we have also performed a least square fit with the additional fitting parameter @xmath184 for the valley splitting . we get values between @xmath185 and 2k for @xmath184 . these values do not depend systematically on @xmath15 , but are rather scattered around in the before mentioned range the significance of the parameter @xmath184 is very low . if we set @xmath171 , as used for the fit above , the deviation of the fitted curve from the data points is nearly the same and the fitted values of @xmath146 and @xmath177 change only marginally . this means that we do not get a reasonable estimate for @xmath184 from the fit of our magnetoresistivity data for in - plane magnetic field . + figure [ fig : f_0sig_comparison ] shows a comparison of the @xmath146 values as obtained from the fits of the @xmath1- and the @xmath72-dependence . in the electron density range of 2 to @xmath186@xmath14 , the values ( marked by circles and squares ) are close together . this means that two independent measurements and fits lead to very similar fermi liquid parameters @xmath146 . this observation supports the validity of the ballistic interaction theory . for @xmath187@xmath14 , the @xmath146 values suddenly increase strongly . it seems that in this regime the description of the experiment by the interaction corrections in the ballistic regime breaks down . this is in good agreement with the observation in si - mos , that for @xmath187@xmath14 , the strong decrease in @xmath23 towards lower @xmath1 is in the diffusive regime , so that the ballistic regime is active only for higher electron densities.@xcite for comparison , @xmath146 was also calculated and plotted for high - mobility si - mos structures ( solid line ) as obtained @xcite from the effective @xmath138-factor by the relation @xmath188 . the @xmath146 values obtained by the two different methods show the same trend in that they increase towards negative values for decreasing electron density , although the value appears to be more negative for si - mos . this discrepancy is not clear at the moment . if ones compares the fit of our soi magnetoresistance data by the theory on interaction corrections at intermediate temperatures ( zna ) @xcite with the theory on temperature dependent screening ( dolgopolov and gold ) @xcite , it can be seen that the former theory gives a better description of the experimental data at low magnetic fields . in the first case , the nearly parabolic dependence of @xmath71 around @xmath63 is described well , whereas in the second case the fit gives a nearly linear @xmath72-dependence . but as discussed earlier in this work , this is not astonishing , as the theory on screening in parallel fields is worked out so far for @xmath189 , where the screening function has the well defined kink at @xmath190 . the first suggestion ( i ) on a new quantum mechanical ground state in the metallic regime should be discussed once more in the light of screening and coherent electron transport in the ballistic regime . in a recent work @xcite , we concluded that the metallic state in si - mos samples is not caused by quantum coherent effects . we discussed screening as a prominent candidate for the metallic effect . at this time screening was not considered as a quantum coherent effect , primarily . in the new sight of zna , screening can be described by coherent back scattering of electrons at the perturbation centers and it s friedel oscillations . it seems like suddenly the effect has changed it s habits . if we now try to describe the strong resistivity decrease in a combined picture of screening and coherent ballistic contributions ( let s denote it coherent screening regime ) , one should say that the correlation ( fock ) contribution may give an important contribution and was missed in the former screening descriptions @xcite . only in some later work , the correlation contributions were included in the local field contribution @xcite . on the other hand , the effect still reflects the behavior of the screening function with it s sharp kink at @xmath93 which is rounded for @xmath94 . in any case , the coherent screening effects leads to a linear @xmath1 behavior and no new or extraordinary effects will emerge from itself . the important realization is that the coherent conductivity corrections in the diffusive regime can be understood as a natural continuation of the coherent screening . it becomes important as soon as the coherent scattering extends over more than one scattering center at a time ( which is known since many years ) . the diffusive regime describes the weak localization effect and the coherent disorder - induced electron - electron interaction corrections . @xcite only if the latter contributions lead to strong and dominant conductivity corrections , one should speak of a new quantum mechanical ground state . such a scenario was described by punnoose and finkelstein @xcite in the case of two degenerate valleys where the delocalizing effect of the 15 triplet channels can be significantly enhanced due to renormalization effects . but this is not the case for higher electron densities of @xmath191@xmath14 as was shown in the recent work on the borders for quantum effects in the diffusive regime . @xcite only for lower electron densities , such a renormalization effect could significantly increase the delocalizing contributions and lead to a new ground state . in conclusion , we have shown that silicon - on - insulator structures show a very similar 2d - metallic state as si - mos structures with comparable peak mobility . we discussed and compared several models on the physical origin of the metallic state . the theory on coherent interaction corrections at intermediate temperatures is able to describe the observed temperature and in - plane magnetic field behavior over a wide density range . under the assumption of a two - fold valley degeneracy , the fermi - liquid parameter @xmath146 is between @xmath181 and @xmath153 . the derivated fermi - liquid parameters from the temperature and the in - plane magnetic field dependence are in very good agreement to each other . this observation points to the validity of the ballistic interaction theory . we do not get a significant influence from the valley splitting parameter @xmath184 . the earlier screening approach gives also a similarly good description for intermediate temperatures . this indicates that the correlation ( or fock ) terms may not be very significant in the investigated regime . the charged trap state model , as given in the literature , is not able to describe the resistivity versus density behavior further away from the critical density of the metal - to - insulator transition . also the two - band transport model seems to be not applicable for electrons in si - layers . it is not clear yet whether there exists a true metallic state for low temperatures at low electron densities or not . we thank b.l . altshuler , i. aleiner , v.t . dolgopolov , k. ensslin , a. finkelstein , a. gold , b. narozhny , d.l . maslov , and v.m . pudalov for stimulating discussions . the work was supported by the austrian science fund ( fwf ) on the `` metallic state in 2d systems '' . analytic results for the metal - insulator transition in gated semiconductors according to trap states were calculated recently by altshuler and maslov ( am ) @xcite . in this work , am assume that a hole trap exists in the oxide layer , which can easily be charged and discharged . the trap state is homogeneously distributed in space but has a distinct energetical position . some of the equations used by am were not derived in that work according to the length limits of the letter publication and in addition several approximations were used there . we find it therefore useful to discuss the calculations and our numerical results in more detail here . figure [ fig : trap_scheme ] shows a simple scheme of the hole trap states . if the chemical potential @xmath10 ( we use here now the same notation as by am @xcite ) is below the trap state energy @xmath192 ( fig . [ fig : trap_scheme]a ) , a hole is trapped , it is positively charged and acts as an efficient scattering center . for the case that @xmath10 is above @xmath192 , the center is neutral and is inefficient for scattering . for the scattering centers , a dipole approximation is used by am . it is assumed that a charge in the oxide layer will introduce an opposite mirror charge in the 2d layer due to screening on in - plane electric fields . for the energetical position of the trap state , a term @xmath193 has to be taken into account , with @xmath194c@xmath195nm@xmath196 the vacuum permeability , @xmath197 the relative dielectric constant of the oxide and @xmath49 the distance of the charge from the 2d electron layer ; the calculations are performed in si units here . the energetical position of the charged trap states also depends on the applied gate voltage via @xmath198 , with @xmath199 the thickness of the oxide layer . figure [ fig : trap_energy ] depicts the dependence of the total energy of the charged trap states in the dipole approximation as a function of the distance @xmath49 for different gate voltages @xmath16 . only in the spacial region where @xmath200 , the centers are positively charged . the expression for the resistivity @xmath89 follows in the drude - boltzmann approximation by calculating the effective transport scattering time @xmath201 for the conductivity @xmath202 . the weighting of @xmath203 with the kinetic electron energy @xmath204 in eq . [ eq : tau ] fundamentally follows from the drude - boltzmann approximation @xcite , as the fermi velocity @xmath205 and the shift of the fermi surface in k - space are both proportional to @xmath206 , which enter into the expression for the current @xmath207 . the integral in the denominator of eq . [ eq : tau ] is in 2d just equal to the fermi energy @xmath46 also for elevated temperatures . the transport scattering time @xmath208 has to be calculated from the scattering rate @xmath209 with @xmath210 the three dimensional density of charged traps , @xmath211 the electron velocity and @xmath212 the cattering ross section for omentum relaxation . according to am , the scattering cross section for a dipol is given by @xmath213 . the effective relative permitivity @xmath214 is the mean value of the si and the sio@xmath22 layer . the density @xmath215 in the am notation with @xmath216 a constant density of traps per volume and @xmath217^{-1}$ ] , the thermodynamic probability for a trap state to be charged@xcite . by inserting the above expressions into eq . [ eq : scat_rate ] one gets @xmath218 with the factor @xmath219 , an effective number of positive trap states per area @xmath220 , the effective width of positive charge layer @xmath221 and the position @xmath222 of the energetical maximum of the trap energy @xcite . by inserting eq.[eq : scat_rate2 ] into eq . [ eq : tau ] , one gets @xmath223 for the drude resistivity @xmath224 , an effective energy @xmath225 can be defined so that a relation equivalent to eq . [ eq : scat_rate2 ] can be written for the effective values , i.e. @xmath226 . a simple calculation gives @xmath227^{-6 } . \label{eq : eeff}\ ] ] the first derivative of the fermi - dirac function @xmath228 can be expressed as @xmath229 and one obtains the same expression as eq.8 in ref . @xcite with these relations , the resistivity can exactly be written in terms of the effective energy @xmath225 as @xmath230 we have calculate the resistivity @xmath231 by numerical integration for the effective thickness @xmath232 of the layer with charged trap states and the effective energy @xmath225 . in the work of am @xcite , a parabolic approximation was used for the dependence of the trap energy @xmath192 on @xmath49 and an analytic expansion was given for the behavior of @xmath231 . with this , am were able to show that in principle a critical behavior in the density dependence of @xmath23 can appear . as in our work , we want to test whether the metallic behavior can be explained for a wide density range or not , we use the numerical calculation method in order to be able to perform a precise evaluation of the model behavior for large deviations of @xmath15 from the critical density @xmath0 . am proposed in their model that the dependence of @xmath233 in the vicinity of @xmath0 comes from the strong change of the number of positively charged centers , as the fermi energy is just above or below the energy maximum @xmath234 of the trap states , where they have a singularity in the density of states . the sensitivity on temperature comes from the broadening of the thermodynamic probability distribution @xmath235 with increasing temperature . in order to test this behavior for the strong temperature dependence of the resistivity , i.e. the most characteristic feature of the metallic state , we calculated the temperature dependence within the model at an intermediate temperature of @xmath236k . for densities @xmath237@xmath14 , @xmath23 does not saturate yet at its low-@xmath1 value . the saturation behavior is not described by the am - model . the low-@xmath1 value has to be cause by other mechanisms like residual impurities which do not change their charge state in the same manner . for the evaluated value of @xmath238k , the @xmath1-dependence should behave according to the am - model if it is able to describe the metallic state . the result of our numerical calculation is shown in fig.[fig : rho_trap_soi ] . it is clearly visible there , that the calculated @xmath89-dependence changes by many more orders of magnitude than the measured one . in order to analyze the origin of the strong change in the calculated @xmath233 behavior , we show in fig . [ fig : n+eff(n ) ] the dependence of the effective number of charged trap states per area @xmath239 on the electron density ( i.e. the gate voltage @xmath16 ) . as long as @xmath240 a part of the trap states is positively charged even for @xmath93 and the variation in effective positive charge layer width @xmath232 is small . but for @xmath241 , the trap states are only charged according to the exponentially decaying part of the thermodynamic probability function @xmath242 and thus the strong exponential @xmath1-dependence occurs . in the same density range , the other values which enter into the resistivity calculation ( [ eq : rho(e ) ] ) change only by about a factor of two . it is thus clear that within the proposed am model , the strong shift of the energetical position of the trap states with the applied gate voltages leads to too strong changes in @xmath233 further away from the critical density @xmath0 . kravchenko , et al . b * 50 * , 8039 ( 1994 ) ; phys . b * 51 * , 7038 ( 1995 ) . e. abrahams , p. w. anderson , d. c. licciardello , and t. v. ramakrishnan , phys . * 42 * , 673 ( 1979 ) . d. popovic , a.b . fowler , and s. washburn , phys . lett . * 79 * , 1543 ( 1997 ) . e. abrahams , s.v . kravchenko , and m.p . sarachik , rev . * 73 * , 251 ( 2001 ) . g. zala , b.n . narozhny , and i.l . aleiner , phys . b * 64 * , 214204 ( 2001 ) . g. zala , b.n . narozhny , and i.l . aleiner , phys . b * 65 * , 020201(r ) ( 2001 ) . a. gold and v.t . dolgopolov , phys . b * 33 * , 1076 ( 1986 ) . m. bruel , mrs bull . * 23 * , 35 ( 1998 ) . pudalov , g. brunthaler , a. prinz , and g. bauer , cond - mat/0103087 . a. gold , phys . lett . * 54 * , 1079 ( 1985 ) . g. brunthaler , a. prinz , g. bauer , and v.m . pudalov , phys . * 87 * , 096802 ( 2001 ) pudalov , g. brunthaler , a. prinz , and g. bauer , physica e * 3 * , 79 ( 1998 ) . pudalov , g. brunthaler , a. prinz , and g. bauer , jetp lett . * 65 * , 932 ( 1997 ) . d. simonian , s.v . kravchenko , m.p . sarachik , and v.m . pudalov , phys . * 79 * , 2304 ( 1997 ) . t. okamoto , k. hosoya s. kawaji , and a. yagi , phys . * 82 * , 3875 ( 1999 ) . j. yoon et al . , phys . rev . lett . * 84 * , 4421 ( 2000 ) . a.m. finkelstein , z. phys . b : condens . matter * 56 * , 189 ( 1984 ) . c. castellani , c. di castro , p.a . lee , and m. ma , phys . b * 30 * , 527 ( 1984 ) . e. abrahams and g. kotliar , science * 274 * , 1853 ( 1996 ) . v. dobrosavljevi , e. abrahams , e. mirinda , and s. chakravarty , phys . * 79 * , 455 ( 1997 ) . a. punnoose and a.m. finkelstein , phys . * 88 * , 016802 ( 2002 ) . f. stern , phys . lett . * 44 * , 1469 ( 1980 ) . s. das sarma , phys . b * 33 * , 5401 ( 1986 ) . a. gold , phys . b * 35 * , 723 ( 1987 ) . a. gold , phys . b * 38 * , 10798 ( 1988 ) , and phys . b * 41 * , 8537 ( 1990 ) . s. das sarma and e.h . hwang , phys . lett . * 83 * , 164 ( 1999 ) . dolgopolov and a. gold , jetp lett.*71 * , 27 ( 2000 ) . pudalov et al . , phys . lett . * 88 * , 196404 ( 2002 ) . a.gold , jetp lett.*72 * , 401 ( 2000 ) . b.l . altshuler and d.l . maslov , phys . lett . * 82 * , 145 ( 1999 ) . altshuler and a.g . aronov , in _ electron - electron interaction in disordered systems _ , edited by a.l . efros and m. pollak , north holland , amsterdam , 1985 , p. `` _ _ physics of semiconductor devices _ _ '' , s.m . sze , john wiley & sons , new york 1981 , p. 379 . t. ando , a.b . fowler , and f. stern , rev . phys . * 54 * , 437 ( 1982 ) , see p. 478 . pudalov , jetp lett . * 66 * , 175 ( 1997 ) . p. pfeffer and w. zawadzki , phys . b * 59 * , r5312 ( 1999 ) . w. zawadzki and p. pfeffer , phys . b * 64 * , 235313 ( 2001 ) . pudalov , g. brunthaler , a. prinz , and g. bauer , phys . lett . * 88 * , 076401 ( 2002 ) ; see ref . 26 therein . papadakis et al . , science * 283 * , 2056 ( 1999 ) . r. winkler , s.j . papadakis , e.p . de poortere , and m. shayegan , phys . lett . * 84 * , 713 ( 2000 ) . y. yaish et al . , phys . lett . * 84 * , 4954 ( 2000 ) . `` _ _ semiconductors _ _ '' , r.a . smith , cambridge university press , cambridge 1964 . s.a . vitkalov et al . , cond - mat/0204566 . pudalov et al . , cond - mat/0205449 . a. gold , j. phys . : condens . matter * 13 * , 11641 ( 2001 ) .

we define a hierarchical clustering method :-unchaining single linkage or . the input of this algorithm is a finite metric space and a certain parameter . this method is sensitive to the density of the distribution and offers some solution to the so called chaining effect . we also define a modified version , , to treat the chaining through points or small blocks . we study the theoretical properties of these methods and offer some theoretical background for the treatment of chaining effects . keywords : hierarchical clustering , single linkage , chaining effect , weakly unchaining ,-bridge - unchaining . msc : 62h30 , 68t10 .

1210.6292

a _ clustering method _ is an algorithm that takes as input a finite space with a distance function ( typically , a finite metric space ) @xmath3 and gives as output a partition of @xmath4 . kleinberg discussed in @xcite the problem of clustering in an axiomatic way and proposed a few basic properties that a clustering scheme should hold . then , he proved that no standard clustering scheme satisfying this conditions simultaneously can exist . this does not imply the impossibility of defining a consistent standard clustering algorithm . kleinberg s impossibility only holds when the unique input in the algorithm is the space and the set of distances . it can be avoided including , for example , the number of clusters to be obtained as part of the input . see @xcite and @xcite . also , ackerman and ben - david , see @xcite , showed that these axioms are consistent in the setting of clustering quality measures . carlsson and mmoli , see @xcite , studied the analogous problem for clustering schemes that yield hierarchical decompositions instead of a certain partition of the space . see also @xcite and @xcite . hierarchical clustering methods also take as input a finite metric space but the output is a hierarchical family of partitions of @xmath4 . they approach the subject focusing on a theoretical basis for the study of hierarchical clustering ( @xmath5 ) . in the spirit of kleinberg s result , they define a few reasonable conditions that a @xmath5 method should hold . they prove that the unique @xmath5 method satisfying three basic conditions is ( the well - known ) single linkage hierarchical clustering , @xmath6 @xmath5 . ackerman and ben - david , see @xcite , proved also a characterization of the class of linkage - based algorithms , including @xmath6 . see also @xcite . in the setting of partitional ( standard ) clustering see @xcite . carlsson and mmoli also study the theoretical properties of @xmath6 @xmath5 obtaining some interesting results . the main advantage seems to be that this method enjoys some sort of stability which is defined by means of the gromov - hausdorff distance . however , the main weakness of @xmath6 @xmath5 is the so called _ chaining effect _ which may merge clusters that , in practice , _ should be detected _ by the algorithm and kept separated . one way to address this difficulty is to take into account the density . in a preprint entitled _ multiparameter hierarchical clustering methods _ the same authors do this by including in the input of the algorithm a function that provides that information . our aim is to define a @xmath5 method which offers some solution to this particular weakness without including any extra information . the first challenge is that the concept of _ clusters that should be detected _ by the algorithm depends on the characteristics of the problem under study . the same happens with what we may consider the _ undesired _ chaining effect . the definition from lance and williams , @xcite , makes reference to the higher tendency of the points to add to a pre - existing group rather than defining the nucleus of a new group or joining to another single point . our algorithm is oriented to another aspect of the chaining effect which is the tendency to merge two clusters when the minimal distance between them is small ( even though they may have dense cores which are clearly distant apart ) . this is typically the problem of @xmath6 @xmath5 . also , we include as an undesired chaining effect the case of two big clusters joined by a chain of points or small clusters . these isolated points or small clusters may be interpreted as noise in the sample and we might want to distinguish the big picture and ignore their effect . this idea is closer to the type of chaining effect considered by wishart in @xcite . there exist other linkage - based methods that enjoy some sort of sensitivity to density and offer some resistance to these chaining effects as average linkage , @xmath7 , or complete linkage , @xmath8 . these methods are extensively used in practice . however , although the main problem of the chaining effect of @xmath6 @xmath5 is reduced , there appears another effect that might be unwanted too . in these methods the distance between a point and a cluster is greater than the minimal distance . therefore , they have a tendency to merge isolated points before joining them to pre - existing big clusters . also , these methods are proved to be extremely unstable in the sense that small perturbations on the data yield very different dendrograms . herein , we define a new @xmath5 method on the basis of @xmath6 : _ @xmath0-unchaining single linkage _ or @xmath1 . the definition of @xmath1 is based in the dimension of the rips complexes defined by the points of @xmath4 . these complexes contain some information about the density distribution of the sample . this allows us to define a density sensitive algorithm such that the input is just the set of distances between the points and a fixed parameter @xmath9 . the parameter determines how sensitive the method will be to the chaining effect . to treat the chaining through single points or smaller blocks , we define another version of the method , @xmath2 , by adding an extra condition on @xmath1 . it is worth mentioning that ester , kriegel , sander and xu , see @xcite , defined a standard clustering method called dbscan . dbscan is also a density - sensitive algorithm where the input is just the data set and some parameters . although this is not hierarchical clustering algorithm we include a short discussion comparing the type of clusterings detected by this method with the clusterings that @xmath1 and @xmath2 may detect in the levels of the dendrogram this paper intends to give a theoretical basis to the study of the problem . so , instead of checking the algorithm on examples of real data we rather try to find general properties characterizing what would be an undesired chaining of two blocks and how good is the algorithm detecting and unchaining them . however , we include several examples where the unchaining properties can be explicitly seen in the resulting dendrogram . to the study the chaining effect we define the concepts of _ chained subsets _ and _ subsets chained by smaller blocks _ as situations of minimal chaining so that they contain what we consider the problematic examples . nevertheless , there may be many examples of chained subsets which should be clearly merged and there is margin to be more restrictive . in such context , a @xmath5 method is _ strongly chaining _ if every pair of chained subsets are always merged before they appear contained in different clusters . a @xmath5 method is _ completely chaining _ if , in addition , every pair of subsets chained through smaller blocks are merged before they appear contained in different clusters . thus , strongly chaining methods and completely chaining methods are extremely sensitive to these effects . this is the case , for example , of @xmath6 @xmath5 . see theorem [ th : complete chaining ] . we define also precise conditions to define what we consider two blocks that should necessarily appear as independent blocks at some point . the definition considers two blocks that have dense cores and such that the minimal distance between them is small only because of a single pair of points . in particular , this pair of points creates a chaining between their dense cores . see figure [ ejp chained edge ] . this is a particular , more restrictive , example of chained subsets . we say that a @xmath5 method is _ weakly unchaining _ if , at least , it distinguishes that pair of blocks . then , we prove that @xmath1 satisfies this condition while other methods which are not strongly chaining as @xmath7 and @xmath8 @xmath5 fail to be weakly unchaining . see theorem [ th : weakly ] , corollary [ cor : weakly ] and example [ ejp : al - cl ] . we also define a minimal condition of two subsets chained by single points that should be detected . we say that a @xmath5 method is _ @xmath0-bridge - unchaining _ if it is able to separate two blocks in that situation . @xmath2 is proved to be more sensitive than that . it also detects some classes of chaining through smaller blocks as it is proved in proposition [ prop : moderate_2 ] . in particular , @xmath2 is @xmath0-bridge - unchaining . see corollary [ cor : unchaining ] . the structure of the paper is the following : section [ section : background ] contains the basic definitions and notation involved . it may be skipped by the experts . in section [ section : hc ] we recall some well known hierachical clustering methods and include some different ways to formulate them . in section [ section : modified sl ] we first introduce the idea of what are we considering undesired chaining effect . formal definitions are left to the last sections to enhance readability . in this section we are just trying to give the reader some notion of what is @xmath1 trying to detect . then , we present @xmath1 . we include a short explanation of the role of each step of the method and check it on a few examples . section [ section : sl^ * ] deals with the problem of chaining through smaller blocks . again , we leave formal definitions for the last sections and we only discuss the intuitive idea . then , we introduce a further step in the algorithm to define @xmath2 and check it on some examples . section [ section : unchaining properties ] studies the unchaining properties of @xmath1 . first we fix the theoretical background to study the chaining effect . we define the concepts of _ chained subsets _ and subsets _ chained by a single edge_. we define the property of being _ strongly chaining _ for @xmath5 methods which are extremely sensitive to the chaining effect . we prove that @xmath6 is strongly chaining while @xmath7 and @xmath8 @xmath5 are not . then , we say that a @xmath5 method is _ weakly unchaining _ if it is at least capable of detecting a certain clustering when the subsets are chained by a single edge . we prove that @xmath1 is weakly unchaining while other methods , partially sensitive to the chaining effect as @xmath7 and @xmath8 , are not . we also compare the results obtained by our method and the results offered by the standard clustering algorithm dbscan . section [ section : unchaining properties 2 ] studies the unchaining properties of @xmath2 . we define the concept of subsets _ chained through smaller blocks_. we say that a @xmath5 method is _ completely chaining _ if it is strongly chaining and also unable to detect clusters if they are chained through smaller blocks . we prove that @xmath6 is completely chaining . then , we define the property of being _ @xmath0-bridge - unchaining _ for algorithms which are able to detect , at least , some type of subsets chained through single points . we prove that @xmath2 is @xmath0-bridge - unchaining . we compare this case , also , with the treatment of the same input by dbscan . section [ section : conclusions ] includes a short discussion about the main advantages of the methods defined and some comments about future research . a dendrogram over a finite set is a nested family of partitions . this is usually represented as a rooted tree . let @xmath10 denote the collection of all partitions of a finite set @xmath11 . then , a dendrogram can also be described as a map @xmath12 such that : * @xmath13 , * there exists @xmath14 such that @xmath15 for every @xmath16 , * if @xmath17 then @xmath18 refines @xmath19 , * for all @xmath20 there exists @xmath21 such that @xmath22 for @xmath23 $ ] . notice that conditions 2 and 4 imply that there exist @xmath24 such that @xmath25 for every @xmath26 , @xmath27 and @xmath28 for every @xmath29 . for any partition @xmath30 , the subsets @xmath31 are called _ blocks_. let @xmath32 denote the collection of all possible dendrograms over a finite set @xmath4 . given some @xmath33 , let us denote @xmath34 . therefore , the nested family of partitions is given by the corresponding partitions at @xmath35 , this is , @xmath36 , @xmath37 . an _ ultrametric space _ is a metric space @xmath3 such that @xmath38 for all @xmath39 . given a finite metric space @xmath4 let @xmath40 denote the set of all ultrametrics over @xmath4 . there is a well known equivalence between trees and ultrametrics . see @xcite and @xcite for a complete exposition of how to build categorical equivalences between them . in particular , this may be translated into an equivalence between dendrograms and ultrametrics : thus , a hierarchical clustering method @xmath41 can be presented as an algorithm whose output is a a dendrogram or an ultrametric space . let @xmath42 denote the dendrogram obtained by applying @xmath41 to a metric space @xmath3 and @xmath43 denote the corresponding ultrametric space . let us define the map @xmath44 as follows : given a dendrogram @xmath45 , let @xmath46 be such that @xmath47 . * theorem 9 ) @xmath48 is a bijection such that @xmath49 . * notation * : for any @xmath5 method @xmath41 and any finite metric space @xmath3 , let us denote @xmath50 and @xmath51 . if there is no need to distinguish the metric space we shall just write @xmath52 and @xmath53 . let us recall the definition of some well - known hierarchical clustering methods . we include here the description of _ single linkage _ based on its @xmath54-connected components . also the recursive description of _ single linkage _ , _ complete linkage _ and _ average linkage _ as presented in @xcite . we introduce also an alternative description of these methods , based in the recursive one . in our description we define a graph , @xmath55 , which will be the key to build our new method , @xmath1 . we think that this approach might be useful to define other algorithms which might be better adapted to other specific problems . an @xmath56-_chain _ is a finite sequence of points @xmath57 that are separated by distances less or equal than @xmath56 : @xmath58 . two points are @xmath56-_connected _ if there is an @xmath56-chain joining them . any two points in an @xmath56-_connected set _ can be linked by an @xmath56-chain . an @xmath56-_component _ is a maximal @xmath56-connected subset . is the partition of @xmath4 in its @xmath56-components . ] clearly , given a metric space and any @xmath59 , there is a partition of @xmath4 in its @xmath56-components @xmath60 . let @xmath4 be a finite metric set . the _ single linkage _ @xmath5 is defined by the map @xmath61 such that @xmath62 is the partition of @xmath4 in its @xmath54-components . see figure [ ejp epsilon - chains ] . for other uses of @xmath56-connectedness on computational topology see @xcite and @xcite . in @xcite there is also an alternative formulation of @xmath6 @xmath5 . in fact , the authors use a recursive procedure to redefine @xmath6 @xmath5 , average linkage ( @xmath7 ) and complete linkage ( @xmath8 ) hierarchical clustering . the main advantage of this procedure is that it allows to merge more than two clusters at the same time . therefore , @xmath7 and @xmath8 @xmath5 can be made _ permutation invariant _ , meaning that the result of the hierarchical clustering does not depend on the order in which the points are introduced in the algorithm . we reproduce here , for completeness , their formulation . let @xmath3 be a finite metric space where @xmath11 and let @xmath63 denote a family of linkage functions on @xmath4 : @xmath64 where @xmath65 denotes the collection of all non - empty subsets of @xmath4 . some standard choices for @xmath66 are : * single linkage : @xmath67 * complete linkage : @xmath68 * average linkage : @xmath69 where @xmath70 denotes the cardinality of the set @xmath4 . fix some linkage function @xmath71 . then , the recursive formulation is as follows * for each @xmath72 consider the equivalence relation @xmath73 on blocks of a partition @xmath74 , given by @xmath75 if and only if there is a sequence of blocks @xmath76 in @xmath77 with @xmath78 for @xmath79 . * consider the sequences @xmath80 and @xmath81 given by @xmath82 , @xmath83 , and recursively for @xmath84 by @xmath85 where @xmath86 until @xmath87 . * finally , let @xmath88 be such that @xmath89 with @xmath90 . [ remark : components ] we can also reformulate the recursive algorithm as follows . * let @xmath83 and @xmath82 . * for every @xmath84 , while @xmath91 , let @xmath86 . then , let @xmath92 be a graph whose vertices are the blocks of @xmath93 and such that there is an edge joining @xmath94 and @xmath95 if and only if @xmath96 . * consider the equivalence relation @xmath97 if and only if @xmath98 are in the same connected component of @xmath55 . then , @xmath85 . * finally , let @xmath88 be such that @xmath89 with @xmath90 . this formulation allows us to consider other properties of the graph @xmath55 , along with the connected components , to define the equivalence relation @xmath97 . this means that we may introduce further conditions to merge the blocks and , this way , reduce some indesired results as the chaining effect . see also @xcite . the chaining effect is usually mentioned as one of the problems to solve in clustering . however , there are different approaches to define `` chaining effects '' . in @xcite , the authors refer to the chaining effect from @xcite which is the one defined by williams , lambert and lance in @xcite . this version of the `` chaining effect '' takes account on the tendency of a group to merge with single points or small groups rather than with other groups of comparable size . thus , williams , lambert and lance study and measure it by comparing the cardinality of the groups . herein , we are focusing on another aspect . we want to deal with the tendency of two clusters to be merged when the minimal distance between them is small independently of their distribution . this can be a problem when the clusters have dense cores distant apart . see figure [ ejp chained edge ] . this is , typically , the chaining effect one finds in @xmath6 @xmath5 . and @xmath99 is @xmath56 . the clustering @xmath100 would not be detected by @xmath6 @xmath5 . ] given a finite metric space @xmath3 , let @xmath101 be the rips ( or vietoris - rips ) complex of @xmath3 . let us recall that the rips complex of a metric space @xmath3 is a simplicial complex whose vertices are the points of @xmath4 and @xmath102 $ ] is a simplex of @xmath101 if and only if @xmath103 for every @xmath104 . given any subset @xmath105 , by @xmath106 we refer to the subcomplex of @xmath107 defined by the vertices in @xmath108 . a simplex @xmath102 $ ] has dimension @xmath109 . the dimension of a simplicial complex is the maximal dimension of its simplices . notice that densely packed points produce high - dimensional simplices in the rips complex . is detected by @xmath1 for @xmath110 . ] [ ejp chain1 ] let @xmath3 be the graph from figure [ ejp chain_1b ] . the edges in @xmath111 have length 1 and the rest have length 3 . the distances between vertices are measured as the minimal length of a path joining them . consider @xmath112 . then , the vertices of @xmath113 and @xmath114 define 3-dimensional simplices . for any vertex @xmath115 , there is no vertex @xmath116 with @xmath117 . therefore , they define 0-dimensional simplices and they are not part of any 1-dimensional simplex of @xmath112 . we define a modified single linkage hierarchical clustering method , @xmath1 , on the basis of @xmath6 introducing a parameter @xmath9 . this method allows us to take into account density without having to provide any additional input to the algorithm apart from @xmath0 and the distances between the points . let @xmath3 be a finite metric space with @xmath11 . notice that in section [ section : hc ] the recursive definition of @xmath6 , @xmath8 and @xmath7 used the distances , @xmath118 , between the blocks from the previous step . for technical reasons , to define our method we need to use the ordered set of distances in the data set , @xmath119 . let @xmath120 and @xmath121 with @xmath122 @xmath123 where `` < '' denotes the order of the real numbers . clearly , @xmath124 . let the dendrogram defined by @xmath1 , @xmath125 or simply @xmath126 , be as follows : * let @xmath127 and @xmath128 @xmath129 . now , for every @xmath130 , given @xmath131 , we define recursively @xmath126 on the interval @xmath132 as follows : * let @xmath133 be a graph with vertices @xmath134 and edges @xmath135 such that the following conditions hold : * * @xmath136 . * * there is a simplex @xmath137 such that @xmath138 , @xmath139 and @xmath140 . * let us define a relation , @xmath141 as follows . + let @xmath142 if @xmath143 belong to the same connected component of the graph @xmath133 . then , @xmath141 induces an equivalence relation . * for every @xmath144 , @xmath145 . condition @xmath146 is the condition used in @xmath6 @xmath5 to define the graph . see remark [ remark : components ] . condition @xmath147 is used to account for the chaining effect between two adjacent blocks . suppose we have two adjacent blocks , densely packed , which are close to each other as sets but whose dense cores are distant apart as in figure [ ejp chained edge ] . then , the dense cores will produce high dimensional simplices in the rips complex while the connection between the blocks might be a low dimensional simplex . in this case , condition @xmath147 will not be satisfied and the edge between the corresponding blocks in @xmath133 is not defined ( although condition @xmath146 holds ) . hence , these blocks will not be merged . [ remark : necessary chain ] notice that if two points @xmath148 belong to the same block of @xmath149 then , necessarily , there exists a @xmath150-chain , @xmath151 joining them . in particular , if @xmath152 , @xmath153 , the corresponding edges @xmath154 , @xmath155 , satisfies condition @xmath147 . this is immediate by construction . for the graph in figure [ ejp chain_1b ] compared with the corresponding dendrogram obtained applying @xmath6 or @xmath1 with @xmath156 [ ejp : unchain 1 ] let @xmath3 be the graph from figure [ ejp chain_1b ] and let @xmath157 . notice that there are eight @xmath158-components , six of them are singletons and two of them , @xmath113 , @xmath114 , with @xmath159 . furthermore , @xmath160 , @xmath161 and @xmath162 for @xmath163 . then , let us check that applying @xmath164 on @xmath3 the clustering @xmath100 is detected . let @xmath165 . @xmath166 if @xmath167 . it @xmath168 , @xmath169 . for @xmath170 , there are edges in @xmath171 between every pair of clusters in @xmath172 . similarly , there are edges between every pair of clusters @xmath173 . @xmath174 and @xmath175 have dimension 3 while the unique simplex in @xmath176 intersecting both @xmath111 has dimension 1 . therefore , condition @xmath147 induces a separation of the blocks @xmath177 . thus , @xmath178 . for @xmath179 , @xmath180 while the maximal dimension of a simplex intersecting both clusters is @xmath181 . then , by @xmath147 , there is no edge in @xmath182 joining @xmath183 . for @xmath184 , @xmath185 . since the diameter of @xmath186 is 5 , these vertices define a simplex @xmath187 in @xmath188 such that @xmath189 . clearly , this simplex intersects @xmath190 and @xmath99 . hence , @xmath191 . therefore , the dendrogram obtained , @xmath192 , is the one from figure [ chain_dendrogram_1 ] . modifying the parameter @xmath0 we can adjust the method to be more or less sensitive to the chaining effect . increasing @xmath193 we would need higher dimensions in @xmath194 to unchain blocks by condition @xmath147 . suppose @xmath195 . in this case , @xmath183 would be joined by an edge in @xmath196 . thus , for @xmath195 , @xmath197 if @xmath167 , @xmath198 if @xmath199 and @xmath200 if @xmath201 . . ] and @xmath126 with @xmath202 for example [ ejp sl ] . ] [ ejp : unchain 2 ] consider the set represented in figure [ ejp sl ] . let us consider three distances , @xmath203 which are represented , respectively , by a short segment , a dots line and a thick long segment . let us assume that the sets @xmath204 are @xmath205-connected and that @xmath206 , @xmath207 . also , we can see that there exist 3-dimensional simplices in @xmath208 inside @xmath190 , @xmath99 and @xmath209 . in figure [ ejp sl(1 ) ] , we represent the corresponding dendrograms for @xmath164 and @xmath6 . ( notice that , in this case , @xmath210 for any @xmath211 . ) it is clear that , @xmath6 @xmath5 generates a dendrogram where it is impossible to detect the clustering @xmath212 because of the chaining effect . introducing the parameter @xmath213 , in this example , we obtain a hierarchical clustering which is consistent with the distribution of the sample . is detected by @xmath214 . ] [ ejp : chained_edge_2 ] let @xmath4 be the set from figure [ ejp chained_edge_2 ] . suppose @xmath215 and let us see what happens in the application of the algorithm @xmath214 until @xmath216 . in the second square of the figure we can see the 1-dimensional skeleton of the rips complex @xmath217 . as it is shown in the figure , there is no edge joining @xmath190 and @xmath99 yet and the dense cores inside @xmath183 produce high dimensional simplices in @xmath217 . condition @xmath147 has not been applied yet and , therefore , the blocks are just the @xmath218-components . let us call @xmath113 , @xmath114 the corresponding blocks in @xmath219 defined by the nontrivial @xmath218components . for @xmath220 there is a single edge joining @xmath113 and @xmath114 as we can see in the third square . in particular , @xmath221 . however , the dimension of @xmath222 , @xmath163 is greater that 3 . then , by condition @xmath147 , there is no edge in the graph @xmath223 joining @xmath113 and @xmath114 and @xmath224 refines the clustering @xmath100 . in fact , the clustering from @xmath225 is given by the connected components in the third square of the figure when the edge of length @xmath56 joining @xmath183 is eliminated . let @xmath226 . in the example , by condition @xmath147 , for every @xmath227 and for any pair of blocks @xmath228 with @xmath229 and @xmath230 , there is no edge between them in @xmath231 . thus , @xmath232 are not merged in @xmath233 . therefore , @xmath234 . it may be noticed that @xmath1 does not detect the possible clustering @xmath100 in the graph represented in figure [ ejp chain_1 ] . see example [ ejp : 1 - 1-chained ] below . this illustrates the fact that our method does not consider directly the distribution of the points for a certain @xmath150 . instead of that , it focuses on the relations between the blocks from @xmath235 . have no dense cores . @xmath1 does not separate them . ] [ ejp : 1 - 1-chained ] let @xmath236 be the graph from figure [ ejp chain_1 ] where every edge has length @xmath158 . let us fix @xmath157 . if @xmath167 , @xmath237 . if @xmath238 , in @xmath147 we consider the dimension of the complexes defined from the blocks in @xmath239 which are singletons . therefore , every edge in the graph defines an edge in @xmath240 . since @xmath240 is connected @xmath241 . for @xmath242 also and @xmath99 are chained through the smaller block @xmath244 . ] the method @xmath1 is defined to prevent two adjacent blocks with dense cores to be chained too soon when the minimal distance between them is small . thus , condition @xmath147 considers the dimension of the rips complex restricted to both blocks . however , any cluster @xmath245 and any isolated point @xmath246 such that @xmath247 are going to be merged in @xmath248 . therefore , if two clusters are at a certain distance @xmath56 from a single point then both blocks will be merged with this point in @xmath249 . consequently , those clusters will be merged together in @xmath249 . consider the graph represented in figure [ ejp chain_2b ] . suppose that the edges in @xmath113,@xmath114 have length @xmath158 , the edges @xmath250 and @xmath251 have length @xmath252 and the rest have length @xmath253 . as we can see in the picture , there are 9 @xmath158-components , two of them , @xmath113 and @xmath114 , have four points and the rest are singletons . the whole space is @xmath253-connected . let us fix any @xmath254 . it is trivial to check that , @xmath255 . now , for @xmath256 , since @xmath244 is a single point and @xmath257 , condition @xmath147 is trivially satisfied . therefore , there exist edges in @xmath258 between @xmath113 and @xmath259 and between @xmath259 and @xmath113 . hence , @xmath260 . similarly , since every block in @xmath261 except from @xmath262 is a single point , condition @xmath147 always holds and @xmath263 . in general , for any pair of clusters @xmath264 such that @xmath265 and @xmath266 the dimension of the rips complex restricted to @xmath99 is at most @xmath267 . therefore , condition @xmath147 will not apply and the clusters will be merged . thus , for any chain of clusters @xmath268 such that @xmath269 for every @xmath270 and @xmath271 for every @xmath272 , @xmath268 are merged together in @xmath249 . we call this chaining effect : * chaining through smaller blocks*. now , we are going to modify the algorithm so that it may distinguish the case when to blocks are chained by isolated points or small blocks . these points or small blocks might be considered as noise in the sample . see , for example , the point @xmath259 in figure [ ejp chain_2b ] . to treat this effect we are going to focus in the `` big '' blocks . the selection is done depending on the parameter @xmath0 ( which defines the sensitivity of the whole method to chaining ) and on the cardinality of the blocks involved . we use ( [ big blocks ] ) to fix the distinction between big blocks and small blocks . let @xmath11 . let @xmath273 be the ordered set of distances between points of @xmath4 . let the dendrogram defined by @xmath2 , @xmath274 or simply @xmath275 , be as follows : * let @xmath276 and @xmath277 @xmath129 . + now , given @xmath278 , we define recursively @xmath275 on the interval @xmath132 as follows : * let @xmath133 be a graph with vertices @xmath134 and edges @xmath135 such that the following conditions hold : * * @xmath136 . * * there is a simplex @xmath137 such that @xmath138 , @xmath139 and @xmath140 . + by an abuse of the notation , we may write @xmath94 to refer both to the block of @xmath235 and to the vertex of @xmath133 . * let us define a relation , @xmath141 between the blocks as follows . + let @xmath279 be the set of connected components of the graph @xmath133 . let @xmath280 with @xmath281 . + let us call * big blocks * of @xmath282 those blocks such that @xmath283 + the rest of blocks of @xmath282 are called * small blocks*. + let @xmath284 be the subgraph of @xmath282 whose vertices are the big blocks and @xmath285 be the subgraph of @xmath282 whose vertices are the small blocks . + then , @xmath286 if one of the following conditions holds : * * @xmath287 such that @xmath288 . * * @xmath289 , @xmath290 and there is no big block in @xmath291 adjacent to any block in @xmath292 . + then , @xmath141 induces an equivalence relation whose classes are contained in the connected components of @xmath133 . * for every @xmath144 , @xmath293 . step @xmath294 and @xmath295 are the same as in @xmath1 . by @xmath296 , if two big blocks , @xmath98 , are joined by and edge in @xmath133 , then @xmath297 . thus , the connected components of big blocks are merged . by @xmath298 , a connected component of small blocks @xmath292 is merged with a component of big blocks @xmath299 if @xmath299 is the unique component of big blocks adjacent to @xmath292 . otherwise , the blocks of @xmath292 stay as separated blocks in @xmath300 . this can be seen also as follows . by @xmath298 , if a small block is connected by chains of small blocks to two different components of big blocks we will consider it as a block apart in @xmath301 . see example [ ejp : big_small ] . with three connected components of big blocks , @xmath302 , and six small blocks @xmath303 . ] [ ejp : big_small ] suppose @xmath280 is as represented in figure [ big_small ] : @xmath284 has three connected components , @xmath302 , @xmath304 and @xmath305 consists of six small blocks @xmath303 , @xmath306 . the components @xmath307 are merged by @xmath296 . the edges in the figure represent the resulting edges from @xmath133 after identifying the components @xmath302 by @xmath296 . now , the component of small blocks formed by @xmath183 is only adjacent to @xmath308 . therefore , by @xmath298 , @xmath309 is contained in some block of @xmath310 . the same happens with @xmath311 which is a component in @xmath285 which is only adjacent to the component @xmath312 . thus , @xmath313 is contained in some block of @xmath310 . however , the component of small blocks given by @xmath314 is adjacent to two different components of big blocks , @xmath308 and @xmath312 . therefore , @xmath314 are independent blocks in @xmath248 . the same happens with @xmath315 . thus , @xmath316 . [ remark : connected ] at step @xmath296 , if @xmath317 is connected , then @xmath318 defines a block of @xmath248 . [ remark : necessary chain 2 ] notice that remark [ remark : necessary chain ] still applies . in fact , if two points @xmath148 belong to the same block of @xmath319 then , necessarily , there exists a @xmath150-chain , @xmath151 joining them so that if @xmath320 , @xmath153 , the corresponding edges @xmath154 , @xmath155 , satisfy condition @xmath147 . for the graph in figure [ ejp chain_2b ] ] [ ejp : unchain 1b ] let @xmath3 be the graph from figure [ ejp chain_2b ] and let @xmath157 . then , let us check that applying @xmath164 on @xmath3 the dendrogram generated is the one from figure [ chain_dendrogram_2 ] . clearly , @xmath321 if @xmath167 . if @xmath322 , @xmath323 . there are nine @xmath158-components , seven of them are singletons and two of them , @xmath113 , @xmath114 , with @xmath159 . furthermore , @xmath160 , @xmath161 and @xmath162 for @xmath163 . for @xmath256 , conditions @xmath146 and @xmath147 induce edges in @xmath324 between @xmath113 and @xmath244 and between @xmath244 and @xmath114 . then , @xmath324 has one component , @xmath282 , which is not a single point : @xmath325 . @xmath244 is a single point and @xmath326 . then , there are two big blocks in @xmath282 , @xmath113 and @xmath114 , and one small block , @xmath244 . since the small block is connected to both big blocks , by condition @xmath298 , these blocks are not merged . thus , for every @xmath168 , @xmath327 . for @xmath170 , there are edges in @xmath171 between @xmath113 and @xmath328 for every @xmath329 and between @xmath114 and @xmath330 for every @xmath329 . thus , @xmath171 is connected . now , there are two big blocks , @xmath113 , @xmath114 and 7 small blocks . by conditions @xmath296 and @xmath298 , it is readily seen that @xmath331 . @xmath332 for every @xmath333 . the minimal distance @xmath54 such that @xmath183 are connected by an edge in @xmath334 is @xmath335 . in this section we try to give some theoretical background to the treatment of the chaining effect . our intention , as it was mentioned above , is to define some concrete element to evaluate the sensitivity of a method to the type of chaining effect we are treating . first , we define the concept of _ chained subsets _ and subsets _ chained by a single edge_. let @xmath4 be a finite metric space . we say that two @xmath337-connected subsets of @xmath4 , @xmath183 , are @xmath338-*chained * if they hold that * @xmath339 , * there exist @xmath340 and @xmath341 such that @xmath342 . if the parameters @xmath343 are not relevant , we say simply that @xmath183 are * chained*. notice that not every case of chained subsets is going to induce an `` undesired '' chaining effect . see figure [ ejp chained_2 ] . the idea is that chained subsets include the cases of undesired chaining effect we are going to treat in this section . and @xmath99 are @xmath338-chained subsets although there is no undesired chaining effect if these blocks are merged . ] inside the cases of chained subsets let us consider some specific type which are the subsets chained by a single edge . we define this as the prototypical case of chained subsets on which the algorithm should be tested to check its unchaining properties . let @xmath4 be a finite metric space . we say that two @xmath337-connected subsets of @xmath4 , @xmath183 , are @xmath338-*chained by a single edge * if they hold that * @xmath344 , * there exist @xmath340 and @xmath341 such that @xmath342 * @xmath345 , @xmath346 . if the parameters @xmath347 are not relevant , we say simply that @xmath183 are * chained by a single edge*. suppose that @xmath348 and @xmath183 are @xmath338-chained by a single edge . notice that given the rips complex @xmath349 and the edge @xmath350 , then @xmath351 has exactly two connected components : @xmath190 and @xmath99 . [ ejp : chain_0 ] the subsets @xmath183 from figure [ ejp chained edge ] are @xmath352-chained by a single edge ( @xmath353 ) . @xmath183 are @xmath150-connected . there is a pair of points @xmath340 , @xmath341 with @xmath354 and for any pair of points @xmath355 , @xmath356 , if @xmath357 , then @xmath358 . [ ejp : chain ] consider the graph represented in figure [ ejp chain_1b ] . suppose the edges in @xmath111 have length @xmath158 and the rest have length @xmath253 . the distance between the vertices are measured as the length of the minimal path joining them . the whole space is @xmath253-connected with @xmath359 . thus , @xmath190 and @xmath99 are @xmath360-chained by a single edge . let @xmath41 be a hc method and @xmath361 . we say that @xmath41 is * strongly chaining * if for any set @xmath4 , any pair of chained subsets @xmath183 of @xmath4 and any @xmath362 , if @xmath190 is contained in some block @xmath94 of @xmath363 , then @xmath364 . it is immediate to check that @xmath6 @xmath5 is strongly chaining . moreover , given any pair of @xmath338-chained subsets @xmath183 such that @xmath348 with @xmath365 , @xmath353 is contained in some block of @xmath366 while @xmath190 is not contained in any block of @xmath62 for any @xmath367 . in particular , @xmath363 does not refine @xmath100 for any @xmath368 . [ th : strongly chaining ] let @xmath41 be a hierarchical clustering method . if for every metric space @xmath4 and every @xmath369 , @xmath370 implies that @xmath371 , then @xmath41 is strongly chaining . in particular , @xmath6 @xmath5 is strongly chaining . first , let us see that @xmath41 is strongly chaining . consider two @xmath338-chained subsets @xmath183 . by hypothesis , there exist @xmath340 , @xmath341 such that @xmath372 . also , there exist @xmath373 with @xmath374 . thus , @xmath375 . if @xmath190 is contained in some block @xmath94 of @xmath363 , then @xmath376 and @xmath364 . @xmath7 and @xmath8 @xmath5 are not strongly chaining : [ ejp : not strong chaining ] consider the graph from figure [ ejp chain_1b ] . suppose that , in addition , we include edges of length @xmath253 from @xmath377 to every vertex in @xmath113 and from @xmath378 to every vertex in @xmath114 . also , suppose that @xmath379 . thus , every pair of points in @xmath113 ( resp . @xmath114 ) are at distance @xmath158 , @xmath380 ( resp . @xmath381 ) for every @xmath382 , @xmath383 , @xmath384 for every @xmath385 and every @xmath329 , @xmath381 for every @xmath382 , @xmath383 , @xmath386 for every @xmath387 and every @xmath388 , @xmath389 and @xmath390 for every @xmath391 . then , @xmath190 and @xmath99 are @xmath392-chained subsets . however , @xmath393 and @xmath394 . as we have seen in the examples above , @xmath1 is able to detect some partitions of chained blocks . in particular , we have seen that it detects chained blocks when they have dense cores whose distance is greater than the minimal distance between the blocks . this is detected in the rips complex ( for some @xmath362 ) because there exist high dimensional simplices in both blocks while there is no high dimensional simplex intersecting both . herein , we give sufficient conditions for blocks chained by a single edge to be detected . we also show how the hierarchical clustering is going to recover them . to formalize this we introduce the definition of _ weakly unchaining clustering method _ and prove that @xmath1 is weakly unchaining . we also check that other methods as @xmath8 or @xmath7 @xmath5 are not weakly unchaining although , as we mentioned in example [ ejp : not strong chaining ] , they are not strongly chaining either . let @xmath41 be a hc method and @xmath361 . we say that @xmath41 is * weakly unchaining for the parameter @xmath0 * if the following implication holds : let @xmath4 be a finite metric space such that @xmath348 , with @xmath183 a pair of subsets @xmath395-chained by a single edge @xmath353 . suppose there exist @xmath396 , @xmath397 such that * @xmath398 is contained in some block @xmath399 of @xmath400 , @xmath163 , * @xmath401 , @xmath163 , * @xmath160 , @xmath161 , * @xmath402 and @xmath403 . then , there exists @xmath362 such that @xmath404 . we say that @xmath41 is * weakly unchaining * if it is weakly unchaining for some parameter @xmath0 . notice that in the definition above we consider two chained subsets with further conditions . therefore , if a @xmath5 method is strongly chaining , in particular , it is not weakly unchaining . [ th : weakly ] let @xmath4 be a finite metric space such that @xmath348 , with @xmath183 a pair subsets of @xmath395-chained by a single edge @xmath353 . suppose there exist @xmath396 , @xmath397 such that * @xmath398 is contained in some block @xmath399 of @xmath400 , @xmath163 , * @xmath401 , @xmath163 , * @xmath160 , @xmath161 . then , @xmath149 refines @xmath100 . if , in addition , @xmath402 and @xmath403 , then @xmath405 . let us recall that , by definition , @xmath406 . for the first part it suffices to check that for every pair @xmath407 , @xmath408 is not contained in any block of @xmath310 , this is , @xmath409 . let @xmath407 . first , notice that for any @xmath410 , there is no @xmath54-chain joining @xmath411 to @xmath412 . thus , @xmath413 . let us check that @xmath414 , @xmath415 . for @xmath416 , since @xmath417 and @xmath418 , condition @xmath147 implies that there is no edge in @xmath419 between @xmath420 and @xmath421 . since @xmath422 for every @xmath423 there is no @xmath424-chain joining @xmath411 to @xmath412 which does not contain the edge @xmath353 . in particular , there is no @xmath425-chain joining @xmath411 to @xmath412 which does not contain the edge @xmath353 . therefore , by remark [ remark : necessary chain ] , it follows that @xmath426 . the same argument works for every @xmath427 . thus , @xmath428 and @xmath301 refines @xmath100 . suppose , in addition , that @xmath402 and @xmath403 . we already proved that @xmath310 refines @xmath100 . clearly , since @xmath402 ( respectively , for @xmath99 ) , all the blocks of contained in @xmath190 ( resp . @xmath99 ) are joined by an edge in @xmath429 . therefore , @xmath190 ( resp . @xmath99 ) is a block of @xmath310 . [ cor : weakly ] @xmath1 is weakly unchaining for the parameter @xmath0 . see example [ ejp : unchain 1 ] . [ cor : weakly_2 ] @xmath2 is weakly unchaining for the parameter @xmath0 . [ ejp : al - cl ] @xmath7 and @xmath8 @xmath5 are not weakly unchaining . consider the graph in figure [ ejp chain_1b ] . to check that @xmath8 @xmath5 is not weakly unchaining suppose that we add some edges between @xmath113 and @xmath114 so that @xmath430 , @xmath431 . notice that this graph satisfies the conditions in the definition of weakly unchaining . then , it suffices to check that @xmath432 is never @xmath100 . it is immediate to check that @xmath433 . then , it is readily seen that @xmath434 for every @xmath435 and @xmath436 . to check that @xmath7 @xmath5 is not weakly unchaining suppose that in figure [ ejp chain_1b ] , we made @xmath437.let us see that @xmath438 is never @xmath100 . first , notice that this graph satisfies the conditions in the definition of weakly unchaining . also , it is immediate to check that @xmath439 , @xmath440 . thus , it is readily seen that @xmath441 for every @xmath442 and @xmath443 . dbscan is a density - based algorithm for clustering . see @xcite . although this is not a hierarchical clustering method it is worth analyzing the chaining effect and comparing its results with @xmath1 . dbscan requires two parameters : some distance @xmath59 and a minimal number of points @xmath444 . a point is a _ core point _ of a cluster if there are at least @xmath444 in its @xmath56-neighborhood . then , the _ density - reachable _ points from a core point define a cluster . let us recall here the formal definition . a _ @xmath56-neighbourhood _ of a point @xmath445 , @xmath446 . a minimum number of points , @xmath444 is defined so that if @xmath447 has at least @xmath444 , then @xmath445 is a core point of a cluster . a point @xmath445 is _ directly density - reachable _ from a point @xmath448 with respect to @xmath56 , @xmath444 if \1 ) @xmath449 and \2 ) @xmath450 a point @xmath445 is _ density - reachable _ from a point @xmath448 with respect to @xmath56 , @xmath444 if there is a chain of points @xmath451 such that @xmath452 is directly density - reachable from @xmath453 . a point is _ density - connected _ to a point @xmath448 with respect to @xmath56 , @xmath444 if there is a point @xmath454 such that both , @xmath445 and @xmath448 are density - reachable from @xmath454 with respect to @xmath56 , @xmath444 . a cluster is defined to be a set of density connected points which is maximal with respect to density - reachability . the points which do not belong to any clusters are considered noise . one of the advantages of dbscan is that it is capable of detecting isolated points and eliminating them as noise . also , the chaining effect through a chain of points is reduced . in fact , the type of unchaining dbscan does is more related to the chaining through smaller blocks . it is not so effective to detect the chaining effect produced between two blocks when the minimal distance between them is small . in general , suppose @xmath348 with @xmath183 two clusters @xmath338-chained by a single edge @xmath353 . let us assume that @xmath455 and that @xmath456 are core points . then , @xmath456 are density connected and they belong to the same cluster in the output of dbscan . therefore , the clustering @xmath100 is not detected by dbscan . see example [ ejp : dbscan ] . however , if @xmath456 belong to simplices with dimension at least @xmath0 in @xmath457 , @xmath458 respectively for some @xmath459 , then @xmath1 detects this clustering . in fact : @xmath460 . [ ejp : dbscan ] let us analyze the case of example [ ejp : chain ] . let @xmath461 and @xmath462 . then , notice that @xmath463 , @xmath464 are core points . therefore , since @xmath465 , they are density - connected . thus , dbscan does not detect the clustering @xmath100 . let @xmath4 be a finite metric space , @xmath466 be @xmath337-connected subsets of @xmath4 and @xmath467 . we say that @xmath468 and @xmath469 , are @xmath338-*chained through @xmath0-smaller blocks * if the following conditions hold * @xmath470 , * there exists a @xmath471-chain @xmath472 with @xmath473 for every @xmath474 * @xmath475 , @xmath346 . * @xmath476 for every @xmath477 . if the parameters @xmath478 are not relevant , we simply say that @xmath183 are * chained through smaller blocks*. let @xmath41 be a strongly chaining @xmath5 method and @xmath361 . we say that @xmath41 is * completely chaining * if for any set @xmath4 , any pair of components @xmath479 of @xmath4 chained through smaller blocks and any @xmath362 , if @xmath468 is contained in some block @xmath94 of @xmath363 , then @xmath480 . [ th : complete chaining ] let @xmath41 be a hierarchical clustering method . if for every metric space @xmath4 and every @xmath369 , @xmath370 implies that @xmath371 , then @xmath41 is completely chaining . in particular , @xmath6 @xmath5 is completely chaining . by theorem [ th : strongly chaining ] , we already know that @xmath41 is strongly chaining . let @xmath479 two @xmath337-connected subsets @xmath338-chained through smaller blocks . let @xmath472 be the corresponding chain . then , @xmath481 for every @xmath482 and there exist @xmath483 such that @xmath484 . thus , @xmath485 for every @xmath482 . now , suppose @xmath362 such that @xmath468 is contained in some block @xmath94 of @xmath363 . then , @xmath486 for every @xmath482 and @xmath480 . satisfies the conditions from definition [ def : moderate ] with equalities on conditions @xmath487 and @xmath488 , then @xmath133 is the graph above and @xmath489 is as indicated . ] [ def : moderate ] @xmath41 is @xmath0-*bridge - unchaining * if it is weakly unchaining for the parameter @xmath0 and the following implication holds : let @xmath4 be a finite metric space , @xmath361 and let @xmath490 with @xmath491 , @xmath492 , @xmath493 single points for every @xmath494 . suppose that * @xmath495 for every @xmath496 , * @xmath497 for every @xmath498 , * @xmath499 for every @xmath20 * @xmath500 for every @xmath501 * @xmath502 and @xmath503 * @xmath504 * @xmath505 * @xmath506 , @xmath507 and @xmath508 . then , there exists @xmath362 such that @xmath509 @xmath41 is * bridge - unchaining * if it is @xmath0-bridge - unchaining for some parameter @xmath0 . notice that in the conditions above , if @xmath510 , then @xmath190 and @xmath99 are @xmath511-chained through the @xmath0-smaller blocks @xmath512 . and @xmath513 satisfies the conditions from proposition [ prop : moderate_2 ] with equalities on conditions @xmath487 and @xmath488 , then @xmath133 is the graph above and @xmath489 is as indicated . ] [ prop : moderate_2 ] let @xmath4 be a finite metric space and let @xmath514 with @xmath515 . suppose that * @xmath516 for every @xmath517 , * @xmath518 for every @xmath519 , * @xmath520 for every @xmath20 * @xmath521 for every @xmath501 * @xmath522 for every @xmath20 and every @xmath523 * @xmath524 for every @xmath501 and every @xmath525 * @xmath526 , @xmath527 and @xmath528 . * @xmath529 for every @xmath530 , @xmath531 , @xmath532 for every @xmath533 . then , @xmath534 let @xmath514 satisfying the conditions above . for @xmath535 let us apply conditions @xmath146 and @xmath147 of @xmath1 . since , @xmath536 for every @xmath530 , we obtain edges @xmath537 for every @xmath517 . since @xmath531 , @xmath532 for every @xmath533 , we also obtain edges @xmath538 and @xmath539 for every @xmath533 such that the distance is less or equal than @xmath540 . thus , the blocks @xmath541 , @xmath542 are in the same connected component of @xmath419 . since @xmath526 , by @xmath298 , @xmath541 is an independent block in @xmath543 for every @xmath530 . also , by @xmath298 , the blocks @xmath544 joined by an edge to @xmath468 ( resp . the blocks @xmath545 joined by an edge to @xmath546 ) are merged with @xmath468 ( resp . @xmath546 ) . the same argument holds for every @xmath547 . thus , @xmath534 renaming the corresponding blocks , it is immediate to obtain the following : [ cor : unchaining ] @xmath2 is @xmath0-bridge unchaining . see example [ ejp : unchain 1b ] . [ al cl isolated ] @xmath8 and @xmath7 are not bridge - unchaining . as we mentioned in the introduction , @xmath8 and @xmath7 show a tendency to merge isolated points before joining them to a pre - existing cluster . this tendency can be seen aplying the algorithms in the following cases . let @xmath548 and suppose the conditions from definition [ def : moderate ] hold with equalities on conditions @xmath487 and @xmath488 . see figure [ moderate ] . assuming that @xmath549 and @xmath550 , then in @xmath551 , @xmath552 is a cluster . similarly , let @xmath553 and suppose the conditions from definition [ def : moderate ] hold with equalities on conditions @xmath487 and @xmath488 . assuming that @xmath554 and @xmath555 , then in @xmath556 , @xmath552 is a cluster . in either case , the point @xmath259 ( resp . @xmath557 ) would appear to be far away from @xmath190 ( resp . @xmath99 ) while @xmath259 , @xmath557 appear to be very close to each other . this type of chaining through points or through smaller blocks may be also avoided by dbscan . in the conditions from definition [ def : moderate ] , if @xmath558 and the points @xmath512 are not core points ( with @xmath559 ) , dbscan would not merge @xmath190 and @xmath99 either . however , it would not necessarily return the clustering @xmath560 . the result obviously depends on the density distribution of the points in @xmath190 and @xmath99 and the parameters @xmath56 , @xmath444 involved . let us assume that every point in @xmath190 , @xmath99 is a core point and @xmath561 . in this case , the output of dbscan will be two clusters , @xmath98 with @xmath562 , @xmath563 and some single points ( noise ) from the sequence @xmath564 herein , we treat a particular type of chaining effect which is characteristic from single linkage . this effect is reduced if the algorithm shows some sensitivity to the density distribution of the data set . this is why average linkage or complete linkage are usually preferred by practitioners . our aim was to define an algorithm such that it encodes information about the density distribution with a very simple input . @xmath1 is able to detect clusters affected by this kind of chaining effect . we also provide some theoretical background to the study of the chaining effect . thus , a hierarchical clustering method is _ strongly chaining _ if every pair of chained clusters is automatically merged in one cluster . this is the case of single linkage . on the contrary , a hierarchical clustering method is _ weakly unchaining _ if at least it detects some type of chained clusters when they have dense nuclei distant apart . we prove that @xmath1 is weakly unchaining while complete linkage and average linkage are not . also , compared with dbscan , our method seems to have a more natural and powerful treatment of this problem . one weakness of @xmath1 is that it fails to detect when two blocks are chained by a single point or a small block . @xmath2 deals with that weakness . we prove that @xmath2 is @xmath0-_bridge - unchaining _ showing that @xmath2 is capable of detecting this kind of chaining . we focused on the theoretical problem of chaining so we did not consider the computational problem involved . it would be interesting to study how efficient is @xmath1 compared with other methods . one of the main advantages of @xmath6 is that it is stable in the gromov - hausdorff sense which is a really interesting property for a clustering algorithm . if the algorithm is too sensitive to small perturbation of the data the output may be easily meaningless . unfortunately , our method does not share with @xmath6 the good stability properties . modifying the algorithm to deal with the chaining effect we lost that advantage . the problem of stability of linkage - based clustering methods and the difficulties to define algorithms , other than @xmath6 , stable in the gromov - hausdorff sense is studied in @xcite . the author would like to express his gratitude to bruce hughes for pointing out such a nice paper to read . ester m , kriegel h , sander j , xu x. _ a density - based algorithm for discoreing clusters in large spatial databases with noise_. in evangelos simoudis , jiawei han , usama m. fayyad . proceedings of the second international conference on knowledge discovery and data mining ( kdd-96 ) . aaai press . pp.226 - 231 . isbn 1 - 57735 - 004 - 9 .

using gemini quickstart infrared observations of the central of m33 , we analyze the stellar populations in this controversial region . based on the slope of the giant branch we estimate the mean metallicity to be , and from the luminosities of the most luminous stars , we estimate that there were two bursts of star formation and gyr ago . we show that the stellar luminosity function not only has a different bright end cutoff , but also a significantly different slope than that of the galactic bulge , and suggest that this difference is due to the young stellar component in m33 . we combine our infrared gemini data with optical hst - wfpc2 measurements revealing a cmd populated with young , intermediate , and old age stellar populations . using surface brightness profiles from to , we perform simple decompositions and show that the data are best fit by a three - component , core + bulge + disk model . finally , we find no evidence for radial variations of the stellar populations in the inner of m33 based on a spatial analysis of the color - magnitude diagrams and luminosity functions .

astro-ph0206344

with the advent of space - based telescopes , such as the hubble space telescope ( hst ) , and large aperture ground - based telescopes with adaptive optics ( ao ) , such as gemini and vlt , the number of galaxies beyond the milky way ( mw ) and its dwarf companions for which detailed studies can be made is gradually increasing . of the nearby galaxies , m33 ( ngc 598 , the triangulum nebula ) is one of the best for studying the stellar content of spiral galaxies . outside of the mw , it is one of the closest and brightest spirals visible , surpassed only by m31 . m31 is more luminous and slightly closer , but it s higher inclination angle of 77 ( compared to 56 for m33 ) makes it more difficult to separate the contributions from different stellar populations . however , is m33 the prototype or the exception when it comes to late - type spiral galaxies ? it contains a healthy population of halo globular clusters @xcite , yet it is not certain whether there is a matching bulge . this simple point is crucial for understanding the role of a bulge in galaxy formation , and its relationship to the halo and globular clusters . as of 1991 , the status of m33 s bulge was `` controversial '' according to a review of literature by @xcite . there have since been many papers on m33 , but the number of authors who find a bulge seems balanced by an equal number who do not . @xcite argued against a traditional bulge , based on the lack of a power - law contribution to his 12 surface brightness measurements . although his @xmath7-band observations do show an excess of light inside @xmath8 , he finds no satisfactory @xmath9 fit . based on infrared ( ir ) observations of the central @xmath10 which show a clustering of stars around the nucleus , @xcite take the opposite side . they find a de vaucouleurs profile fits all but the inner @xmath11 of their surface photometry . @xcite are also in favor of a bulge in m33 . their infrared surface brightness measurements of the central @xmath12 seem reasonably well fit by an exponential disk plus @xmath9 profile . in contrast , @xcite cite an unchanging ir luminosity function from @xmath13 to @xmath14 as evidence against a significant bulge population beyond @xmath13 . thus , a decade after van den bergh s review , the controversy remains unresolved . in this paper we use ir gemini - north observations with quirc / hokupaa to study the stellar populations in the inner regions of m33 . in section [ sec : observations ] we describe our observations and observing procedure . section [ sec : data_reduction ] details the data reduction , photometric techniques , and calibration . we give the azimuthally averaged surface brightness profile in section [ sec : decomposition ] and , after combining our observations with those of @xcite , present simple two- and three- component model decompositions . we describe artificial star tests performed to understand completeness and observational effects as a function of a star s position and luminosity in section [ sec : artstartests ] . we use the color - magnitude - diagram ( cmd ) to estimate the metallicities and ages of the stars in our field in section [ sec : cmds ] . we present the stellar luminosity function ( lf ) in section [ sec : lfs ] , and compare it to the lf measured in the galactic bulge . in section [ sec : comparison ] we compare our observations to the recent work in the infrared by @xcite with the cfht ao system , and combine our data with optical hst - wfpc2 measurements made by @xcite . we look for radial variations in the stellar properties in section [ sec : radial_variations ] . in section [ sec : blending ] we perform a theoretical analysis of blending on our own and previous observations following the procedures of @xcite . finally we give a summary of our conclusions in section [ sec : conclusions ] . the observations upon which this paper is based were taken as part of the gemini north quickstart service observing program using hokupaa & quirc . @xcite is a natural guide star , 36-element curvature - sensing adaptive optics system built by the university of hawaii . quirc is a near - infrared imager on loan from the university of hawaii and mounted at the exit focus of hokupaa . quirc has a @xmath15 hawaii hgcdte array with a plate scale of @xmath16 pixel@xmath17 , giving a @xmath18 field of view . the array is linear to @xmath19 adu , and saturates at @xmath20 adu , with a gain of 1.85 @xmath21/adu . we observed the central regions of m33 ( @xmath2201:33:50.9 , @xmath2330:39:37 , j2000 ) through three filters , @xmath24 , @xmath25 , and @xmath26 @xcite . we used the nucleus of m33 as the ao wavefront reference source . observations were carried out over three separate nights spanning three months . table [ tab : observations ] lists the number of successful frames which were obtained in each band on each night . exposure times were 180 seconds except on october 25 when observations were made without a shutter ; on that night the exposure times were approximately 10.8 seconds longer due to the detector readout time . the relative size , location , and orientation of our field is illustrated in figure [ fig : field_locator ] . on the left of this figure is a @xmath27 digitized sky survey image centered on m33 . the box drawn in the center of the galaxy represents the placement of our infrared gemini - north image . the infrared image shown on the right is our combined @xmath25-band image , and is @xmath28 across . we obtained sky observations for background subtraction in a blank field @xmath29 se of the nucleus ( @xmath2201:34:57 , @xmath2329:57:20 , j2000 ) . the typical observing pattern was to take two sky dithers , four dithers on the nucleus , two skys , etc . with this technique the sky frame is constructed from the combination of the pairs of sky frames before and after each observation , and hence each two - dither sky set is used for both the preceding and following galaxy observations . the sky sets used @xmath30 dithers , while the galaxy sets used @xmath31 dithers . ukirt photometric standard stars were observed each night , although possibly not quite as frequently as the authors would have liked . the september night had one standard before and one after our observations ; the october night had two before ; and the december night had two before and one after . these standards were typically observed with 5 second exposure times , implementing a five position dither pattern moving the star in @xmath32 steps . cccc 2000 - 09 - 27 & 8 & 0 & 12 + 2000 - 10 - 25 & 8 & 0 & 8 + 2000 - 12 - 22 & 0 & 19 & 0 + [ tab : observations ] image quality was extremely good . we used the iraf gemseeing procedure to estimate the full - width at half - maximum ( fwhm ) on all 55 images . the results shown in figure [ fig : fwhm ] are the average fwhms measured on the same @xmath33 hand - selected stars on all frames . the @xmath26-band images have a mean seeing of @xmath34 , @xmath25-band obtained @xmath35 , while the @xmath24-band images are slightly worse and more variable with a mean seeing of @xmath36 . the superior image quality at longer wavelengths is a consequence of the increased coherence length of the atmosphere ( @xmath37 ) . the image size does not vary significantly across the field , although the psf morphology has a systematic dependence on the azimuthal angle around the nucleus of m33 , which was used as the ao reference . for comparison , the diffraction limits of an 8 m telescope are @xmath38 , @xmath39 , and @xmath40 at @xmath24,@xmath25 , and @xmath26 respectively . the fwhm , however , is not the best measurement of image quality since most ao images consist of a tight core superimposed on much broader emission . we therefore also recorded the gemseeing task s estimates of various encircled energy diameters ( eed ) . we found that the 50% eed is approximately twice the fwhm , while the average 95% eeds are @xmath41 , @xmath42 , and @xmath43 at @xmath24 , @xmath25 , and @xmath26 respectively . we reduced our data using the gemini package of tools in iraf . the procedure involves division by a normalized lamps on lamps off dome flat field , and subtraction of a sky frame constructed from sky observations before and after each galaxy observation . a constant equal to the median of the sky frame was then added back to the image to restore the original sky level and maintain the noise characteristics of the frame . photometry was performed with peter stetson s suite of software . object detection was done on a combined image made up of all the dithers of all the bands . we then used daophot @xcite to determine point spread functions ( psfs ) for each dither of each band using isolated stars . we used a quadratically variable model psf with a 25 pixel ( @xmath44 ) radius ; an alternate run using a larger 50 pixel ( @xmath45 ) radius psf , yielded little difference in the photometry , while taking a significantly larger amount of processing time . instrumental magnitudes were measured using the allframe psf fitting routine @xcite which simultaneously fits psfs to all stars on all dithers . daogrow @xcite was then used to determine the best magnitude in a @xmath44 radius aperture . our first attempt at calibrating the data used the photometric standards which were observed for us . we measured each standard using simple aperture photometry with a @xmath44 radius aperture , and a sky annulus from @xmath46 to @xmath47 . due to the small number of observed standards , we combined the measurements from all three nights . this gave us three @xmath24 standards and four @xmath26 standards for the september and october nights , and three @xmath25 standards for the december night . unfortunately , there are no published @xmath26 measurements for ukirt standards . we therefore estimated @xmath26 for each of the observed standards using the ( @xmath48 ) color and the transformation of @xcite . using the median combination of these measurements , we estimated our photometric zero point with uncertainties of 0.02 magnitudes in @xmath25 and @xmath26 , and 0.04 magnitudes in @xmath24 . we then converted our science @xmath26 measurements to @xmath49 based on each star s @xmath50 color using a transformation derived from the data published by @xcite . the last step was to convert our measurements to the cit / ctio photometric system using the transformation equations of @xcite . this initial calibration seemed to work acceptably for the @xmath24 and @xmath25 bands , but _ not _ for @xmath49 . evidence of the failure was in the disagreement ( @xmath51 magnitudes ) with previously published data for the stars in the inner region of m33 . specifically , with the @xmath49-band measurements of bright stars measured by @xcite , and an equal offset in the @xmath49-band surface brightness measurements published by @xcite . one possible cause of the problem is an inadequate transformation between @xmath49 and @xmath26 . the @xcite transformation was published in a paper describing the original @xmath26 filter . yet even a cursory comparison of the @xmath26 filter transmission curve published for quirc and that in the @xcite article reveals that they are very different . as an example , the original filter had a peak transmission of @xmath52% , while the quirc @xmath26 filter peaks over 95% . to overcome this problem we had to find alternate `` standards '' , preferably ones which had been observed before . for this purpose , we matched observations of 31 stars in common with the photometry published by @xcite . we examined the difference between his @xmath49 band observations ( ukirt photometric system ) and our instrumental @xmath26 for any dependence on @xmath53 or @xmath54 color , but found no significant trend . we therefore adopted the mean difference as our photometric zero point and conversion to the @xmath49-band . the uncertainty in this zero point is 0.02 magnitudes . a more detailed discussion of the comparison between these two datasets can be found in section [ sec : davidge ] . with the measurements finally calibrated we could construct our final photometry list . since our transformation to the cit / ctio system requires both @xmath53 and @xmath54 color information , we had already selected out only stars which have measurements in all three bands . in addition , each star had to have been detected on at least five dithers in each band , and the final error had to be less than 0.25 magnitudes . this gave us a list of 3716 objects over the entire frame . after ignoring objects detected within @xmath47 of the center of m33 because of severe crowding and incompleteness ( see section [ sec : completeness ] ) , we were left with a final photometry list of 3308 measurements , covering an area of 416 square arcseconds . when converting to absolute or bolometric magnitudes , we will use the distance and reddening to m33 as determined by @xcite from observations of cepheid variables . they find a true distance modulus of @xmath55 ( 850 kpc , @xmath56 pc ) , assuming 18.5 for the lmc . they also measure the m33 reddening to be @xmath57 , assuming 0.1 for the lmc , and a ratio of total - to - selective extinction of @xmath58 . these values translate to an infrared extinction and reddening of @xmath59 and @xmath60 . we note , however , that @xcite have recently calculated a slightly farther distance of 930 kpc , @xmath61 , using rr lyrae s in two halo clusters . in order to obtain accurate surface brightness measurements , we started with our reduced frames , and subtracted off the constant sky value which had been added previously to maintain each frame s noise characteristics . we then converted each frame to count rates , since some of the frames had slightly different exposure times . the resulting set of images in each band were averaged together to create the final images . we used the xvista annulus routine to compute azimuthally averaged radial surface brightness profiles . assuming an inclination of 56@xcite and a position angle of 23 for m33 , the routine defines elliptical annuli to be used as the averaging paths , in effect calculating the deprojected surface brightness profile assuming a circular disk . with 2 pixel wide annuli ( @xmath38 ) , we were able to calculate the surface brightness profile out to @xmath62 . we then combined our surface brightness measurements with those of @xcite which span from @xmath63 to @xmath5 . the resulting @xmath49-band profile is shown in figure [ fig : decomposition ] . there are three @xcite points in the region of overlap ( @xmath64 ) , and in this region the average difference between the two datasets is less than 0.01 magnitudes arcsecond@xmath65 . note that since the surface brightness information is extracted directly from pixel values , the profile is unaffected by image blending , even in the most crowded regions . using this combined surface brightness profile ( @xmath66 ) , we have performed a simple bulge / disk decomposition . we assume an exponential disk ( eqn . [ eqn : disk ] ) with a central intensity @xmath67 and a disk scale length @xmath68 , and a de vaucouleurs @xmath9 bulge ( eqn . [ eqn : bulge ] ) with an effective radius @xmath69 and an intensity @xmath70 at @xmath69 . the factor @xmath71 is a correction factor for deprojecting a spherical bulge . the best fitting 2-component model is illustrated in the left panel of figure [ fig : decomposition ] with a solid line . the disk has a central surface brightness of 17.16 magnitudes arcsec@xmath65 , and a scale length @xmath72 . the bulge is very compact , with an effective radius @xmath73 , where the surface brightness is 13.87 mag arcsec@xmath65 . the results of the fit are not very encouraging for this oversimplistic disk + bulge model . it is very clear that m33 has a strong compact spheroid component which becomes dominant over the disk at @xmath74 . however , the best fit parameters are neither a good fit to the inner bulge , nor to the disk between @xmath75 . @xmath76 @xmath77 \ } \label{eqn : bulge}\ ] ] @xmath78 \ } \label{eqn : core}\ ] ] a simple alternative is to add a third component to our model . we thus try an exponential disk + @xmath9 spheroid + sersic core @xcite . the sersic function ( eqn . [ eqn : core ] ) is parameterized with an effective radius @xmath79 , an intensity @xmath80 at @xmath79 , and the shape index @xmath81 ( @xmath82 ) . the variable power - law index @xmath81 gives the sersic function an extra degree of freedom over the more constrained exponential or @xmath9 profiles . the results of this three component model are shown on the right side of figure [ fig : decomposition ] and the parameter values are listed in table [ tab : decomposition ] . not surprisingly , this yields a significantly better fit than for the two component model . the central peak in surface brightness is well described by the sersic function , the disk is @xmath3 magnitude fainter , and the spheroid explains the excess luminosity seen inside @xmath11 . note that we list only the parameters derived from the @xmath25- and @xmath49-band images . the @xmath24-band image quality was inadequate to accurately model the very narrow core component . lcc disk scale length , @xmath68 & @xmath83 & @xmath84 + disk central sb , @xmath67 & 18.13 & 17.75 + spheroid effective radius , @xmath69 & @xmath85 & @xmath86 + spheroid sb @ @xmath69 , @xmath70 & 24.08 & 23.40 + core effective radius , @xmath79 & @xmath87 & @xmath88 + core sb @ @xmath79 , @xmath80 & 13.03 & 12.73 + core shape index , @xmath81 & 2.21 & 2.07 + [ tab : decomposition ] these disk and spheroid parameters are similar to those derived by @xcite based on their @xmath89 data alone . they found a disk scale length of @xmath90 at @xmath25 and @xmath91 at @xmath49 , with a central surface brightnesses of @xmath92 and @xmath93 mag / arcsec@xmath94 . however , since they were only using a two - component model ( and excluding the inner @xmath95 ) , they derived a brighter , more compact spheroid with @xmath96 , @xmath97 , and @xmath98 magnitudes arcsecond@xmath65 . their more compact spheroid is also in agreement with the @xmath99 spheroid effective radius found by @xcite . @xcite studied the surface brightness profile of m33 in the visible , and found a larger exponential disk scale length of @xmath100 . he also realized that the profile was too high for just an exponential disk inside @xmath8 , but he dismissed it due to the presence of visible spiral structure , concluding that the disk is non - exponential in the center . if our three component model is taken literally , it implies that virtually our entire field is spheroid dominated . at @xmath101 the spheroid and central core have equal surface brightness , which is 0.7 magnitudes brighter than that of the disk . at the edges of our field , @xmath102 , the disk and spheroid surface brightness become nearly equal . if this is correct , and the spheroid and disk are composed of different stellar populations , we may be able to detect a very small change in the composite population as a function of radius ( see section [ sec : radial_variations ] ) . we have also estimated the size of the core from our combined @xmath26 image . this image has better image quality than the @xmath24 band , and is not saturated like the @xmath25 images . on this image the nucleus has a fwhm of @xmath103 , compared to the stars which have a width of @xmath104 . thus we estimate that the nucleus has a width in @xmath26 of less than @xmath105 . we give this value as an upper limit since we do not attempt to correct for the nonlinearity of the array in the very core ( @xmath106 ) where the count rate is too high for the exposure time . this value seems in line with the increasing core size with wavelength measured by @xcite using hst . they find a core fwhm of @xmath107 with the f300w filter , @xmath108 with the f555w filter , and @xmath109 with the f1042 filter . although we could not measure the nuclear luminosity because the central pixels of our images were saturated , we can make an estimate based on our surface brightness fits . integrating the `` core '' component from the best - fitting three component model , we find a total luminosity of @xmath110 and @xmath111 . assuming @xmath112 , this corresponds to @xmath113 . in order to understand the completeness of our observations and estimate the importance of blending , we have performed a series of tests using artificial data . the traditional completeness tests inject artificial stars into the observed frames ; the efficiency and accuracy of their recovery give information about the completeness of the observations @xcite . this method is straightforward and easy to use , but is limited to primarily uncrowded fields . the more complex artificial field technique creates an entire frame containing millions of artificial stars to match the observations @xcite . this method is especially useful in very crowded fields , providing important constraints on the underlying stellar population . however , since photometry of the entire set of 55 @xmath114 pixel frames can be quite time consuming , taking many days to complete , our artificial star tests were instead run on the final combined @xmath115 images . to validate this shortcut , we have compared photometry obtained from the combined images with that from the simultaneous measurements of the 55 individual dithers . tests show that these two techniques are nearly identical , except for the faintest stars , where the photometry performed on the individual frames tends to be more complete . this can be seen in figure [ fig : dither_vs_combined_lfs ] , which shows luminosity functions derived from each of the two methods . the lf obtained from the individual frame photometry ( solid line ) rolls over approximately one magnitude fainter than the lf from the combined frames ( beaded line ) , even though they both go to about the same depth . starting with the three ( @xmath116 ) final combined images , we used the daophot addstar routine @xcite to insert 361 artificial stars into each image . these stars were arranged in a grid of @xmath117 , with a spacing of @xmath118 pixels ( @xmath119 ) , to avoid self - crowding . each star included random poisson noise , and random sub - pixel coordinate offsets from the aforementioned grid . all stars were input with the same magnitude starting with @xmath120 , and a color based on the mean colors observed in the real frame . we then repeated this process , with the grid of input stars shifted by 26 pixels ( @xmath121 ) . after four trials we had added a total of 1444 artificial stars , each sampling different positions in our frame . after the addition of the artificial stars , we analyzed these frames in the same manner as our original frames . stars were detected with daofind and measured on each of the four sets of @xmath116 images simultaneously with allframe . once the final photometry list was completed , we extracted the colors and magnitudes recovered for the input artificial stars using daomaster to match recovered coordinates with input coordinates , requiring less than a 2 pixel ( @xmath38 ) difference in position . we repeated this 4-trial procedure for six different integer input magnitudes ( @xmath122 ) . we have thus accumulated statistics on a total of 8664 artificial stars . figure [ fig : deltamag ] shows the difference between the recovered and input magnitudes as a function of distance from the center of m33 for three of the six input magnitudes ( @xmath123 ) . as expected the brightest stars are recovered very accurately , with only a few outliers caused by blends with other bright stars on the frame . note that an artificial star has to fall almost exactly on top of another star to be brightened but not lost since its position can not be perturbed by more than @xmath38 to be considered the same input star . near the center of m33 ( @xmath124 ) , where the density of stars is very high , figure [ fig : deltamag ] shows that almost all of the artificial stars are measured significantly brighter than their input magnitude . since the surface brightness varies across the field , our photometric completeness will also be a function of position on the frame . in figure [ fig : comp_vs_rad ] we plot the completeness as a function of radius for three of the six input magnitudes . to estimate the completeness , we have counted up the number of recovered stars in radial bins around the center of m33 , requiring that the difference between input and recovered magnitudes be less than 0.25 . we then divide by the number of artificial stars input into these same bins . the results are averaged over the four trials , and the errorbars show the one sigma dispersion in the range of values obtained over the four trials . this plot shows several important things about the completeness across our frame . first , the completeness drops off very quickly near the center of m33 . the completeness of the brighter stars drops from almost 100% to near zero across a range of only @xmath45 . the fainter stars do not exhibit quite as precipitous a drop , but still make the transition from maximum to minimum values quite quickly ; albeit at a larger central distance . second , the completeness away from the nucleus is relatively constant . thus it is safe to simply throw away measurements inside some critical radius , where the completeness is low ( and the effects of blending are high ) , and afterwards not worry about spatially varying completeness . finally , we can measure no stars accurately inside @xmath45 . while some stars are recovered in this region ( see figure [ fig : deltamag ] ) their measured brightnesses are all discrepant by more than 0.25 magnitudes . figure [ fig : comp_vs_mag ] shows the photometric completeness as a function of the brightness of the input artificial stars . here again we require that the recovered magnitude be no more than 0.25 magnitudes different than the input magnitude . since figure [ fig : comp_vs_rad ] shows that the negative effects of the core of m33 are constrained to the inner @xmath125 , we split this figure into two components . the solid line shows the completeness measured for all stars outside @xmath47 , which are again lower limits on the completeness since measured on the combined frames , while the science photometry was performed on all 55 individual frames simultaneously . the dashed line is for all objects measured closer than @xmath47 to the center of m33 , and shows that the photometry inside @xmath47 is severely degraded at all magnitudes . we have created a completely artificial field to match our m33 observations . we processed and measured this field in exactly the same manner as the real observations . since we know both the measured and true magnitude and color of every star in the artificial field , we can quantify our errors and try to estimate the true properties of the observed stellar population being modeled , free from observational effects . starting with blank frames with the appropriate amount of poisson noise , we add 8 million stars using spatially variable psfs modeled from the real data . the stars follow the observed m33 radial distribution , specifically the sum of the three component @xmath49-band decomposition of [ sec : decomposition ] . the input stellar population is chosen to match the colors and luminosity function observed in our m33 field . the input colors are the mean colors observed for stars with @xmath126 , and calculated at 0.5 magnitude intervals . the input luminosity function follows a power law distribution with a slope @xmath127 ( as is measured in [ sec : lfs ] ) , extending from @xmath128 . we add stars until the recovered lf of the simulation approximately matches the recovered lf of the real observations . in this case we added 8 million stars , which gives us slightly fewer recovered stars than in the real frame , 2677 artificial vs 2802 real , however as figure [ fig : simlfs ] shows , we get a very good match over most of the range of luminosities . the artificial frames were then processed and measured in the same manner as the real data , namely finding stars on a combined image with daofind , modeling the psf from isolated stars , and simultaneous psf - fitting photometry with allframe . the resulting @xmath129,@xmath53 and @xmath129,@xmath130 cmds are shown in figure [ fig : simcmd ] . here we have included all objects measured on the frame to show how well the simulation reproduces the observations , including blended objects located within @xmath131 ( half - size points ) . we will discuss the implications of these simulations when we attempt to quantify the effects of crowding on our photometry . the cmd of our m33 field is shown in figure [ fig : cmd ] . here we have omitted objects measured within @xmath47 of the nucleus which are mostly blends of fainter , bluer stars as indicated by their characteristic blueward and upward shift from the main giant branch locus ( see figure [ fig : simcmd ] ) . we have shown in section [ sec : completeness ] that nearly all objects detected close to the nucleus are blended , and that @xmath47 is approximately the boundary outside of which accurate photometry can be achieved . we thus take @xmath47 as the dividing radius between `` good '' and `` bad '' photometry . the resulting cmd for @xmath126 is relatively clean ; we interpret the gap at @xmath132 as the tip of the red giant branch ( rgb ) , and stars extending to @xmath133 as the asymptotic giant branch ( agb ) . we compare our cmd with measurements by others in section [ sec : comparison ] . the slope of the rgb is very sensitive to the metallicity of the population . @xcite and @xcite have used this fact to develop a metallicity indicator using the slope @xmath134 of the infrared rgbs of the upper five magnitudes ( @xmath135 ) of galactic globular clusters . their relation is given in equation [ eqn : rgb_slope ] . @xmath136 } = -2.98 - 23.84 \times slope_{gb } \label{eqn : rgb_slope}\ ] ] to apply this relation , we measure the rgb slope in the range @xmath137 , iteratively throwing away @xmath138 outliers ; the faint limit of @xmath139 is where the measured lf begins to turn over due to incompleteness ( see [ sec : lfs ] ) . we find a slope of @xmath140 , which implies a mean metallicity of [ fe / h]@xmath141 . this is our estimate for the average stellar population in the central regions of m33 ( @xmath142 ) . this determination is in agreement with the estimate of [ fe / h]@xmath143 by @xcite using population spectral synthesis of the central @xmath144 of m33 . however , as we will show in the next section , there is clearly a young stellar component present in the region imaged . since the gb slope is also sensitive to age @xcite , in the sense that at a constant metallicity the slope increases ( gets less negative ) with decreasing age . thus if all the stars ( old and young ) have the same metallicity , the gb slope will be slightly more positive than for a purely old population , and the kuchinski relation will give a metallicity which is too low . however , the younger stars are probably more metal rich than the mean of the old population , and hence their gb slope will be closer to that of the older , metal - poor population than if they were of the same metallicity . in short , our metallicity estimate is dependent on both the metallicities and ages of all the stars in the population . to estimate the age of the youngest stars in our field , we convert our ( cit / ctio ) @xmath49-band measurements to bolometric luminosities using the corrections of @xcite . these corrections were derived for m - giants in baade s window and depend on the dereddened @xmath145 color . the resulting bolometric cmd is shown in figure [ fig : cmd_mbol ] . assuming that the few bright stars at the tip of the agb are members of a young population , we can estimate the age of this population using the relationship between agb tip luminosity and age . first used by @xcite , this relation makes use of the monotonically decreasing maximum luminosity of the agb tip with increasing age . due to the limited number of stars in our field , and the short lifetime of stars on the agb , this estimate is only an upper limit to the age of the youngest stars . @xcite have recalculated the relation between the agb tip luminosity and age using the zvar synthetic cmd code of @xcite . this program makes use of the stellar evolution models of @xcite , and the mass - loss prescription of @xcite . averaging three different models with different metallicities and binary fractions , they derived equation [ eqn : age ] for the ages of stars on the agb tip at different luminosities . thus by measuring the bolometric luminosity of the agb tip , one can estimate the age of the youngest stars in the field . @xmath146 the simplest estimate of the age comes from the luminosity of the brightest star in the field . this star has @xmath147 , which , with equation [ eqn : age ] , gives an age of 0.51 gyr . if we assume that the intermediate age stars on the agb can be explained by a single burst of star formation , we can apply the averaging technique of @xcite to overcome small number statistics when estimating the tip of the agb . their method is to find the average luminosity of stars on the agb over the agb tip of an old population ( @xmath148 ) . by assuming that the distribution of stars along the agb is uniform , the peak bolometric luminosity for a fully populated agb should be twice the mean ( plus @xmath149 ) . using this procedure on the 28 stars brighter than @xmath150 we estimate the agb tip luminosity to be @xmath151 for a fully populated agb . this luminosity implies an age of @xmath152 gyr for this intermediate age population . however , looking at the bolometric cmd in figure [ fig : cmd_mbol ] , it is fairly clear that the stars above @xmath153 are not uniformly populated along its entire extent . this is probably due to multiple bursts of star formation contributing to the agb . a careful inspection shows that the agb is well represented from @xmath154 , with a bright tail of stars up to @xmath155 ( see also the luminosity function in figure [ fig : lf_mbol ] ) . the predominant agb is most likely due to a large burst of star formation around 2 gyr ago , while the less populated secondary agb is probably due to an even more recent burst of star formation @xmath3 gyr ago . a similar multi - generation model was proposed for the nucleus of m33 by @xcite to explain its blue ( @xmath156 ) and red ( @xmath157 ) colors and strong spectral absorption features such as h@xmath158 and h@xmath159 . @xcite also argued for multiple epochs of star formation based on the integrated optical spectrum of the inner @xmath32 , and concluded that @xmath33% of the @xmath160-band light originates in stars younger than 1 gyr , with the youngest generation about @xmath161 years old . ultraviolet observations agree , @xcite used iue spectra together with @xmath162 photometry of the nucleus to model the stellar population , and found the best - fit model required both a young component with an age of @xmath163 years , and an old component with an age of @xmath164 years . while combining photometric and spectroscopic observations , @xcite found that the nucleus of m33 is best fit by a @xmath165 myr old single burst of star formation enshrouded by a significant amount of dust . most recently , @xcite used stellar spectral synthesis to model hst stis spectroscopy of the nucleus and found that the best fit is with two bursts of star formation 50 myr and 1 gyr ago . we can reject the possibility of field star contamination as an explanation of the bright stars observed in our field using the field star density tabulated by @xcite which was calculated from the bahcall and soneira galaxy model . to estimate the number of field stars brighter than @xmath166 ( @xmath167 ) , we assume that potential contaminants are field m - dwarfs , they could be as red as @xmath168 . thus they could be as faint as @xmath169 . @xcite predicts that the number of field stars with @xmath160 brighter than 25 is @xmath170 arcmin@xmath65 . multiplying by the area of our field 0.116 arcmin@xmath171 gives an upper limit of 0.36 stars brighter than @xmath166 in our field . this means that there is a @xmath172 chance of finding _ one _ star in our field as bright as @xmath166 , while we see 27 . the luminosity functions ( lfs ) derived for our field are shown in figure [ fig : lfs ] and listed in table [ tab : lfs ] . these include only objects measured farther than @xmath47 from the center of m33 . this region has an area of 416 arcsec@xmath171 . the data have been binned into 0.25 magnitude bins , and the center of each bin is given in column 1 . these lfs all show dips at the location of the tip of the rgb ( @xmath173 ) . crrr 15.875 & 0 & 0 & 3 + 16.125 & 0 & 1 & 3 + 16.375 & 0 & 2 & 4 + 16.625 & 0 & 5 & 13 + 16.875 & 0 & 9 & 5 + 17.125 & 2 & 9 & 16 + 17.375 & 0 & 15 & 21 + 17.625 & 5 & 23 & 27 + 17.875 & 11 & 26 & 14 + 18.125 & 7 & 19 & 22 + 18.375 & 16 & 25 & 38 + 18.625 & 25 & 43 & 68 + 18.875 & 19 & 80 & 85 + 19.125 & 29 & 92 & 95 + 19.375 & 22 & 106 & 105 + 19.625 & 49 & 105 & 118 + 19.875 & 78 & 142 & 162 + 20.125 & 110 & 198 & 208 + 20.375 & 116 & 213 & 249 + 20.625 & 139 & 256 & 310 + 20.875 & 190 & 258 & 359 + 21.125 & 226 & 310 & 388 + 21.375 & 310 & 357 & 427 + 21.625 & 350 & 364 & 349 + 21.875 & 385 & 293 & 176 + 22.125 & 318 & 207 & 38 + 22.375 & 330 & 94 & 5 + 22.625 & 286 & 45 & 0 + 22.875 & 175 & 10 & 0 + 23.125 & 83 & 1 & 0 + 23.375 & 27 & 0 & 0 + [ tab : lfs ] we show the absolute @xmath49-band luminosity function in figure [ fig : lfcompbw ] , assuming a distance modulus to m33 of 24.64 . for comparison we have also plotted a composite galactic bulge luminosity function constructed from @xcite and @xcite measured in baade s window ( bw ) . there are two very obvious differences between these two lfs : their bright end extents and their slopes . the m33 lf extends over a magnitude brighter than what is observed in bw . these bright stars , as discussed in section [ sec : age ] , are a result of a young population of stars in the central regions of m33 . the slopes of the m33 and baade s window rgb lfs are also significantly different . we fit each with a single power law , m33 over @xmath174 , and bw over @xmath175 . in m33 we measure a lf slope of @xmath176 , while in bw we find @xmath177 . one possible reason for this difference is the difference in mean ages of the stars in these two regions . m33 contains a mix of young and old stars ( e.g. section [ sec : age ] and the optical - ir cmd in figure [ fig : cmd_vk ] ) , while the galactic bulge is predominantly an old population . however , @xcite has also suggested that the slope of the lf may vary throughout the galactic bulge . looking at dereddened lfs of 17 bulge fields , he fits a power law between @xmath178 and @xmath179 and finds a range in lf slope of @xmath180 , with a mean value , obtained by coadding the lfs , of @xmath181 . this mean lf slope obtained by davidge agrees with what we have measured in the central region of m33 , however involves many uncertainties , such as the reddening and the small luminosity range over which the lfs were fit . if indeed galaxy lfs are variable , as davidge s data suggests , deeper high - resolution imaging of m33 may prove one of the most robust ways to verify this effect . finally we present the bolometric luminosity function derived from our observations in figure [ fig : lf_mbol ] . the data cover 416 square arcseconds , avoiding the central @xmath47 . we have used the bolometric corrections from @xcite which are based on each star s @xmath53 color . here the dip in the lf at the rgb - agb transition @xmath182 is especially apparent . the termination of the lf at @xmath183 is in agreement with the youngest stars on the agb having an age @xmath3 gyr ( see section [ sec : age ] ) . fitting a power - law to the lf of the rgb ( @xmath184 ) we find a slope of @xmath185 . if instead we fit the entire luminosity function from the bright end dropoff at @xmath186 to the completeness limit at @xmath187 , we find a slope @xmath188 . @xcite measured the slope between @xmath189 , a much smaller range than ours , and found a slope of @xmath190 , in good agreement with our determination . recently @xcite has made ao observations of the nuclear regions of m33 with the 3.6 m canada - france - hawaii telescope ( cfht ) . his @xmath191 field had a total of 20 minutes exposure time per @xmath115 filter , and his images have a fwhm of @xmath192 , about twice the size of the gemini psf , which ranged from @xmath107 to @xmath108 . using his published list of photometry of stars with @xmath193 , we have matched up observations of 34 stars . the difference between these measurements is illustrated in figure [ fig : dm_compare_davidge ] . several of the fainter stars are obviously blended in davidge s images , however the agreement with the bulk of the sample is quite good . throwing out three sigma outliers , the average difference between the samples ( davidge gemini ) is @xmath194 , @xmath195 , and @xmath196 . the perfect agreement in the @xmath49-band is of course because we used the same sample of good measurements to determine our @xmath49-band photometric zero point . the comparison between our @xmath49-band lfs is shown in figure [ fig : lf_compare_davidge ] . the two lfs agree very well , both in slope and normalization . the most obvious difference is that our lf extends to @xmath197 , while the davidge lf rolls over at @xmath198 due to incompleteness . the davidge lf also has one star @xmath199 magnitude brighter than any star in our field , however this is most likely a result of our smaller area not completely sampling the brightest star population . another difference is that while we see a small dip in the lf due to the tip of the rgb at @xmath173 , this feature is not visible in the davidge lf , probably because it is smeared out due to his reduced photometric accuracy at this level . @xcite have observed the nuclear region of m33 using the wide field planetary camera 2 ( wfpc2 ) on the hubble space telescope ( hst ) . they have obtained images through the f555w ( @xmath200 ) and f814 ( @xmath201 ) filters . their optical cmd reveals a bright blue main sequence , a red supergiant plume , a very red agb , and a wide rgb . they interpret this complex cmd as indicative of a mix of young ( @xmath202 myr ) , intermediate age ( @xmath199 gyr ) , and old stellar populations ( @xmath203 gyr ) . we have matched up their optical wfpc2 observations with our infrared gemini measurements using daomaster ( part of the daophot package , * ? ? ? * ) and show the resulting @xmath204 cmd in figure [ fig : cmd_vk ] . for comparison we have overplotted the gb of the old galactic globular cluster 47 tuc ( [ fe / h ] @xmath205 ) from @xcite . this cluster gb exemplifies the region of this cmd which will be occupied by an old intermediate - metallicity stellar population . the color is fairly red , and the gb tip is well below @xmath206 . we have also overplotted the mean ridge line of 12 intermediate age magellanic cloud clusters from @xcite . the component clusters have an average swb - type of iv @xcite , an average @xmath207 value of 36 @xcite , and a mean metallicity of [ fe / h]@xmath208 based on the slope and placement of their composite @xmath209 gb . this gb exemplifies the location of an intermediate age metal poor stellar population , with the tip of the gb extending over a magnitude brighter than that of 47 tuc . lastly , we overplot a 100 myr , @xmath210 solar metallicity ( @xmath211 , @xmath212 ) isochrone from @xcite , which uses a detailed tp - agb treatment . this isochrone extends even bluer and brighter than the other two rgbs , and can explain the large number of very blue stars with @xmath213 . figure [ fig : cmd_vk ] makes it clear that , as concluded by @xcite , there is a lot going on in the central regions of m33 . there appear to be young blue main sequence stars and red supergiants , an intermediate age gb similar to the mc clusters , and an evolved gb like that of the galactic cluster 47 tuc . for a discussion of previous work on this topic see the end of section [ sec : age ] . unfortunately , degeneracies between age and metallicity make it very difficult to quantitatively describe the mix of stellar populations based on our photometric dataset alone . in section [ sec : decomposition ] we showed that the best 3-component fit to the surface brightness profile includes spheroid and disk components whose relative contributions vary significantly across our field . if this model is correct , and these two populations are sufficiently different , we may be able to detect this variation by looking for a radial dependence of the slope and extent of the luminosity function , and in the morphology of the color - magnitude diagram . to test this hypothesis , we have divided our field into four equal - area rings around the center of m33 . the results are summarized in table [ tab : radial_variations ] which lists the properties measured in each ring as well as in the entire field . the first two columns of table [ tab : radial_variations ] give the limits of each ring , chosen so that each has an area of 50 arcseconds@xmath171 ( also illustrated in figure [ fig : stellar_dist ] ) . the third column ( @xmath214 ) lists the number of stars measured in each ring . the fourth and fifth columns give the giant branch slope ( @xmath215 ) and the luminosity function power - law slope ( @xmath216 ) , both measured from @xmath217 . the last column gives the brightest star measured in each annulus as an estimate of the tip of the agb . the results for the entire frame are listed in the last row of the table , and marked @xmath126 . cccccc 3.0 & 5.0 & 484 & @xmath218 & @xmath219 & @xmath220 + 5.0 & 6.4 & 467 & @xmath221 & @xmath222 & @xmath223 + 6.4 & 7.5 & 444 & @xmath224 & @xmath225 & @xmath226 + 7.5 & 8.5 & 424 & @xmath227 & @xmath228 & @xmath229 + & 3308 & @xmath230 & @xmath176 & @xmath231 + [ tab : radial_variations ] in this section we analyze both the radial distribution of stars , as indicated by the kolmogorov - smirnov statistic ( ks - test ) , and the distribution of stellar luminosities as measured by fitting power - law luminosity functions as well as the ks - test . the radial distribution of measured @xmath49-band magnitudes are shown in figure [ fig : stellar_dist ] , which illustrates the continuum of stellar luminosities over the field . first we perform the ks - test on the radial positions of stars plotted in the optical - ir cmd ( figure [ fig : cmd_vk ] ) . here we have divided the stars into two groups : blueward of the intermediate age magellanic cloud cluster locus , which we call `` young '' , and redward of the mc cluster locus , which we call `` old '' . for both groups we consider stars only brighter than @xmath232 to minimize the effects of incompleteness . when considering all stars from @xmath47 to @xmath233 , the ks - test shows a marginal difference ( @xmath234 ) between the radial distribution of the `` young '' and `` old '' stars , however this difference arises in the very central regions , and is most likely due to incompleteness in this more crowded region . when we limit the test to stars outside @xmath235 , the radial distributions of the two groups become nearly indistinguishable ( @xmath236 ) . thus based on the stars from @xmath235 to @xmath233 in the optical - ir cmd , we would conclude that the radial distributions of `` young '' and `` old '' stars are identical . if instead , we consider our whole sample of infrared measurements , and are more selective in our choice of `` young '' stars , namely bluer than @xmath237 and brighter than @xmath238 , these stars appear to lie at larger radii than the remainder of the population . performing a ks - test on this small subsection of the cmd , from @xmath235 to @xmath233 , gives a very low probability ( @xmath239 ) that these stars have the same radial distribution as the other stars . however the observed difference arises at primarily large radii ( @xmath240 ) . if the ks - test is run from @xmath235 to @xmath32 the significance completely disappears . we mention this result as an aside because it is only evident in the outermost regions of the field , where the completeness due to dithering is difficult to calculate and the probability for systematic effects caused by the large distance from the wavefront reference source ( nucleus ) is highest . @xcite did find significantly different radial distributions of young and old stars in the central @xmath241 of m33 based on optical hst - pc observations . specifically they found that the younger pop i stars preferentially lie farther from the nucleus than the more centrally concentrated older pop ii stars . next we look at the luminosity functions measured in each of the four rings defined in table [ tab : radial_variations ] and illustrated in figure [ fig : stellar_dist ] . we fit a power - law to the rgb from @xmath242 and find that there is no significant change across the field , and that the slopes determined are all consistent with that measured for the entire field : @xmath243 . a more rigorous analysis using the ks - test verifies this general result . these tests show that based on the distribution of stellar luminosities in each of the four annuli , as well as on the radial distribution of stars of different luminosities ( binned into 1-magnitude bins ) , all stars are consistent with being drawn from the same population . @xcite obtained similar results for the inner disk of m33 . comparing agb stars brighter than @xmath173 in two regions , @xmath244 and @xmath245 , they found no significant difference between the luminosity functions except for the presence of luminous supergiants in the outer region . the cmds of each of the four rings are displayed in figure [ fig : cmd_4radii ] . the radial limits of each are at the top of each cmd , and the number of objects found in each ring is printed in the bottom right corner . using the same technique which we used to estimate the mean metallicity of the stellar population in section [ sec : cmds ] , we now measure the slope of the gb in each ring . as before , the iterative linear fit is only to data with @xmath246 , and ignores three sigma outliers . the results of the gb fitting are listed in column 4 of table [ tab : radial_variations ] . we find that based on the slope of the giant branch , the mean metallicity of each annulus is consistent with that of the entire field . this is similar to what has been observed in the inner bulge of our milky way ( @xmath247 pc ) , which shows no evidence for a gradient along the major or minor axes @xcite . although we have not detected a gradient in the stellar properties in the inner @xmath248 , we have nonetheless helped fill in the gap between the metallicity gradient observed in the disk of m33 @xcite , and low abundance measurements for the nuclear region , most recently [ fe / h]@xmath249 ( @xmath250 ) based on the co index by @xcite . as @xcite has pointed out , meaningful photometry can only be obtained for stars brighter than the luminosity contained in each resolution element . we have calculated the enclosed luminosity in m33 for five different imaging resolutions ( @xmath104 , @xmath192 , @xmath251 , @xmath252 , and @xmath47 ) , assuming that the size of the resolution element is defined by the fwhm . figure [ fig : blendinglimits ] shows the corresponding limiting @xmath49-band magnitude as a function of surface brightness . this plot shows that at @xmath253 , the surface brightness over most of our field , the faintest stars which we can accurately measure with our @xmath254 resolution have @xmath255 . this is in very good agreement with the observed limit of our photometry ( e.g. figure [ fig : cmd ] ) . even more useful perhaps , is a plot of the limiting magnitude due to blending as a function of spatial position . for symmetric surface brightness distributions , such as exist around m33 , the conversion is relatively simple . using our three component surface brightness model ( section [ sec : decomposition ] ) , we have transformed from generic surface brightness to the radial distance from the center of m33 measured in arcminutes . the resulting relations for the same five different resolutions are shown on the right side of figure [ fig : blendinglimits ] . thus for our data , with a resolution of @xmath256 , the faintest stars which we can accurately measure ( against blending ) at @xmath95 arcseconds from the center of m33 have @xmath257 . if we try to measure stars at a distance of @xmath47 , blending will limit our measurements to only stars brighter than @xmath258 , a full magnitude brighter than we could accurately measure at @xmath95 . another simple calculation advocated by @xcite is the estimation of the number of stars in each evolutionary stage per resolution element . this allows one to estimate the severity of blending in any observation . this calculation has several parameters which have a weak dependence on the age , metallicity , and imf of the stellar population . the ratio of total ( @xmath259 ) to @xmath49-band luminosity ( @xmath260 ) , and the specific evolutionary flux , @xmath261 , are two such parameters . for our calculation we used @xmath262 , and @xmath263 stars yr@xmath17 l@xmath264@xmath17 , numbers suitable for a 15 gyr old , solar - metallicity population . the results of this calculation for stars within one magnitude of the rgb tip ( rgbt ) are displayed in figure [ fig : nrgbt_reselement ] . in the left panel we show the number of rgbt stars per resolution element as a function of surface brightness for five different imaging resolutions . the number of blends of two rgbt stars on a frame can be estimated as the square of the number per resolution element ( for @xmath265 ) multiplied by the number of resolution elements in the frame @xcite . if the number of rgbt stars per resolution element is greater than one , it should be clear that photometry of stars at or below the level of the rgbt is impossible . at this point only stars several magnitudes brighter than the rgbt can be measured with any accuracy , but then one must face the question of whether the objects measured are real or just blends of many stars , each of which can be as bright as the tip of the rgb . @xcite have performed simulations of the blending in their nicmos observations of globular clusters in m31 . they show that severe blending can easily create objects which are several magnitudes brighter than any star in the parent population . thus one must be very careful when interpreting bright objects measured in a very crowded fields , i.e. where n(rgbt ) per resolution element is greater than one . it might be illuminating to examine a few previous studies of the central regions of m33 keeping figures [ fig : blendinglimits ] and [ fig : nrgbt_reselement ] in mind . the most critical , at least in terms of our photometric calibration , are the observations of @xcite discussed in section [ sec : data_reduction ] . plotting his @xmath192 resolution on both figures , we see that his observations are not significantly affected by blending . the right side of figure [ fig : nrgbt_reselement ] shows that davidge stays below 1 rgbt star per resolution element until only @xmath46 from the nucleus , which he ( for the most part ) steers clear of . assuming the average distance of his field from the nucleus is @xmath266 , figure [ fig : blendinglimits ] shows that the limit of accurate photometry ( against blending ) is @xmath267 . since his estimated completeness limit is @xmath268 , his photometry is therefore limited by photon noise and not blending . checking the observations of @xcite , we see that their infrared @xmath115 observations of the central @xmath269 of m33 had @xmath252 seeing . they divided their observations into a `` central core region '' going from @xmath13 to @xmath14 , and an `` inner disk '' region , spanning from a radius of @xmath14 to @xmath270 . looking at the @xmath252 resolution curve on figure [ fig : blendinglimits ] , we find that at @xmath14 , their photometric limit imposed by blending is @xmath271 , and figure [ fig : nrgbt_reselement ] shows that at this distance there is @xmath199 rgbt star per resolution element . since the number of detected stars in their core region abates at @xmath272 , they are not trying to go much deeper than blending allows . however , the high density of stars as bright as the rgb tip , and the fact that out to @xmath62 we only find stars as bright as @xmath133 , suggests that some of the brighter ( @xmath273 ) , bluer [ @xmath274 objects they detected in the central region may be blends . the observations of @xcite are in a similar situation in terms of blending . with @xmath252 resolution they measured stars as bright as @xmath275 in the inner @xmath276 of m33 . however , as figure [ fig : nrgbt_reselement ] shows , at @xmath276 with @xmath252 resolution , there are @xmath277 stars within one magnitude of the tip of the rgb per resolution element . this certainly does not guarantee that these bright objects are blends , but merely shows that the potential for blending is high . we have used gemini - north to study the stellar populations in the central regions of m33 . the surface brightness profiles from @xmath4 to @xmath5 , formed from the combination of our data and those of @xcite , show that the data need to be modeled using a three - component , core + spheroid + disk model . the best - fit parameters are listed in table [ tab : decomposition ] . these high - resolution observations allow us to accurately measure individual stars to @xmath278 . artificial star tests ( [ sec : artstartests ] ) show that our completeness is relatively uniform across the field ( 50% at @xmath232 ) , although within @xmath47 from the nucleus the completeness is dramatically lower due to severe crowding . artificial fields are used to understand the observational effects associated with adaptive optics measurements in crowded fields . based on the slope of the giant branch in the infrared color magnitude diagram , we estimate the mean metallicity to be @xmath1 ( [ sec : metallicity ] ) . using the bolometric luminosities and density of stars on the agb , we hypothesize two bursts of star formation ; at @xmath3 and @xmath2 gyr ago ( [ sec : age ] ) . we note however , that this component of young stars may have influenced our metallicity estimate due to the sensitivity of the gb slope on age . the stellar luminosity function in m33 is shown to be significantly different from that measured in the galactic bulge as viewed through baade s window ( [ sec : lfs ] ) . the difference in their maximum luminosities is due to differences in ages of the two regions . we speculate that this is also the origin of their different slopes as well . in section [ sec : comparison ] we compare our data with previous observations . recent work by @xcite is in good agreement with our data , although our observations go @xmath279 magnitudes deeper ( [ sec : davidge ] ) . we also combine our data with optical hst - wfpc2 measurements @xcite , and present the optical - ir cmd in [ sec : mighell ] . this cmd clearly shows that the central regions of m33 are composed of young , intermediate , and old aged stellar populations . dividing the inner @xmath280 ( @xmath281 pc ) into equal area rings around the nucleus , we look for radial variations in the stellar properties ( [ sec : radial_variations ] ) . however , based on the distribution of stellar luminosities and the morphology of the cmd , we find that all stars are consistent with being drawn from a single population . in the last section ( [ sec : blending ] ) we perform calculations to estimate the severity of blending at various imaging resolutions and locations in m33 . using the formulations of @xcite and our composite surface brightness profile , these calculations call into question some previous claims of very luminous stars in the central regions of m33 . this paper was based on observations obtained at the gemini observatory , which is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the particle physics and astronomy research council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) and conicet ( argentina ) . observations used the adaptive optics system hokupaa / quirc , developed and operated by the university of hawaii adaptive optics group , with support from the national science foundation . many thanks to the gemini queue observers , peter stetson for supplying and helping us with his photometry package , allframe , and to ken mighell for providing us with his hst - wfpc2 images and photometry lists before publication . thanks to michael regan for helping with his surface brightness decomposition . valuable comments from ken mighell , darren depoy , dante minniti , and an anonymous referee were also greatly appreciated .

we prove a generalization to the context of real geometry of an intersection formula for the vanishing cycle functor , which in the complex context is due to dubson , l , ginsburg and sabbah ( after a conjecture of deligne ) . it is also a generalization of similar results of kashiwara - schapira , where these authors work with a suitable assumption about the micro - support of the corresponding constructible complex of sheaves . we only use a similar assumption about the support of the corresponding characteristic cycle so that our result can be formulated in the language of constructible functions and lagrangian cycles .

math0201300

in this paper we give a proof ( of a generalization to the context of _ o - minimal structures _ on the real field , or to _ analytic ( nash ) geometric categories _ ) of the following intersection formula : [ thm : int1 ] let @xmath0 be an m - dimensional real analytic manifold and @xmath1 a subanalytic @xmath2-function . consider on @xmath0 a bounded subanalytically constructible complex @xmath3 of sheaves of vector spaces ( over a base - field @xmath4 ) , with finite dimensional stalks @xmath5 ( @xmath6 ) . suppose that the intersection of @xmath7 and the support @xmath8 of the characteristic cycle of @xmath3 is contained in a compact subanalytic subset @xmath9 , with @xmath10 . then one has @xmath11\cap [ cc({\cal f})]\ ; \bigr ) \:.\ ] ] here we use the following notations : 1 . @xmath12 is the usual _ euler characteristic _ ( note that @xmath13 has finite dimensional cohomology , since @xmath14 , with @xmath15 the natural projection , is a compact subanalytic subset ) . 2 . @xmath16 \in h^{m}_{|cc({\cal f})|}(t^{*}m,\pi^{-1}or_{m})$ ] is the _ characteristic cycle _ of @xmath3 ( as in @xcite , and see the next section ) , with @xmath17 the _ orientation sheaf _ on @xmath0 . @xmath18\in h^{m}_{\sigma_{f}}(t^{*}m , or_{t^{*}m / m})$ ] corresponds to @xmath19 ( as in ( * ? ? ? 9.3.5 ) , with @xmath20\;$ ] ) . 4 . the _ intersection number _ @xmath21 is defined by @xmath22 here we use @xmath23 , and the above cup - product is the composition of the usual cup - product with support and the natural maps @xmath24 this intersection formula is an important and useful generalization of ( * ? ? ? * thm.9.1 , p.207 ) and ( * ? ? ? * thm.9.5.6 , p.386 ) . note that we only assume , that the intersection of @xmath25 and the _ support _ @xmath8 of the characteristic cycle of @xmath3 is contained in a compact subanalytic set , and this is the reason why our formula is a result about _ lagrangian cycles_. the results of loc.sit . are formulated under suitable assumptions on the _ micro - support _ of @xmath3 , which is in general much bigger than @xmath8 . the characteristic cycle @xmath26 of @xmath3 depends only on the _ constructible function _ @xmath27 ( compare with ( * ? ? ? * thm.9.7.11 ) and ( * ? ? ? * subsec.5.0.3 ) ) . therefore our results can also be stated in terms of the associated constructible function @xmath28 . in this context , one can only use the support of the characteristic cycle @xmath29 as an invariant of the constructible function @xmath28 ( the micro - support @xmath30 depends on the sheaf complex @xmath3 , and _ not _ ( ! ) only on the corresponding constructible function ) . our proof of theorem [ thm : int1 ] is a modification of the proof of ( * ? ? ? * thm.9.5.6 ) . we use the result of @xcite ( due in the subanalytic context to bierstone , milman and pawlucki ) , that @xmath31 is the zero - set of a non - negative subanalytic @xmath2-function @xmath32 . then there exists a relatively compact open neighborhood @xmath33 of @xmath34 in @xmath0 , and a @xmath35 such that @xmath36 is proper ( since @xmath37 is compact ) . after restriction to @xmath33 we can assume @xmath38 . then @xmath39 calculates for @xmath40 ( i.e. for @xmath41 and @xmath42 sufficiently small , with @xmath42 small compared to @xmath41 ) the euler characteristic of the relative cohomology of a `` tube '' @xmath43 modulo the _ left milnor fiber _ @xmath44 ( compare @xcite ) . in terms of the constructible function @xmath28 , this euler characteristic can also be rewritten as ( compare @xcite and @xcite ) : @xmath45 here we use for a _ compact _ subanalytic subset @xmath46 the notation @xmath47 the _ unique _ @xmath48-linear map on the group @xmath49 of @xmath48-valued subanalytically constructible functions with _ compact _ support such that @xmath50 for @xmath51 a compact subanalytic subset . + for the special case @xmath52 we get @xmath53 so that theorem [ thm : int1 ] implies the following generalization of the classical _ poincar - hopf index formula _ , which corresponds to the case @xmath54 for a @xmath0 a compact manifold ( with @xmath55 the zero - section of @xmath56 ) : [ cor : globalindex ] consider a constructible complex @xmath3 as in theorem [ thm : int1 ] and denote by @xmath28 the corresponding constructible function @xmath57 . assume @xmath3 or @xmath28 has a compact support . then @xmath58\cap [ cc({\cal f})]\ ; \bigr ) \quad \text{or } \quad \int_{m}\ , \alpha \,d\chi = \sharp\bigl(\ ; [ t^{*}_{m}m]\cap [ cc(\alpha)]\ ; \bigr ) \:.\ ] ] this global index formula goes back to dubson ( * ? ? ? * thm.2 , p.115 ) in the complex , and to kashiwara ( * ? ? ? * thm.8.1 , p.205 ) in the real context . other presentations can be found in ( * ? ? ? * thm.1.5 , p.832 ) , ( * ? ? ? * thm.9.1 , p.382 ) , ( * ? ? ? * cor.9.5.2 , p.384 ) and ( * ? ? ? * eq.(5.30 ) , p.298 ) . + another important special case of our intersection formula ( [ eq : int1 ] ) is the case @xmath59 given by a point @xmath60 , with @xmath61 and @xmath62 . in this case we can use for @xmath32 the usual distance function ( in suitable coordinates @xmath63 ) . then we get ( cf . * thm.9.5.6 , p.386 ) ) : @xmath64\cap [ cc({\cal f})]\ ; \bigr)\ ] ] and @xmath65\cap [ cc(\alpha)]\ ; \bigr ) \:,\ ] ] with @xmath66 the corresponding _ local intersection number _ and @xmath67 the _ local left milnor fiber _ of @xmath68 in @xmath69 . we get in particular : @xmath70 in another paper , we will use the formula ( [ eq : int3 ] ) ( together with a specialization result for lagrangian cycles ) for a proof of the non - characteristic pullback formula ( * ? ? ? * prop.9.4.3 , p.378 ) under the _ weaker _ assumption , that the map is only non - characteristic with respect to the _ support _ @xmath8 of the characteristic cycle of @xmath3 . as a further example , let us consider ( locally ) the case @xmath71 a ( real analytic ) distance function to the point @xmath69 , i.e. @xmath72 with @xmath73 . by the _ curve selection lemma _ there exists an open neighborhood @xmath33 of @xmath69 in @xmath0 , with @xmath74 this will be explained later on in terms of stratification theory ( and compare with ( * ? ? ? * prop.8.3.12 , p.332 ) ) . so if we work on the manifold @xmath33 , then we can apply ( [ eq : int2 ] ) and ( [ eq : int2cf ] ) with @xmath75 : [ cor : localindex ] @xmath76\cap [ cc({\cal f})]\ ; \bigr ) \quad \text{and } \quad \alpha(x ) = \sharp_{dr_{x}}\bigl(\ ; [ dr(u)]\cap [ cc(\alpha)]\ ; \bigr ) \:.\ ] ] so ( [ eq : localindex ] ) describes an inversion formula for reconstructing the constructible function @xmath28 out of the lagrangian cycles @xmath16 $ ] and @xmath77 $ ] ( compare ( * ? ? ? * thm.2 , p.115 ) , ( * ? ? ? * thm.8.3 , p.205 ) , ( * ? ? ? * eq.(9.5.8 ) , p.386 ) and ( * ? ? ? * eq.(5.30 ) , p.298 ) ) . + theorem [ thm : int1 ] also gives a purely _ real _ proof of the corresponding intersection formula in complex geometry : [ cor : int2 ] let @xmath0 be an m - dimensional complex analytic manifold and @xmath78 a holomorphic function . consider on @xmath0 a bounded complex analytically constructible complex @xmath3 of sheaves of vector spaces , with finite dimensional stalks @xmath5 ( @xmath6 ) . suppose that the intersection of @xmath79 and the support @xmath8 of the characteristic cycle of @xmath3 is contained in a compact analytic subset @xmath9 , with @xmath80 . then one has @xmath81\;{\cal f})\bigr ) = \sharp\bigl(\ ; [ dh(m)]\cap [ cc({\cal f})]\ ; \bigr ) \:.\ ] ] here we use the notation @xmath56 for the _ holomorphic _ cotangent bundle , and @xmath82 is the _ vanishing cycle functor _ of deligne . the holomorphic section @xmath83 of @xmath56 corresponds under the natural isomorphism @xmath84 to the section @xmath85 of @xmath86 , with @xmath86 the _ real _ cotangent bundle of the underlying real manifold @xmath87 . if we use the induced _ complex orientation _ of @xmath86 , then the class @xmath88\in h^{2m}_{\sigma_{f}}(t^{*}m , or_{t^{*}(m^{r})/m^{r}})\simeq h^{2m}_{\sigma_{h}}(t^{*}m,{\mbox{$\mathbb{z}$}})\ ] ] for @xmath89 corresponds under poincar duality to the _ fundamental class _ in borel - moore homology of the complex manifold @xmath90 . + then the corollary follows from theorem [ thm : int1 ] and the isomorphism ( cf . * lem.1.3.2 , p.70 ) and ( * ? ? ? * cor.1.1.1 , p.32 ) ) : @xmath91\;{\cal f } ) \simeq r\gamma(k , r\gamma_{\{re(h)\geq 0\}}{\cal f})\:.\ ] ] or in terms of the constructible function @xmath28 : @xmath92 and @xmath93\;{\cal f})\bigr ) = \int_{k}\ , \alpha \,d\chi - \int_{m_{h}}\ , \alpha \,d\chi \:,\ ] ] with @xmath94 the _ milnor fiber _ of the holomorphic function @xmath95 . + this holomorphic intersection formula for the vanishing cycle functor is due to dubson ( * ? ? ? * thm.1 , p.183 ) , ginsburg ( * ? ? ? * prop.7.7.1 , p.378 ) , l ( * ? ? ? * thm.4.1.2 , p.242 ) and sabbah ( * ? ? ? * thm.4.5 , p.174 ) . for a discussion of the history of this holomorphic intersection formula we recommend the paper @xcite . but most of these references are in the language of ( regular ) _ holonomic d - modules _ , or _ perverse sheaves _ ( with respect to middle perversity ) . so the assumption on the intersection for a holomorphically constructible complex of sheaves corresponds to an assumption on the micro - support ( ( * ? ? ? * thm.11.3.3 , p.455 ) ) . only the result of sabbah ( * ? ? ? * thm.4.5 , rem.4.6 , p.174 ) is in terms of the underlying ( complex analytic ) lagrangian cycle , which corresponds therefore to an assumption on the intersection of @xmath90 and the support of the corresponding characteristic cycle ! similarly , only this reference @xcite , and in the real subanalytic context also ( * ? ? ? * thm.9.1 , p.207 ) consider the case of an intersection in a _ subset @xmath96 . + the other references deal only with the case @xmath59 given by a point @xmath60 . in this special case we get back a formula conjectured by deligne ( with @xmath61 and @xmath62 ) : @xmath97\;{\cal f})_{x}\bigr ) = \sharp_{dh_{x}}\bigl(\ ; [ dh(m)]\cap [ cc({\cal f})]\ ; \bigr)\ ] ] and @xmath98\cap [ cc(\alpha)]\ ; \bigr ) \:,\ ] ] with @xmath99 the _ local milnor fiber _ of @xmath95 in @xmath69 . we get in particular : @xmath100\;{\cal f})_{x}\bigr ) = 0\:.\ ] ] we used this formula in our paper @xcite for a short proof of a formula of brasselet , l and seade for the _ euler obstruction _ ( ( * ? ? ? * thm.3.1 ) ) . + as another application of the formula ( [ eq : vcyc ] ) let us consider the following example : @xmath101 , with @xmath102 a _ closed _ complex analytic submanifold . consider a holomorphic function germ @xmath103 such that @xmath104 has in @xmath105 a _ complex morse critical point_. then we get @xmath106 $ ] , with the orientation convention of @xcite . if we denote by @xmath107_{c}\in h^{2m}_{t^{*}_{n}m}(t^{*}m,{\mbox{$\mathbb{z}$}})\ ] ] the class , which corrresponds under poincar duality to the fundamental class of the complex manifold @xmath108 , then we get of course @xmath109\cap [ t^{*}_{n}m]_{c}\ ; \bigr)\;= 1\:.\ ] ] but @xmath110 has in @xmath69 a _ real morse critical point _ of index @xmath111 . we therefore get by ( * ? ? ? * eq.(9.5.18 ) , p.388 ) for @xmath89 : @xmath112\cap [ cc(k_{n})]\ ; \bigr)\;= ( -1)^{n}\:.\ ] ] and this implies @xmath107_{c } = ( -1)^{n}\cdot [ t^{*}_{n}m ] \in h^{2m}_{t^{*}_{n}m}(t^{*}m,{\mbox{$\mathbb{z}$}})\:.\ ] ] so one has to be very careful about _ orientation conventions _ and the definition of _ characteristic cycles _ that one uses . here we follow the notations and conventions of @xcite . note that there are many approaches to this subject , often using quite different techniques and conventions . we recall in the next section a detailed comparison , which is worked out in ( * ? ? ? * subsec.5.0.3 ) . + let us only recall , that the theory of characteristic cycles has its origin in the theory of _ holonomic d - modules _ and the _ local index formula _ of kashiwara ( * ? ? ? * thm . on p.804 ) ( compare ( * ? ? ? * thm.6.3.1 , p.127 ) , ( * ? ? ? * thm.2 , p.574 ) and ( * ? ? ? * thm.11.7 , p.393 ) ) . this corresponds to corollary [ cor : localindex ] for @xmath113 the _ solution complex _ of a holonomic @xmath114-module @xmath115 . later on , the theory of characteristic cycles was extended to constructible functions and sheaves in the context of real geometry . first in the _ subanalytic _ context by kashiwara @xcite , and independently also by fu @xcite . a simple approach in the _ semialgebraic _ context is sketched in @xcite , and the extension to _ o - minimal structures _ and _ analytic geometric categories _ has been worked out in @xcite . all these approaches in real geometry are based on a suitable _ morse theory _ , e.g. the _ micro - local sheaf theory _ of kashiwara - schapira @xcite , or the _ stratified morse theory _ of goresky - macpherson @xcite . + for a detailed comparison and translation of these two different theories in the framework of _ morse theory for constructible sheaves _ , including a _ geometric _ introduction to characteristic cycles of constructible functions and sheaves , we refer to our book @xcite . this language will be used in the proof of theorem [ thm : int1 ] . moreover , we use the following important result about the behaviour of the _ support _ @xmath8 of the characteristic cycle of @xmath3 under a suitable intersection : [ thm : int2 ] let @xmath0 be a real analytic manifold and @xmath116 a subanalytic @xmath2-function . consider on @xmath0 a bounded subanalytically constructible complex @xmath3 of sheaves of vector spaces , with finite dimensional stalks @xmath5 ( @xmath6 ) . assume @xmath3 is constructible with respect to a subanalytic whitney b - regular stratification ( i.e. the cohomology sheaves of @xmath3 are locally constant on the subanalytic strata @xmath117 ) such that the set of nondegenerate covectors ( with respect to this stratification ) is dense in all fibers of the projection @xmath118 ( for all strata @xmath117 ) . let @xmath41 be a regular value of @xmath32 such that @xmath119 is _ transversal _ to all strata @xmath117 . then one has for the open inclusion @xmath120 the following estimate for the support @xmath121 of the characteristic cycle of @xmath122 : @xmath123 the assumption on our stratification is for example satisfied for a subanalytic _ @xmath124-stratification _ in the sense of ( * ? ? ? * def.8.3.19 , p.334 ) ( see ( * ? ? ? * cor.8.3.24 , p.336 ) ) . this _ @xmath124-condition _ is by @xcite equivalent to the _ w - condition _ of verdier , and such a _ w - regular _ stratification can always be used in the context of `` geometric categories '' as in the last section of this paper ( cf . but it is not known ( at least to the author ) , if the _ w - condition _ implies the assumption on our stratification ( used in the theorem ) in this more general context . kashiwara - schapira used in their proof of ( * ? ? ? * thm.9.5.6 , p.386 ) a similar result in terms of the _ micro - support _ of the corresponding constructible complex of sheaves ( cf . * prop.8.4.1 , p.338 ) and ( * ? ? ? * prop.5.4.8.(a ) , p.233 ) ) . we will give a proof based on our results on _ morse theory for constructible sheaves _ @xcite , especially on an explicit result about suitable `` stratified spaces with boundary '' ( * ? ? ? * thm.5.0.2 , p.286 ) . this approach works also , if one uses the supports of the corresponding characteristic cycles . it implies a similar estimate for @xmath125 ( with @xmath126 ) , from which we deduce ( [ eq : est * ] ) by _ duality_. in this section we explain the construction of the characteristic cycles . as an application we give a proof of theorem [ thm : int2 ] in a much more general context , which applies especially to complexes of sheaves constructible with respect to suitable `` geometric categories '' . + let @xmath127 be a closed subset of a smooth manifold @xmath0 . in this paper , a smooth manifold @xmath0 has by definition a countable topology . we also assume ( for simplicity ) , that the dimension of the connected components of @xmath0 is bounded . a _ stratification _ of @xmath127 is a filtration @xmath128 of @xmath127 : @xmath129 by closed subsets such that @xmath130 ( @xmath131 ) is a smooth submanifold of @xmath0 . the connected components of the @xmath132 are by definition the _ strata _ of this stratification . here we fix a degree @xmath133 of smoothness ( with @xmath134 real analytic ) . ( sub)manifold and smooth map or function always means a @xmath135-(sub)manifold and @xmath135-smooth map or function . we assume that the stratification @xmath128 is _ whitney b - regular _ : 1 . if @xmath136 ( for @xmath137 ) and @xmath138 are sequences converging to @xmath139 such that the tangent planes @xmath140 converge to some limiting plane @xmath141 , and the secant lines @xmath142 ( with respect to some local coordinates ) converge to some limiting line @xmath143 , then @xmath144 . consider on @xmath127 a bounded complex @xmath3 of sheaves of vector spaces ( over a base - field @xmath4 ) , with finite dimensional stalks @xmath5 ( @xmath145 ) . we assume that @xmath3 is _ constructible _ with respect to the stratification @xmath128 , i.e. the cohomology sheaves of @xmath3 are _ locally constant _ on all @xmath132 . + from now on we assume @xmath146 and @xmath147 . then we have by ( * ? ? ? * cor.4.0.3 , p.216 ) and ( * ? ? ? * prop.4.1.2 , p.227 ) the following estimate for the _ micro - support _ @xmath30 of @xmath3 in the sense of ( * ? ? ? * def.5.1.2 , p.221 ) : @xmath148 here @xmath149 denotes the conormal bundle of a locally closed submanifold @xmath117 of @xmath127 . by whitney regularity , @xmath150 is a closed subset of @xmath151 ( this is in fact equivalent to _ a - regularity _ ) . + moreover , we have by ( * ? ? ? * prop.4.0.2 , p.219 ) that @xmath3 is _ cohomologically constructible _ in the sense of ( * ? ? ? * def.3.4.1 , p.158 ) . therefore we can define as in @xcite the following chain of morphisms : @xmath152 @xmath153 @xmath154 with @xmath155 the projection , @xmath156 the diagonal embedding and @xmath157 the dualizing complex on @xmath127 . since @xmath127 is a manifold , one has by ( * ? ? ? * prop.3.3.2 , p.152 ) : @xmath158\ ; , \quad \text{with $ d:=dim(x)$.}\ ] ] here we assume that @xmath127 is _ pure - dimensional _ ( or one should interpret @xmath159 as a locally constant function ) . + for the definition of the above chain of morphisms we use the following properties : 1 . @xmath160 ( ( * ? ? ? * prop.4.4.2(i ) , p.202 ) ) . @xmath161 has its support in @xmath30 ( ( * ? ? ? * cor.5.4.10.(ii ) , p.234 ) ) , and therefore by ( [ eq : est.musupp ] ) also in @xmath150 . 3 . @xmath162 ( ( * ? ? ? * def.4.4.1.(iii ) , p.202 ) and ( * ? ? ? * prop.3.4.4 , p.159 ) ) . here we can apply ( * ? ? ? * prop.3.4.4 ) , since @xmath3 is cohomologically constructible . we finally use the natural morphisms @xmath163 , and @xmath164 , together with @xmath165 . [ def : cc ] the image of @xmath166 hom@xmath167 in @xmath168 is called the _ characteristic cycle _ @xmath26 of @xmath3 . this definition extends ( * ? ? ? * def.9.4.1 , p.377 ) to our context . the same construction works for a whitney _ a - regular stratification _ @xmath128 , if we have the estimate ( [ eq : est.musupp ] ) , and the property that @xmath3 is cohomologically constructible , e.g. @xmath128 is ( locally ) _ c - regular _ in the sense of bekka @xcite ( compare with @xcite ) . we now explain the calculation of @xmath26 in terms of our ( stratified ) _ morse theory for constructible sheaves _ @xcite . + consider the set of _ nondegenerate _ covectors with respect to our whitney b - regular stratification @xmath128 ( see ( * ? ? ? * def.5.1.2 , p.308 ) and ( * ? ? ? * def.1.8 , p.44 ) ) : @xmath169 choose a smooth function germ @xmath170 , with @xmath171 and @xmath172 . take a _ normal slice _ @xmath173 at @xmath69 , i.e. a locally closed submanifold @xmath173 of @xmath127 , with @xmath174 such that @xmath173 intersects @xmath175 transversally in @xmath69 ( cf . * def.5.0.2(2 ) , p.277 ) and ( * ? ? ? * def.1.4 , p.41 ) ) . then the isomorphism - class of @xmath176 is the ( sheaf theoretic ) _ `` normal morse datum '' _ of @xmath68 in @xmath69 with respect to @xmath3 , i.e. it is the cohomological counterpart of the corresponding _ normal morse data _ of goresky - macpherson ( * ? ? ? * def.3.6.1 , p.65 ) . by ( * thm.5.0.1(2 ) , p.278 ) , this isomorphism - class depends _ only _ on @xmath177 and is _ locally constant _ on @xmath178 . by ( * prop.5.0.2 , p.279 ) we have a distinguished triangle @xmath179 @ > > > nmd({\cal f},f , x ) @ > > > { \cal f}_{x } @ > [ 1 ] > > \ : , \end{cd}\ ] ] with @xmath180 the _ `` lower halflink '' _ of @xmath68 in @xmath69 ( ( * ? ? ? * def.3.9.1 , p.66 ) ) . in the notation of this paper , this @xmath180 is just the intersection of @xmath127 with a _ local left milnor fiber _ of @xmath181 in @xmath69 . but @xmath182 is constructible with respect to an induced whitney b - regular stratification of @xmath180 . then @xmath183 has _ finite dimensional _ cohomology , since @xmath180 is compact ( cf . * rem.4.2.2 , p.245 ) and ( * ? ? ? * thm.3.5 , p.70 ) ) . by the above distinguished triangle , this is also true for @xmath184 , with @xmath185 so we can associate to a connected component @xmath186 of @xmath178 the integer ( for @xmath187 ) : @xmath188 if we work in the context of `` geometric categories '' as in the next section , with @xmath68 a _ definable _ function germ , then this multiplicity @xmath189 can also be expressed in terms of the _ constructible function _ @xmath27 ( compare ( * ? ? ? * eq.(5.16 ) , p.290 ) ) : @xmath190 take now an open subset @xmath191 in @xmath151 , with @xmath192 ( e.g. the complement of the set @xmath193 of _ degenerate _ covectors ) . then one gets for the image of @xmath26 under the natural map @xmath194 the formula @xmath195\:.\ ] ] here @xmath196 $ ] is defined as in @xcite , i.e. it is the image of the class @xmath197 \in h^{d}_{t_{x^{s}}^{*}x}(\omega,\pi^{-1}or_{x})$ ] of loc.sit . ( with @xmath198 ) under the projection @xmath199 where the product is over all @xmath4 with @xmath200 . + this follows as in @xcite from our identification ( * ? ? ? * eq.(5.52 ) , p.318 ) of the _ `` normal morse datum '' _ @xmath184 of @xmath68 in @xmath69 with the _ `` local type of @xmath3 in @xmath177 '' _ in the sense of kashiwara - schapira ( compare @xcite , p.274 ) and ( * ? ? ? * def.7.5.4 , p.311 ) ) . + assume @xmath201 so that the above map of local cohomology groups is injective . then the characteristic cycle @xmath26 is uniquely determined by the euler characteristics @xmath189 of the `` normal morse data '' of @xmath3 . this applies for example , if @xmath202 has a stratification with all strata of dimension @xmath203 ( e.g. in the context of `` geometric categories '' as in the next section ) . it would also be enough ( compare with the proof of ( * ? ? ? * prop.9.2.2.(i ) , p.367 ) ) , that @xmath202 has locally this property ( i.e. each point in @xmath202 has an open neighborhood @xmath33 in @xmath151 such that @xmath204 has such a stratification ) . by ( [ eq : multi ] ) and ( [ eq : chinmd2 ] ) we get in this case an easy _ geometric _ description of the characteristic cycle @xmath205 , or more precisely , of @xmath206 . and one should ask if one can use this description as a _ definition _ of the characteristic cycle @xmath205 . the main problem is then to show that this is a _ cycle _ coming from @xmath207 the sophisticated definition [ def : cc ] seems to be the only one working for a general _ whitney b - regular _ stratification . but in the _ subanalytic _ context , other more _ geometric _ approaches to this question are due to kashiwara ( * ? ? ? * thm.4.1 , p.199 ) and fu ( compare with ( * ? ? ? * def.4.1 , p.856 ) and ( * ? ? ? * thm.4.7 , p.865 ) ) , where @xcite uses the language of `` geometric measure theory '' . a `` translation '' of the last approach into `` geometric categories '' and a _ specialization result in homology _ is worked out in ( * ? ? ? * thm.5.0.3 , p.296 ) and ( * ? ? ? * subsec.5.2.2 ) . suppose that @xmath202 has ( locally ) such a stratification , and that the set @xmath178 of nondegenerate covectors is _ dense _ in @xmath150 . then one gets the following `` explicit '' description of the _ support _ @xmath8 of the characteristic cycle of @xmath3 : @xmath208 the micro - support @xmath30 contains by definition the closure @xmath209 of all @xmath186 such that the corresponding `` normal morse datum '' @xmath210 is not isomorphic to @xmath211 . moreover , in the context of `` geometric categories '' , @xmath30 is _ exactly _ given as the union of these @xmath209 ( ( * ? ? ? * prop.5.0.1 , p.279 ) ) . we see in particular , that in general @xmath8 is much smaller than @xmath30 . suppose that the set @xmath178 of nondegenerate covectors is dense in @xmath150 , and that @xmath202 has ( locally ) a stratification with all strata of dimension @xmath203 such that the frontier @xmath212 of @xmath186 is a union of strata of @xmath202 ( for all @xmath213 ) . then all morphisms in the following commutative diagram are injective : @xmath214 moreover , there exists a _ unique _ class @xmath16\in h^{d}_{|cc({\cal f})|}(t^{*}x,\pi^{-1}or_{x})$ ] , with image @xmath215 , and image @xmath216 \in h^{d}_{\omega\cap |cc({\cal f})|}(\omega,\pi^{-1}or_{x})\:.\ ] ] this is the cohomology class that we used in the introduction ( see the next section , and compare with @xcite ) . for the convenience of the reader , we also compare our definition of the characteristic cycle @xmath205 with the other conventions used in the literature ( compare with ( * ? ? ? * subsec.5.0.3 . ) for the details ) . + for simplicity , we assume that @xmath0 is _ oriented _ ( as in most of the following references ) , i.e. an isomorphism @xmath217 has been chosen . moreover , we _ orient _ @xmath56 by using first this orientation of the base @xmath0 , and then the induced ( real dual ) orientation of the fibers @xmath218 ( as in @xcite ) . this differs by the factor @xmath219 from the _ symplectic _ orientation of @xmath56 ( used in @xcite ) . in particular , for @xmath0 a complex analytic manifold of complex dimension @xmath220 , this agrees only up to the factor @xmath221 with the complex orientation of @xmath56 ! let @xmath222 $ ] be the _ fundamental class _ of this oriented manifold in _ borel - moore homology _ , and denote by @xmath223 the _ antipodal map _ ( i.e. multiplication by @xmath224 on the fibers ) . * by definition @xmath26 corresponds to the characteristic cycle used in @xcite . * @xmath225 $ ] corresponds to the characteristic cycle used in @xcite . * @xmath226\ ; \bigr)$ ] corresponds to the characteristic cycle used in @xcite . assume @xmath0 is a complex analytic manifold of complex dimension @xmath220 , with @xmath222_{c}$ ] the _ fundamental class _ of the complex manifold @xmath56 . let @xmath3 or @xmath28 be complex analytically constructible . * @xmath225_{c}$ ] corresponds to the characteristic cycle used in @xcite . * @xmath227_{c}$ ] corresponds to the characteristic cycle used in @xcite . * @xmath225_{c}$ ] , with @xmath228 or @xmath229 , corresponds to the characteristic cycle of the _ holonomic d - module _ @xmath115 used in @xcite . * @xmath230_{c}$ ] , with @xmath231 $ ] , corresponds to the characteristic cycle of the _ holonomic d - module _ @xmath115 used in @xcite . now we prove the first main result of this section : [ thm : int3 ] let @xmath232 be a @xmath2-function on the smooth manifold @xmath127 . consider a whitney @xmath233-regular stratification @xmath128 of @xmath127 such that the set of nondegenerate covectors ( with respect to this stratification ) is dense in all fibers of the projection @xmath234 ( for all strata @xmath117 ) . let @xmath41 be a regular value of @xmath32 such that @xmath119 is transversal to all strata @xmath117 . assume that the set of degenerate covectors of @xmath128 and also of the induced whitney @xmath233-regular stratification of @xmath43 has ( locally ) a stratification with all strata of dimension @xmath203 . consider on @xmath127 a bounded complex @xmath3 of sheaves of vector spaces ( over a base - field @xmath4 ) , with finite dimensional stalks @xmath5 ( @xmath145 ) . assume @xmath3 is constructible with respect @xmath128 . then one has for the open inclusion @xmath235 the following estimate for the support of the characteristic cycles of @xmath236 and @xmath122 : @xmath237 @xmath238 1 . ) by transversality , @xmath43 gets an induced whitney @xmath233-regular stratification with strata @xmath239 of the form @xmath240 ( for @xmath117 a stratum of @xmath128 ) . the assumptions imply , that also the set of nondegenerate covectors with respect to this induced stratification of @xmath43 is _ dense _ in all fibers of the projection @xmath241 ( for all strata @xmath239 , compare with the proof of ( * ? ? ? * thm.5.0.2 , p.286 ) ) . @xmath236 and @xmath122 have also finite dimensional stalks , and are constructible with respect to the induced stratification of @xmath43 ( ( * ? ? ? * prop.4.0.2 , p.219 ) ) . more precisely , they have their support in @xmath43 , and their restrictions to @xmath43 are constructible with respect to the induced stratification . therefore the characteristic cycles of these complexes of sheaves are defined , and we can apply the description ( [ eq : suppcc ] ) for their support . first we consider @xmath236 . let @xmath242 _ nondegenerate _ with respect to the induced stratification of @xmath43 , and @xmath61 a point in a `` boundary '' stratum @xmath243 . we can approximate @xmath244 by a covector @xmath245 ( with @xmath246 and @xmath247 ) such that @xmath248 is _ nondegenerate _ with respect to @xmath128 , and the following properties hold ( see for instance ( * ? ? ? * thm.5.0.2 , p.286 ) ) : 1 . @xmath249 implies that the `` normal morse datum '' with respect to @xmath236 at the covector @xmath244 is isomorphic to @xmath211 . @xmath250 implies that the `` normal morse datum '' with respect to @xmath236 at the covector @xmath244 is isomorphic to the `` normal morse datum '' with respect to @xmath3 at the covector @xmath251 , up to a shift by @xmath252 $ ] . compare with @xcite for the details ( and note , that we used in loc.cit . the function @xmath253 for the description of @xmath254 ) . these results can also be deduced from ( * ? ? ? * prop.6.1.9(ii ) , p.256 ) and ( * ? ? ? * prop.7.5.10 , p.314 ) . so if the first `` normal morse datum '' ( at @xmath244 ) has a non - zero euler characteristic , then the same is true for the second `` normal morse datum '' ( at @xmath251 ) . by the description ( [ eq : suppcc ] ) , this implies our claim for @xmath236 ( since @xmath236 and @xmath3 have the same `` normal morse data '' for the strata @xmath239 of the form @xmath255 ) . the claim for @xmath122 follows from the case before by the _ duality _ isomorphism ( with @xmath256 the duality functor as in ( * ? ? ? * def.3.1.16(ii ) , p.148 ) ) : @xmath257 because @xmath258 ( for @xmath259 ) is equal to the image of @xmath260 under the antipodal map @xmath261 ( i.e. multiplication by @xmath224 on the fibers of @xmath262 ) . this follows from the description ( [ eq : suppcc ] ) together with ( * ? ? ? * eq.(5.25 ) , p.296 ) ( compare also with ( * ? ? ? * prop.9.4.4 , p.380 ) ) . suppose that we are in the context of `` geometric categories '' ( as in the next section ) . then we can use ( * ? ? ? * prop.5.0.1 , p.279 ) for the description of the _ micro - support _ ( i.e. the description of remark [ rem : musupp ] ) , and the above proof gives also the corresponding estimate in terms of the micro - support ( instead of the support of the characteristic cycles ) : @xmath263 @xmath264 @xmath265 @xmath266 in this way one can get in particular a proof of ( * ? ? ? * thm.5.3 , p.293 ) in terms of our ( stratified ) _ morse theory for constructible sheaves _ @xcite , without the use of the general _ microlocal theory _ of kashiwara - schapira . for the proof of the definable counterpart of theorem [ thm : int1 ] , we need in the next section also ( the second part of ) the following result ( compare with ( * ? ? ? * thm.4.2 , thm.4.3 , p.199 ) and ( * ? ? ? * thm.9.5.3 , cor.9.5.4 , p.385 ) ) : [ thm : int4 ] let @xmath267 be a @xmath2-function on the smooth manifold @xmath127 ( with @xmath268 ) . consider a whitney @xmath233-regular stratification @xmath128 of @xmath127 such that the set @xmath178 of nondegenerate covectors ( with respect to this stratification ) is dense in @xmath150 . assume that the set @xmath193 of degenerate covectors of @xmath128 has ( locally ) a stratification with all strata of dimension @xmath203 such that the frontier @xmath212 is a union of strata of @xmath202 ( for all connected components @xmath186 of @xmath178 ) . let @xmath3 be a bounded complex of sheaves of vector spaces on @xmath127 , with finite dimensional stalks @xmath5 ( @xmath145 ) . assume that @xmath3 is constructible with respect to @xmath128 , and that @xmath269 is proper . 1 . suppose that @xmath270 is compact . then @xmath271 is a cohomologically bounded complex with finite dimensional cohomology , and @xmath272\cap [ cc({\cal f})]\:)\:.\ ] ] 2 . suppose that @xmath273 is compact . then @xmath274 is a cohomologically bounded complex with finite dimensional cohomology , and @xmath275\cap [ cc({\cal f})]\:)\:.\ ] ] here we use the definition of the _ intersection product _ of the introduction , with @xmath276 . note , that we have the inclusion @xmath277 . the proof is by _ morse theory _ , and is almost the same as the proof of ( * ? ? ? * thm.5.0.4 , p.297 ) given in @xcite ( compare also with @xcite and the proof of ( * ? ? ? * thm.10.3.8 , p.429 ) ) . by approximation ( which does nt change the intersection number by a homotopy argument ) , we can assume that @xmath68 is a _ proper stratified morse function _ , i.e. all critical points of @xmath68 with respect to @xmath128 are _ stratified morse critical points _ in the sense of ( * ? ? * def.5.0.2 , p.277 ) and ( * ? ? ? * def.2.1 , p.52 ) : if @xmath171 is a _ critical _ point of @xmath278 , then @xmath177 is a _ nondegenerate _ covector and @xmath278 has in @xmath69 a classical _ morse critical point _ ( i.e. its hessian is nondegenerate ) . + by remark [ rem : musupp ] and the assumption on @xmath279 ( or @xmath280 ) , there are then only _ finitely _ many such critical points @xmath69 , whose `` normal morse datum '' with respect to @xmath3 are non - trivial at @xmath177 ( or @xmath281 ) . by ( * lem.5.1.1 , p.303 ) and ( * ? ? ? * lem.5.1.2 , p.306 ) we get for @xmath282 big enough ( by a _ mittag - leffler argument _ as in ( * ? ? ? * cor.6.1.2 , p.430 ) ) : @xmath283 in the first case , and @xmath284 in the second case . by induction , it is enough to consider the case that @xmath68 has at most one critical point @xmath285 , with @xmath171 , and @xmath286 . then we get by ( * ? ? ? * thm.5.0.1(1 ) , p.278 ) and ( * ? ? ? * lem.5.1.2 , p.306 ) : @xmath287 \quad \text{or } \quad r\gamma_{c}(\{f < r\},{\cal f})\simeq nmd({\cal f},-f , x)[-\tau ] \:,\ ] ] with @xmath141 the _ morse index _ of @xmath288 ( i.e. its hessian in @xmath69 has @xmath141 negative eigenvalues ) . but the corresponding `` normal morse datum '' @xmath289 is finitely dimensional ( as explained before ) . therefore our claim follows from the description ( [ eq : cc ] ) for @xmath26 , and the _ local intersection formula _ ( see ( * ? ? ? * eq.(9.5.18 ) , p.388 ) and ( * ? ? ? * eq.(5.20 ) , p.292 ) ) : @xmath290\cap [ t^{*}_{x^{s}}x]\ : ) = ( -1)^{\tau}\:.\ ] ] we used in the above proof the fact , that we can approximate @xmath68 by a stratified morse function . this follows from our assumption about the ( local ) existence of a suitable _ stratification _ of the set of _ degenerate _ covectors . by using a proper @xmath2-embedding @xmath291 , we can assume @xmath292 . then our claim follows for example from ( * ? ? ? * thm.1 , thm.3 ) . in this section we work in one of the following `` geometric categories '' : 1 . @xmath127 is an affine space @xmath293 , and @xmath294 is an _ o - minimal structure _ on the real field @xmath295 ( @xcite , e.g. the structure of _ semialgebraic _ subsets of real affine spaces ) . we could also assume , that @xmath127 is a real analytic @xmath294-manifold as in @xcite ( e.g. a real analytic nash manifold ) . 2 . @xmath127 is a real analytic manifold , and @xmath294 is an _ analytic geometric category _ ( @xcite , e.g. the structure of _ subanalytic _ subsets of real analytic manifolds ) . 3 . @xmath127 is a real analytic nash manifold , and @xmath294 is a _ nash geometric category _ ( @xcite and @xcite , e.g. the structure of _ locally semialgebraic _ subsets of real analytic nash manifolds ) . the subsets of @xmath296 are by definition the _ definable _ subsets of @xmath127 . a _ definable map _ @xmath297 between definable subsets @xmath298 and @xmath299 ( with @xmath300 ambient manifolds as in the above cases ) is a continous map with _ definable graph_. a complex of sheaves @xmath3 on @xmath127 is called _ @xmath294-constructible _ , if it is constructible with respect to a filtration @xmath128 : @xmath129 by closed _ definable _ subsets of @xmath127 ( i.e. the cohomology sheaves of @xmath3 are _ locally constant _ on all @xmath130 ) . by @xcite we can then assume , that @xmath3 is constructible with respect to a _ definable whitney b - regular @xmath301-stratification _ , with @xmath302 ( i.e. a definable filtration as before , with all @xmath132 @xmath301-submanifolds of @xmath127 , which is b - regular ) . for the following discussion , the reader should also compare with @xcite . for the basic results about definable sets and maps , we refer to @xcite . note , that their results about _ analytic geometric categories _ are also true in the context of _ nash geometric categories _ ( with the obvious modifications , as explained in @xcite , where we developed the basic theory of @xmath294-constructible complexes of sheaves ) . let @xmath128 be a definable whitney stratification . then @xmath303 are _ definable _ subsets of @xmath151 , and the dimension of the set @xmath193 of _ degenerate _ covectors is @xmath304 . therefore , the set of _ nondegenerate _ covectors @xmath178 is _ dense _ in @xmath150 , and we can find ( @xcite ) a _ whitney b - regular stratification _ @xmath305 of @xmath150 of the form @xmath306 in particular , the frontier @xmath212 of each connected component @xmath186 of @xmath178 is a union of strata of @xmath202 ( ( * ? ? ? * prop.4.0.2(1 ) , p.219 ) ) . + therefore we can apply the description of the first section for @xmath307 and @xmath16 $ ] , and theorem [ thm : int3 ] implies theorem [ thm : int2 ] . + now we are ready for the proof of the _ main theorem _ of this paper : [ thm : int5 ] let @xmath308 be a definable @xmath2-function . consider on @xmath127 a bounded @xmath294-constructible complex @xmath3 of sheaves of vector spaces , with finite dimensional stalks @xmath5 ( @xmath145 ) . suppose that the intersection of @xmath25 and the support @xmath8 of the characteristic cycle of @xmath3 is contained in a compact definable subset @xmath9 , with @xmath10 . then one has @xmath309\cap [ cc({\cal f})]\ ; \bigr ) \:.\ ] ] 1 . ) since @xmath34 is a compact definable subset of @xmath127 , there exists by @xcite a definable @xmath2-function @xmath232 , with @xmath37 . we can also assume that @xmath32 is non - negative ( otherwise use @xmath310 ) . + choose a definable whitney @xmath233-regular @xmath2-stratification @xmath128 of @xmath127 with the following properties ( @xcite ) : 1 . @xmath3 is _ constructible _ with respect to @xmath128 . 2 . the definable sets @xmath311 and @xmath34 are _ unions of strata_. 3 . the set of _ nondegenerate _ covectors ( with respect to this stratification ) is _ dense _ in all fibers of the projection @xmath312 ( for all strata @xmath117 ) . the last property can be achieved inductively by @xcite and the fact , that the dimension of the set of _ degenerate _ covectors @xmath202 is @xmath304 . note , that the set of nondegenerate covectors is _ dense _ in such a fiber @xmath313 , if and only if @xmath314 . but the set @xmath315 is a definable subset of @xmath117 with dimension @xmath316 ( compare ( * ? ? ? * prop.1.5 , p.65 ) ) . there exists a relatively compact open neighborhood @xmath33 of @xmath34 in @xmath0 , and a @xmath35 with the following properties : 1 . @xmath317 is _ proper _ ( since @xmath37 is compact ) . after restriction to @xmath33 we can and will assume @xmath38 . @xmath32 has no _ critical values _ in @xmath318 with respect to @xmath128 ( this follows from a ) and the @xmath319-version of the _ curve selection lemma _ the natural morphism @xmath320 is for _ all _ @xmath321 an _ isomorphism_. this follows from the fact , that the restriction of @xmath322 to @xmath323 , with @xmath324 as in a ) , is @xmath294-constructible ( see ( * ? ? ? * thm.2.0.1 , p.85 ) and ( * ? ? ? * cor.2.2.1 , p.105 ) ) . one can also use b ) and ( a cohomologically version of ) the _ first isotopy lemma of thom _ ( as in ( * ? ? ? * subsec.4.1.1 ) ) . 4 . @xmath325 . otherwise there exists ( @xcite ) a stratum @xmath175 and a @xmath319-curve @xmath328 , with @xmath329 such that @xmath330 for @xmath331 , and @xmath332 but this implies @xmath333 , since @xmath334 , and @xmath335 since @xmath336 is a _ conic isotropic _ submanifold of @xmath151 ( cf . @xcite ) . but by the _ monotonicity theorem _ @xcite we have that @xmath337 and @xmath338 are _ constant or strictly monotonic _ on @xmath3390,a[\;$ ] for @xmath340 small enough . by the assumptions , this implies the contradiction @xmath341 and @xmath342 for @xmath343 . fix a @xmath41 with @xmath321 , and consider the inclusion @xmath235 . by the choice of @xmath128 and b ) we can apply theorem [ thm : int3 ] . let @xmath344 , which is _ constructible _ with respect to the induced definable b - regular stratification of @xmath43 . by the _ curve selection lemma _ we can find @xmath345 so that @xmath68 has in @xmath346 _ no critical points _ with respect to this induced stratification . + note that @xmath347 is _ compact_. therefore we get by ( * ? ? ? * lem.5.0.1 , p.276 ) and ( * ? ? ? * lem.5.1.1(2 ) , p.303 ) ( compare also with ( * ? ? ? 5.4.19.(ii ) , p.239 ) ) : @xmath348 together with c ) and the general isomorphism @xmath349 this implies the isomorphism @xmath350 choose an @xmath351 with @xmath352 , and consider @xmath353 . note that @xmath354 is _ proper_. since @xmath68 has in @xmath346 no critical points with respect to the induced stratification of @xmath355 , we get @xmath356 in particular , @xmath357 is a _ compact _ subset of @xmath358 . therefore we can apply theorem [ thm : int4].2 to the manifold @xmath33 , the function @xmath95 and the sheaf complex @xmath359 , and get the intersection formula : @xmath360\cap [ cc({\cal f}'|u)]\:)\:.\ ] ] by equation ( [ eq : compact2 ] ) , the proof of the theorem is complete , if we show @xmath361 for @xmath362 , we get @xmath363 by ( [ eq : compact3 ] ) . + assume @xmath364 , with @xmath365 and @xmath366 . by theorem [ thm : int3 ] we get @xmath367 , with @xmath368 and @xmath369 . in the case @xmath370 we get a contradiction to d ) : @xmath371 so we can assume @xmath372 . in the case @xmath373 we get @xmath374 but this is by @xmath375 impossible , since the intersection of @xmath25 and @xmath8 is contained in the compact subset @xmath9 , with @xmath376 . in the case @xmath377 we get @xmath378 since @xmath379 , and @xmath380 for @xmath372 ( because @xmath311 is by b. a union of strata of @xmath128 ) . but this is a contradiction to b ) . sch02b bekka , k . : _ c - rgularit et trivialit topologique . _ in : warwick singularity symposium ( 1989 ) , lect . notes in math . 1462 , 42 - 62 ( 1991 ) borel , a . et al : _ intersection cohomology . _ progress in math . 50 , birkhuser , basel ( 1984 ) brasselet , j.p . , l d.t . , seade , j . : _ euler obstruction and indices of vector fields . _ topology 39 , 1193 - 1208 ( 2000 ) brylinski , j.l . , dubson , a . , kashiwara , m . : _ formule dindice pour les modules holonomes et obstruction deuler locale . _ c. r. acad . paris 293 , 573576 ( 1981 ) dubson , a . : _ formule pour lindice des complexes constructibles et d - modules holonomes . _ c. r. acad . paris 298 , 113116 ( 1984 ) dubson , a . : _ formule pour les cycles vanescents . _ c. r. acad . paris 299 , 181 - 184 ( 1984 ) fu , j.g.h . : _ curvature measures of subanalytic sets . 116 , 819880 ( 1994 ) ginsburg , v . : _ characteristic cycles and vanishing cycles . _ 84 , 327 - 402 ( 1986 ) goresky , m . , macpherson , r . : _ stratified morse theory . _ springer , berlin heidelberg ( 1988 ) grinberg , m.,macpherson , r . : _ euler characteristics and lagrangian intersections . _ in : symplectic geometry and topology . ias / park city math . ser . 7 , 265293 . soc . , providence ( 1999 ) kashiwara , m . : _ index theorem for maximally overdetermined systems of linear differential equations . japan acad . 49 , 803804 ( 1973 ) kashiwara , m . : _ systems of microdifferential equations . _ progress in mathematics 34 , birkhuser , basel ( 1983 ) kashiwara , m . : _ index theorem for constructible sheaves . _ astrisque 130 , 193 - 209 ( 1985 ) kashiwara , m . , schapira , p . : _ sheaves on manifolds . _ springer , berlin heidelberg ( 1990 ) l d.t . : _ morsification of d - modules . soc . mat . mexicana 3 , 229 - 248 ( 1998 ) loi , t.l . : _ verdier and strict thom stratifications in o - minimal structures . _ ill . jour . of math . 42 , 347356 ( 1998 ) massey , d . : _ a little microlocal morse theory . _ math . 321 , 275 - 294 ( 2001 ) orro , p . : _ espaces conormaux et densit des fonctions de morse . _ c. r. acad . paris 305 , 269 - 272 ( 1987 ) sabbah , c . : _ quelques remarques sur la gomtrie des espaces conormaux . _ astrisque 130 , 161 - 192 ( 1985 ) schmid , w . , vilonen , k . characteristic cycles for constructible sheaves . _ 124 , 451 - 502 ( 1996 ) schrmann , j . : _ on the comparison of different notions of geometric categories . _ preprint , math.ag/0201283 schrmann , j . : _ a short proof of a formula of brasselet , l and seade for the euler obstruction . _ preprint , math.ag/0201316 schrmann , j . : _ topology of singular spaces and constructible sheaves . _ to appear in monografie matematyczne ( mathematical monographs - new series ) , birkhuser , basel ( 2003 ) trotman , d . une version microlocale de la condition ( w ) de verdier . fourier grenoble 39 , 825829 ( 1989 ) van den dries , l . : _ tame topology and o - minimal structures . _ london math . notes ser . 248 , cambridge univ . press ( 1998 ) van den dries , l . , miller , c . : _ geometric categories and o - minimal structures . _ duke math . journal 84 , 497 - 540 ( 1996 )

the recently initiated arecibo legacy fast alfa ( alfalfa ) survey aims to map 7000 of the high galactic latitude sky visible from arecibo , providing a hi line spectral database covering the redshift range between -1600 and 18,000 with resolution . exploiting arecibo s large collecting area and small beam size , alfalfa is specifically designed to probe the faint end of the hi mass function in the local universe and will provide a census of hi in the surveyed sky area to faint flux limits , making it especially useful in synergy with wide area surveys conducted at other wavelengths . alfalfa will also provide the basis for studies of the dynamics of galaxies within the local and nearby superclusters , will allow measurement of the hi diameter function , and enable a first wide - area blind search for local hi tidal features , hi absorbers at and oh megamasers in the redshift range . although completion of the survey will require some five years , public access to the alfalfa data and data products will be provided in a timely manner , thus allowing its application for studies beyond those targeted by the alfalfa collaboration . alfalfa adopts a two - pass , minimum intrusion , drift scan observing technique which samples the same region of sky at two separate epochs to aid in the discrimination of cosmic signals from noise and terrestrial interference . survey simulations , which take into account large scale structure in the mass distribution and incorporate experience with the alfa system gained from tests conducted during its commissioning phase , suggest that alfalfa will detect on the order of 20,000 extragalactic hi line sources out to , including several hundred with hi masses .

astro-ph0508301

the first 21 cm line detection of an extragalactic source ( the magellanic clouds ) was achieved by kerr & hindman ( 1953 ) with a 36foot transit telescope just over half a century ago . the construction of large , single dish radio telescopes produced seminal discoveries in the decade of the 1960s , as illustrated in the fundamental paper of that period ( roberts 1975 ) . a decade later , the completion of the very large array ( vla ) and of the westerbork synthesis radio telescope , the resurfacing of the arecibo dish and rapid progress in detector and spectrometer technology made it possible for hi spectroscopy to achieve order of magnitude improvements in terms of sensitivity and resolution . new scientific problems became accessible , and extragalactic hi line research underwent a phase of rapid growth . the study of rotation curves led to the discovery of dark matter in spiral galaxies ; the potential of the luminosity linewidth relation as a cosmological tool became apparent ; the impact of tidal interactions and of the intracluster medium on galaxy evolution was illustrated in great detail through measures of the hi emission . the 21 cm line was found to be an expedient tool to determine accurate galaxy redshifts , playing an important role in confirming the filamentary nature of the large scale structure of the universe . the application of the luminosity linewidth relation led to accurate estimates of cosmological parameters and to the characterization of the peculiar velocity field in the local universe . highly sensitive measurements in the peripheries of disk galaxies revealed edges in their visible components , and a number of optically inert objects was discovered . until a few years ago , however , comprehensive wide angle surveys of the extragalactic hi sky were unavailable . at the close of the last decade , the advent of multifeed front end systems at l band finally made possible the efficient coverage of large sections of the extragalactic sky . the first such system to be used for that purpose was installed on the 64 m parkes telescope in australia , and has produced the excellent results of the hipass survey ( barnes 2001 ; meyer 2004 ) . a second 4-feed system on the 76 m lovell telescope at jodrell bank produced the hijass ( lang 2003 ) survey . the 1990s upgrade of the arecibo telescope , which replaced its line feeds with a gregorian subreflector system , made it possible for that telescope to host feed arrays , as proposed by kildal ( 1993 ) . eventually built and installed at arecibo in 2004 , this 7-beam radio `` camera '' , named alfa ( arecibo l band feed array ) , is now operational , enabling large scale mapping projects with the great sensitivity of the 305m telescope . a diverse set of mapping projects are now underway , ranging from extragalactic hi line , to galactic line and continuum , to pulsar searches . here , we introduce one of these newly - initiated surveys , specifically designed to map approximately one fifth of the sky in the hi line , out to a distance of 250 mpc . the survey , currently underway at arecibo , is referred to as alfalfa , the arecibo legacy fast alfa survey . as illustrated in figure [ skycov ] , alfalfa aims to cover 7074 of the high galactic latitude sky between @xmath6 and @xmath7 in declination , requiring a total of 4130 hours of telescope time . exploiting the large collecting area of the arecibo antenna and its relatively small beam size ( @xmath8 ) , alfalfa will be nearly eight times more sensitive than hipass with @xmath0four times better angular resolution . furthermore , its spectral backend provides 3 times better spectral resolution ( 5.3 at @xmath9 ) over 1.4 times more bandwidth . these advantages , in combination with a simple observing technique designed to yield excellent baseline characteristics , flux calibration and hi signal verification , offer new opportunities to explore the extragalactic hi sky . data taking for alfalfa was initiated in february 2005 , and , in the practical context of time allocation at a widely used , multidisciplinary national facility like arecibo , completion of the full survey is projected to require 56 years . as discussed in section [ simul ] , simulations predict that alfalfa will detect some 20,000 extragalactic hi line sources , from very nearby low mass dwarfs to massive spirals at @xmath4 . the survey is designed specifically to determine robustly the faint end of the hi mass function ( himf ) in the local universe at masses @xmath10 @xmath11 , and will at the same time provide a census of hi in the surveyed sky area , making it especially useful in synergy with other wide area surveys such as sdss , 2mass , galex , astro f , etc . in conjunction with optical studies of comparable volumes , alfalfa will help determine the true census of low mass satellites and the widely distributed dwarf galaxy population in the local and surrounding groups . its dataset will also provide the basis for studies of the dynamics of galaxies within the local and nearby superclusters , will allow measurement of the hi diameter function , and will enable a first wide area blind search for local hi tidal features , hi absorbers at @xmath2 and oh megamasers in the redshift range @xmath3 . survey details and status can be found by visiting its website . survey efforts of this scale and scope require careful optimization of their operational strategy towards achieving the science objectives within the constraints imposed by practical observing conditions and requirements . in this paper , we introduce the science objectives of alfalfa , the principal constraints which set its strategy and the results of survey simulations which allow prediction of its eventual results . in a companion paper ( giovanelli 2005 ; paper ii ) , we present results obtained during a precursor observing run , designed to allow us to test and optimize the alfalfa strategy during the alfa commissioning phase in fall 2004 . we summarize , in section [ science ] , the main scientific motivations of the survey . technical details of the hardware are given in section [ alfa ] , while criteria leading to the design of the survey , in the form of scaling laws and survey simulations are described in section [ strategy ] . observing modes , sky tiling and data processing plans are presented in section [ tiles ] , while section [ sensitivity ] summarizes sensitivity numbers at various stages of the survey . we elaborate on the treatment of candidate detections and follow up observations in section [ followup ] and summarize in section [ summary ] . throughout the paper , we assume @xmath12 mpc@xmath13 . -6 cm -12 cm following on the results of a number of important and sucessful previous blind hi surveys , extragalactic hi surveys with alfa will exploit arecibo s huge collecting area to explore larger volumes of the universe with greater sensitivity and higher angular and spectral resolution if they are to break new science ground . with arecibo s tremendous sensitivity and beam size advantages , alfalfa is designed for wide areal coverage , thereby increasing the volume sampled locally , yielding a deep , precise census of hi in the local universe to the lowest hi masses . alfalfa aims to survey 7074 of sky at high galactic latitudes which lie within the declination limits of the arecibo telescope , @xmath14 decl . @xmath15 , as illustrated in figure [ skycov ] . based on simulations described in section [ simul ] and now verified by the results of the alfalfa precursor observations presented in paper ii , alfalfa is expected to yield on the order of 20,000 hi line detections , sampling a wide range of sources from local , very low hi mass dwarfs to gas - rich massive galaxies seen to @xmath16 ( @xmath0250 mpc ) . hi spectra provide redshifts , hi masses and rotational widths for normal galaxies , trace the history of tidal events with high kinematical accuracy and provide quantitative measures of the potential for future star formation via comparative hi contents . as a blind hi survey , alfalfa will not be biased towards the high surface brightness galaxies typically found in optical galaxy catalogs and moreover , in contrast to hipass and hijass , will have adequate angular and spectral resolution to be used on its own , generally without the need for follow up observations to determine identifications , positions and , in many cases , hi sizes . the wide areal coverage of alfalfa overlaps with several other major surveys , most notably the sloan digital sky survey ( sdss ) , 2mass and the nvss . the catalog products of alfalfa will be invaluable for multiwavelength data mining for a wide spectrum of purposes , and a key element of our overall collaborative program is to provide broad application , legacy data products that will maximize the science fallout of the alfalfa survey . a primary objective of alfalfa is the robust determination of the faint end of the hi mass function ( himf ) . the himf is the cosmic number density , per bin of hi mass , of detectable hi line signals in a survey sensitive to the global neutral hydrogen within a system . the most recent estimates of the himf based on significant numbers of galaxies have been presented by zwaan ( 1997 , hereafter z97 ) , rosenberg & schneider ( 2002 ; hereafter rs02 ) , zwaan ( 2004 , hereafter z04 ) , zwaan ( 2005 , hereafter z05 ) and springob ( 2005a ) . the z04 and z05 himfs are based on the hipass survey , while the rs02 and z97 himfs are both based on surveys conducted at arecibo during the period of its recent upgrade . the faint end slope of those determinations of the himf vary between @xmath17 and @xmath18 , yielding extrapolations below @xmath19 that disagree by an order of magnitude near @xmath20 , the rs02 himf having the steeper slope . all of the previous hi blind surveys sample a lower mass limit just below @xmath21 . no extragalactic hi sources were detected by rs02 or z97 with @xmath22 , while 3 are claimed by z04 , and only a small number of detections have @xmath23 . we note that the distances of those detections are highly uncertain , for they are very nearby and the impact of peculiar velocity on the observed redshift is quite large , as pointed out by masters ( 2004 ) . thus current inferences on the behavior of the himf at low mass levels are quite unreliable , as they are based on very few objects of highly uncertain distance . with the aim of exploring the himf at masses @xmath23 , alfalfa will cover a very large solid angle in order to survey an adequate volume at d @xmath24 20 mpc , a distance within which the low hi mass systems are detectable . as shown by the simulations described in section [ simul ] , alfalfa will detect several hundred objects with @xmath5 . in addition , its extensive catalog of more massive objects will allow comparison of the high mass end , @xmath25 , in the diverse range of environments found in the volume out to 250 mpc . the arecibo sky to be surveyed by alfalfa , as shown in figure [ skycov ] includes the rich central regions of the local supercluster and the nearby low density anti - virgo region as well as a number of more distant large scale features , most notably the main ridge of the pisces - perseus supercluster and the great wall connecting the abell 1367 coma and hercules superclusters . at the virgo distance , alfalfa should detect galaxies with hi masses as low as @xmath26 . alfalfa will cover more than one thousand around virgo , yielding a database of unprecedented breadth for combination with the sdss , galex and other surveys to construct a complete census of baryon bearing objects in the cluster and its full infall region . the combination of hi content , hi distribution and the derived kinematical information with other multiwavelength studies will enable detailed modelling of the relative efficiency of gas stripping mechanisms such as tides , ram pressure or galaxy harassment as the origin of gas deficiency in virgo . alfalfa s hi maps will trace intriguing hi features like the virgo `` dark cloud '' ( davies 2004 ; minchin 2005 ) , the hi `` plume '' around ngc 4388 ( oosterloo & van gorkom 2005 ) and the huge envelope surrounding ngc 4532 and ddo137 ( hoffman 1992 ) . in more quiescent regions than virgo , extensive tidal features such as the leo triplet ( haynes , giovanelli & roberts 1979 ) , and enigmatic systems such as the 200 kpc `` leo ring '' ( schneider 1983 ) may be found . alfalfa will enable the first truly blind survey for hi tidal remnants with both sufficient angular resolution and wide areal coverage to verify their nature . while hi appendages uncover past disruptive events in galaxy evolution , extended gas disks around galaxies represent a reservoir for future star formation activity . in contrast to hipass and hijass which were limited by much poorer angular resolution ( 15.5 and 12 , respectively ) , the 3.5 beam of alfa will resolve the hi disks of @xmath0500 gas - rich galaxies , allowing a quantitative measure of their hi sizes ( hewitt 1984 ) and the derivation of the hi diameter function . in combination with optical photometry , alfalfa will determine the fraction of galaxies with extended gas disks and enable studies of their host galaxies , their environments , morphologies and the role of gas in their evolution . more extremely extended gas disks , such as those found in ddo 154 ( krumm & burstein 1984 ) , ugc 5288 ( van zee 2004 ) and ngc 3741 ( begum 2005 ) may lurk yet unidentified . because of its wide sky coverage , alfalfa will trace important high - velocity cloud ( hvc ) structures in and around the milky way , such as the northern portions of the magellanic stream and complex c at several times better spatial and spectral resolution than hipass , particularly important advantages in the case of narrow linewidth hvc cores ( giovanelli & brown 1973 ) . because of its high flux sensitivity , alfalfa will be eight times more sensitive than hipass to unresolved small clouds , or ultra - compact hvcs . while arecibo can not reach as far north as m31 , alfalfa will cover part of the region containing the clouds in its periphery identified by thilker ( 2004 ) and their possible extension toward the region around m33 ( westmeier , braun & thilker 2005 ) . in addition to the study of hi in emission , alfalfa will provide a dataset well - suited for a blind survey of hi absorption out to @xmath27 . the background continuum source counts in the alfalfa survey region at 1.4 ghz yield over 2000 sources brighter than 0.4 jy and more than 10000 brighter than 0.1 jy . the major practical difficulty with hi line absorption studies is spectral baseline determination in the presence of standing waves . the large number of continuum sources present in the alfalfa dataset and the adopted `` drift '' technique ( section [ drift ] ) will aid in the assessment of whether a given spectral feature is real absorption . by its combination of studies of hi emission and absorption in the local universe , alfalfa will allow a robust estimate of the local hi cross section , as well as a measure of its clustering correlation amplitude and scale . in addition to hi line studies , the frequency range of the alfalfa survey will also include , serendipitously , lines from oh megamasers ( ohm ) arising from the nuclear molecular regions in merging galaxy systems . approximately 100 such sources are known to date , half of which were discovered recently at arecibo ( e.g. darling & giovanelli 2002 ) . observations of ohms hold the potential for tracing the merger history of the universe since the sources are associated with merging galaxies . an essential tool in this exercise is the ohm luminosity funtion at low @xmath28 . alfalfa should detect several additional dozen ohms in the redshift interval 0.160.25 , and allow a more robust determination of the low @xmath28 ohm luminosity function than currently available . the construction of the gregorian subreflector system for the arecibo telescope , completed in the late 1990s , made possible the development of focal plane feed arrays ( effectively , _ creating _ a focal plane ) . this development was foreseen during the planning phases of the gregorian upgrade ( kildal 1993 ) . a seven feed array was commissioned at the observatory during 2004 . six of the seven feeds ( numbered 1 through 6 ) are physically arranged on the corners of a regular hexagon , while the seventh ( feed 0 ) is at its center , as shown in figure [ alfabeams ] . the feeds can receive dual , linear polarizations and their spectral response is optimized for the range 12251525 mhz . they are stepped te@xmath29 mode horns of 25 cm aperture , as described in cort ' es medelln ( 2002 ) . because the optical design of the gregorian subreflectors maximizes the illuminated area of the primary by sacrificing its circular symmetry ( the illuminated area is elliptical ) , a circular pattern in the sky maps on the focal plane as an ellipse of axial ratio 1.15 ; reciprocally , the footprint of the centers of the outer beams of the alfa array on the sky is that of a hexagon inscribed in an ellipse of that axial ratio . similarly , the seven beams have an elliptical shape of the same axial ratio and orientation as the array pattern . the major axis of the ellipse is linked to the azimuth of the receiver , so its orientation on the sky changes with telescope configuration . in figure [ alfabeams ] the relative location of the beams is shown when the array is positioned at the meridian and rotated about its symmetry axis by @xmath30 . in this sketch , the outlines of the beams are shown at the half power response , for which the beam sizes are 3.3 along the azimuth direction and 3.8 along the zenith angle directon , with small variations from one beam to the other . the central beam 0 has higher gain ( @xmath31 k / jy ) than the peripheral beams 16 ( gain of @xmath32 k / jy ) , which is illustrated in the sketch by the brighter contours . the dotted lines indicate the tracks of constant declination made by each of the beams , when data is acquired in drift mode . projected on the sky , the alfa footprint in this configuration is such that beam 1 points farthest to the north and beam 2 farthest to the west , for observations south of the zenith . for observations north of the zenith , beam 1 points farthest to the south and beam 2 farthest to the east . figures [ sidelobe0 ] and [ sidelobe6 ] show the pattern for each of the alfa beams , obtained by mapping the radio source 3c 138 near transit . sidelobe levels are very different for each of the beams of alfa . located at the center of the array , beam 0 has the most symmetric beam pattern , with a first sidelobe ring near 15 db below the response at beam center , as shown in figure [ sidelobe0 ] . contour levels are plotted at intervals of 3 db . the outer beams have a very marked comatic aberration , as shown in figure [ sidelobe6 ] . the first sidelobe ring of the outer beams is strongly asymmetric , reaching levels near 78 db below peak response , on the section away from the array center . this feature of the system will require careful attention , especially in the analysis of data obtained in the vicinity of strong and/or extended sources . the system temperature ranges between 26 and 30 k for all beam / polarization channels , when pointing away from strong continuum sources . -5 cm -2 in -3 cm -5 cm the array can be rotated about its axis , centered on beam 0 , and thus the relative position of the beams on the sky can be rotated along the elliptical perimeter . in the case of drift mode observations , it is desirable to position the array in such a manner that the beam tracks are equally spaced in declination . because of the ellipticity of the array pattern on the sky , the separation between beam tracks depends on both the array rotation angle as well as on the array azimuth . when the telescope feed arm is stationed along the local meridian , the optimal array rotation angle is @xmath30 , as shown in figure [ alfabeams ] . in that case , beam tracks are spaced 2.1 in declination . a single drift will thus sweep seven nearly equidistant tracks covering 14.5 in declination , at slightly below the nyquist sampling rate . alfalfa will map most of the extragalactic sky in drift mode with alfa stationed along the local meridian , at a local azimuth of either @xmath6 ( for observations south of zenith ) or @xmath33 ( for north of zenith ) . only declination tracks transiting within @xmath34 of the zenith will be mapped with alfa at an azimuth near @xmath35 or @xmath36 in order to avoid impractically small zenith angles . the beam separation for equidistant tracks is 1.8 at these azimuths , and thus the declination sampling will be denser , as the elliptical pattern of the sky footprint of both the array and the individual beams will have its major axis oriented nearly parallel to the drift direction . our survey will thus be done with alfa in only two sets of configurations : one for all observations between declinations @xmath6 and @xmath37 , as well as between @xmath38 and @xmath7 , with alfa on the meridian , and a second for observations between declinations @xmath37 and @xmath38 . spectra will be recorded every second , yielding approximately 14 samples per beam in the right ascension direction . this sampling rate , which largely exceeds nyquist , is principally motivated by the advantages deriving in the identification of radio frequency interference ( rfi ) . further details on alfa can be found at the naic website . alfalfa uses the feed array connected through a fiber optics if lo system to a spectral line , digital backend consisting of a set of processors each individually referred to as a wapp ( wideband arecibo pulsar processor ) . the full spectral backend consists of four wapp units , each capable of processing the two polarization signals from two alfa beams . the wapp set can thus produce 16 autocorrelation spectra , each with a maximum bandwidth of 100 mhz , over 4096 lags . fourteen of those are matched to the seven polarization pairs from the alfa beams , and a spare pair duplicates the signal of the seventh beam . at the offline processing stage , the extra pair of spectra are used for rfi monitoring purposes . each data record thus consists of 65,536 spectral samples ( 16 @xmath39 4096 ) . since , as mentioned above , alfalfa records data every second , the generation of raw data by the survey is slightly over 1 gb per hour , including headers . the strategy for the alfalfa survey has been developed over the last few years , balancing the practical realities involved in using the arecibo telescope , the constraints of telescope time availability , and the principal science objectives outlined in section [ science ] . here we review the considerations that enter into the survey design and numerical simulations that have been used to refine it . the hi mass of an optically thin hi source at distance @xmath40 , in solar units , is m_hi / m_= 2.35610 ^ 5 d^2_mpc s(v ) dv , where @xmath41 is the hi line profile in jy and @xmath42 is the doppler velocity in . to first order , m_hi / m_2.410 ^ 5 d^2_mpc s_peak w_kms , where @xmath43 is the line peak flux and @xmath44 its velocity width in . for detection , the signal to noise ratio @xmath45 must exceed some threshold value ; @xmath46 quantifies the fraction of the source flux detected by the telescope s beam . the parameter @xmath47 for a point source , while for resolved sources , it decreases roughly like the ratio between the beam solid angle and the solid angle subtended by the source . an estimate of @xmath48 can be obtained from the radiometer equation for the rms figure s_rms = ( t_sys / g ) , where @xmath49 is the system temperature divided by the system gain ( for the alfa feeds , @xmath49 will vary between 2.65 and 3.40 jy ; here we adopt a flat value of 3.25 jy ) ; @xmath50 is the channel bandwidth in hz and @xmath51 the integration time in seconds . the factor 2 under the square root indicates that two independent polarization channels are averaged . for alfalfa , @xmath52 khz , which at the rest frequency of the hi line , is equivalent to 5.3 . the factor @xmath53 accounts for spectral smoothing of the signal , @xmath54 , the switching technique applied for bandpass subtraction , @xmath55 , and other observational details , such as autocorrelation clipping losses , i.e. @xmath56 . for the data taking scheme of alfalfa , @xmath57 . the signal to noise of a feature of width @xmath44 is best rendered when the noise is measured after smoothing the signal to a spectral resolution on order of @xmath58 . in practice , however , the smoothing of l band spectra of @xmath59 several hundred does not reduce the noise in proportion to @xmath60 and , moreover , @xmath43 is depressed by such smoothing , for spectral shapes are by no means boxlike . the fact that the detection criterion described above applies well to narrow lines but not so to wider ones was also noted by rosenberg & schneider ( 2002 ) . we assume here that spectral smoothing will increase signal to noise up to a maximum @xmath61 , and that smoothing beyond that width will be ineffective in increasing @xmath62 . for a conservative signal to noise threshold of 6 , we can then write : 12.3 f_t_s^1/2(m_hi10 ^ 6 m _ ) d^-2_mpc ( w_kms200)^ > 6 , [ eq : sngt5 ] where @xmath63 for @xmath64 and @xmath65 for @xmath66 . by inverting , we can obtain a minimum detectable hi mass ( m_hi10 ^ 6)_min = 0.49 f_^-1 d_mpc^2 t_s^-1/2 ( w_kms/200)^- . [ eq : mhi ] with an integration time of 30 sec per pixel solid angle ( see section [ sensitivity ] ) or 48 sec per beam solid angle , alfalfa should thus detect an hi mass of @xmath20 @xmath11 , @xmath67 , at a distance of @xmath68 mpc , and a source of @xmath69 @xmath11 and of the same width out to @xmath70 mpc . it is useful to review some of the basic scaling relations relevant to the design of a survey : @xmath71 the minimum integration time required to detect a source of hi mass @xmath72 and width @xmath44 at @xmath73 , at the distance @xmath40 with alfa is , from eqn . [ eq : mhi ] , t_s 0.023 f_^-2 ( t_sysg)^2 ( m_hi10 ^ 6 m_)^-2 ( d_mpc)^4 ( w_kms200)^-2 , [ eq : ts ] i.e. * the depth of a survey increases only as @xmath74*. with equality of back ends , the @xmath51 required to detect a given @xmath72 at a given distance decreases as the square of @xmath75 , i.e. as the 4th power of the reflector diameter . arecibo offers a tremendous advantage because of its huge collecting area . @xmath71 the beam of a telescope of collecting area @xmath76 is @xmath77 , while the maximum distance at which a given hi mass can be detected is @xmath78 . since @xmath79 , the volume sampled by one beam to the maximum distance @xmath80 is @xmath81 , i.e. in a fixed time , a radio telescope samples a * volume that scales with the reflector diameter * , yielding a significant comparative advantage for a large aperture like arecibo . @xmath71 assuming that clouds of mass @xmath72 are randomly distributed in space out to the maximum distance at which they are detectable , @xmath82 , the number of clouds detected by a survey increases linearly with the sampled volume @xmath83 , where @xmath84 is the solid angle mapped by the survey . we can thus increase the number of detections either by sampling a larger solid angle @xmath84 or by increasing @xmath82 . now , the total time required to complete the survey is t_survey ( _ survey/_b ) t_s , [ eq : tsurvey ] where @xmath85 is the telescope beam . since @xmath86 , as shown in equation [ eq : ts ] , we can write v_survey(m_hi ) _ survey [ d_max(m_hi)]^3 _ survey t_s^3/4 t_survey t_s^-1/4 , and inverting : t_survey v_survey(m_hi ) d_max(m_hi ) v_survey(m_hi ) t_s^1/4 . [ eq : tsurvey2 ] to achieve a given surveyed volume @xmath87 , once @xmath72 is detectable at an astrophysically interesting distance , * it is more advantageous to maximize @xmath88 than to increase the depth of the survey @xmath82*. the scaling relations described above provide only general guidelines in the design of a survey . other considerations can and will play important roles in the survey strategy . for example , the growing impact of rfi on hi spectroscopy dictates increased attention to signal identification and corroboration , recommending a survey with more than a single pass over a given region of sky , as we discuss in section [ 2pass ] . the determination of specific properties of galaxies or systems may drive towards deeper surveys of narrow solid angle regions , as planned for other alfa surveys with the arecibo telescope , the goals and products of which will be complementary to alfalfa . the scaling relations described above dictate that alfalfa cover a very large solid angle . in practice , the survey design must weigh the desire to cover a wide area with the need for sensitivity . an indispensable aid in the design of a survey is a thorough examination of expectations , _ vis a vis _ variance over the survey parameter space . to this end , we have carried out an extensive set of survey simulations to help in the design of alfalfa and present a sample of the results in this section . the main ingredients for our survey simulation are : ( i ) the survey mode and sensitivity parameters , deriving from the instrument configuration ; ( ii ) an estimate of the space density of sources given by an adopted himf ; ( iii ) an understanding of the clustering properties and deviations from smooth hubble flow in the local universe . sensitivity considerations were presented in section [ scaling ] . for the himf , we use two recent estimates which differ strongly from each other at the low mass end : that of z97 and that of rs02 . the more recent himfs by z04 and springob ( 2005a ) are bracketed by those of z97 and rs02 . we use a density map of the local universe provided by branchini ( 1999 ) , which is a density reconstruction derived from the pscz catalog . the grid we used has a spacing of 0.9375 @xmath89 mpc in the inner 60 @xmath89 mpc , and a spacing twice that value between 60 and 120 @xmath89 mpc , where @xmath90 ; the map is smoothed with a gaussian filter of @xmath91 mpc . for distances larger than 120 @xmath89 mpc , we assume a constant density . sources are seeded using the density map and , separately , each of the two himfs . the hi gas is assumed to be optically thin . because many of the sources will be resolved by the arecibo beam , an estimate of hi sizes is necessary . assuming that the hi distribution is disk like , inclinations to the line of sight and linewidths need to be assigned to each source . we use empirical scaling relations obtained from our own hi survey data ( springob 2005b ) and broeils & rhee ( 1997 ) , we add random inclinations , realistic scatter and broad band spectral baseline instability . with these recipes , we have inspected a wide grid of survey parameters in arriving at the adopted alfalfa survey strategy . -3 cm -4 cm as an example , figure [ him_d ] displays the variation with distance of the hi mass of expected detections within the region of the alfalfa sky coverage , as described in section [ tiles ] , adopting as input the two different himfs : the z97 himf ( panel a ) and the rs02 ( panel b ) . only expected detections out to 150 mpc are shown . the simulation corresponds to an integration time @xmath92 seconds per map pixel solid angle . the number of detections expected with rs02 is 22,200 , while the number expected with z97 is 15,022 , with a detection threshold of @xmath93 , as defined in section [ scaling ] . the difference in expected detections is more dramatic when the hi mass of the source is restricted to @xmath94 . in that case , we expect 1400 detections with rs02 and only 249 with z97 . it is interesting to point out that in the course of alfalfa precursor observations , reported in the companion paper ii , three objects with @xmath95 were detected . albeit of still marginal statistical value , that rate is consistent with the high end of the expectations ( rs02 himf ) obtained from the simulations reported here . three curves are inset in the panels of figure [ him_d ] : the two solid lines are the loci of constant integrated hi line flux of 0.9 and 1.25 jy , respectively . the lowest of the two corresponds to the completeness limit of the survey for sources of @xmath96 ( detections below that line correspond to sources of smaller width ) . the second curve corresponds to completeness limit for sources of the same width , for an integration time per pixel of @xmath97 seconds . such an integration time applies to the analysis of individual drift tracks , without the corroborating support ( and higher resulting integration ) of spectra in beam tracks at neighboring declinations . this detection limit would result if source extraction would be carried out , for example , right after data taking , and before an entire data cube ( spatially two dimensional , plus one spectral dimension ) is available . in this case , the expected number of detections would be 13,804 for the rs02 himf , and 9601 for the z97 case , a drop of respectively 38% and 36% from the previous set of numbers . the decrease in the number of detections with small hi masses would be more severe if signal extraction were applied to individual tracks only , rather than to full maps : in that case , only fewer than half of the sources would be detected in the rs02 case , and just above half in the z97 case . the topmost ( dashed ) curve inset in figure [ him_d ] corresponds to a flux integral of 6.8 jy , the hipass completeness limit at the 6@xmath98 level , for detection of sources of width @xmath99 ; this is the hipass analog of the lowest of the two solid lines for alfalfa . it uses a hipass limit of 13.3 mjy per map pixel area , as reported by barnes ( 2001 ) . this provides a good graphical illustration of the comparison between the two surveys . figure [ aitoffa ] shows the sky distribution of the detected sources by an alfalfa like survey , in the simulation with the z97 himf . sources at all right ascensions are plotted , albeit alfalfa will only cover 60% of the full r.a . range . in the lower panel , only the detections with @xmath94 are plotted . figure [ aitoffb ] shows the analogous graphs for the rs02 himf . -5 cm -15 cm -5 cm -15 cm simulations results such as those presented here aid us in the estimate of the statistical efficacy of the survey data , most importantly in the determination of the faint end of the hims and the clustering properties of the new detections . given the science objectives outlined in section [ science ] and the scaling relations and simulations presented in the preceding section , the final consideration of the survey design strategy takes into account more telescope - related practicalities . in this section , we review those issues which have led us to adopt a very simple observing strategy , a two - pass drift scan mode , covering the sky with `` tiles '' extending from @xmath100 . the arecibo telescope is an altitude azimuth system located at a latitude near 18@xmath101 . its gregorian dome can be steered within @xmath102 of the zenith , but the system gain and performance degrades rapidly at zenith angles above @xmath103 . in general , the performance characteristics including beamwidth , pointing accuracy , sidelobe levels , spectral baseline stability and susceptibility to rfi vary with both azimuth , zenith angle and feed rotation angle . furthermore , the alfa footprint on the sky and its beams structure vary likewise in a complicated manner . the design of alfa surveys is thus strongly constrained by this variance , and with it , the corresponding degree of calibration complexity a particular observing program can endure . alfalfa aims to minimize the impact of these factors on performance through a choice of maximum simplicity in the observing mode and minimal electronic intrusion at the detection level . as mentioned previously in section [ alfa ] , the alfalfa survey is thus being carried out in a fixed azimuth drift mode . for most of the survey , the azimuth arm of the telescope is stationed along the local meridian , the zenith angle of alfa determining the declination to be mapped . a tiny elevation readjustment is periodically applied to maintain drift tracks at constant j2000.0 , rather than current declination . without such adjustment , drift tracks taken few years apart would noticeably diverge from one another . no firing of noise calibration diodes is done during normal data acquisition . rather , data taking is interrupted very briefly every 10 minutes while a calibration noise diode is fired for 1 second . this interruption produces data gaps of 5 seconds approximately 1/3 of a beamwidth in r.a . in each 600 second drift data stream . the central frequency of the bandpass is set and never changed during an observing session , i.e. no doppler tracking of the local oscillator frequency is applied ( realignment to a common heliocentric reference frame is applied to each spectrum off line , using high precision time , position and earth s motion stamps updated every second in data headers ) . with no moving telescope parts , constant gain and nearly constant system temperature along a drift are obtained ; standing waves will change slowly , as driven by the sidereal rate ; beam characteristics remain fixed ; bandpass subtraction is optimized . the solid angle mapped by the survey is subdivided into `` tiles '' of @xmath104 in declination ( see below and figure [ skycov ] ) extending from dec.=@xmath6 to @xmath7 . for 8 of the 9 bands of tiles , the azimuth of the feed array is along the local meridian ( at azimuth either @xmath6 or @xmath33 ) while the rotation angle of the feed array is fixed at 19@xmath101 , yielding tracks for the seven array beams that are equally spaced in declination . in order to map the telescope s `` zone of avoidance '' near zenith , the band of tiles centered at dec.=@xmath105 will require a different orientation : with the azimuth arm nearly e - w . this strategy greatly simplifies the disentangling of main beam and sidelobe contributions to the maps : characterization of alfa parameters thus needs to be made on a greatly reduced volume of telescope configuration parameter space . drift mode observations , combined with the calibration scheme described above , yield maximally efficient use of telescope time , providing high photometric quality with very small overhead . we expect that , bar instrumental malfunctions , telescope time usage for science data will approach 97% . as discussed in section [ scaling ] , the volume sampled at any hi mass limit , for a survey of fixed total duration , varies with the integration time per point as @xmath106 . once a threshold sensitivity is reached , it is more advantageous to increase the solid angle of the survey than its depth . because of the spacing of the alfa beam tracks in drift mode discussed in section [ alfa ] , coverage of the sky in a one pass drift survey is slightly worse than nyquist . for a fixed amount of observing time , a single pass strategy would appear to maximize the number of detected sources . for a fixed total survey time , the loss of survey volume sampled by going from a one pass to a two pass drift survey is , according to section [ scaling ] , 19% . several advantages of a two pass strategy offset that loss , however : ( 1 ) a second pass will greatly aid the separation of cosmic emission from rfi which is unlikely to affect each pass identically ; ( 2 ) the denser sky sampling will allow statistical separation of spurious signals from cosmic ones , thus allowing reliable detection to lower values of s / n ; ( 3 ) if the two passes are made when the earth is at very different phases of its orbit , confirmation of cosmic nature for detection candidates can be obtained by verifying that they are separated in topocentric radial velocity by @xmath107 , where @xmath108 is the change in the angle between the line of sight to the detection candidate and the velocity vector of earth on its heliocentric orbit ; this requires that the second pass be undertaken 3 to 9 months after the first pass , modulo a year ; ( 4 ) the variability of radio continuum sources can be measured , and radio transients can be identified , allowing commensality with other science teams interested in studies of those phenomena ; ( 5 ) given design features of the alfa hardware , maintenance will be difficult and , as a result , alfa may operate at less than 100% capacity ( i.e. one or more beams may be unusable ) during some fraction of the time . loss of a beam in a single pass survey would result in grievous holes in sky coverage , whereas a two pass strategy would greatly attenuate the resulting damage to the survey . for all these reasons , the high galactic latitude arecibo sky to be mapped by alfalfa will be covered in two drift passes . the resulting effective integration time of the survey , per beam area , will be about 48 seconds . another way of expressing the sensitivity of the survey is in terms of the integration time per , which will be about 14,700 seconds . as shown in figure [ skycov ] , the sky to be mapped by alfalfa extends between @xmath109 and over two blocks of right ascension , respectively @xmath110 to @xmath111 and @xmath112 to @xmath113 , although the vagaries of telescope time allocation will produce some irregularities in the survey solid angle boundaries . the exclusion of the low galactic latitude regions within the telescope s horizon is driven by ( a ) the realistic assessment that pulsar and other galactic alfa surveys will greatly increase the pressure on low galactic latitude lsts and ( b ) the expectation that part of the low galactic latitude , extragalactic sky will be surveyed commensally with pulsar and other galactic surveys . for bookkeeping and data release purposes , the sky mapped by alfalfa will be subdivided into 378 _ tiles _ , each of @xmath114 in r.a . and @xmath104 in dec . mapping a tile in single pass drift mode requires 17 drifts of alfa , spaced @xmath115 in dec . and each yielding 7 drift tracks ; equally as many additional drifts are required to complete the second pass at a later time . for the second pass , beam tracks will be interleaved with those of the first pass , so that the final declination sampling will be @xmath116 , better than nyquist . in order to minimize `` scalloping '' of the gain over the map introduced by the higher gain of central beam relative to the outer ones , the second pass drifts are offset by 718 relative to the first pass tracks . the data processing environment chosen for alfalfa is idl . a substantial body of spectral line software generated by one of us ( pp ) already exists at the arecibo observatory . further development specific to alfalfa has been grafted on this fertile base . the tile size was chosen to constitute a data block that can reasonably be handled for data processing in an efficient manner by current desktop computers . the generation of raw data proceeds at the rate of @xmath117 gb / hr , and upon conversion from its raw fits format to an idl structure , a single 600 sec drift is @xmath118 mb . such a data block is well suited for one of the most computer intensive parts of the reduction pipeline , that of bandpass subtraction . the data for a full tile , after polarization averaging and regridding , can fit within the 24 gb memory of current low cost desktops . the data processing path for alfalfa data can be summarized as follows : * one fits file per 600record drift is generated by the data taking software at the arecibo observatory . by the end of each observing session , each of those is converted into an idl structure , a ` drift ' structure , and stored for further analysis at the observatory and the observers institutions . * within weeks , all data of an observing session is noise calibrated and a bandpass solution is computed . the `` bandpassed '' , calibrated and baselined spectral data for each beam / polarization configuration are obtained as output of an automated pipeline that is designed to preserve not only small angular scale features such as external galaxies , but also large structures such as hvcs and galactic hi . * the first detailed visual inspection of the data follows , in the course of which the observer flags regions of each position velocity map for rfi and other occurrences of data corruption . it is anticipated that as much as about half of all sources to be detected by alfalfa will be visible to the eye at this stage . a first automated signal extraction algorithm pass will produce a list of candidate detections . noise diode calibrated , bandpass corrected , baselined and rfi flagged spectra as obtained to this stage constitute what we shall refer to as _ level i data products_. * upon completion of the second pass through a given sky tile , data will be re calibrated using the continuum sources present within the tile , regridding of the sky sampling will take place , after smoothing by a homogenoeus resolution kernel and conversion into data cubes will follow . the output of this processing stage shall be referred to as _ level ii data products_. because telescope scheduling is a dynamic process which responds to proposal pressure at a national center , the scheduling of data releases far in advance is not possible . however , the alfalfa observing status is continuously updated at the survey website . observing plans foresee completion by bands of tiles and , when an accurate prediction of the completion is available , data release plans for that band will be posted . data release will take place through an alfalfa / hi node connected to the u.s . national virtual observatory . a preliminary example of web based data presentation is linked to the aforementioned website as well as directly reachable : it allows access to the spectral line data , images of optical counterparts and parameter tabulation of the precursor observations results discussed in paper ii . as an example of a level i data product and the potential of the alfalfa survey , figure [ n3628 ] shows a single pass drift across the galaxy ngc 3628 , one of the leo triplet galaxies , and its tidal tail . the displayed position velocity image consists of a constant declination drift of 600 1-sec records at dec . ( j2000 ) = @xmath1193645 , corresponding to dec . ( b1950 ) = @xmath1195309 for comparison with figure 1 of haynes , giovanelli & roberts ( 1979 ) . the single drift rms noise in the image is 3.5 mjy . contours are linearly spaced by 6 mjy and the lowest contour is plotted at 2 mjy per beam . the tidal tail is traced by the alfalfa data as far as the earlier point - by - point map , but a countertail at earlier right ascension than ngc 3628 is also clearly visible in the new map . after the second pass alfalfa data become available , the leo triplet region will be mapped with a sensitivity twice as deep as that shown in figure [ n3628 ] , allowing a detailed study of the system over a wide area . -5 cm -20 cm here we summarize the sensitivity of alfalfa at several survey levels : * a 1second record of a drift scan , after accumulation of both polarizations , will yield a spectrum of @xmath120 mjy , where @xmath121 is the spectral resolution in . * a single drift , position frequency map spatially smoothed to the spatial resolution of the telescope beam will yield @xmath122 mjy . * a spatially two dimensional map of two pass alfalfa data , smoothed with a kernel of 2 at half power , will have @xmath123 mjy per pixel . * the rms sensitivity per beam area , after a two pass survey , will be @xmath124 . * the 6@xmath98 hi column density limit will be @xmath125 atoms @xmath126 , for a spectral line of width @xmath127 , observed with a spectral resolution of @xmath121 . column density sensitivity is , in general , independent of telescope size , and thus alfalfa will not reach deeper @xmath128 levels than previous wide angle surveys such as hipass . in fact , given the shorter integration time per beam area , alfalfa will have lower sensitivity to @xmath128 than did hipass for very extended sources . it may be argued that only surveys with longer integration times per beam area than hipass can break new ground . however , this argument holds true only if sources are well resolved by the telescope beam . the beam area of arecibo is nearly _ 20 times smaller _ than that of the parkes telescope . if sources are unresolved by the beam , the telescope can only detect total flux , and the observation can not be used for any inference on source column density . in fact , very few extragalactic hi sources were resolved by the parkes beam ; the smaller arecibo beam size gives alfalfa a major advantage . to illustrate this point , figure [ sizehis ] shows two histograms of the angular size distribution of optically selected , catalogued galaxies : the upper one for galaxies known to be within the alfalfa survey region , the lower one for galaxies in the whole of the southern hemisphere . the optical size used for this comparison is d@xmath129 as catalogued in the third revised catalog of galaxies ( de vaucouleurs 1991 ) . it is well established that , on average , the hi size ( measured at the level near 1 m@xmath130 pc@xmath131 ) of optically selected galaxies is about 1.6 times that blue size ( broeils & rhee 1997 ) , although for dwarf irregular systems that ratio may rise significantly ( swaters 2002 ) . even allowing for a small number of extremely large hi - to - optical size ratios , the total number of galaxies resolved by the parkes telescope beam over the whole of the southern sky is on the order of a dozen or two . only for those , the magellanic stream and high velocity clouds is column density sensitivity of any relevance for hipass . the alfalfa survey should , on the other hand , resolve several hundred galaxies and high velocity clouds , and map their peripheries to a column density limit of the order of @xmath132 atoms @xmath126 . through careful analysis of the hi mapping datasets including consideration of the impact of sidelobe contamination in the two or three selected telescope configurations adopted for the drift survey , the alfalfa survey will address the issue of whether a column density regime below @xmath133 @xmath126 is commonly found in the local universe ( corbelli & bandiera 2002 ) . alfalfa will produce a catalog of tens of thousands of candidate detections ; on order of 20,000 will be cosmic sources . the number of candidate detections per bin of signal to noise @xmath62 will increase steeply , and the probability that the candidates are real sources decreases rapidly with diminishing @xmath62 value . internal ( i.e. within the survey data set ) corroboration of candidate detections will rely on ( a ) comparison of independent polarization samples and ( b ) comparison of spatially adjacent survey samples . the effectiveness of part ( b ) will depend on the spatial sampling density and , in the case of multiple drifts through the same region , on the temporal consistency of the data . these comparisons will help exclude many marginal candidate detections of non cosmic origin , which we shall refer to as `` false '' candidates . post survey , corroborating observations will be desirable to confirm candidate detections just below the signal to noise threshold above which signal corroboration can be internally possible . this may allow significant expansion of the survey ` catch ' with modest additional amounts of telescope time . the usual compromise is necessary in setting a @xmath62 threshold : too high a threshold will lose many valuable potential detections ; too low a threshold will require impractical amounts of post survey telescope time ; a haphazard criterion may corrupt the completeness of acquired samples . we expect that the bulk of follow up observations to corroborate the cosmic nature of detection candidates will be carried out at arecibo , using a single pixel feed , hopping from one candidate to the next , minimizing slew and setup time . two modes may lead to low efficiency usage of telescope time : too dense a set of follow up targets may be comprised of too large a fraction of false candidates , and thus produce a low yield per unit of telescope time ; too sparse a set may lead to a large fraction of the time spent slewing and in other overhead . careful optimization will be required . we consider some of these issues in this section . visual inspection or automated signal extraction algorithms identify detection candidates that can be assigned to three classes : ( i ) cosmic sources , ( ii ) extreme statistical fluctuations of noise and ( iii ) spurious signals due to rfi or other instrumental and data analysis causes . _ cosmic sources . _ figure [ stnhisto ] shows a histogram of signal to noise @xmath62 of expected sources , as obtained in one of the simulations described in section [ simul ] . we emphasize that the number of sources rises steeply as @xmath62 decreases , illustrating the well known fact that most of the survey candidate detections will occur near the detection limit . because of the massive bulk of alfalfa data sets , signal extraction will largely rely on automated procedures . amelie saintonge has coded a matched filter , cross correlation signal extraction algorithm , described elsewhere ( saintonge , in preparation ) . detection probability simulations have been carried out with this algorithm , by randomly injecting a gaussian signal in simulated spectra with gaussian noise , and monitoring the effectiveness of the signal extraction algorithm in recovering the injected signal . the detection probability is computed as the fraction of all trials in which the signal extraction algorithm positively identifies the injected signal ; such probability is monitored as a function of @xmath62 and of signal width . when the noise is measured after spectrally averaging over 1/2 the spectral width of the injected signal , the detection probability is largely independent of the signal width . _ statistical noise fluctuations . _ for gaussian noise , the probability that a single spectral channel yield a fluctuation of signal to noise between @xmath134 and @xmath135 is p_1 ds_1 = 1 e^-s_1 ^ 2/2 ds_1 , where @xmath136 , with @xmath43 the peak flux density and @xmath137 the rms noise , with single channel spectral resolution . similarly , the probability for a @xmath138 channels wide spectral feature to exhibit a deviation between @xmath139 and @xmath140 is p_n ds_n = 1 e^-s_n^2/2 ds_n , where @xmath141 . in a survey of @xmath142 line of sight samples , taken with a spectrometer of @xmath143 channels , the number of samples @xmath138 channels wide , with signal to noise between @xmath62 and @xmath144 is n_s , n_w ds = n_los n_cn_w p_n ds = n_los n_cn_w 1 e^-s^2/2 ds and the total number of statistical fluctuations of that width with @xmath62 larger than a threshold @xmath145 is n_s_th,1 = n_los n_cn_w 1 _ s_th^ e^-s^2/2 ds = n_los n_cn_w [ f()-f(s ) ] , where , again , the noise is computed with a spectral resolution of @xmath146 channels , and @xmath147 is the familiar cumulative distribution of the normal error function : @xmath148 , @xmath149 and @xmath150 . the total number of purely statistical noise fluctuations between @xmath151 and @xmath152 , with widths between @xmath153 and @xmath154 , appearing in the survey will then be n_[a , b],[1,2 ] = _ n_w = n_w1,n_w2 n_los n_cn_w [ f(s_a)-f(s_b ) ] . to first order , and ignoring the fact that in the expression above a high @xmath62 , broad feature gets overcounted as several , lower @xmath62 , narrower ones , we can approximate n_[a , b],[1,2 ] n_los n_c [ f(s_a)-f(s_b ) ] . [ nab12 ] for example , for a survey that samples @xmath69 lines of sight , with a spectrometer usefully covering 85 mhz with @xmath155 spectral channels ( so that a velocity width range between 25 and 500 translates into @xmath156 and @xmath157 , one should expect @xmath158 features with @xmath159 , @xmath160 features with @xmath161 , and @xmath162 features with @xmath163 , with width anywhere between 25 and 500 . these are ominously large numbers when compared to expected numbers of cosmic sources between @xmath164 and @xmath165 , for that range of @xmath62 . _ rfi or other spurious signals . _ the above assumption of gaussian noise is heuristic . the true nature of the noise will be ascertained after a significant fraction of the survey data will have been collected and the `` normal '' characteristics of both equipment and rfi environment will have been measured . as we discuss in paper ii , the precursor observations carried out in 2004 were taken in commissioning mode for the alfa hardware , and several internal bugs have been found and fixed since those observations were completed . those observations are not well suited for a careful analysis of the problem . for the moment , we will ignore the impact of rfi and other non gaussian sources of noise , and restrict our analysis to the discrimination between cosmic sources and gaussian noise fluctuations . assuming that the majority of weak detection candidates will be unresolved by the telescope beam , the most important means of discriminating between cosmic sources and noise fluctuations will result from comparison of contiguous drift tracks and different polarizations of the same beam . consider a single pass alfalfa survey , whereby contiguous drift tracks are separated by 2.1 in declination on the sky . a point source swept by one of the feeds will appear also in contiguous tracks , at lower @xmath62 . for an alfa beam averaging 3.5 width at half power , the response 2.1 off the beam center is about 0.37 of that on beam center . in a two pass drift survey , the track separations are 1.05 , and at that distance from the beam center , the beam response is 0.78 of that on beam center . a point source will thus be far more easily confirmed in a two pass survey . similarly , since the hi line is unpolarized , the comparison of the two independent polarization spectra of the same beam will deliver equal signals , plus noise , for cosmic sources , and completely uncorrelated results for noise . we shall refer to the exclusion of detection candidates made possible by comparison of adjacent drift tracks and polarization channels as _ vicinity trimming _ , and distinguish three sets of detection candidates : ( 1 ) that obtained without any vicinity trimming , and those ( smaller ones ) obtained after ( 2 ) vicinity trimming in a single pass survey and ( 3 ) vicinity trimming in a double pass survey . figure [ cumdets ] displays the cumulative number of candidate detections plotted as a function of signal to noise , expected for an alfalfa like survey . the simulation assumes a rs02 himf and a sensitivity corresponding to a double pass drift survey . the thick line corresponds to the detection candidates associated with cosmic ( ` real ' ) sources . the three thin lines correspond to the expected number of noise fluctuations with the given @xmath62 in the three ` vicinity trimming ' cases described above . at @xmath166 , most of the detection candidates are cosmic sources . at @xmath167 , the number of ` real ' cosmic sources is 25% higher than at @xmath73 , but the candidate detections resulting from noise fluctuations is several times higher than that of real sources . vicinity trimming can however drastically reduce the number of candidate detections deserving attention . only at signal to noise levels @xmath168 does the number of noise fluctuations overwhelm that of cosmic sources , after two pass vicinity trimming . at this level , however , the impact of low power level rfi will play an important and yet quantitatively unknown role . candidate detections with a comfortably high signal to noise threshold @xmath145 will not need corroborating follow up observations in order to confirm their reality as cosmic sources . without considering the impact of rfi , @xmath145 may be in the vicinity of 6 ; consideration of the impact of rfi may raise @xmath145 to higher values , in a variable manner depending on the frequency of the candidate signal . candidates with @xmath169 or slightly below that level will be reobserved with the arecibo telescope . to how low a level of @xmath62 should re observations be considered ? a simple criterion would be that corroborating observations should be requested only for candidates of @xmath62 such that the expectation of confirmation , expressed in terms of detections per unit of telescope time , is at least as high as for the full survey . if a corroborating observation is to require an increase in the @xmath62 from a value of , say , 5 for the survey data , to about 7 for the corroborating observation , integration times of at least 1 minute per candidate will be necessary for corroborating observations . observing runs to corroborate several hundred candidate detections at a time will thus be the norm . if the set of candidate detections to be checked is very sparse say one candidate every several square degrees slew times will be very substantial and bandpass correcting observations will be required for each candidate source , more than doubling the required telescope time . in that case , the on source @xmath170 of order of 1 minute may be a small fraction of the overall time required to observe each source . the arecibo telescope slew times are respectively @xmath171 s@xmath13 in azimuth and @xmath172 s@xmath13 in elevation . a 1@xmath101 change in elevation will require 25 seconds . it will thus be observationally advantageous if the sky density of tentative sources to be corroborated is high , e.g. on order of one per square degree or higher . not only will that reduce the overhead of slew motions and settle time , but it will also allow for a running mean bandpass to be accumulated over a few contiguous staring observations , as the telescope configuration would change little between adjacent source candidates . in that case , allowing for slew and settle time , a corroborating observation of a single source will require on order of one to two minutes of telescope time . the steeply rising fraction of ` false ' sources with decreasing @xmath62 suggests that corroborating observations requesting single pixel telescope time at the level of approximately 10% of the request proposed for the alfa observations will deliver optimal returns . alfalfa uses the new 7-beam arecibo l - band feed array to carry out a wide area survey of the high galactic latitude sky visible from arecibo . in addition to the all important sensitivity advantage that accrues from using arecibo , the world s most sensitive radio telescope at l band , alfa offers important and significant improvements in angular and spectral resolution over the available major wide area extragalactic hi line surveys such as hipass and hijass . alfalfa is intended to produce an extensive database of hi spectra that will be of use to a broad community of investigators , including many interested in the correlative mining of multiwavelength datasets . it is specifically designed to probe the faint end of the himf in the very local universe . as a result of practical considerations and simulations of survey efficiency , alfalfa exploits a simple fixed - azimuth drift scanning _ minimum intrusion _ technique . a two pass strategy will greatly aid in the rejection of spurious signals and rfi , thus minimizing the need for follow - up confirmation observations , evening out the scalloping in the maps that arises from unequal pixel gain , and offering the opportunity to use the same dataset for the statistical characterization of continuum transients . initial tests of the hardware , software and survey observing mode , conducted in fall 2004 during the alfa commissioning phase as described in paper ii confirm the efficacy of the planned approach . the basic parameters of the alfalfa survey can be summarized accordingly as follows : * sky coverage of 7074 , between @xmath6 and @xmath173 in declination , 7.5 to 16.5 and 22.0 to 3.0 hrs in right ascension , with 3.5 spatial resolution . * frequency coverage between 1335 and 1435 mhz , yielding coverage of extragalactic hi in redshift out to @xmath174 , with 5.3 maximum spectral resolution . * sensitivity of @xmath175 mjy per beam area , where @xmath121 is the spectral resolution in . * on order of 20,000 hi sources are expected to be detected by the survey . extragalactic hi sources with @xmath176 will be detectable to a distance of 6.5 mpc , while hi masses @xmath177 , will be detectable throughout most of the local supercluster , including the virgo cluster and out to 20 mpc . several hundreds will have @xmath5 , thus allowing a robust determination of the faint end of the himf . * public access data products will be produced on a continuing basis as subsets ( tiles ) of the overall survey are completed . observations for alfalfa started in february 2005 , and completion of the survey is expected to take five to six years . cataloguing a complete census of hi locally , pinning down the himf to the lowest masses and conducting the blind hi absorption and oh megamaser surveys will require completion of the full 5-year program , but even the initial 2005 allocation promises early science results in several important areas including : the mapping of nearly 1600 , more than 3 times the coverage with twice the sensitivity of the arecibo dual beam survey ( rosenberg & schneider 2000 ) ; a first blind census across the virgo cluster region with a detection limit of m@xmath178 at the cluster distance ( assuming a width w = 30 ) ; a complete search for hvcs around m33 ; the identification of gas - rich galaxies in the ngc 784 and leo i groups ; the mapping of the environments of 12 gas - rich galaxies with d@xmath179 7 ; and a first attempt at a large blind survey for hi absorbers . rg and mph acknowledge the partial support of naic as visiting scientists during the period of this work . this work has been supported by nsf grants ast0307661 , ast0435697 , ast0347929 , ast0407011 , ast0302049 ; dgict of spain grant aya200307468c0301 ; and by a brinson foundation grant . we thank the director of naic , robert brown , for stimulating the development of major alfa surveys , enzo branchini for providing a pscz density grid in digital form , h ' ector hern ' andez for a sensible and friendly approach to telescope scheduling , the director , telescope operators and support staff of the arecibo observatory for their proactive assistance .

we consider two distinct coupled cluster ( cc ) perturbation series that both expand the difference between the energies of the ccsd ( cc with single and double excitations ) and ccsdt ( cc with single , double , and triple excitations ) models in orders of the mller - plesset fluctuation potential . we initially introduce the series , in which the ccsd amplitude equations are satisfied at the expansion point , and compare it to the recently developed series [ j. chem . phys . , 064108 ( 2014 ) ] , in which not only the ccsd amplitude , but also the ccsd multiplier equations are satisfied at the expansion point . the computational scaling is similar for the two series , and both are term - wise size extensive with a formal convergence towards the ccsdt target energy . however , the two series are different , and the series is found to exhibit a more rapid convergence up through the series , which we trace back to the fact that more information at the expansion point is utilized than for the series . the present analysis can be generalized to any perturbation expansion representing the difference between a parent cc model and a higher - level target cc model . in general , we demonstrate that , whenever the parent parameters depend upon the perturbation operator , a perturbation expansion of the cc energy ( where only parent amplitudes are used ) differs from a perturbation expansion of the cc lagrangian ( where both parent amplitudes _ and _ parent multipliers are used ) . for the latter case , the bivariational lagrangian formulation becomes more than a convenient mathematical tool , since it facilitates a different and faster convergent perturbation series than the simpler energy - based expansion .

1512.06552

coupled cluster ( cc ) theory @xcite is perhaps the most powerful method for describing dynamical electron correlation effects within the realm of modern quantum chemistry . the cc singles and doubles ( ccsd ) model @xcite , in which the cluster operator is truncated at the level of double excitations , is a robust and useful model , but it is well - known that the effects of triple ( and higher - level ) excitations need to be taken into account in order to obtain highly accurate results that may compete with the accuracy of experiments @xcite . however , the steep scaling of the cc singles , doubles , and triples @xcite ( ccsdt ) and cc singles , doubles , triples , and quadruples @xcite ( ccsdtq ) models limits their use to rather modest molecular systems . for this reason , a computationally tractable alternative to the expensive iterative ccsdt and ccsdtq models is to develop approximate models , for which the important triples and/or quadruples contributions are determined from a perturbation analysis , and hence included in a cheap and non - iterative fashion . a plethora of different models for the approximate treatment of triples and/or quadruples effects have been suggested , and we refer to ref . for a recent theoretical overview of approximate non - iterative triples and quadruples models and refs . and for a numerical comparison of many of these . in the present work , we focus on perturbation theory within a cc framework , where a mller - plesset ( mp ) partitioning of the hamiltonian is performed @xcite , and the energy difference between a zeroth - order ( parent ) cc model and a higher - level ( target ) cc model is expanded in orders of the perturbation ( the fluctuation potential ) . in particular , we will base the perturbation analysis on a bivariational cc lagrangian obtained by adding to the cc target energy the cc amplitude equations with associated lagrange multipliers . we note that the linearly parametrized state formally spanned by the lagrange multipliers is often referred to as the cc @xmath3-state @xcite , and that this is in general different from the exponentially parametrized cc state . as pointed out by arponen @xcite , extensively exploited in the cc lagrangian formulation@xcite , and recently discussed by kvaal @xcite , the cc energy may be interpreted as a cc functional in both the cc amplitudes and the @xmath3-state parameters . we will show that the fastest convergence is obtained when these two sets of state parameters are treated on an equal footing in the perturbative expansion of the energy difference between a parent and a target cc model . thus , we will distinguish between a perturbation expansion of the cc energy , for which only parent amplitudes are used at the expansion point , and a perturbation expansion of the cc lagrangian , for which both parent amplitudes _ and _ parent multipliers are used at the expansion point . at first sight , the bivariational lagrangian formulation might seem like an unnecessary complication , since , for the target model , the lagrangian formally equals the energy . the purpose of this work is , however , to highlight and exemplify that not only is the lagrangian formulation a convenient mathematical tool that may simplify the derivation of various perturbation expansions ; in many cases , the lagrangian formulation will actually lead to different perturbation series than the corresponding energy formulation . to exemplify this difference , we consider two perturbation series which both expand the difference between the energies of the ccsd and ccsdt models . we initially introduce the @xmath0 series , in which the ccsd amplitude equations are satisfied at the expansion point , and next compare it to the recently developed @xmath1 series @xcite , in which not only the ccsd amplitude equations are satisfied at the expansion point , but also the ccsd multiplier equations . despite depending on the fluctuation potential to infinite order in the space of all single and double excitations , the ccsd amplitudes are formally considered as zeroth - order parameters in both the @xmath0 and @xmath1 series , since the ccsd model represents the expansion point . similarly , the ccsd multipliers , which too depend on the fluctuation potential , are considered as zeroth - order parameters for the @xmath1 series , but not so for the @xmath0 series . the path from the ccsd expansion point towards the ccsdt target energy , as defined by a perturbation expansion , is thus different within the @xmath0 and @xmath1 series , and , as will be shown in the present work , the @xmath1 series is the more rapidly converging of the two , since more information is utilized at the expansion point . we finally reiterate that the lowest - order contribution of the @xmath1 series ( that of the @xmath4 model ) is identical to the triples - only part of the ccsd(2 ) model of gwaltney and head - gordon @xcite and the second - order model of the cc(@xmath5)pt(@xmath6 ) series of hirata _ et al . _ @xcite , the @xmath7 model is identical to the triples - only part of the third - order cc(@xmath5)pt(@xmath6 ) model , while for fourth and higher orders , the @xmath1 and cc(@xmath5)pt(@xmath8 ) series are different @xcite . the @xmath0 series , however , differs from the aforementioned perturbation series for all corrections . + the present study is outlined as follows . in [ sec : theory ] , we derive the @xmath0 series and compare it to the @xmath1 series in order to illustrate the importance of treating parent amplitudes and multipliers on an equal footing . in [ sec : results ] , we present numerical results for the @xmath0 and @xmath1 energies , while some concluding remarks are given in [ sec : conclusion ] . in this section , we consider two perturbation series that expand the difference between the ccsd and ccsdt energies in orders of the perturbation . in [ sec : eccsdt ] , we develop a new energy - based perturbation series denoted the @xmath0 series , which can be formulated in terms of cc amplitudes without the need for invoking a cc lagrangian . next , in [ sec : general ] , we develop a common bivariational framework , in which we recast the @xmath0 and @xmath1 series . finally , we present a theoretical comparison between the two series in [ sec : comparison ] in order to exemplify how the cc lagrangian framework may lead to a perturbation series that is inherently different from that which arise from the corresponding energy formulation . in this work , we use a mp partitioning of the hamiltonian @xmath9 where @xmath10 is the fock operator and @xmath11 is the fluctuation potential . we consider first the ccsd model , which we choose as the common reference point for the perturbation expansions to follow . the ccsd energy , @xmath12 , and associated amplitude equations may be written as @xmath13 and @xmath14 where @xmath15 and @xmath16 represent a singly and a doubly excited state with respect to the hf determinant , @xmath17 , and the ( non - hermitian ) ccsd similarity - transformed hamiltonian is given by @xmath18 with @xmath19 and @xmath20 being the ccsd singles and doubles cluster operators . throughout the paper , we will use asterisks to denote ccsd quantities and generally use the generic notation @xmath21 for a cluster operator at excitation level @xmath22 , where @xmath23 is an excitation operator and @xmath24 is the associated amplitude . we now parametrize the difference between the ccsd and ccsdt energy in terms of correction amplitudes , @xmath25 , which represent the difference between the ccsd and ccsdt amplitudes . the correction amplitudes are expanded in orders of the fluctuation potential , and the ccsdt amplitudes , @xmath24 , may thus be written as @xmath26 where @xmath27 for @xmath28 and @xmath29 . we emphasize that , since we have chosen to expand the ccsdt amplitudes around the ccsd reference point , the ccsd amplitudes , @xmath30 , are zeroth - order by definition . the @xmath31 amplitudes thus represent corrections to the ccsd amplitudes , while @xmath32 are the ccsdt triples amplitudes . the ccsdt cluster operator may now be written as @xmath33 , where @xmath34 contains the correction amplitudes , and the ccsdt energy may be obtained by projecting the ccsdt schrdinger equation , @xmath35 , against @xmath36 @xmath37 + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_1],\delta\hat{t}_1 ] \ket { { \mathrm{hf } } } \label{etf3}\end{aligned}\ ] ] where we have carried out a baker - campbell - hausdorff ( bch ) expansion , while the ccsdt amplitude equations are obtained by projection against the combined excitation manifold of all single , double , and triple excitations out of the hf reference @xmath38 the order analysis of [ ampeq ] is identical to the one performed in ref . ( orders are counted in @xmath11 ) , and the resulting amplitudes are thus the same . compactly , these are given by [ amp_pert_expand ] @xmath39 + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg)^{(n ) } \qquad ( i=1,2 ) \\ \delta t_{\mu_3}^{(n ) } & = - \epsilon_{\mu_3}^{-1 } \bigg ( \bra{\mu_3 } { \hat{\phi}^ { { { ^{*}}\hat{t } } } } + [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg)^{(n)}\end{aligned}\ ] ] where @xmath40 is the orbital energy difference between the virtual and occupied spin - orbitals of excitation @xmath41 , and the right - hand sides of the equations contain all terms of order @xmath8 , i.e. , the sum of the orders of all @xmath42 operators plus one ( for the fluctuation potential ) equals @xmath8 . for example , the first - order singles and doubles corrections are zero , @xmath43 , while the first - order triples corrections are given as @xmath44 . by collecting terms in orders of the fluctuation potential , the ccsdt energy in [ etf3 ] may now be expanded as [ eccsdtn ] @xmath45 \ket { { \mathrm{hf } } } \\ & + \tfrac{1}{2 } \sum_{m=2}^{n-3 } \bra { { \mathrm{hf } } } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_1^{(m ) } ] , \delta\hat{t}_1^{(n - m-1 ) } ] \ket { { \mathrm{hf } } } \label{eccsdtn2}\end{aligned}\ ] ] where we have used the fact that the first - order singles and doubles amplitudes vanish to restrict the summations . we denote the perturbation series defined by [ eccsdtn ] as the _ @xmath0 series _ to emphasize that it is based on a perturbation expansion of the ccsdt energy around the ccsd _ energy _ point , at which the ccsd amplitude equations are satisfied . we note that this notation is not to be confused with ecc , which is usually used as an acronym for extended coupled cluster theory in the literature @xcite from [ eccsdtn ] , it follows that the first non - vanishing energy correction to the @xmath0 series is of third order . the two lowest - order corrections are given in [ e3alt_app_n_1 ] and [ e4alt_app_n_1 ] of [ appendix_comparison ] . the @xmath0 series is evidently different from the @xmath1 series developed in ref . , which starts at second order . both series , however , describe the difference between the ccsd and ccsdt energies using a mp partitioning of the hamiltonian , and the correction amplitudes are identical . the only apparent difference is that the ccsdt energy is the central quantity for the @xmath0 series , while the ccsdt lagrangian forms the basis for the @xmath1 series . in [ sec : general ] , we develop a general lagrangian framework to enable a direct comparison of the @xmath0 and @xmath1 series in [ sec : comparison ] . the ccsdt lagrangian may be obtained by adding to the ccsdt energy the amplitude equations in [ ampeq ] with associated ( undetermined ) multipliers @xmath46 where we have chosen the following parametrization of the ccsdt multipliers in analogy with [ ccsdt_ccsd_exp ] @xmath47 if the expansion of the ccsdt multipliers in [ multpara ] is left untruncated , these will equal the parameters of the linearly parametrized ccsdt @xmath3-state . the notation for the lagrangian , @xmath48 , highlights that the amplitudes and multipliers at the expansion point are @xmath49 and @xmath50 , respectively , while @xmath51 and @xmath52 represent the correction amplitudes and correction multipliers , respectively . by setting @xmath53 , we arrive at an mp - like perturbation expansion , albeit one that has the ccsdt energy as the target energy instead of the full configuration interaction ( fci ) energy . in this work , however , we focus on the case for which the ccsd amplitudes are used as zeroth - order amplitudes ( @xmath54 , @xmath55 , @xmath29 ) , while considering different choices of zeroth - order multipliers . in particular , we show below how the @xmath0 series may be recovered by choosing @xmath56 ( i=1,2,3 ) , while the @xmath1 series corresponds to using ccsd multipliers as zeroth - order multipliers ( @xmath57 , @xmath58 , @xmath59 ) . equations for the ccsd multipliers are obtained by requiring the ccsd lagrangian @xmath60 to be stationary with respect to variations in the ccsd amplitudes @xmath61 \ket { { \mathrm{hf } } } + \sum_{j=1}^2 \sum_{\nu_j } { { ^{*}}\bar{t}}_{\nu_j } \bra{\nu_j } [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \hat{\tau}_{\mu_i } ] \ket { { \mathrm{hf } } } \qquad ( i=1,2 ) \ .\end{aligned}\ ] ] in the following , we let @xmath62 represent a general zeroth - order multiplier to treat the two series on an equal footing . equations for the correction amplitudes are determined by requiring @xmath63 to be stationary with respect to variations in the correction multipliers @xmath64 which reproduces the ccsdt amplitude equations in [ ampeq ] . it follows that the equations for the correction amplitudes are independent of the choice of zeroth - order multipliers , and the correction amplitudes for the @xmath0 and @xmath1 series are therefore identical to all orders . equations for the ccsdt multipliers are obtained by requiring @xmath63 to be stationary with respect to variations in the amplitudes @xmath65 unlike for the correction amplitudes in [ ampeq ] , the precise form of the multiplier equations will depend upon the choice of zeroth - order multipliers , and the correction multipliers for the @xmath0 and @xmath1 series will therefore be different . to keep this distinction clear , we will refer to the correction multipliers associated with the choices @xmath66 and @xmath67 as @xmath68 and @xmath69 , respectively , while the generic notation @xmath52 may refer to either of the series . we note that at infinite order , the multipliers as defined within either the @xmath0 or @xmath1 series are identical ( assuming that both expansions converge ) , i.e. @xmath70 where @xmath71 is the final set of converged ccsdt multipliers ( @xmath3-state parameters ) . + to simplify the comparison of the two series , it proves convenient to expand the lagrangian in [ lt1 ] in a form that emphasizes the dependence on the ccsd multiplier equation in [ ccsdmult ] @xmath72 \ket { { \mathrm{hf } } } + \sum_{j=1}^2 \sum_{\nu_j } { \bar{t}}_{\nu_j}^{(0 ) } \bra{\nu_j } [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \hat{\tau}_{\mu_i } ] \ket { { \mathrm{hf } } } \bigg ) \\ \nonumber & + \tfrac{1}{2 } \bra { { \mathrm{hf } } } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_1 ] , \delta\hat{t}_1 ] \ket { { \mathrm{hf } } } \\ \nonumber & + \sum_{j=1}^2 \sum_{\nu_j } { \bar{t}}_{\nu_j}^{(0 ) } \bigg ( \bra{\nu_j } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_3 ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \\ \nonumber & + \sum_{j=1}^2 \sum_{\nu_j } \delta { \bar{t}}_{\nu_j } \bigg ( \bra{\nu_j } [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \\ & + \sum_{\nu_3 } \delta { \bar{t}}_{\nu_3 } \bigg ( \bra{\nu_3 } { \hat{\phi}^ { { { ^{*}}\hat{t } } } } + [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \label{lt2}\end{aligned}\ ] ] where we have set @xmath73 and used the ccsd amplitude equations in [ ccsdamp ] . + by choosing @xmath74 in [ lt2 ] , we arrive at the lagrangian @xmath75 , which reads @xmath76 + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_1 ] , \delta\hat{t}_1 ] \ket { { \mathrm{hf } } } \\ \nonumber & + \sum_{j=1}^2 \sum_{\nu_j } { \delta \bar{t}^e}_{\nu_j } \bigg ( \bra{\nu_j } [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \\ & + \sum_{\nu_3 } { \delta \bar{t}^e}_{\nu_3 } \bigg ( \bra{\nu_3 } { \hat{\phi}^ { { { ^{*}}\hat{t } } } } + [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \ . \label{lt3}\end{aligned}\ ] ] since the @xmath68 multipliers multiply the ccsdt amplitude equations in [ ampeq ] , they may be eliminated from [ lt3 ] , and it follows that @xmath75 equals the ccsdt energy in [ etf3 ] @xmath77 by evaluating @xmath75 to different orders in the fluctuation potential , we thus arrive at the @xmath0 series in [ eccsdtn ] , for which the energy corrections may be expressed exclusively in terms of correction amplitudes . the evaluation of the @xmath0 energy corrections using [ eccsdtn ] corresponds to using the @xmath78 rule , where amplitudes to order @xmath8 determine the energy to order @xmath78 . alternatively , by exploiting that @xmath75 is variational in the amplitudes as well as the multipliers , it may be evaluated to different orders in the fluctuation potential by using the @xmath79 and @xmath80 rules @xcite for the amplitudes and multipliers ( i.e. , the amplitudes / multipliers to order @xmath8 determine the lagrangian to order @xmath81/@xmath80 ) . these two approaches are of course equivalent , but the use of the @xmath79 and @xmath80 rules allows for an easy comparison of the @xmath0 series to the @xmath1 series ( cf . [ appendix_comparison ] ) . + by setting @xmath82 in [ lt2 ] , the resulting lagrangian @xmath83 becomes @xmath84 , \delta\hat{t}_1 ] \ket { { \mathrm{hf } } } \\ \nonumber & + \sum_{j=1}^2 \sum_{\nu_j } { { ^{*}}\bar{t}}_{\nu_j } \bigg ( \bra{\nu_j } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_3 ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \\ \nonumber & + \sum_{j=1}^2 \sum_{\nu_j } { \delta \bar{t}^l}_{\nu_j } \bigg ( \bra{\nu_j } [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \\ & + \sum_{\nu_3 } { \delta \bar{t}^l}_{\nu_3 } \bigg ( \bra{\nu_3 } { \hat{\phi}^ { { { ^{*}}\hat{t } } } } + [ { \hat{h}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] + \tfrac{1}{2 } [ [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t } ] , \delta\hat{t } ] + \ldots \ket { { \mathrm{hf } } } \bigg ) \label{lt5}\end{aligned}\ ] ] where we have used the ccsd multiplier equations in [ ccsdmult ] . an expansion of @xmath83 in orders of the fluctuation potential defines the recently proposed @xmath1 series @xcite . the @xmath0 series begins at third order ( see [ eccsdtn ] ) , while the @xmath1 series starts already at second order . the series are evidently different , and in [ sec : comparison ] we perform an explicit comparison of them . in this section , we compare the @xmath0 and @xmath1 series first , we discuss the origin of the difference between the series from a formal point of view , and next , we compare the two lowest - order multipliers and energy corrections for the two series . as shown in [ sec : general ] , both series can be derived from [ lt2 ] with different choices of parent multipliers , resulting in the lagrangians , @xmath75 and @xmath83 , for the @xmath0 and @xmath1 series , respectively . the @xmath75 lagrangian reduces to the standard ccsdt energy expression , because no zeroth - order multipliers enter the lagrangian and because the correction multipliers , @xmath85 , are multiplied by the ccsdt amplitude equations ( cf . [ lt3 ] and [ lt4 ] ) . however , the @xmath83 lagrangian can not be subject to a similar reduction , since certain terms in the ccsdt amplitude equations were cancelled when the ccsd multiplier equations were used to manipulate [ lt2 ] to arrive at [ lt5 ] . consequently , the ccsd multipliers can not be removed from [ lt5 ] , and @xmath83 can therefore not be reduced to the standard ccsdt energy expression . in the same way as the ccsd amplitudes formally depend on the fluctuation potential to infinite order in the space of all single and double excitations , so do the ccsd multipliers ( or ccsd @xmath3-state parameters ) . thus , since the ccsd multipliers are not counted as zeroth - order parameters in the @xmath0 series ( unlike in the @xmath1 series ) , orders are necessarily counted differently in the perturbative expansions of @xmath83 and @xmath75 ; however , we may compare the two series by comparing their leading - order , next - to - leading - order , etc . , corrections to one another , which we will do numerically in [ sec : results ] . on that note , we have theoretically compared the lowest- and next - to - lowest - order corrections of the two series in [ appendix_comparison ] . in summary , we find that for the @xmath83 lagrangian , a good zeroth - order description of the ccsdt @xmath3-state in the space of all single and double excitations is used ( the ccsd @xmath3-state ) , and the lowest - order multiplier correction therefore occurs in the triples space ( cf . [ ccsdmult_order_exp_3_1 ] ) . for the @xmath75 lagrangian , on the other hand , there is no representation of the ccsdt @xmath3-state at zeroth order , and the leading - order contributions to this state ( in the form of the first - order multipliers , @xmath86 ) hence occur within the singles and doubles space , while triples multipliers first enter the @xmath0 series at the following ( second ) order ( cf . [ ccsdmult_order_exp_2 ] ) . in fact , we find that the first- and second - order singles and doubles multipliers , @xmath87 for @xmath28 , are nothing but the two lowest - order contributions to the ccsd @xmath3-state parameters , if the ccsd multiplier equations in [ ccsdmult ] are solved perturbatively ( cf . [ tbs_pert_expansion ] ) . furthermore , the lowest - order @xmath0 and @xmath1 energy corrections are found to have the same structural form , however , they are expressed in terms of different sets of multipliers ( cf . [ e3alt_plus_e4alt_app ] and [ l2alt_plus_l3alt_app ] ) . thus , the @xmath0 series may be viewed as attempting to compensate for the poor ( non - existing ) guess at the ccsdt @xmath3-state by mimicking the @xmath1 series as closely as possible within a perturbational framework . for both series , a perturbative solution of the ccsdt @xmath3-state is embedded into the energy corrections , and the @xmath0 series is thus trailing behind the @xmath1 series from the onset of the perturbation expansion . again , this motivates the claim that it is advantageous to consider the cc energy as the stationary point of an energy functional in both the cc and @xmath3-state parameters , and hence that perturbative expansions are optimally carried out whenever the two states are treated on an equal footing @xcite . for higher orders in the perturbation , a direct comparison of the @xmath0 and @xmath1 series is more intricate , but we note that they will differ to all orders . only in the infinite - order limit are the two bound to agree ( @xmath75 and @xmath83 both equal the ccsdt energy ) , but this is a natural consequence of the fact that the cc energy may be described in terms of fully converged cc amplitudes alone . + in conclusion , the @xmath75 and @xmath83 lagrangians have the same parent energy ( ccsd ) and the same target energy ( ccsdt ) , and all contributions of both series will be trivially term - wise size extensive to all orders , as they are all expressed in terms of ( linked ) commutator expressions . however , the paths between these two cc energies , as defined by a perturbation expansion , are obviously very different for the two series . conceptually , the expansion point for the @xmath0 series is formally the _ ccsd energy _ ( only the ccsd amplitude equations are satisfied ) , while the expansion point for the @xmath1 series is the _ ccsd lagrangian _ ( the ccsd amplitude _ and _ ccsd multiplier equations are satisfied ) . for the energy - based @xmath0 series , the lagrangian is thus merely a mathematical tool that allows for correction energies to be obtained using amplitude and multiplier corrections that satisfy the @xmath81 and @xmath80 rules . on the contrary , the @xmath1 series is deeply rooted within a bivariational lagrangian formulation and has no energy - based analogue . the performance of the models of the @xmath1 series has recently been theoretically as well as numerically compared to a variety of alternative triples models for two sets of closed - shell @xcite and open - shell species @xcite , and the formal convergence of the series ( through sixth order in the perturbation ) has been confirmed . furthermore , the performance of the higher - order @xmath1 models with respect to the target ccsdt model was found to be essentially independent of the hf reference used , and thus , independent of the spin of the ground state . in this section , we assess the numerical performance of the @xmath0 models ( once again measured against results obtained with the target ccsdt model ) in order to compare the rate of convergence throughout the series to that of the @xmath1 series . we here use the two test sets previously used in refs . , , , and : @xmath88 17 closed - shell molecules , all optimized at the all - electron ccsd(t)/cc - pcvqz level of theory , and @xmath89 18 open - shell atoms and radicals , all optimized at the all - electron ccsd(t)/cc - pvqz level of theory . for a specification of the members of the closed- and open - shell test sets as well as tabularized molecular geometries , cf . refs . and the papers describing the heat thermochemical model @xcite , respectively . all of the closed - shell calculations are based on a restricted hf ( rhf ) reference , while unrestricted hf ( uhf ) as well as restricted open - shell hf ( rohf ) trial functions have been used for the open - shell calculations . the correlation - consistent cc - pvtz basis set @xcite is used throughout for all of the reported valence - electron ( frozen - core ) results , and the aquarius program @xcite has been used for all of the calculations . in [ e_t_n_figure ] , we consider the performance of the five lowest - order models of the @xmath0 and @xmath1 hierarchies . mean recoveries ( in @xmath90 ) of the triples correlation energy , @xmath91 , are presented in [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] , while mean deviations from @xmath92 ( in kcal / mol ) are presented in [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] . in all cases , we report statistical error measures generated from the individual results , cf . the supplementary material . as noted in [ sec : comparison ] , the @xmath0 and @xmath1 series start at third and second order , respectively , but we may group these together like @xmath93/@xmath4 , @xmath94/@xmath7 , etc . the results in [ e_t_n_figure ] show that the @xmath1 models in general yield smaller mean and standard errors than their @xmath0 counterparts , and the @xmath1 series furthermore exhibits a more stable convergence than the @xmath0 series . in other words , the rate of convergence is improved in the @xmath1 series over the @xmath0 series . for most of the considered molecules , the @xmath0 corrections are negative / positive for uneven / even orders , which leads to the oscillatory convergence behavior observed for the @xmath0 series in [ e_t_n_figure ] . some oscillatory behavior is also observed for the @xmath1 series , but this is much less prominent , and primarily observed beyond fourth order . the superior stability of the @xmath1 series compared to the @xmath0 series is also manifested in the smaller standard deviations for the former ( in [ e_t_n_figure ] represented in terms of standard errors of the mean ) . some of the molecules , however , differ considerably from the mean trends in [ e_t_n_figure ] . for example , methylene ( ch@xmath95 ) and ozone ( o@xmath96 ) are notoriously difficult cases due to significant multireference character@xcite , and , for both , we observe a rather slow convergence throughout either of the series . while for o@xmath96 , the convergence towards the ccsdt limit is oscillatory , for ch@xmath95 , the convergence is stabile , yet slow , cf . table s4 of the supplementary material for the @xmath0 results . similar , but significantly less pronounced problems , are observed for the @xmath1 series @xcite . finally , we note how the results obtained using the two open - shell references ( uhf and rohf ) are similar , and also that the general behavior for the mean deviations are similar to the rhf results . 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] 0.47 , [ e_t_n_recoveries_rhf_figure ] , [ e_t_n_recoveries_uhf_figure ] , and [ e_t_n_recoveries_rohf_figure ] ) and mean deviations from ( in kcal / mol , [ e_t_n_abs_diff_rhf_figure ] , [ e_t_n_abs_diff_uhf_figure ] , and [ e_t_n_abs_diff_rohf_figure ] ) the triples energy @xmath92 for the @xmath1 and @xmath0 series using rhf , uhf , and rohf references . the error bars show the standard error of the mean . , title="fig : " ] from an application point of view , the @xmath1 series is practically converged onto the ccsdt limit at the @xmath97 model ( robust for closed- and open - shell systems ) , while two additional corrections ( two additional orders in the perturbation ) are needed in the @xmath0 series in order to match these results ( the @xmath98 model ) . however , even for the @xmath98 model , the standard deviations ( for both recoveries and deviations ) are larger than for the @xmath97 model . if results more accurate than those provided by the @xmath97/@xmath98 models are desired , it is in general necessary to also account for the effects of quadruple excitations , as the quadruples energy contribution may easily exceed the difference between the ccsdt and @xmath97 energies . in such cases , the recently proposed ccsdt(q@xmath8 ) models @xcite may offer attractive alternatives to the iterative ccsdtq model . these models are theoretically on par with the @xmath1 models , but expand the ccsdtq ccsdt energy difference , rather than the ccsdt ccsd difference , in orders of the mp fluctuation potential . in conclusion , a number of similarities exist between the @xmath0 and @xmath1 series , but both the magnitude of the ( individual ) errors as well as the oscillatory convergence pattern are significantly reduced in the @xmath1 series , as compared to the @xmath0 series . this improvement is in line with the theoretical analysis in [ sec : comparison ] . more information about the expansion point is used in the @xmath1 series , where the ccsd amplitudes and ccsd multipliers are both built into the perturbative corrections , to yield a faster and more balanced rate of convergence than that observed for the @xmath0 series , in which only the ccsd amplitudes are used to construct the energy corrections . based on the results for the five lowest - order models in [ e_t_n_figure ] , the @xmath0 and @xmath1 series both appear to converge for all of the considered molecules , although slowly for some of the notoriously difficult cases . however , a formal convergence analysis is required to firmly establish whether the series indeed converge ( i.e. , establish the radius of convergence for the series ) . this will be the subject of a forthcoming paper . we have developed the @xmath0 perturbation series and compared it to the recently proposed @xmath1 series in order to gain new insights into the importance of treating amplitudes and multipliers ( parameters of the @xmath3-state ) on an equal footing whenever perturbation expansions are developed within cc theory . both series represent a perturbation expansion of the difference between the ccsd and ccsdt energies , and they share the same common set of correction amplitudes . the @xmath0 series formally describes an expansion around the ccsd energy point ( ccsd amplitude equations are satisfied ) , while the @xmath1 series may be viewed as an expansion around the ccsd lagrangian point ( ccsd amplitude equations _ and _ ccsd multiplier equations are satisfied ) . the two series are therefore different , and the @xmath1 series is found to converge more rapidly towards the ccsdt target energy , since all available information at the ccsd expansion point is utilized . the presented analysis may be generalized to any perturbation expansion representing the difference between a parent cc model and a higher - level target cc model . for developments of cc perturbation expansions , we thus generally advocate the use a bivariational lagrangian cc formulation to ensure an optimal rate of convergence in terms of term - wise size extensive corrections towards the target energy . for example , two perturbation series formulated around the ccsdt energy ( e - ccsdt(q@xmath8 ) ) and ccsdt lagrangian ( ccsdt(q@xmath8 ) ) expansion points , respectively , to describe an expansion towards the ccsdtq target energy in order the mller - plesset fluctuation potential , are also bound to exhibit different rates of convergence , following a similar line of arguments . + in quantum chemistry , a lagrangian energy functional has traditionally been viewed merely as a convenient mathematical tool for deriving perturbative expansions , however , one that would give rise to expansions that are identical to those based on the standard energy . the present work highlights how this equivalence between energy- and lagrangian - based perturbation theory only holds whenever the zeroth - order parameters do not depend on the perturbing operator , as is , for example , the case for standard mp perturbation theory where the zeroth - order parameters vanish . when the zeroth - order parameters are independent of the perturbing operator , a lagrangian formulation is merely of mathematical convenience , but , for perturbation - dependent zeroth - order parameters ( e.g. , like those of the right- ( cc ) and left - hand ( @xmath3 ) eigenstates of a non - hermitian cc similarity - transformed hamiltonian ) , a bivariational lagrangian formulation is in general expected to lead to a faster and more stable convergence than a corresponding energy formulation . this is an important point to keep in mind for future developments and applications involving perturbation expansions . k. k. , j. j. e. , and p. j. acknowledge support from the european research council under the european union s seventh framework programme ( fp/2007 - 2013)/erc grant agreement no . . j. o. acknowledges support from the danish council for independent research , dff-4181 - 00537 , and d. a. m. acknowledges support from the us national science foundation ( nsf ) under grant number aci-1148125/1340293 . in the present appendix , we compare the lowest- and next - to - lowest - order corrections of the @xmath0 and @xmath1 series . for this , we need a closed - form expression for the ccsdt multipliers ( i.e. , the ccsdt @xmath3-state parameters ) , which , from [ generalmulteq ] , reads @xmath99 \ket { { \mathrm{hf } } } + \sum_{j=1}^3 \sum_{\nu_j } \bar{t}_{\nu_j}\bra{\nu_j } [ \phi^{\hat{t } } , \hat{\tau}_{\mu_i } ] \ket { { \mathrm{hf } } } \big)\end{aligned}\ ] ] where we will again partition the ccsdt cluster operator , @xmath100 , as @xmath101 . [ ccsdt_mult_explicit ] may now be expanded in orders of the fluctuation potential ( cf . [ multpara ] ) @xmath102 if @xmath103 , the multiplier corrections in [ ccsdt_mult_order_exp_1 ] will be those belonging to the @xmath0 series , and the two lowest - order corrections are given by [ ccsdmult_order_exp_2 ] @xmath104 \ket { { \mathrm{hf } } } \label{ccsdmult_order_exp_2_1 } \\ { \delta \bar{t}^{e(2)}}_{\mu_i } & = -\epsilon^{-1}_{\mu_i}\sum_{j=1}^2 \sum_{\nu_j } { \delta \bar{t}^{e(1)}}_{\nu_j}\bra{\nu_j } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \hat{\tau}_{\mu_i } ] \ket { { \mathrm{hf } } } \ . \label{ccsdmult_order_exp_2_2}\end{aligned}\ ] ] [ e3alt_e4alt_app_n_1 ] @xmath107 \ket { { \mathrm{hf } } } \label{e3alt_app_n_1 } \\ e^{(4 ) } & = \sum^2_{j=1 } \bra { { \mathrm{hf } } } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_j^{(3 ) } ] \ket { { \mathrm{hf } } } \ . \label{e4alt_app_n_1 } \ ] ] [ e3alt_e4alt_app_2n_1 ] @xmath110 \ket { { \mathrm{hf } } } \label{e3alt_app_2n_1 } \\ e^{(4 ) } & = \sum^3_{k=1 } \big\{\sum_{i=1}^2 \sum_{\mu_i } { \delta \bar{t}^{e(1)}}_{\mu_i}\bra{\mu_i } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_k^{(2 ) } ] \ket { { \mathrm{hf } } } \big\ } \ . \label{e4alt_app_2n_1 } \ ] ] the expressions in [ e3alt_e4alt_app_n_1 ] and [ e3alt_e4alt_app_2n_1 ] are of course equivalent as may be verified by explicit comparison . finally , the @xmath109 energy in [ e4alt_app_2n_1 ] may be further recast by expanding the second - order correction amplitudes , @xmath111 , given in [ amp_pert_expand ] @xmath112 \ket { { \mathrm{hf } } } \bra{\nu_k } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}^{(1)}_{3 } ] \ket { { \mathrm{hf } } } \big\ } \nonumber \\ & = \sum_{k=1}^3 \sum_{\nu_k } { \delta \bar{t}^{e(2)}}_{\nu_k}\bra{\nu_k } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_3^{(1 ) } ] \ket { { \mathrm{hf } } } \ . \label{e4alt_app_final}\end{aligned}\ ] ] by taking the sum of the third- and fourth - order @xmath0 energies in [ e3alt_app_n_1 ] and [ e4alt_app_final ] , respectively , we can write @xmath113 \ket { { \mathrm{hf } } } \nonumber \\ & \phantom{= } \ + \sum_{\mu_3}{\delta \bar{t}^{e(2)}}_{\mu_3 } \bra{\mu_3 } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_3^{(1 ) } ] \ket { { \mathrm{hf } } } \ .\label{e3alt_plus_e4alt_app}\end{aligned}\ ] ] to evaluate the two lowest - order energy corrections of the @xmath1 series ( @xmath114 ) , we only need to consider the first - order multipliers , which read@xcite by applying the @xmath81/@xmath80 rules to [ lt5 ] , the two leading corrections of the @xmath1 series are given as @xmath116 \ket { { \mathrm{hf } } } \label{l2alt_app}\end{aligned}\ ] ] and @xmath117 \ket { { \mathrm{hf } } } \label{l3alt_app}\end{aligned}\ ] ] the sum of which becomes @xmath118 \ket { { \mathrm{hf } } } \nonumber \\ & \phantom{= } \ + \sum_{\mu_3 } { \delta \bar{t}^{l(1)}}_{\mu_3}\bra{\mu_3 } [ { \hat{\phi}^ { { { ^{*}}\hat{t } } } } , \delta\hat{t}_3^{(1 ) } ] \ket { { \mathrm{hf } } } \ . \label{l2alt_plus_l3alt_app}\end{aligned}\ ] ] we may now compare the energy sums in [ e3alt_plus_e4alt_app ] and [ l2alt_plus_l3alt_app ] . since the two leading - order multiplier corrections of the @xmath0 series are independent of the triple excitations in the ccsdt ansatz , these will equal the two lowest - order contributions to the ccsd @xmath3-state parameters , i.e. @xmath119 where @xmath120 denotes terms of third and higher orders in the fluctuation potential . for this reason , the first term on the right - hand side of [ e3alt_plus_e4alt_app ] may be viewed as mimicking the first term on the right - hand side of [ l2alt_plus_l3alt_app ] , and the same applies for the second term on the right - hand side of the two equations , by noticing that the @xmath121 multipliers of [ ccsdmult_order_exp_2_2 ] are similar to the @xmath122 multipliers of [ ccsdmult_order_exp_3_1trip ] , with the notable exception that @xmath123 in moving from the @xmath0 to the @xmath1 series . ek , j. _ j. chem . phys . _ * 1966 * , _ 45 _ , 4256ek , j. _ adv .. chem . phys . _ * 1969 * , _ 14 _ , 15 paldus , j. ; ek , j. ; shavitt , i. _ phys . * 1972 * , _ 5 _ , 50 shavitt , i. ; bartlett , r. j. _ many body methods in chemistry and physics : many body perturbation theory and coupled cluster theory _ ; cambridge university press : cambridge , uk , 2009 purvis , g. d. ; bartlett , r. j. _ j. chem . * 1982 * , _ 76 _ , 1910 helgaker , t. ; jrgensen , p. ; olsen , j. _ molecular electronic structure theory _ , 1st ed . ; wiley & sons , ltd . : west sussex , england , 2000 noga , j. ; bartlett , r. j. _ j. chem . phys . _ * 1987 * , _ 86 _ , 7041 scuseria , g. e. ; schaefer , h. f. _ chem . * 1988 * , _ 152 _ , 382 oliphant , n. ; adamowicz , l. _ j. chem . phys . _ * 1991 * , _ 95 _ , 6645 kucharski , s. a. ; bartlett , r. j. _ j. chem . phys . _ * 1992 * , _ 97 _ , 4282 eriksen , j. j. ; kristensen , k. ; kjrgaard , t. ; jrgensen , p. ; gauss , j. _ j. chem . phys . _ * 2014 * , _ 140 _ , 064108 eriksen , j. j. ; jrgensen , p. ; gauss , j. _ j. chem . _ * 2015 * , _ 142 _ , 014102 eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ j. chem . phys . _ * 2015 * , _ 143 _ , 041101 mller , c. ; plesset , m. s. _ phys . rev . _ * 1934 * , _ 46 _ , 618622 handy , n. c. ; schaefer , h. f. _ j. chem . phys . _ * 1984 * , _ 81 _ , 5031 scheiner , a. c. ; scuseria , g. e. ; rice , j. e. ; lee , t. j. ; schaefer , h. f. _ j. chem . phys . _ * 1987 * , _ 87 _ , 5361 arponen , j. _ ann . phys . _ * 1983 * , _ 151 _ , 311 arponen , j. ; bishop , r. ; pajanne , e. _ phys . rev . * 1987 * , _ 36 _ , 2519 helgaker , t. ; jrgensen , p. _ adv . quantum chem . _ * 1988 * , _ 19 _ , 183 kvaal , s. _ j. chem . phys . _ * 2012 * , _ 136 _ , 194109 kvaal , s. _ mol . phys . _ * 2013 * , _ 111 _ , 1100 gwaltney , s. r. ; head - gordon , m. _ chem . phys . _ * 2000 * , _ 323 _ , 21 gwaltney , s. r. ; head - gordon , m. _ j. chem . phys . _ * 2001 * , _ 115 _ , 2014 hirata , s. ; nooijen , m. ; grabowski , i. ; bartlett , r. j. _ j. chem . _ * 2001 * , _ 114 _ , 3919 hirata , s. ; fan , p .- d . ; auer , a. a. ; nooijen , m. ; piecuch , p. _ j. chem . phys . _ * 2004 * , _ 121 _ , 12197 shiozaki , t. ; hirao , k. ; hirata , s. _ j. chem . phys . _ * 2007 * , _ 126 _ , 244106 eriksen , j. j. ; jrgensen , p. ; olsen , j. ; gauss , j. _ j. chem . phys . _ * 2014 * , _ 140 _ , 174114 , ; t. helgaker , _ j. chem . phys . _ * 1988 * , _ 89 _ , 1560 helgaker , t. ; jrgensen , p. _ theor . chim . acta _ * 1989 * , _ 75 _ , 111 kristensen , k. ; jrgensen , p. ; thorvaldsen , a. j. ; helgaker , t. _ j. chem . phys . _ * 2008 * , _ 129 _ , 214103 eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ arxiv:1512.02846 _ * 2015 * eriksen , j. j. ; matthews , d. a. ; jrgensen , p. ; gauss , j. _ arxiv:1601.06379 _ * 2016 * tajti , a. ; szalay , p. g. ; csszr , a. g. ; kllay , m. ; gauss , j. ; valeev , e. f. ; flowers , b. a. ; vzquez , j. ; stanton , j. f. _ j. chem . phys . _ * 2004 * , _ 121 _ , 11599 bomble , y. j. ; vzquez , j. ; kllay , m. ; michauk , c. ; szalay , p. g. ; csszr , a. g. ; gauss , j. ; stanton , j. f. _ j. chem . _ * 2006 * , _ 125 _ , 064108 harding , m. e. ; vzquez , j. ; ruscic , b. ; wilson , a. k. ; gauss , j. ; stanton , j. f. _ j. chem . phys . _ * 2008 * , _ 128 _ , 114111 dunning jr . , t. h. _ j. chem . phys . _ * 1989 * , _ 90 _ , 1007 solomonik , e. ; matthews , d. ; hammond , j. r. ; stanton , j. f. ; demmel , j. j. _ j. j. parallel distrib . _ * 2014 * , _ 74 _ , 3176 see supplementary material at [ aip url ] for individual recoveries and deviations . individual @xmath1 results are given in ref . .

in this paper , we consider the dynamics of a system of hot super earths or neptunes such as hd 40307 . we show that , as tidal interaction with the central star leads to small eccentricities , the planets in this system could be undergoing resonant coupling even though the period ratios depart significantly from very precise commensurability . in a three planet system , this is indicated by the fact that resonant angles librate or are associated with long term changes to the orbital elements . in hd 40307 , we expect that three resonant angles could be involved in this way . we propose that the planets in this system were in a strict laplace resonance while they migrated through the disc . after entering the disc inner cavity , tidal interaction would cause the period ratios to increase from two but with the inner pair deviating less than the outer pair , counter to what occurs in hd 40307 . however , the relationship between these pairs that occurs in hd 40307 might be produced if the resonance is impulsively modified by an event like a close encounter shortly after the planetary system decouples from the disc . we find this to be in principle possible for a small relative perturbation on the order of a few but then we find that evolution to the present system in a reasonable time is possible only if the masses are significantly larger than the minimum masses and tidal dissipation is very effective . on the other hand we found that a system like hd 40307 with minimum masses and more realistic tidal dissipation could be produced if the eccentricity of the outermost planet was impulsively increased to we remark that the form of resonantly coupled tidal evolution we consider here is quite general and could be of greater significance for systems with inner planets on significantly shorter orbital periods characteristic of for example corot 7 b. planetary systems : formation planetary systems : protoplanetary discs

1002.2125

neptune mass extrasolar planets around main sequence stars were first detected five years ago . since then , 27 planets with a projected mass lower than 25 earth masses have been discovered , the lightest one having a projected mass of 1.94 m@xmath2 . among these objects , 17 have a semi major axis smaller than 0.1 astronomical unit ( au ) . they are called hot neptunes or hot super earths . so far , two multiple planet systems with at least two such objects have been observed . one of them is a four planet system around the m dwarf gj 581 ( bonfils et al . 2005 , udry et al . 2007 , mayor et al . the projected masses of the planets are 1.9 , 15.6 , 5.4 and 7.1 m@xmath2 and the periods are 3.15 , 5.37 , 12.93 and 66.8 days , respectively . note that although the eccentricities of the two innermost planets are compatible with zero , those of the next outermost and outermost planets are 0.16 and 0.18 respectively . the second multiple system which we shall focus upon in this paper is that around the k dwarf hd 40307 ( mayor et al . it comprises three planets with projected masses of 4.2 , 6.9 and 9.1 m@xmath2 and periods of 4.31 , 9.62 and 20.46 days , respectively . the eccentricities are all compatible with zero . if we label the planets in a system with successive integers starting from the innermost labelled , 1 , and moving outwards , then , in the system gl 581 , the ratio of the periods of planets 2 and 1 is 2.41 and that of the periods of planets 3 and 2 is 1.70 . we note that these numbers depart from 5/2 and 5/3 only by 4% and 2% , respectively.a in the hd 40307 system , the ratio of the periods of planets 2 and 1 is 2.23 and that of planets 3 and 2 is 2.13 , which depart from 2 by 11.5% and 6.5% , respectively . because departure from exact mean motion resonances in these systems is significant , the importance of resonances for their dynamical evolution has been ruled out ( mayor et al . 2009a , 2009b , barnes et al . we shall address this aspect for the hd 40307 system in this paper . migration due to tidal interaction with the disc is probably the mechanism by which planets end up on short period orbits , as _ in situ _ formation requires very massive discs ( e.g. , raymond et al . planet scatterings may also lead to short period orbits , but only when tidal circularization is efficient . according to kennedy & kenyon ( 2008 ) , scatterings are not a likely way of producing hot super earths , because of the long circularization timescales involved . in a previous paper ( terquem & papaloizou 2007 , see also brunini & cionco 2005 ) , we proposed a scenario for forming hot super earths on near commensurable orbits , in which a population of cores that assembled at some distance from the central star migrated inwards due to tidal interaction with the disc while merging on their way in . we found that evolution of an ensemble of cores in a disc almost always led to a system of a few planets , with masses that depended on the total initial mass of the system , on short period orbits with mean motions that frequently exhibited near commensurabilities and , for long enough tidal circularization times , apsidal lines that were locked together . starting with a population of 10 to 25 planets of 0.1 or 1 m@xmath3 , we ended up with typically between two and five planets with masses of a few tenths of an earth mass or a few earth masses inside the disc inner edge . interaction with the central star led to the initiation of tidal circularization of the orbits which , together with possible close scatterings and final mergers , tended to disrupt mean motion resonances that were established during the migration phase . the system , however , often remained in a configuration in which the orbital periods were close to commensurability . apsidal locking of the orbits , if established during migration , was often maintained through the action of these processes . this scenario has been questioned ( mayor et al . 2009a , 2009b ) because the multiple planet systems of hot super earths detected so far do not exhibit close enough mean motion commensurabilities . in this paper , we argue that _ the system around hd 40307 does actually exhibit the effects of resonances _ through secular effects produced by the action of the resonant angles coupled with the action of tides from the central star , and that such a configuration could be produced if the system formed as described in terquem & papaloizou ( 2007 ) , but with the addition of relatively small perturbations to the orbits arising from close encounters or collisions after the system enters the disc inner cavity . the reason that what are regarded as resonant effects can arise even when departures from strict commensurability are apparently large is that tidal circularization produces small eccentricities which , for first order resonances , can be consistent with resonant angle libration under those conditions ( see murray & dermott 1999 ) . we consider in detail the evolution of this system and other putative similar systems with scaled masses and orbital periods after they form a strict three planet laplace resonance as a result of convergent inward orbital migration . we go on to consider the onset of tidal circularization and how it leads to a separation of the semi - major axes and consequent increasing departure from commensurability . we find that if the strict laplace resonance is broken by a fairly small relative perturbation corresponding to a few parts in a thousand , continuing evolution in a modified form of the three planet resonance could in principle in the case of hd 40307 lead to a system like the observed one . however , this would require masses significantly larger than the minimum estimates and possibly unrealistically efficient tidal dissipation with the tidal dissipation parameter @xmath4 however , weaker tidal effects could be responsible for some of the deviation from strict commensurability , with the rest being produced by a larger perturbation resulting from the processes mentioned above inducing eccentricities of the order @xmath5 note however that tidal evolution could play a more significant role in similar systems with shorter orbital periods . in section [ sec : model ] , we describe the numerical model we use to simulate the evolution of a system of three planets migrating in a disc and evolving under the tidal interaction with the star after they enter the disc inner cavity . in section [ sec : hd40307 ] , we simulate the system around hd 40307 using either the minimum or twice the minimum masses for the planets under varying assumptions about initial eccentricities and the strength of the tidal interaction . we show that three of the four possible resonant angles associated with 2:1 commensurabilities either librate or have long term time averages , indicating evolution of the system towards a resonant state driven by the tidal influence of the star . in a related appendix , we give a semi analytic model for a system in a three planet resonance undergoing circularization with departures from strict mean motion commensurabilities . we note in passing that the discussion may also be applied to a two planet system near a 2:1 resonance . we show that the departure from commensurability increases with time being @xmath6 and derive an expression for the timescale required to attain a given departure that can be used to interpret extrapolate from and generalize the numerical simulations . in section [ sec : simulations ] , we present results of numerical simulations of a three planet system migrating in a disc , establishing a 4:2:1 laplace like resonance , and evolving under tidal effects induced by the star after entering the disc inner cavity . as expected , as the orbits are being tidally circularized , the planets move away from exact commensurabilites , and the period ratios of the outer and inner pairs of planets increases . the system is still resonant in the sense that some of the resonant angles continue to librate . to get a larger period ratio for the inner pair of planets , as in hd 40307 , some disruption of this laplace resonant state is needed . in our simulations , this is achieved by applying an impulse ( that could result from an encounter or collision ) to the system . this leads to departures from exact commensurabilites of the form seen in hd 40307 while still retaining libration of three of the four resonant angles . finally , in section [ sec : discussion ] we discuss and summarize our results . the model we use has been described in terquem & papaloizou ( 2007 ) . we recall here the main features . we consider a system consisting of a primary star and n planets embedded in a gaseous disc surrounding it ( for hd 40307 , n = 3 ) . the planets undergo gravitational interaction with each other and the star and are acted on by tidal forces from the disc and the star . the evolution of the system is studied by means of the solution of @xmath7body problems , in which the tidal interactions are included as dissipative forces . the equations of motion are : @xmath8 where @xmath9 , @xmath10 and @xmath11 denote the mass of the central star , that of planet @xmath12 and the position vector of planet @xmath12 , respectively . the acceleration of the coordinate system based on the central star ( indirect term ) is : @xmath13 and that due to tidal interaction with the disc and/or the star is dealt with through the addition of extra forces as in papaloizou & larwood ( 2000 ) : @xmath14 where @xmath15 , @xmath16 and @xmath17 are the timescales over which , respectively , the angular momentum , the eccentricity and the inclination with respect to the unit normal @xmath18 to the assumed fixed gas disc midplane change . evolution of the angular momentum and inclination is due to tidal interaction with the disc , whereas evolution of the eccentricity occurs due to both tidal interaction with the disc and the star . we have : @xmath19 where @xmath20 and @xmath21 are the contribution from the disc and tides raised by the star , respectively . relativistic effects are included through @xmath22 ( see papaloizou & terquem 2001 ) . the way it is implemented , interaction with the star does not modify the angular momentum of a single orbit ( the orbital decay timescale due to tidal interaction with the star , which as indicated below is estimated to be much longer than the circularization timescale , has been ignored here ) . however , in the formulation above , eccentricity damping causes radial velocity damping , which results in energy loss at constant angular momentum . as a consequence , both the semi major axis and eccentricity are reduced together ( at least for a single planet ) . the circularization timescale due to tidal interaction with the star can be estimated , from expressions given by goldreich & soter ( 1966 ) , to be : @xmath23 where @xmath24 is the semi major axis of planet @xmath25 here we have adopted a mass density of 1 g @xmath26 for the planets ( uncertainties in this quantity could be incorporated into a redefinition of @xmath27 ) . the parameter @xmath28 where @xmath29 is the tidal dissipation function and @xmath30 is the love number . equation ( [ teccs ] ) applies to the situation where tides raised by the central star are dissipated in the planet in the limit of small eccentricity which is appropriate here . tides raised by the planet in the star have been estimated to be unimportant for the low mass planets of interest ( eg . barnes et al 2009 ) . for solar system planets in the terrestrial mass range , goldreich & soter ( 1966 ) give estimates for @xmath29 in the range 10500 and @xmath31 , which correspond to @xmath27 in the range 502500 . of course this parameter must be regarded as being very uncertain for extrasolar planets , and accordingly we have considered values of @xmath27 in the range @xmath32 in this paper . for an earth mass planet at 0.05 au and @xmath33 , we get @xmath34 years . this indicates that circularization processes will be important for close in extrasolar planets in this mass range over the entire range of @xmath27 considered . but note that although circularization operates on each orbit according to equation ( [ teccs ] ) , because the planets interact they may enter into a decay mode with a decay timescale that is not straightforwardly estimated as above ( see section [ initecc ] ) . nonetheless the above estimate is found to be indicative . we have pointed out above that , in our formulation , tidal interaction with the star does not change the angular momentum of a single orbit but modifies the semi major axis . the physical basis for this is that the planets rapidly attain pseudo - synchronization ( e.g. , ivanov & papaloizou 2007 ) , after which they can not store significant angular momentum changes through modifying their intrinsic angular momenta . we restrict consideration for the moment to a single planet . since the orbital angular momentum is conserved , @xmath35 for small eccentricities , where @xmath36 is the eccentricity of the planet we consider . therefore @xmath24 changes only by about 5 percent in @xmath37 years for @xmath38 @xmath39 m@xmath2 and @xmath40 however , larger changes may occur for larger planet masses . we investigate the dependence of this form of evolution on the system configuration in detail below . finally we remark that all the simulation results reported here were obtained assuming that @xmath27 is the same for all the planets . however , tests were carried out with this assumption relaxed and these are alluded to in section [ sec : simulations ] below . in the local linear treatment of type i migration ( e.g. , tanaka et al . 2002 ) , if the planet is not in contact with the disc , there is no interaction between them so that @xmath15 , @xmath20 and @xmath17 are taken to be infinite . when the planet is in contact with the disc , disc planet interactions occur leading to orbital migration as well as eccentricity and inclination damping ( e.g. , ward 1997 ) . in that case , away from the disc edge , for some simulations we have adopted : @xmath41 \left [ 1- \left ( \frac { e_i } { 1.1 h / r}\right)^4 \right]^{-1 } \ ; \left ( \frac{h / r}{0.05 } \right)^2 \ ; \frac{{\rm m}_\odot}{m_d } \ ; \frac{{\rm m}_\oplus}{m_i } \ ; \frac{a_i}{{\rm 1~au } } \ ; \ ; \ ; { \rm years } , \label{tm}\ ] ] @xmath42 \ ; \left ( \frac{h / r}{0.05 } \right)^4 \ ; \frac{{\rm m}_\odot}{m_d } \ ; \frac{{\rm m}_\oplus}{m_i } \ ; \frac{a_i}{{\rm 1~au } } \ ; \ ; \ ; { \rm years } , \label{te}\ ] ] and @xmath43 ( eq . [ 31 ] and [ 32 ] of papaloizou & larwood 2000 with @xmath44 ) . here @xmath36 is the eccentricity of planet @xmath45 , @xmath46 is the disc aspect ratio and @xmath47 is the disc mass contained within 5 au . we have assumed here that the disc surface mass density varies like @xmath48 . for a 1 earth mass planet on a quasi circular orbit at 1 au , we get @xmath49 yr and @xmath50 yr for @xmath51 m@xmath52 and @xmath53 . the timescales given by equations ( [ tm ] ) and ( [ te ] ) can be used not only for small values of @xmath36 , but also for eccentricities larger than @xmath46 . we have also carried out simulations with @xmath15 and @xmath20 taken to be constants and thus independent of eccentricity . the commensurabilities that are formed in the system depend on the ratio of the migration rate to the orbital frequency , with commensurabilities of low order and low degree forming when this ratio is small . therefore , using equations ( [ tm ] ) and ( [ te ] ) or constant timescales give very similar results provided the eccentricity damping is ultimately effective and the magnitudes of the migration rates are comparable in both cases at zero eccentricity . the type i migration rates to be used are uncertain even when the surface density is known , largely because of uncertainties regarding the effectiveness of coorbital torques ( e.g. , paardekooper & melema 2006 , pardekooper & papaloizou 2008 , 2009 ) . in this context there are indications from modelling the observational data that the adopted type i migration rate should be significantly below that predicted by ( [ tm ] ) and ( [ te ] ) or tanaka et al ( 2002 ) ( see schlaufman et al 2009 ) . hence we explore a range of possible values . three super earths have been detected orbiting the star hd 40307 through the radial velocity variations they produce ( mayor et al . the planets have projected masses of 4.2 , 6.9 and 9.2 m@xmath54 and the reported semi major axes are 0.047 , 0.081 and 0.134 au , respectively , which correspond to period ratios of 2.26 and 2.13 for the inner and the outer pairs of planets , respectively . the periods reported by mayor et al . ( 2009a ) are 4.31 , 9.62 and 20.46 days , which correspond to period ratios of 2.23 and 2.13 . the derived eccentricities are compatible with zero . we will refer to the planets as 1 , 2 and 3 , planet 1 being the innermost object . because the period ratios differ significantly from 2 , resonances have been ruled out ( mayor et al . 2009a , barnes et al . 2009 ) . however , the period ratios are not far from 4:2:1 , being characteristic of a three planet resonance . this indicates that such a resonance might play a role at some evolutionary phase of the system . to investigate this hypothesis , we have simulated the system by means of @xmath7body calculations using the method described in the previous section . the equations of motion are integrated using the bulirsch stoer method ( e.g. , press et al . 1993 ) . the planets were set up on circular orbits around a 0.77 m@xmath55 star ( the estimated mass of hd 40307 ) with both the minimum masses indicated above and also masses a factor of two larger . the initial semi major axes were taken to be those indicated above , except in two simulations described below , where the innermost semi major axis was adjusted to make the period ratio of the innermost pair @xmath56 instead of 2.26 . the results of a calculation with the tidal parameter @xmath57 and minimum masses are displayed in fig . [ figjp1 ] , which shows the evolution of the eccentricities and runnimg time averages of the cosines of the resonant angles @xmath58 @xmath59 @xmath60 and @xmath61 over a timescale of 3.6 million years . here @xmath62 and @xmath63 are the mean longitude and the argument of pericentre of planet @xmath45 , respectively . the semi major axes do not vary significantly with time while the eccentricities remain small , below @xmath64 . however , tidal circularization processes are important for this run as the eccentricity of the outermost planet and the amplitude of the oscillations of the eccentricity of the inner two planets tends to decrease . interestingly , the running time averages of the cosines of the resonant angles are not zero , as would be expected for a non resonant system . they are such that the values for @xmath65 and @xmath66 approach unity , while the value for @xmath67 moves towards @xmath68 , corresponding to the angle approaching @xmath69 the angle @xmath70 exhibits non resonant behaviour . ( full line ) , @xmath71 ( dotted line ) , @xmath72(dashed line ) , @xmath73 ( dot dashed line ) . for all plots time is measured in yr.,width=604 ] we have also investigated the dependence of the evolution on @xmath74 small variations of the initial semi - major axes and the magnitudes of the planet masses . figure [ figjp2 ] shows the evolution for @xmath75 , minimum masses and the initial innermost semimajor axis adjusted to make the innermost period ratio 2.23 . in this case , the mean values of the eccentricities show evidence only of limited evolution . nonetheless , running time averages of the cosines of @xmath76 @xmath77 and @xmath78 quickly attain steady non zero values . this is consistent with the evolution of these runs being driven by tidal effects and towards a resonant state in which the resonant angles are associated with secular evolution of the orbital elements . this is able to occur without precise commensurability only because the eccentricities are very small ( see appendix a ) . for comparison in fig . [ figjp2a ] we plot the running means of @xmath79 @xmath80 and @xmath81 for the same run . these also attain non zero values also indicating long term variations associated with the corresponding resonant angles . note that the cosines of the resonant angles are not special in this respect ( see also appendix a ) . but for @xmath75 , minimum masses and initial innermost period ratio 2.23.,width=604 ] the dependence on the planet mass scale is illustrated in fig . [ figjp3 ] which shows the evolution for @xmath57 , twice the minimum masses and without adjustment of the innermost semi - major axis . figure [ figjp4 ] gives the evolution for the same system but with the initial innermost semimajor axis adjusted to make the innermost period ratio 2.23 . both these runs show similar evolution to the minimum mass case illustrated in fig . [ figjp1 ] but scaled to larger eccentricities . thus the results are not sensitive to the choice of minimum masses or small changes in the semi - major axes . ( solid line ) @xmath82 ( dashed line ) and @xmath81 ( dot - dashed line ) for q=1000 and minimum masses . time is measured in yr.,width=604 ] but with q=10 and with masses a factor of two larger than the minimum.,width=604 ] but with initial inner period ratio 2.23.,width=604 ] the first second and third rows of panels correspond to the cases where the inner , middle and outer planets were started with an initial eccentricity ( see text ) respectively and minimum masses were adopted . the fourth row is the same as the third except that the planet masses were twice the minimum values . from left to right for each row , the first panel shows the eccentricity of the innermost planet , the second the eccentricity of the middle planet and the third panel shows the eccentricity of the outermost planet . the fourth panel shows running time averages of @xmath83 ( solid line ) , @xmath84 ( dotted line ) @xmath85 ( dashed line ) and @xmath86 ( dot - dashed line).,width=604 ] and the innermost planet was started with a non zero eccentricity ( see text ) . in the upper right panel the masses were twice the minimum values , @xmath57 and the outermost planet was started with a non zero eccentricity . the lower left panel was as for the upper left panel but @xmath87 the lower right panel was as for the upper right panel but @xmath87 time is measured in yr . note that the shifts occurring in the period ratios are independent of @xmath88,width=604 ] and @xmath75 respectively , minimum masses were adopted and the innermost planet was started with a non zero eccentricity ( see text ) . for the fourth @xmath89 and the masses were twice the minimum values . in this case the outermost planet was started with a non zero eccentricity . from left to right for each row , the first panel shows the eccentricity of the innermost planet , the second the eccentricity of the middle planet and the third panel shows the eccentricity of the outermost planet . the fourth panel shows running time averages of @xmath83 and @xmath85 ( ultimately uppermost).,width=604 ] the calculations described above started with the planets on circular orbits . the motivation for this is that it is expected that tidal circularization will lead to a state closely approaching this provided that initial eccentricities are not so large as to disrupt the system . in order to explore this aspect in more detail , we have carried out a number of simulations which started with non circular orbits . we have explored initial eccentricities @xmath90 and found that the system enters into an eccentricity decay mode that causes the system to evolve towards the state we found starting from circular orbits in which three of the resonant angles move towards libration with non zero values for the running time averages of their cosines . in order to illustrate this we describe here results obtained when just one of the planets was started with a non zero eccentricity for various values of @xmath91 to initiate the system it was started as for the simulation illustrated in figure [ figjp1 ] but with one of the planets set at pericentre in an orbit with the same angular momentum but with an eccentricity @xmath92 this is equivalent to transfer to a less tightly bound orbit with the original specific angular momentum , but with the specific energy multiplied by @xmath93 in all cases the eccentricities of the planets become coupled and start to decay . after a while the system attains a slow decay mode for which , in a time average sense , the eccentricities are in an approximately constant ratio such that @xmath94 and @xmath95 fig . [ figjp6 ] illustrates the evolution of the eccentricities and running time averages of the cosines of the resonant angles for minimum mass simulations with @xmath96 cases for which the innermost , middle and outermost planet respectively had an initial eccentricity are shown . for comparison a case with twice the minimum masses and for which the outermost planet was started with a non zero eccentricity is also shown . the evolution of all of these systems was as described above . however , the introduction of the initial eccentricity while conserving the angular momentum of the system causes the semi - major axes and hence the period ratios in the system to change . this evolution involves energy dissipation at fixed angular momentum and thus the system must spread radially as for an accretion disc ( eg . lynden - bell & pringle 1974 ) . this causes the period ratios to increase ( see fig . [ figjp7 ] ) . when this happens the system attains period ratios that differ from the original ones . however , this is easily compensated for by starting with appropriately shifted initial conditions . the important point is that we get the same type of evolution independently of such shifts . [ figjp7 ] shows that the largest shift occurs when the outermost planet starts with an initial eccentricity . the period ratio of the innermost pair shifts from @xmath56 to @xmath97 a consequence of this is that the forced eccentricity of the innermost planet is @xmath98 smaller in this case than in the case where the inner planet started with a non zero eccentricity ( see fig . [ figjp6 ] ) . we remark that the same evolution is obtained when the masses are increased by a factor of two except that the ultimate eccentricities are a factor between @xmath99 and @xmath100 larger . the above trends are in qualitative agreement with a simplified discussion of the resonant interaction between the planets that we give in appendix a even though only two of the resonant angles @xmath76 and @xmath77 are found to be librating at this point with @xmath86 having a non zero time average of relatively small magnitude . finally we have checked that changing @xmath27 gives the same form of solution but with an evolutionary time that scales with @xmath88 results for @xmath89 and @xmath75 that confirm this view when compared with results already described for @xmath57 are illustrated in fig . [ figjp8 ] . however , practical considerations mean that runs with large @xmath27 can not be followed to libration of the resonant angles . nonetheless when @xmath101 @xmath85 and @xmath83 begin to attain non zero running means at the end of simulations with minimum masses after @xmath102 this indicates that resonance could be effectively entered for @xmath103 within the lifetime of such systems . these results taken together suggest that the planets may be undergoing circularization in a three planet resonance . as the circularization process involves energy dissipation at fixed angular momentum , the system must spread increasing the period ratios of neighbouring planets . although this means an increasing departure from strict commensurability , resonant angles remain librating or circulating in such a way as to enable the required redistribution of angular momentum to take place . a simple model illustrating how this works is given in appendix a. the amount of spreading and increase in the period ratios that can take place in a given time depends on @xmath88 we investigate this in the context of a migration scenario for producing the configuration below . simulations incorporating migration indicate that the planets were likely to have been in a strict laplace resonance with all four angles librating . however , if this persists the period ratios would increase maintaining a larger departure from commensurability for the outer pair than the inner pair in disagreement with observations . thus some perturbation event must have disrupted this resonance enabling the planets to continue circularizing with the correct period ratio relationships . as we have seen above , giving one of the planets a relatively small eccentricity can produce significant changes to the period ratios . to investigate this scenario , we now describe simulations of systems of three planets migrating inwards in a disc that set up a laplace resonance . we go on to study the evolution following from its disruption as the result of a perturbation applied once the planets have decoupled from the disc . in all the runs we have performed , the planets converge on each other forming a three planet 4:2:1 resonance , and migrate in locked in that configuration . whether the resonance can be maintained all the way down to the disc inner cavity depends on how fast the migration is . there is a tendency for the 4:2:1 resonance to be stable for slow migration rates but to be disrupted , with the appearance of higher order resonances once the migration rate becomes sufficiently fast . however , the evolution is difficult to predict precisely , as the transition from the 4:2:1 resonance to higher order resonances happens through an instability which appears to be extended in time and chaotic . small changes to the model parameters in this regime may affect whether this instability occurs or not within a given time interval . similar phenomena have been seen in two planet resonant migration where a sequence of resonances can be formed , maintained and subsequently disrupted in turn ( see papaloizou & szuszkiewicz 2009 ) . figure [ fig3 ] shows the evolution of the semi major axes of the planets and of the period ratios for 3 planets with minimum masses initially located at 0.2 , 0.34 and 0.6 au , respectively . the disc aspect ratio is @xmath104 equations ( [ tm ] ) and ( [ te ] ) for the migration and circularization times are used with @xmath105 m@xmath106 this was equal to the disc mass within 5 au in the original formalism of papaloizou & larwood ( 2000 ) . however , as there is much uncertainty in the migration rates to be used for low mass planets we investigate different possibilities using this parameter to scale the migration and circularization rates ( see section [ typeimigration ] ) . used in this way it no longer corresponds to a disc mass as in the original formulation . the initial migration times for this case are then @xmath107 and @xmath108 years for the innermost and outermost planets respectively . the initial period ratios for the inner and outer pairs are @xmath109 and @xmath110 , respectively . this calculation led to the setting up of a laplace resonance that is maintained during the subsequent evolution of the system , although for a finite period of a time around @xmath111 years a temporary weak and inconsequential instability was present ( see below ) . ( in au , _ plots a _ and _ b _ ) , eccentricity @xmath112 ( _ plot c _ ) , period ratios ( _ plot d _ ) , resonant angles @xmath113 ( in rad . , _ plot e _ ) and @xmath114 ( in rad . , _ plot f _ ) versus time ( in million years ) for 3 planets initially located at 0.2 , 0.34 and 0.6 au . on plots a and b , each solid line represents a planet , whereas the dotted line indicates the location of the disc inner edge . note the different time interval for plot b. on plot c , the solid and dotted lines represent the period ratio of the inner and outer pair of planets , respectively . here @xmath115 m@xmath55 and @xmath116 ( note that the evolution during the disc migration phase is not sensitive to the value of this parameter ) . even though strict commensurability is lost as a result of tidal interaction with the star , the planets stay locked in resonances , as indicated by the behaviour of the resonant angles . , width=604 ] in these calculations , the tidal @xmath27 parameter was set to @xmath117 . this is at least 3 orders or magnitude smaller than the values supposed to apply in earth like planets , but this value was chosen to give evolutionary timescales that could be reasonably handled numerically ( i.e. runs that could be done within weeks on a single processor ) . tests carried out below with more realistic values of @xmath27 indicate that the evolution during the disc migration phase is not affected by this specification . furthermore , when the planets are inside the cavity , the scaling given by equation ( [ hscal ] ) holds , implying that increasing @xmath27 corresponds to scaling to a longer evolutionary timescale . with @xmath116 , the eccentricity damping timescale given by equation ( [ teccs ] ) is @xmath118 years for a 1 m@xmath54 planet at 0.05 au , inducing a timescale on which the semi major axis evolves of about @xmath119 years for an eccentricity of @xmath64 ( see section [ sec : startide ] ) . figure [ fig3 ] shows the evolution of the semi major axes , the eccentricities , the period ratios and the resonant angles @xmath120 and @xmath121 . as pointed out above , the laplace resonance for which all four resonant angles librate is established while the planets migrate in the disc , after about @xmath122 years . the resonant angles @xmath123 and @xmath77 stay close to zero . then , while some of the planets are still in the disc , a weak instability appears for a finite period of time ( after about @xmath111 years ) and the resonance is weakened with the angles driven to large amplitude libration with an occasional circulation . but the instability eventually dies out and the resonance is recovered before the planets decouple from the disc . note that this instability is not always seen in the runs we have performed . as indicated above it seems to be associated with an approach to a transition to a resonance of higher degree . this is indicated by the fact that when the migration rate was speeded up by a factor of @xmath124 the outer pair of planets underwent a transition to a @xmath125 resonance while the inner pair maintained a @xmath126 commensurability at the point at which the semi - major axes had attained the same configuration . during the migration phase , the eccentricities are pumped up to several hundredths . because of the low value of @xmath27 , tidal interaction with the star begins to affect the evolution of the system slightly before the outermost planet penetrates inside the cavity . once the outermost planet reaches the disc inner edge , the disc is removed altogether from the calculation , so that if the planet is pushed out again it will not interact with the disc . this procedure is introduced to take into account the disappearance of the disc on a timescale of a few million years . in this particular run , the outermost planet enters the cavity after @xmath127 years . at that point , changes in semi major axes are due only to tidal interaction with the star . as the eccentricities of the two inner planets are damped from a few @xmath64 to @xmath64 , their semi major axes decrease by about 1 percent ( see plot b of fig . [ fig3 ] ) in such a way that their period ratio increases by somewhat more than a percent , going from 2 to 2.037 ( see plot d of fig . [ fig3 ] ) . in the meantime , the eccentricity of the outermost planet is hardly affected , as it was already very small ( a few @xmath128 ) when the planet entered the cavity . because of the conservation of the total angular momentum though , its semi major axis increases by about 1 percent , which results in the period ratio of the outer pair of planets increasing by a few percent , from 2 to 2.077 . therefore , strict commensurability is lost . however , we note that the resonant angles still librate , so that the laplace resonance is not disrupted . the fact that the deviation of the outer period ratio from 2 is twice the deviation of the inner period ratio from 2 is a consequence of the laplace relation ( equation [ [ lapr ] ] of appendix a ) and this feature is preserved in the subsequent evolution of this and similar cases . as we shall see , these deviations increase approximately @xmath6 , as predicted from equation ( [ hscal ] ) . as noted above , the laplace relation is not satisfied in hd 40307 . as we shall see below , it is possible to obtain the correct relationship between the period ratios provided the laplace resonance is disrupted such that only @xmath129 and @xmath78 librate and contribute to the secular evolution ( see also section [ initecc ] ) . in order to investigate the dependence on the specification of the circularization rates induced by the disc , we have performed simulations for which these depend only on the planet mass and are therefore constant . we here consider examples for which : @xmath130 and : @xmath131 results for cases both with the minimum masses for hd 40307 and twice the mimimum masses are plotted in figure [ figjpmig ] . during the disc induced migration , a three planet resonance is formed in both cases . as expected , the evolution for the higher mass case is twice as fast but otherwise similar . the resonance shows only small signs of instability while the planets migrate inwards . this is probably helped by the fact that the migration rate is here constant rather than increasing inwards as in the previous simulation . as the migration continues , the resonant angles @xmath132 ( @xmath133 ) all enter libration , but after the planets enter the cavity there is an adjustment so that only @xmath123 and @xmath77 librate . the system then evolves back towards a laplace resonance where all angles librate or contribute to long term time averages . this latter aspect is similar to what was seen in the simulations with larger @xmath27 illustrated in figs . [ figjp1][figjp4 ] . ( although evolution during the disc migration phase is not sensitive to this parameter),width=604 ] . the _ upper panels _ give the evolution of the semi major axes and the _ lower panels _ the period ratios . [ figjpmig ] this final evolution is one where the eccentricities all reduce and the planets physically separate with increasing period ratios for the inner and outer pairs of planets . this form of evolution is to be expected in a keplerian system in which energy is dissipated while the total angular momentum is conserved , as is the case for an accretion disc ( e.g. , lynden bell and pringle 1974 ) . it is this form of evolution we investigate below and how far the tidal effects of circularization can bring the system towards the hd 40307 system . as the continuation of the simulations described above maintains a three planet laplace resonance , the period ratios will not have the same form as in hd 40307 . as indicated above the system has a larger period ratio for the inner pair as compared to the outer pair whereas the laplace relation leads to the converse . however , we have found that there are resonant interactions for which only @xmath134 and @xmath78 librate . hereafter we refer to this as the standard state of libration . this can lead to the required form for the period ratios but some disruption of the form of the resonance initially set up within the cavity needs to take place . mechanisms could involve close encounters or collisions occurring shortly after the disc migration phase . we comment that we have found it impossible to disrupt the 2:1 resonance of the inner pair of planets with a slow process . for instance , we have added a fourth planet on an inclined orbit , but found that even a massive perturber ( 3 jupiter masses ) at 5 au could not take the inner pair away from the 2:1 resonance . to do this , the parameters of one or more of the orbits have to be varied suddenly preventing the resonance from responding adiabatically . an encounter with a fourth planet on a parabolic orbit is a possible means to achieve this and there are other possibilities involving one or more direct collisions . to affect the resonance , the semi major axis of one of the planets should change by more than an effective libration width associated with one of the resonant angles . we have performed simulations for which the velocity components of one of the planets were all suddenly increased or decreased by a constant factor . this has the effect of preserving circular orbits which is reasonable under conditions of effective circularization . we found that fractional changes exceeding @xmath135 were sufficient and , provided the changes were not too excessive , simulations approach the same form of evolution . below we describe a simulation where an impulsive change that decreased the velocity components of the innermost planet by @xmath136 was applied just after the outermost planet entered the cavity . the end of the simulation illustrated in fig . [ figjpmig ] ( or its larger planet mass counterpart where appropriate ) was used as a starting point for the simulations presented below , as these are reasonable outcomes for a system having undergone inward migration induced by a disc and then entered an inner cavity . in some cases , the tidal parameter @xmath27 was also adjusted . we remark that we have performed simulations with other types of applied impulse . for example we have augmented the above form by giving the innermost planet an initial eccentricity in a similar manner to that undertaken in section [ initecc ] . when this is small and typically of order @xmath64 this makes no difference on account of rapid effective circularization . we have also considered impulsive changes to the outermost planet giving it an eccentricity @xmath137 as in section [ initecc ] . in this case period ratios can be shifted to be close to the observed ones just after the outermost planet has entered the cavity ( see below ) . in fig . [ figjpmig1 ] , we illustrate the evolution of a model system of three planets with masses twice the minimum masses of hd 40307 with @xmath96 two runs are displayed . the first one continues the evolution of the corresponding case plotted in fig . [ figjpmig ] . in the second run , an impulse was first applied to the innermost planet that reduced all velocities by a factor of @xmath138 . right panels _ correspond to the evolution after an impulse was applied to the inner planet ( see text ) . for both models @xmath139 the period ratios @xmath140 and @xmath141 are plotted as a function of time in the _ upper panels_. the upper curve corresponds to the outer pair ( _ left panel _ ) and the inner pair ( _ right panel _ ) . in this and similar figures below , the triangles correspond to analytic fits of the form given by equation [ anfit ] ) ( and see text ) . the _ lower panels _ show the evolution of the eccentricities of the three planets . in the _ lower left panel _ the curves from top down correspond to the inner , middle and outer planets respectively . in the _ lower right panel _ the curves from top down correspond to the middle , lower and outer planet.,width=604 ] as shown in fig . [ figjpmig1 ] , during the subsequent evolution the system spreads radially with increasing period ratios . the way this happens is illustrated by the simple model described in appendix a. in this discussion the period ratios are replaced by the quantities @xmath142 and @xmath143 that are easily obtained from fig . [ figjpmig1 ] . we have been able to find good fits to the time evolution of the form found in appendix a as given by equation ( [ hscal ] ) @xmath144 where @xmath145 and @xmath146 are constants and @xmath147 is the time in units of @xmath148 yr measured from the start of the relevant simulation shown in fig . [ figjpmig ] . in the case with no initial impulse , @xmath149 and @xmath150 for the outer pair , and @xmath151 and @xmath1521.5 for the inner pair . for the case with an initial impulse , @xmath153 and @xmath154 for the outer pair , and @xmath155 and @xmath154 for the inner pair . for the latter case , a standard libration state occurred with the deviation of the inner period ratio from two being about a factor of two larger than the corresponding deviation for the outer period ratio , similar to the situation in hd 40307 . this is in contrast to the case with no impulse that persisted in a laplace resonance with all angles librating . in that case , the laplace relation persists which leads to the relative deviation of the period ratios from two behaving differently from hd 40307 ( see above ) . despite this , we can estimate the time it would take to obtain the outer period ratio of 2.13 appropriate to hd 40307 . this is @xmath156 yr . the scaling of the circularization time , @xmath157 leading to the scaling of the evolution time , @xmath158 to reach a given state , with mass of the form @xmath159 implies that for a minimum mass system this would be @xmath160 yr . thus , it might just be possible to achieve the outer period ratio in this case but then the inner period ratio would have to be determined subsequently by alternative means . for the case with the applied impulse , we may estimate the time to bring the inner period ratio to the appropriate value @xmath161 this is somewhat uncertain because of the small variation seen in the simulation , but it is about @xmath162 yr . note that the relative period ratios are not precisely identical to those of hd 40307 . in particular the ratio of the deviation of the inner period ratio from two and , the deviation of the outer period ratio from two is about @xmath163 too large . however , this may be adjusted by changing simulation input parameters . for example we recall that the simulations reported here were performed with the same value of @xmath27 for all of the planets . to indicate one possibility we note that tests we have performed indicate that the above ratio may be reduced by taking a smaller value of @xmath27 for the outermost planet than the others . we remark that , while it might be barely possible to obtain an inner period ratio of 2.23 for a larger mass system , it is almost certainly not possible for the minimum mass system as the time required would be @xmath164 yr . note that , as the evolution progresses , the eccentricities of all the planets tend to decrease , as does their variation . this is a general consequence of the spreading that is driven by the energy dissipation in these simulations . to investigate further the form of the evolution and its scaling with run parameters , we plot in fig . [ figjpmig2 ] the results of two simulations of models with minimum masses and for which an impulse of the above form was applied to the innermost planet . we compare cases with @xmath165 and @xmath166 as stated above , we had to consider rather small values in order to perform simulations that produced a reasonable variation in a reasonable time . these resulted in a standard state of libration and relative period ratios very similar to the higher mass simulation with an initial impulse illustrated in fig . [ figjpmig1 ] . , but for models with minimum masses and for which an impulse was applied to the innermost planet ( see text ) . for the _ left panels _ , @xmath165 and for the _ right panels _ , @xmath166 in the _ upper panels _ , the upper curves correspond to the inner period ratio . in the _ lower panels _ , the curves from top down correspond to the middle , inner and outer planets respectively.,width=604 ] in the case with @xmath165 , the fit parameters are @xmath167 and @xmath168 for the inner pair , and @xmath169 and @xmath1521.7 for the outer pair . for the case with @xmath170 , they are @xmath171 and @xmath172 for the inner pair , and @xmath173 and @xmath1521.5 for the outer pair . the values of @xmath145 , being about an order of magnitude larger in the second case , are consistent with the evolution time being an order of magnitude larger , thus scaling with @xmath27 as expected . the time to attain an inner period ratio of @xmath56 in the @xmath165 case is @xmath174 yr , leading to an estimate of @xmath175 yr when @xmath96 this is less than the estimate based on the analytic scaling with planet mass implied by equation ( [ hscal ] ) applied to the higher mass simulations illustrated in fig . [ figjpmig1 ] by a factor @xmath176 which we believe is a reasonable level of consistency in view of the extent of the extrapolation involved . the eccentricities in these simulations decrease with time in a manner analogous to the simulations shown in fig . [ figjpmig1 ] . the long times required for the above cases to reach a configuration similar to that now observed in hd 40307 arise because only a very small perturbation was applied just after the systems decoupled from the disc , requiring a large amount of subsequent period ratio evolution . such a scenario is only feasible for masses exceeding twice the minimum and small @xmath4 however , if a larger perturbation is allowed , it can provide a larger shift in the period ratios requiring less evolution later on ( see fig . [ figjp7 ] ) . this is equivalent to starting on the evolutionary curves shown in figs . [ figjpmig1]-[figjpmig2 ] at a later time . for example we found that if a perturbation of the type adopted in section [ initecc ] was applied to the outer planet a shift of the right type could be obtained . we found that when the imposed eccentricity of the outer planet was taken to be @xmath177 the period ratios were shifted to very similar values to those in hd 40307 . thus if large enough perturbations are allowed , only a small amount of subsequent evolution may be needed . in this paper , we have shown that the planets around the star hd 40307 could be undergoing a resonant interaction , despite departure of the period ratios from very precise commensurability . this is indicated by the fact that three of the four resonant angles librate or are associated with long term changes to the orbital elements . note that such a resonant state can occur without exact commensurability only because the eccentricities are very small , which can be brought about as a result of tidal circularization from a state with initially higher eccentricities ( see section [ initecc ] ) . when this occurs in a system that starts from a near 2:1 commensurability , the difference of the period ratios from two then increases @xmath178 we propose that the planets in this system were in a strict laplace resonance while they migrated through the disc , with all 4 resonant angles librating . exact commensurability was then departed from as indicated above as a result of the tidal interaction with the star , which preserved libration of at least some of the resonant angles , after the planets entered the disc inner cavity . because of the laplace relation , the period ratios evolve in such a way that the ratio for the inner pair of planets is smaller than that of the outer pair of planets . the opposite is observed in hd 40307 . to get the appropriate form for the period ratios , some disruption of the resonance set up in the cavity needs to take place ( this is in contrast to claims by zhou 2009 ) . a close encounter for instance might be a possible cause . an impulsive interaction would be sudden enough that the resonance would not respond adiabatically . if the velocity of the innermost planet is reduced by a factor 0.996 for instance , deviation of the inner period ratio from two gets twice as large as the corresponding deviation for the outer period ratio , while 3 of the 4 resonant angles still librate , similar to the situation in hd 40307 , within the lifetime of the system . however , we found that we could get a period ratio of 2.23 for the inner pair only if the masses are significantly larger than the minimum masses and tidal dissipation is possibly unrealistically effective . from the results obtained in section [ sec : simulations ] , when the laplace resonance is disrupted at an early stage , we estimate that the deviation of the inner period ratio from two is approximately given as a function of time by @xmath179 where @xmath180 @xmath181 and @xmath182 ( @xmath183 and @xmath184 are assumed fixed for the purposes of this discussion ) . accordingly , to attain @xmath185 we would require masses exceeding the minimum estimate by about a factor of two for @xmath186 if the laplace resonance is preserved , from the results obtained in section [ sec : simulations ] , we may also estimate that the deviation of the outer period ratio from two is approximately given as a function of time by @xmath187 . thus it may be possible to get a period ratio of 2.13 for the outer pair of planets with near to minimum masses in a reasonable time if @xmath188 for all the planets , but then the inner period ratio would be only @xmath189 it would thus seem likely that additional stronger dynamical interactions are required to bring the system to its final state . for example we found that if the outer planet was given an eccentricity @xmath190 shortly after decoupling from the disc , period ratios similar to those observed in hd 40307 could be produced on a time scale of a few @xmath191 ( see sections [ initecc ] and [ sec : simulations ] ) . then , the work presented in this paper indicates that tidal circularization would continue through resonant interaction with the period ratios separating after these interactions are over . note that a similar type of scenario may apply to the system of planets around gj 581 , which is currently closer to exact commensurablilty than hd 40307 . indeed , different mean motion resonances can be established while the planets migrate within the disc , depending on the initial relative location of the planets and on the migration timescale . however , as it appears that this system may have been in higher order resonances than hd 40307 , it is possible that these would have been broken completely as a result of circularization . we intend to present a study of gj 581 in a forthcoming paper . in a study of the tidal evolution of the system around hd 40307 , barnes et al . ( 2009 ) have concluded that the planets can not be earth like . according to them , if it were the case , extrapolating back in time the tidal evolution of the eccentricities from current values ( estimated from direct simulations of the observed system ) , the system would have been unstable . the work in this paper does not support such a conclusion . we note that barnes et al . ( 2009 ) neglected the dynamical interactions between the planets , which clearly will have played a role if the current eccentricities are small because resonant interactions coupled with tidal effects have been associated with long term changes to the orbital elements . the results reported in this paper indicate that resonances in multiple low mass planet systems may play more of a role than currently claimed in the literature . significant departure from exact mean motion resonance does not necessarily preclude a resonant state when , for example , tidal effects are important causing eccentricities to be small , as libration of some of the resonant angles in that case may still exist , and be associated with long term evolution of the orbital elements . although the confirmation of resonance effects is problematic for small eccentricities , we expect that further detections and analysis of systems similar to hd 40307 will test the relevance of the scenarios explored in this paper . finally , we comment that , because of the rapid increase of the effectiveness of tidal interactions with decreasing orbital period , the dynamical interactions discussed here would be of much greater importance for systems in which the period of the innermost planet is significantly shorter than in the hd 40307 system . for example , we note that the planet corot7 b , which has a projected mass of 11.12 m@xmath2 , has a period of only 0.85 day ( lger et al . if this short period orbit is a result of migration , it is likely this planet was pushed inwards by a companion which would have been on a commensurable orbit ( terquem & papaloizou 2007 ) . we note in support of this idea that pre - main sequence stars typically have rotation periods of several days ( bouvier et al . 1993 ) , which would imply that the disc inner cavity should extend well beyond the orbit of corot7 b if that were magnetically maintained . in that case , this planet could not have migrated so far inwards without being shepherded by a companion . in this appendix we develop a semi analytic model that shows how a two or three planet system undergoes radial spreading as a result of evolution driven by tidal circularization . the coupling between the planets occurs because resonant angles may librate or be associated with long term changes _ even though there may be significant deviations from exact commensurability_. we begin with the system without tides which is hamiltonian . subsequently , we add the effects of tides . we adopt jacobi coordinates ( sinclair 1975 , papaloizou & szuszkiewicz 2005 ) for which the radius vector of planet @xmath192 @xmath193 is measured relative to the centre of mass of the system comprised of @xmath194 and all other planets interior to @xmath192 for @xmath195 with @xmath196 here and @xmath197 corresponding to the innermost planet . the hamiltonian , correct to second order in the planetary masses , can be written in the form : @xmath198 here @xmath199 and @xmath200 here the orbital angular momentum of planet @xmath45 which has reduced mass @xmath202 with @xmath203 being the actual mass , is @xmath204 and the orbital energy is @xmath205 for motion around a central point mass @xmath194 we have : @xmath206 where @xmath207 @xmath24 denotes the semi - major axis and @xmath36 the eccentricity of planet @xmath25 the mean longitude of planet @xmath45 is @xmath208 where @xmath209 is its mean motion , with @xmath210 denoting its time of periastron passage and @xmath211 the longitude of periastron . we remark that results valid for a two planet system , @xmath212 , may be obtained by setting @xmath213 here we are interested in the effects of two first order @xmath126 commensurabilities associated with the outer and inner pairs of planets . in this situation , we expect that any of the four angles @xmath217 @xmath59 @xmath218 and @xmath61 may be slowly varying . following standard practice ( see , e.g. , papaloizou 2003 , papaloizou & szuszkiewicz 2005 ) , only terms in the fourier expansion involving linear combinations of @xmath219 @xmath220 @xmath77 and @xmath78 as argument are retained . working in the limit of small eccentricities applicable here , only terms that are first order in the eccentricities need to be retained . expanding to first order in the eccentricities , the hamiltonian may be written in the form : @xmath221 where : @xmath222 , \label{hamil}\ ] ] with : @xmath223 here @xmath224 denotes the usual laplace coefficient ( e.g. , brouwer & clemence 1961 ) with the argument @xmath225 using equations ( [ eqnmo1])([eqnmo4 ] ) together with equation ( [ hamil0 ] ) we may first obtain the equations of motion without the effect of circularization tides and then add this in . we obtain correct to lowest order in the eccentricities : @xmath226 the terms on the right hand sides involving the circularization times @xmath21 describe the effects arising from tides raised by the central mass on planet @xmath25 the terms @xmath227 in equations ( [ eqntid0 ] ) , ( [ eqntid1 ] ) and ( [ eqntid2 ] ) account for the orbital energy dissipation occurring as a result of circularization at the lowest order in @xmath36 for which it appears . in the absence of tides , the total energy @xmath228 and angular momentum @xmath229 are conserved . when circularizing tides act , the total angular momentum is conserved but energy is lost according to : @xmath230 when both pairs of planets @xmath231 and @xmath232 are in resonance , a three planet resonance is said to exist . it is convenient to perform a time average of equations ( [ eqne1])-([eqntid5 ] ) taken over a time long compared to both the characteristic orbital period and the libration period , but short enough compared to the tidal time scale ( see eg . sinclair 1975 for a discussion of this aspect ) . the angles @xmath132 may be undergoing libration or circulation . equations ( [ eqne1])([eqntid1 ] ) may be averaged directly , while equations ( [ eqntid2])([eqntid5 ] ) may be averaged directly for small amplitude librations or alternatively after multiplying them by quantities such as either , e.g. , @xmath233 @xmath234 @xmath235 and @xmath236 or @xmath79 @xmath237 @xmath82 and @xmath81 , respectively . the time averages of these quantities as well as @xmath238 and @xmath239 @xmath240 are assumed to vary on the tidal time scale or longer . when these are non zero , long term changes to the orbital elements @xmath24 and @xmath36 may occur through planet - planet interactions even without tides . as illustrated in figures [ figjp2 ] and [ figjp2a ] , six of these quantities are found to exhibit steady long term time averages of the required type that are quickly established for a model of hd 40307 with assumed @xmath241 as this averaging procedure results in equations of essentially the same form as the case where the participating resonant angles ( defined to be those that are associated with non zero time averages ) are assumed to have averages close to but not equal to zero or @xmath242 while undergoing small amplitude librations , leading to the same scaling with physical parameters , but with an adjustment of numerical coefficients . for simplicity we restrict consideration to that case . assuming small amplitude librations of all four angles , equations ( [ eqntid2])([eqntid5 ] ) give relations of the form : @xmath243 where : @xmath244 and the @xmath132 in the above can be set to either zero or @xmath69 we note that , when all four angles participate , equations ( [ lap2 ] ) and ( [ lap3 ] ) imply the strict laplace resonance relation @xmath245 however , simulations of hd 40307 indicate that only @xmath246 and @xmath78 participate . thus equation ( [ lap2 ] ) is absent and the strict laplace relation does not hold . indeed , this relation implies that @xmath247 should be unity . being about @xmath248 , the laplace relation is not satisfied . when @xmath70 does not participate and the laplace relation does not hold , the quantity @xmath249 is not constrained from the outset but is determined as a consequence of the initial conditions and following time dependent evolution . below , we consider the case when either all four angles contribute or only @xmath250 and @xmath78 contribute . we note that a similar discussion applies when only @xmath251 and @xmath77 participate that leads to the same conclusions ( @xmath252 decays independently due to tides in that case and may be taken to be zero ) . in all cases , one can readily verify that the above system provides a complete set of equations for the time averages of the resonant angles and the orbital elements when the form of the tidal forces is specified . however , we note that we can obtain the characteristic form and time scale of the evolution by considering the conservation energy given by ( [ econ ] ) together with the constancy of the total angular momentum . to do this we begin , by noting that equations ( [ lap1 ] ) and ( [ lap4 ] ) provide equations for the mean motion ratios of the form : @xmath253 and : @xmath254 and @xmath255 and @xmath256 are defined implicitly through equations ( [ lap1 ] ) and ( [ lap4 ] ) . in addition , equation ( [ lap3 ] ) gives @xmath257 in terms of @xmath252 , @xmath258 and @xmath259 in the form : @xmath260 when @xmath261 has zero time average and so is replaced by zero , equation ( [ lap2 ] ) does not hold and accordingly the laplace relation also does not hold . we comment that @xmath258 and @xmath259 measure the deviation from strict commensurability of the corresponding mean motions and _ do not have to be small . _ for a system such as hd 40307 these are large , particularly in comparison comparison to the characteristic size of the eccentricities . we remark that according to equations ( [ lap1 ] ) - ( [ lap4 ] ) , other things remaining fixed , the eccentricities are proportional to the planet mass scale and , inversely proportional to the deviation from commensurability . the results presented in section [ initecc ] confirm these trends but with some deviation that is to be expected because all the relevant resonant angles , although contributing to the evolution , are not in a state of small amplitude libration in those simulations . replacing @xmath262 ( @xmath263 ) by @xmath194 to simplify expressions without loss of content ( as @xmath264 ) , the total energy and angular momentum may be written , respectively , as : @xmath265,\ ] ] and : @xmath266.\ ] ] here , we have neglected in @xmath267 terms proportional to the squares of the eccentricities as these can be shown to lead to small effects compared to those arising from the eccentricity dependence of @xmath255 and @xmath268 we now use the above together with equation ( [ econ ] ) to form the quantity @xmath269 and find that this is given by : we may write equation ( [ elcon ] ) in terms of the deviations from commensurabiity : @xmath271 and @xmath272 , obtaining : @xmath273 in addition , the right hand side can be expressed in terms of @xmath274 and @xmath275 using equations ( [ lap1])([lap4 ] ) which give : @xmath276 \frac{gm_3m_2 ^ 2n_3 ^ 2c_{2,3}^2\cos^2\phi_4 } { w_{2,3}^2ma_3n_3 ^ 2 t_{e,3}^s}. \label{elwcon}\ ] ] we may now obtain characteristic form of the time evolution of the departure from strict commensurability . to do this , we may simply assume that @xmath277 @xmath275 are comparable , significantly less than unity and @xmath278 say , as has been found numerically in general and must be at least approximately the case for a strict laplace resonance for which @xmath279 ( see equation [ [ lapr ] ] ) . in addition , we take the planet masses to be comparable and @xmath280 then the characteristic evolution expected from equations ( [ elcon1 ] ) and ( [ elwcon ] ) can be obtained from an equation of the form @xmath281 where we take @xmath282 to be the shortest circularization time @xmath283 thus : @xmath284 where @xmath285 specifies an arbitrary origin of time . we recall that ( [ hscal ] ) applies to a two planet system by appropriately setting @xmath286 as well as a three planet system . the time required to attain a departure @xmath287 is @xmath288 . in general , this is very much longer than the characteristic circularization time .

we analyze ultraviolet ( ) images of the old open clusters m67 , ngc 188 , and ngc 6791 obtained with ultraviolet imaging telescope ( uit ) during the second flight of the _ astro _ observatory in march 1995 . twenty stars are detected on the uit image of m67 , including 11 blue stragglers , seven white dwarf candidates , and the yellow giant white dwarf binary s1040 . the ultraviolet photometry of the blue stragglers f90 ( s975 ) and f131 ( s1082 ) suggests that these stars have hot subluminous companions . we present a semi - empirical integrated ultraviolet spectrum of m67 , and show that the blue stragglers dominate the integrated spectrum at wavelengths shorter than 2600 . the number of white dwarfs in m67 is roughly consistent with the number expected from white dwarf cooling models . eight candidate sdb / sdo stars are detected in ngc 6791 , and two are detected in ngc 188 . the luminosity range l / l , derived from the ultraviolet photometry of the six sdb candidates , is consistent with theoretical models of metal - rich hot horizontal branch ( hb ) stars . the fraction of hot hb stars in both ngc 6791 and ngc 188 is about 30% , implying that the integrated spectra of both clusters should show a uv turnup at least as strong as that observed in any elliptical galaxy .

astro-ph9804164

as a stellar population ages , the main - sequence turnoff becomes cooler and fainter , until , after an age of about 2 gyrs , the ultraviolet ( @xmath4 1500 ) flux is no longer dominated by stars near the main - sequence , but instead must be due to a more exotic population of hot stars , such as blue stragglers , hot subdwarf ( sdb , sdo ) stars , hot post - asymptotic giant branch ( post - agb ) stars , white dwarfs , or interacting binaries . the blue stragglers and hot subdwarfs are of particular interest since they might have significant roles , respectively , in two current problems in extragalactic astronomy : the age - dating of old galaxies from their rest - frame near - uv ( @xmath5 ) spectra , and understanding the origin of the observed uv - upturn in elliptical galaxies . in the absence of blue stragglers , the rest - frame near - uv spectrum of an old galaxy should be dominated by the turnoff population , and thus fitting the spectrum to population synthesis models should be a reliable method for age - dating the galaxy , although metallicity effects also need to be properly disentangled ( @xcite , @xcite ) . however , as noted by spinrad et al . ( 1997 ) , a significant underestimate of the galaxy age could result if blue stragglers contribute a large fraction of the galaxy near - uv flux , but are not included in the adopted population synthesis model . greggio & renzini ( 1990 ) outlined the various candidates for the origin of the observed uv - upturn in elliptical galaxies , which is characterized by a rising uv flux shortward of 1800 ( @xcite , @xcite ) . among these candidates , white dwarfs and hot post - agb stars have the advantage of being relatively well - understood phases of stellar evolution , but lack sufficient fuel to explain the strongest uv - upturn galaxies . the more promising candidates are currently the extreme horizontal - branch ( ehb ) stars and their hot progeny ; the agb - manqu stars , which miss the agb phase entirely , and the post - early agb stars , which leave the agb before the thermally pulsing phase . although the horizontal branch ( hb ) morphology of the globular clusters becomes generally redder with increasing metallicity , a significant ehb population is theoretically expected at metallicities above solar ( and an age larger than about 5 gyr ) , if reasonable assumptions are made concerning the metallicity dependence of the mass loss rate and helium abundance ( @xcite , @xcite ) . spectroscopically , ehb and agb - manqu stars are expected to appear , respectively , as sdb and sdo stars , and the measured effective temperatures and gravities of the field hot subdwarf stars is consistent with this scenario ( @xcite ) . however , the field hot subdwarfs apparently have a large binary fraction ( @xcite , @xcite ) , which instead suggests that binary interactions might be involved in their formation ( @xcite ) . despite their probable contributions to the ultraviolet spectra of galaxies , blue stragglers and hot subdwarfs are either neglected or poorly constrained in existing population synthesis models of old stellar populations . the currently favored explanations for the origin of blue stragglers involve binary interactions , either through mass - transfer processes , binary mergers , or direct stellar collisions ( @xcite , @xcite ) , but the large parameter space of these binary processes limits their inclusion in synthesis models to special cases ( e.g. @xcite ) . some attempts have been made to include ehb stars in synthesis models of metal - rich populations , but these are hampered by large uncertainties , for example , in the dependence of mass loss and helium abundance on metallicity ( @xcite , @xcite ) . the modeling of galaxy ultraviolet spectra should clearly benefit from empirical studies of the ultraviolet content of resolved , old , metal - rich stellar populations . here we report on ultraviolet ( @xmath4 1600 ) images of the three old ( @xmath6 gyr ) open clusters m67 , ngc 188 , and ngc 6791 , which were obtained in march 1995 using the _ ultraviolet imaging telescope _ ( uit , @xcite ) during the second flight of the _ astro _ observatory . the @xmath7 diameter field of view of uit is sufficiently large to allow a complete census of the hot stellar population in the observed clusters . thus , the uit images of these three clusters can be used for an empirical study of the ultraviolet content of old , metal - rich stellar populations . as it turns out , m67 contains blue stragglers , but no hot subdwarfs , whereas ngc 188 and ngc 6791 contain hot subdwarfs , but are too old , distant and reddened for the blue stragglers to be detectable with uit . m67 is also sufficiently nearby that all the white dwarfs hotter than @xmath4 21,000 k should have been detected on the uit image . in the population synthesis models of magris & bruzual ( 1993 ) , white dwarfs supply about 10% of the 1750 flux in elliptical galaxies with a weak uv - upturn , and in section 3.2 we re - examine the white dwarf contribution to galaxy ultraviolet spectra , using more recent stellar atmospheres and evolutionary models . a brief summary of this work was given by landsman & stecher ( 1997 ) . a subsequent paper will discuss the uit observations of the intermediate age ( @xmath4 2 gyr ) clusters ngc 752 and ngc 7789 , which are sufficiently young to allow the turnoff population to be detected in the ultraviolet . table [ tbl-1 ] gives the adopted fundamental parameters of the observed clusters , and table [ tbl-2 ] lists the deepest exposure obtained with each uit filter , along with the coordinates of the uit image center . the uit image center was chosen to provide a target for the co - pointed spectroscopic instruments on the _ astro _ observatory , and could be offset by as much as @xmath8 ( in the case of ngc 6791 ) from the cluster center . m67 was observed with the b1 filter , which has a peak wavelength of 1520 and a bandwith of 350 ( see figure [ fig-3 ] ) . ngc 6791 was observed with the b5 filter , which has a peak wavelength of 1620 and a bandwidth of 230 . ngc 188 was observed with both the b1 and the b5 filter . ( the less sensitive b5 filter was used on the daytime side of the shuttle orbit , because it suppresses 1300 dayglow emission . ) ultraviolet magnitudes on the b1 and b5 images are denoted here as m@xmath9 and m@xmath10 , respectively , and are given on the monochromatic system , where @xmath11 , and @xmath12 is the observed flux in units of erg @xmath13 s@xmath14 @xmath14 . only a few stars in each cluster are detected on the uit images , so that ultraviolet magnitudes could be derived from simple circular aperture photometry . however , the absolute calibration of the uit images is somewhat problematic . as discussed by stecher et al . ( 1997 ) , comparison of fluxes from uit and the _ international ultraviolet explorer _ ( iue ) indicated that the sensitivity of the uit during the _ astro-2 _ mission appeared to decrease with exposure time . the cause of this behaviour has not been determined , and instead an empirical correction to the exposure time was adopted that , for most images , brought the iue and uit fluxes to within 15% of each other . for m67 , this mean calibration could be tested , and an aperture correction determined , by direct comparison with spectra from iue and the _ goddard high - resolution spectrograph _ ( ghrs ) of five stars in the cluster . the ngc 6791 calibration was modified by comparison with a single star ( ngc 6791-b4 ) with a ghrs observation , while for ngc 188 we had to rely on the mean uit calibration . thus while the photometry of all the clusters has a random uncertainty of about 0.1 mag , we estimate an uncertainty in the absolute photometry of about 0.1 mag for m67 , and about 0.15 mag for ngc 6791 and ngc 188 . all iue spectra in this paper used the newsips ( @xcite ) reduction , with the fluxes multipled by 1.06 , as suggested by colina & bohlin ( 1994 ) . m67 is a well - studied , nearby ( 820 pc ) , solar - metallicity open cluster with an age of about four gyr . twenty stars are detected on the uit image of m67 ( figure [ fig-1 ] ) , including eleven blue stragglers , seven white dwarf candidates , the yellow - giant s1040 ( from the catalog of @xcite ) , and the probable non - member s1466 . the integrated 1520 flux of m67 is completely dominated by the blue stragglers , and in particular , by the bright , hot ( v = 10.03 , = 0.09 ) blue straggler f81 . figure [ fig-2 ] shows a color - magnitude diagram ( cmd ) of a @xmath15 field centered on m67 taken from montgomery et al.(1993 , hereafter mmj93 ) , with circles identifying the sources detected in the ultraviolet . in general , the stars that are detected in the ultraviolet are ones with the bluest colors , with the notable exception of the star s1040 , which has v = 11.52 and = 0.82 . s1040 is a known spectroscopic binary and landsman et al . ( 1997 ) used ghrs spectroscopy of s1040 to show that the ultraviolet flux is due to a hot white dwarf companion of the yellow giant . landsman et al . also constructed a mass - transfer history for s1040 that explains its unusual `` red straggler '' location in the cmd ( 0.2 mag blueward of the giant branch ) , relatively long orbital period ( 42d ) , and the unusually low mass ( @xmath16 m@xmath2 ) of the white dwarf companion . in their scenario , s1040 originated as a short - period ( @xmath4 2d ) binary , in which the more massive star ( the donor ) fills its roche lobe while on the lower giant branch . a period of rapid mass transfer reverses the mass - ratio of the donor and accretor ( but without creating a common envelope ) , and is followed by an extended period ( @xmath4 800 myr ) of stable mass transfer . during the mass transfer , the orbital period increases , and the accretor eventually becomes a blue straggler , more massive than the cluster turn - off stars . mass transfer ends when the donor s envelope is depleted , leaving a low mass helium white dwarf , and a blue straggler , which has since evolved redward to its current location in the cmd . eleven of the thirteen blue stragglers in m67 listed by milone & latham ( 1994 ) are detected on the uit image . the first two columns of table [ tbl-3 ] contain , respectively , the fagerholm and the sanders ( 1977 ) identifications of the detected blue stragglers , the third and fourth columns list the v and magnitudes tabulated by mmj93 , and the fifth column gives the observed uit magnitude , m@xmath9 . the brightest and hottest blue straggler is f81 ( v = 10.03 , = 0.07 ) , which is saturated even on the shortest exposure ( 109s ) uit image . optical and iue spectra of f81 were studied by schnberner & napiwotzki ( 1994 ) , who derived t@xmath17 = 12,750 k , @xmath18 = 4.26 and a spectroscopic mass of @xmath19 m@xmath2 . comparing the iue spectra of f81 with the uit photometry of the remaining stars in m67 , we find that f81 supplies more than 90% of the total 1520 flux of m67 . despite the extreme properties of f81 , the evidence is strong from both astrometric ( @xcite ) and radial velocity ( @xcite ) studies that it is a cluster member . in order to search for evidence of a subluminous hot companion , or an unusual ultraviolet spectrum , we compare the uit photometry of the blue stragglers with model atmosphere predictions , using stellar parameters derived from optical photometry . temperatures and gravities for most of the blue stragglers are calculated from the strmgren photometry tabulated by hauck & mermilliod ( 1998 ) , using the code of napiwotzki et al . ( 1993 ) , and assuming e(b y ) = 0.73e ( ) = 0.018 for all stars . no strmgren photometry is available for f90 , and so the values of t@xmath17 and @xmath18 for this star are taken from mathys ( 1991 ) , who derived them using geneva photometry . the predicted m@xmath9 magnitude is then calculated from the derived t@xmath17 and @xmath18 using the 1995 kurucz lte model atmospheres as tabulated by lejeune et al . this version of the kurucz models includes the modification of the convective overshoot algorithm discussed by castelli ( 1996 ) , which eliminates the discontinuities in color indicies for 6500 k @xmath20 t@xmath17@xmath20 8000 k that was present in earlier versions of the kurucz grid . the models are normalized to the stellar v magnitude , and reddened using the parameterization of cardelli et al . ( 1989 ) with r@xmath21 = 3.1 . the calculated values of t@xmath17 and @xmath18 are listed in columns 5 and 6 of table [ tbl-3 ] , and the predicted m@xmath9 magnitudes are given in column 7 . for the range of effective temperatures considered here ( 6500 8500 k ) , the uit 1520 bandpass is located shortward of the blackbody peak , and thus is very sensitive both to the exact value of t@xmath17 and to the accuracy of the model atmosphere . an uncertainty of 200 k in the value of t@xmath17 corresponds to about a 0.5 mag uncertainty in the predicted uit magnitude . archival iue spectra are available for the blue stragglers f280 , f156 , f153 , and f190 , and these provide a much more robust comparison with the model atmospheres . figure [ fig-3 ] compares the iue spectra with kurucz models with [ fe / h ] = 0.1 , and t@xmath17 and @xmath18 from table [ tbl-3 ] , normalized to the stellar v magnitude . no parameters were adjusted to improve the fit . the model spectra provide a good fit to both f280 and f156 , and the model fits for f153 and f190 can be improved by adopting values of t@xmath17 that are lower by 60 k and 140 k , respectively , than the values listed in table [ tbl-3 ] ; a change that is within the uncertainty of the strmgren photometry . in addition , the somewhat poorer fit for f153 may be related to its am star abundance peculiarities ( @xcite ) , while the poorer fit for f190 may be related to its 0.02 mag @xmath22 sct pulsations ( @xcite ) , since the flux variations of @xmath22 sct stars are known to be much larger in the ultraviolet ( @xcite ) . we conclude that the iue spectra provide no evidence for a subluminous companion in any of the four stars . the lack of a hot subluminous companion for f190 is especially noteworthy , because it is a single - lined spectroscopic binary with a 4.2d period , and suspected to be currently undergoing mass transfer ( @xcite ) . we also note that if we let the reddening be a free parameter in the model atmosphere fits , then lower reddening values ( 0.015 @xmath20 e ( ) @xmath20 0.025 ) are consistently favored over the higher values in the range usually considered acceptable for m67 ( 0.015 @xmath20 e ( ) @xmath20 0.052 ; @xcite ) . for the remaining blue stragglers , taking into account the @xmath23 mag uncertainty in the predicted uit flux , and the 0.14 mag uncertainty in the measured uit flux , there is strong evidence for a uv excess only for the two stars f131 ( s1082 ) and f90 ( s975 ) . in the case of f131 , the observed 1520 flux is a factor of six higher than predicted from the optically derived t@xmath17 . mathys ( 1991 ) found evidence for an additional broad component in the d and absorption lines of f131 , which he suggested was due to a hot , fast - rotating secondary . f131 is also a rosat source with an x - ray luminosity of @xmath24 erg s@xmath14 ( @xcite ) . the x - ray emission and the presence of a hot , subluminous secondary together suggest that f131 is an algol - type mass - transfer binary . milone ( 1991 ) reports that f131 is a radial velocity variable , with a small amplitude and a long period , but no orbit has yet been determined . the uit magnitude of f90 is probably uncertain by close to a factor of two , because f90 is located in the wings of the heavily saturated image of f81 . nevertheless , the detection of f90 with uit is quite robust , and likely due to a hot subluminous companion , since the predicted ultraviolet magnitude from the optically - derived t@xmath17 is about 1.9 mag fainter than the uit detection limit . latham & milone ( 1996 ) report that s975 is a binary with a period of 1221 days , and a small eccentricity @xmath25 . the probable presence of a hot subluminous companion in f90 , along with a nearly circular binary orbit , suggests that its blue straggler origin is also due to a binary mass - transfer process . note that while the hot companion of f131 is probably not a white dwarf ( since it is detected in the d and absorption lines ) , the nature of the hot companion of f90 is unknown . the star s1466 occupies the blue straggler region of the hr diagram , and is located @xmath26 from the center of m67 . however , the star is usually considered to be a non - member based on proper motion studies . the most precise proper motion study of m67 is that of girard et al . ( 1989 ) , who give a 21% probability of cluster membership for s1466 . further study of s1466 is perhaps warranted , given that a blue straggler might acquire a peculiar velocity if its formation process is due to stellar interactions ( e.g. @xcite ) . to study the contribution of blue stragglers at other wavelengths , we perform a semi - empirical computation of the integrated m67 ultraviolet spectrum , using the list of stars with at least 80% probability of cluster membership from the astrometric studies of ( in order of preference ) girard et al . ( 1989 ) , zhao et al . ( 1993 ) , and sanders ( 1977 ) . although the list of proper motion members does not extend fainter than about v = 15 ( figure [ fig-2 ] ) , the fainter main - sequence stars should contribute negligible light below 3000 . for all but two of the proper motion members , v and magnitudes could be obtained from the catalogs of mmj93 , sanders ( 1989 ) , girard et al . ( 1989 ) or fan et al . effective temperatures are estimated from the dereddened colors , using the t@xmath17 relation of lejeune et al . gravities are approximated by assuming a mass of 1.25 m@xmath2 for stars brighter than v=13 , and assuming @xmath18 = 4.5 for fainter stars . from the t@xmath17 and @xmath18 of each star , a model synthetic ultraviolet spectrum from the compilation of lejeune et al.(1997 ) is assigned , normalized to the star s dereddened v magnitude . figure [ fig-4 ] shows the predicted integrated ultraviolet spectrum of m67 , with the blue straggler contribution shown separately from all other stars . even neglecting the possibly anomalous star f81 , the blue straggler contribution still dominates the integrated m67 spectrum at wavelengths shorter than 2600 . the lower panel of figure [ fig-4 ] compares the integrated spectra with synthetic spectra derived from the solar - metallicity isochrones computed by bruzual & charlot , and tabulated in leitherer et al . ( 1996 , also see @xcite ) . the synthetic spectra are computed for an instantaneous starburst and a salpeter initial mass function . ( we do not attempt to fit the integrated spectra shortward of 2100 , because the bruzual & charlot models include a contribution from the short - lived hot post - agb stars , which do not exist in the relatively small population of m67 . ) if the blue stragglers are excluded , than a 4.5 gyr model provides the best fit to the semi - empirical integrated spectrum , which is in agreement with the age of m67 estimated from color - magnitude diagrams . however , if all the blue stragglers except f81 are included in the integrated spectrum , then a 2.5 gyr isochrone provides the best fit , and , in particular , provides a good fit of the spectral break at 2600 . thus , in this case , the neglect of blue stragglers in a population synthesis model of the unresolved ultraviolet spectrum would lead to a significant underestimate of the cluster age . there are at least two ways in which the blue straggler population of m67 might differ from that of galaxies of similar age . first , dynamical interactions of m67 with the galaxy have enhanced the relative contribution of blue stragglers to the integrated spectrum , because the low - mass stars have evaporated from the cluster , while blue stragglers are concentrated toward the cluster center ( mmj93 ) . a more important question concerns the origin of blue stragglers themselves ; whether they arise from primordial binaries either via mass transfer processes or stellar mergers , or whether they originate from stellar collisions enhanced by the moderately high stellar density ( 20 pc@xmath27 ; @xcite ) in the m67 center . while this stellar density is often exceeded in the cores of many elliptical galaxies ( @xcite ) , the typical star density in elliptical galaxies or stellar bulges will be at least an order of magnitude lower . in fact , there is good evidence that both mass - transfer and stellar collision processes contribute to the m67 blue straggler population . the short - period ( 4.2d ) binary f190 has long been a good candidate for ongoing mass transfer ( @xcite ) , and the present work provides evidence that both f90 and f131 are algol - type mass - transfer systems . as noted above , landsman et al . ( 1997 ) showed that the yellow giant s1040 is almost certainly a blue straggler descendant , and a product of early case b mass transfer . on the other hand , the bright blue straggler f81 has a mass larger than twice the turnoff mass , and the blue stragglers f55 , f238 , and f555 ( s997 ) have orbits with appreciable eccentricities ( @xcite ) . neither of these properties can be easily understood with a mass - transfer scenario , or mergers of primordial binaries , but are easily accommodated with binary - binary or binary - single star collisions ( @xcite ) . is the large contribution of blue stragglers to the integrated ultraviolet spectrum of m67 typical of other open clusters ? spinrad et al . ( 1997 ) estimate that the blue straggler contribution to the cluster ultraviolet spectrum is higher for m67 than for any of the seven clusters they examined . ahumada & lapasset ( 1995 ) have compiled a catalog of blue stragglers in 390 open clusters , in which they tabulate both nbs , the number of blue stragglers in the cluster , and n2 , the number of stars on the main sequence to two magnitudes below the turnoff . the ratio nbs / n2 is 0.15 for m67 , about twice the mean of the entire open cluster population . we do not pursue this comparison any further here , but instead make the following two points . first , the best way to detect the presence of blue stragglers in an unresolved population is to have wide spectral coverage , including both ultraviolet and visible bandpasses . second , evolutionary models of close binaries including mass transfer ( e.g. @xcite ) would be a useful addition to galaxy population synthesis models , especially for those spectral regions expected to be dominated by the stellar turn - off population . figure [ fig-2 ] shows that three uit sources are coincident with faint hot sources in mmj93 , and occupy the white dwarf region in the lower left - hand corner of the hr diagram . four other uit sources are outside of the region studied by mmj93 , but are coincident with faint , hot stars in the batc ( beijing - arizona - taipei - connecticut ) intermediate - band spectrophotometric survey of fan et al . the uit photometry of these seven white dwarf candidates , along with that of the yellow giant white dwarf binary s1040 is given in column 4 of table [ tbl-4 ] . the bv photometry is given in columns 2 and 3 , where the intermediate - band photometry of the batc survey is converted to v and magnitudes using equations ( 3 ) and ( 4 ) in fan et al . column 5 gives the distance of the white dwarf candidates from the star s1023 , assumed to be the cluster center . the brightest of the white dwarf candidates ( and the fifth brightest source in m67 at 1520 ) is mmj 5670 ( v @xmath28 ) , which is also star g152 in gilliland et al . this star was suspected to be responsible for the soft component of a blended rosat x - ray source in m67 discovered by belloni et al . subsequent optical spectroscopy by pasquini et al . ( 1994 ) and fleming et al.(1997 ) demonstrated that mmj5670 is a hot da white dwarf . fleming et al.derive t@xmath17 = 68,230 @xmath29 k , and @xmath18 = @xmath30 from a model atmosphere fit to the balmer lines . using a pure hydrogen model atmosphere with this t@xmath17 and @xmath18 and normalizing to v = 18.71 , and assuming e ( ) = 0.025 , yields a predicted uit magnitude of m@xmath9 = 14.11 , in reasonable agreement with the observed value of m@xmath9= 13.92 . only 10% of hot ( @xmath31 20,000 k ) white dwarfs are x - ray luminous ( @xcite ) , and this fraction is much lower for white dwarfs with t@xmath17 @xmath31 50,000 k , because the x - ray emission in hot white dwarfs can be strongly suppressed by the presence of radiatively supported trace metals in the photosphere ( @xcite , @xcite ) . thus , mmj 5670 must have have a remarkably pure hydrogen atmosphere . the high t@xmath17 of mmj 5670 also indicates a young age ; according to the 0.6 m@xmath2 white dwarf cooling models of wood ( 1995 ) , mmj 5670 began its descent on the white dwarf cooling curve within the past @xmath32 yr . fleming et al . ( 1997 ) also obtained an optical spectrum of mmj 5973 ( v @xmath33 ) , after learning of its detection on the uit image . they find that mmj 5973 has a db spectrum , and they derive t@xmath17 = 17,150 @xmath34 k , and @xmath18 = @xmath35 from a model atmosphere fit . presuming cluster membership ( see below ) , mmj 5973 is the first db white dwarf found in an open cluster , and shows that both db and da white dwarfs can be produced in the same cluster . the origin of the distinction between da and db white dwarfs is still a matter of some debate . the high surface helium abundance of db white dwarfs might arise if the db stars are descended from stars that left the asymptotic giant branch ( agb ) during a helium shell flash ( e.g. @xcite ) . as opposed to this `` primordial '' scenario , surface phenomena such as mass loss , gravitational settling and convective overturn might provide an evolutionary path between db and da white dwarfs , provided that the hydrogen surface layer is sufficiently thin ( @xcite ) . remarkably , all three of the spectroscopically studied white dwarfs in m67 have turned out to have unusual characteristics ; mmj 5670 is a very hot x - ray - luminous da , mmj 5973 is a db , and the s1040 companion is a very low mass core - helium white dwarf . no spectra exist for the five faintest white dwarf candidates in table [ tbl-4 ] . the three faintest sources ( mmj 6061 , batc 2776 , batc 3337 ) are near the uit detection limit , and are best considered as three sigma detections . the sources batc 4672 (= 0.46 ) and batc 3009 (= 0.24 ) have redder optical colors than expected for a hot white dwarf , and thus might be white dwarf red dwarf binaries . the approximate sensitivity limit of the uit image of m67 at 1520 is m@xmath9 = 16.6 or @xmath36 erg s@xmath14 @xmath37 @xmath14 . using the field white dwarf luminosity function and scale height ( 275 pc ) of boyle ( 1989 ) , only 1.26 white dwarfs brighter than this limit are predicted to be within the uit @xmath7 diameter field of view at the galactic latitude and reddening of m67 . ( since the luminosity function is given in terms of m@xmath21 , we used the @xmath18 = 8.0 pure hydrogen model atmospheres of bergeron et al . ( 1995 ) to compute absolute magnitudes , m@xmath21 , and m@xmath9 v colors as a function of t@xmath17 . ) this low contamination rate is consistent with a general lack of white dwarf candidates in other random uit fields . in view of this low expected contamination rate , and the clustering of the white dwarf candidates toward the cluster center , most of the stars in table [ tbl-4 ] are likely cluster members . fleming et al . do raise a caveat about the cluster membership of mmj 5670 and mmj 5973 . they compute absolute magnitudes for the two stars using a mass - radius relation , and the t@xmath17 and @xmath18 values derived from the best - fit model atmosphere , and find that the very hot da star mmj 5670 is 1 mag too bright , and the db star mmj 5973 is 0.7 mag too faint to be consistent with a cluster distance modulus of m however , there are several large uncertainties in these absolute magnitude determinations . the s / n of the white dwarf spectra are low ( @xmath38 ) , and there are known problems with model atmospheres for both db stars and very hot da stars . the optical photometry of mmj 5973 is near the detection limit of mmj93 , and there is a 0.4 mag difference between mmj93 and gilliland et al . ( 1991 ) in the photometry of mmj 5670 . nevertheless , the absolute magnitude differences are still uncomfortably large , and proper motion studies of the two stars would be very desirable to verify cluster membership . m67 has a core radius of @xmath39 and a tidal radius of about @xmath40 ( @xcite ) , whereas the uit image has a diameter of @xmath7 and is offset by @xmath41 from the center of m67 . to estimate the incompleteness of the uit white dwarf sample , we note that 85% of the 222 stars with membership probabilities greater than 90% in the wide field astrometric study of zhao et al . ( 1993 ) , are located within uit field of view . thus , one or two white dwarf cluster members brighter than the uit detection limit might be missing from table [ tbl-4 ] . how many white dwarfs brighter than the uit detection limit are expected in m67 , based on the cluster size and age ? using pure hydrogen model atmospheres with @xmath18 = 8.0 , and the mass - radius relation of wood ( 1995 ) , the temperature of a 0.6 m@xmath2 wd at the distance and reddening of m67 at the uit detection flux limit is computed to be 21,300 k. from the white dwarf cooling models of wood ( 1995 ) , the time for a 0.6 m@xmath2 wd to cool to this temperature is about 60 myr . thus , all the white dwarfs in m67 less than 60 myr old should detectable on the uit image , including those that would be optically hidden in binary systems with late - type companions . ( only possible white dwarf companions of the hottest blue stragglers would remain undetected on the uit image . ) the question now becomes how many white dwarfs should have been created in m67 in the last 60 myrs ? a quick estimate of about seven white dwarfs can be made by simply comparing with the seven observed red clump stars , which are expected to have a similar lifetime ( @xmath42 myr , @xcite ) . alternatively , dorman et al . ( 1995 ) estimate that there should be 0.076 stars leaving the main sequence per gyr per l@xmath43 for a 4 gyr solar metallicity isochrone . multiplying this number by 0.060 gyr and the cluster visible luminosity of l / l@xmath43 = 1660 ( @xcite ) gives an estimate of 7.6 wds . thus , the predicted number of wds is in reasonable agreement with the eight detected white dwarf candidates . we conclude that the stellar evolution timescale and the hot white dwarf cooling timescale are not in major disagreement , and that most of the young ( @xmath20 60 myr ) white dwarfs have not yet evaporated from the cluster . note that although there is a marked deficit of 0.6 m@xmath2 _ main - sequence _ stars in m67 ( mmj93 ) , no evaporation of the white dwarfs is expected on a time scale of 60 myr ( @xcite ) . stronger assertions will be possible once spectra and proper motion studies have been obtained for all the white dwarf candidates . because the relatively small population of m67 does not include any post - agb or hot hb stars , its integrated ultraviolet spectrum will be dominated by the hot white dwarfs at wavelengths shortward of lyman @xmath44 . would white dwarfs be detectable in the integrated ultraviolet spectra of galaxies ? bica et al . ( 1996a , b ) pointed out the presence of features at 1400 and 1600 in iue spectra of some early - type galaxies , and suggested that these might correspond to the lyman @xmath44 satellite lines observed at these wavelengths in intermediate temperature ( 9000 - 18,000 k ) da white dwarfs ( @xcite ) . however , these features are not present in spectra of the same galaxies observed with the hopkins ultraviolet telescope ( hut : @xcite ) . in addition , the satellite lines at 1400 and 1600 are not unique to white dwarfs , but also appear in the spectra of low - metallicity a - type and horizontal branch stars ( @xcite ) . nevertheless , the population synthesis models of magris & bruzual ( 1993 ) do suggest that white dwarfs are a small but non - negligible contributor to the ultraviolet spectrum of ellipticals with a small uv upturn . they find that , at 1750 , the contribution of white dwarfs is 10% of the contribution of hot post - agb stars in galaxies older than 8 gyr . we have performed our own estimate of the white dwarf contribution to the ultraviolet spectrum of an old stellar population . we use the carbon - core cooling models of wood ( 1995 ) with thick hydrogen and helium envelopes , and the pure hydrogen model atmospheres of bergeron et al . ( 1995 ) . to allow an easy comparison of the contribution of white dwarfs with that of other ultraviolet sources , we follow dorman et al . ( 1995 ) and compute the energy e@xmath45 integrated over the cooling curve in units of l@xmath43 gyr @xmath14 , where l@xmath46 ergs s@xmath14 is the visible solar luminosity . figure [ fig-5]a shows the resultant ultraviolet spectrum for both a 0.6 m@xmath2 and a 0.9 m@xmath2 cooling white dwarf . the time required for the 0.9 m@xmath2model to cool to 20,000 k is 190 myr , whereas only 75 myr are needed for a 0.6 m@xmath2 to cool to this temperature . despite this large difference in cooling age , the integrated fluxes , e@xmath45 , along the two tracks differ by only @xmath47% outside of the lyman lines , because the radii in the more massive model are typically 30% smaller , which partially compensates for the longer cooling time . other complications , such as the presence of db white dwarfs , and variations in the envelope mass , probably do not cause flux variations much larger than 15% . figure [ fig-5 ] also shows that the main contribution to the integrated spectrum at 1500 comes from white dwarfs with t@xmath17 @xmath31 20,000 k , so that the lyman @xmath44 satellite line at 1400 , characteristic of cooler da white dwarfs , is barely detectable . the integrated energy , along the 0.6 m@xmath2 track at 1500 is @xmath48 e@xmath49 l@xmath43 gyr @xmath14 . by comparison , dorman et al . ( 1995 ) estimate @xmath48 e@xmath50 l@xmath43 gyr @xmath14 for their lowest core mass ( 0.546 m@xmath2 ) post - agb track , and @xmath48 e@xmath51 l@xmath43 gyr @xmath14 for typical ehb tracks . since the use of a low core mass probably somewhat overestimates the true post - agb contribution , we confirm the result of magris & bruzual ( 1993 ) that white dwarfs contribute a small , but non - negligible ( @xmath4 10% ) fraction of the ultraviolet flux due to post - agb stars . however , the ehb population will dominate the uv spectrum , even if only a small fraction ( @xmath4 5% ) of core helium - burning stars follow ehb tracks . ngc 6791 is noteworthy for being among the oldest , most populous , and most metal - rich of the galactic open clusters ( @xcite ) . liebert et al . ( 1994 ) discovered another noteworthy property of ngc 6791 with their spectroscopic detection in the cluster of an sdo star and four sdb stars with 24,000 k @xmath20 t@xmath17 @xmath20 32,000 k. they speculated that these were the metal - rich analogue of the hot hb stars found in globular clusters , and that the existence of these stars in ngc 6791 was connected with the super - solar metallicity of the cluster . recent stellar evolution models of high - metallicity , low - mass stars provide support for this idea , and have also been used to show that such metal - rich hot hb stars could be primarily responsible for the uv - upturn observed in the spectra of elliptical galaxies ( @xcite , @xcite , @xcite ) . twenty stars are detected on the uit image of ngc 6791 but twelve of these are certain non - members , either because they are identified with bright ( v @xmath2013 ) field stars , or because they are more than @xmath52 from the cluster center . the detected probable cluster members include the five hot subdwarfs discussed by liebert et al . ( b2 , b3 , b4 , b5 , and b6 ) , plus the additional two sdb / o candidate stars , b9 and b10 , reported by kaluzny & rucinski ( 1995 ) . the star b1 , suspected to be a cooler ( t@xmath17 = 15,000 k ) member blue hb star by green et al . ( 1998a ) , is detected on the uit image at the three sigma level . the other two blue hb stars studied by green et al . , k1943 and k5695 , are too cool to be detected with uit . neither of the cataclysmic variables , b7 and b8 , reported by kaluzny et al . ( 1997 ) are detected with uit . even during their outburst phase , these cataclysmic variables have redder optical colors than the sdb / o candidates , and probably remain below the uit detection limit of m@xmath10 @xmath4 16.1 . table [ tbl-5 ] presents the uit photometry of the possible cluster stars , along with the visible photometry from kaluzny & rucinski ( 1995 ) . figure [ fig-6 ] shows archival ghrs low - dispersion ( g140l ) spectra covering the wavelength region 1300 1600 of the sdo star b2 and the sdb star b4 . also shown are high - dispersion _ iue _ spectra of the field sdo star feige 67 ( swp 20488 ) , and the field sdb star pg 0342 + 026 ( swp 27466 ) , smoothed to the approximate ghrs resolution , and reddened to match the ngc 6791 reddening . from ultraviolet and optical spectrophotometry , theissen et al . ( 1995 ) found t@xmath17 = 25,000 k , @xmath18 = 5.25 and e ( ) = 0.10 , for pg 0342 + 026 , while saffer et al . ( 1994 ) found t@xmath17 = 26,200 k , @xmath18 = 5.63 from fitting the balmer lines . the best - fit kurucz model atmosphere ( normalized to the v magnitude and reddened by e ( ) = 0.17 ) to the ghrs spectrum of b4 has t@xmath17 = 26,800 k , @xmath18 = 5.0 , and [ fe / h ] = 0.2 . while a good fit to b4 can also be achieved using a solar metallicity model , figure [ fig-6 ] shows that a low metallicity ( [ fe / h ] = 0.5 ) model does not match the depressed continuum longward of 1500 , which is primarily due to the lines of and ( @xcite ) . this result is of some importance because abundance depletions are known to exist in sdb stars , due to diffusion and other particle transport processes in the high - gravity atmosphere . for example , liebert et al . ( 1994 ) found the sdb stars in ngc 6791 ( including b4 ) to be helium - poor , as is typical of field sdb stars ( @xcite ) , and lamontagne et al . ( 1987 ) summarizes the evidence for depletions of carbon and silicon in the field sdb stars ( also see brown et al . 1996 ) . nevertheless , the carbon and silicon depletions are smaller for the sdb stars than for hotter ( @xmath31 30,000 k ) stars , and other major ultraviolet opacity sources such as iron are not depleted . thus , our rough analysis of the ghrs spectrum of b4 suggests that ultraviolet population synthesis models which incorporate evolutionary tracks of metal - rich hot hb stars should also adopt stellar atmosphere models of an equal metallicity . a more detailed comparison of the ghrs spectrum of b4 with spectral synthesis models would be desirable to verify this point . liebert et al . ( 1994 ) did not derive atmospheric parameters for the sdo star b2 , but showed that its optical spectrum is similar in appearance to the hot ( t@xmath17 @xmath4 60,000 k ) field sdo stars feige 67 and feige 110 . figure [ fig-6 ] shows that feige 67 and b2 have a similar continuum slope in the 1300 1600 region , and many absorption lines are in common , although the spectra clearly differ in many details . due to the large number of lines of ionized iron and nickel in the ultraviolet spectra of sdo stars , a quantitative analysis requires fully line - blanketed nlte model atmospheres , which have only recently become available ( e.g. @xcite , @xcite ) . becker & butler ( 1995 ) derived t@xmath17 = 70,000 k , @xmath18 = 5.2 for feige 67 , and found it to have a low surface abundance of helium ( n(he / h ) = 0.05 ) , but a high iron abundance ( [ fe / h ] = 1.0 ) . we do not attempt to further estimate the atmospheric parameters of b2 , but limit ourselves to the following remarks . first , the identification of b2 as an agb - manqu star may be suspect , because the evolutionary tracks of agb - manqu stars do not spend much time at such high effective temperatures ( @xmath4 60,000 k ) . second , despite the higher t@xmath17 of the sdo stars in figure [ fig-6 ] , they have flatter continua than the two sdb stars , mostly likely due to the large number of lines in the 1300 1600 region ( @xcite ) . thus , accurate population synthesis of the uv - upturn in elliptical galaxies may need to include nlte line - blanketed model atmospheres to properly model agb - manqu stars . unfortunately , the few field sdo stars which have had a detailed model atmosphere analysis of their ultraviolet spectra have shown a wide variety of strong elemental ( including iron ) under- and over - abundances ( @xcite , @xcite , @xcite ) , most likely due to the effects of radiative levitation and diffusion . the question of the appropriate model atmosphere to use for population synthesis of the ultraviolet spectra of metal - rich agb - manqu stars remains a problematic one . the uit photometry is especially useful for estimating the luminosities of the sdb stars , since the bolometric correction at 1620 for sdb stars is relatively small . the effective temperature and luminosity of the sdb stars b3 , b4 , b5 , b6 , and b9 as derived from the uit photometry is tabulated in columns 5 and 6 of table [ tbl-5 ] . ( no spectra exist for the star b9 , but its ultraviolet and visible magnitudes are similar to those of the four spectroscopically identified sdb stars . ) temperatures are estimated by finding the kurucz model with [ fe / h ] = 0.2 , and @xmath18 = 5.0 , which best matches the observed m@xmath10 v color , for an assumed reddening of e ( ) = 0.17 . luminosities are then derived from the m@xmath10magnitude , using the ( small ) bolometric correction for the adopted kurucz model . the values of t@xmath17 derived from m@xmath10 v are consistently about 1200 k hotter than those derived by liebert et al . ( 1994 ) using ( pure hydrogen ) model fits to the balmer lines . table [ tbl-5 ] shows that the five sdb stars in ngc 6791 have a narrow luminosity range , @xmath1l / l@xmath2 @xmath53 . this luminosity range is consistent with that predicted for metal - rich hot hb stars by dorman et al . ( 1993 ) and yi et al.(1997a ) . for example , the hottest models of yi et al . ( with z = 0.04 and a main - sequence helium abundance y = 0.31 ) , have a luminosity range during core helium burning of @xmath54l / l@xmath2 @xmath20 1.28 . the observed t@xmath17 range of the sdb stars is perhaps slightly higher than predicted , since the t@xmath17 of the hottest hb models decreases slightly with increasing metallicity ( @xcite ) . the hottest models of yi et al . and dorman et al.(with envelope masses less than 0.005 m@xmath2 ) with z = 0.04 have t@xmath17@xmath4 23,000 k. the narrow range of luminosities of the sdb stars in ngc 6791 rules outs certain binary star mechanisms for their origin , such as the merger of helium white dwarfs ( @xcite ) , since these predict a wide range of helium core masses , and hence luminosities however , green et al.(1998a ) found that b1 and the other two blue hb stars in ngc 6791 are spectroscopic binaries . ( the sdb stars are too faint for a radial velocity study , so their binary status is unknown . ) the situation is reminiscent of that of the field sdb stars , which occupy a narrow region in the t@xmath17 @xmath18 diagram , consistent with single star evolution ( @xcite ) , but for which there is increasing evidence for a large binary population ( @xcite , @xcite , @xcite ) . one binary star mechanism consistent with a narrow range of gravity or luminosity is the scenario discussed by mengel et al . ( 1976 , also see @xcite ) , in which an evolving red giant loses mass to a companion , so that it arrives on the horizontal branch with a small envelope mass . in effect , the mass transfer to the companion substitutes for the high mass - loss rate required for the production of sdb stars in single star evolution models . one difficulty with both the high mass - loss rate and mass - transfer scenarios is the amount of fine - tuning required ( @xcite ) , since a red giant which loses too much mass never reaches the helium flash , and instead ends up as a helium white dwarf ( but see @xcite ) . note that the maximum envelope mass for which a star still follows agb - manqu tracks increases with increasing metallicity ( @xcite , @xcite ) , so that the binary star mechanism for the production of sdb stars might still show a metallicity dependence . the brightest source in ngc 6791 at 1620 is b10 , which has m@xmath10 = 13.60 . kaluzny & rucinski ( 1995 ) report that b10 is a blend of two stars with @xmath55v @xmath4 2.0 mag and an angular separation of about 0.7 . the photometry in table [ tbl-5 ] is from the 2.1-m telescope observations of kaluzny & rucinski ( 1995 ) , in which the photometry of the hot star is better isolated . nevertheless , the hot star might still have a composite spectrum because its ultraviolet color ( m@xmath10 v = 2.68 ) is somewhat inconsistent with its relatively red color ( = @xmath56 ) . green et al . ( 1998a ) obtained three spectra of b10 in which the neighboring cool star apparently dominates the spectrum . they find a radial velocity consistent with cluster membership , and no evidence for velocity variations , although it is not clear whether these results also apply to the hot star . minimum values of t@xmath17 @xmath31 22,900 k and @xmath48l / l@xmath2 @xmath57 for b10 can be derived from the uit photometry by neglecting any contribution to the v light from a possible red companion . b10 appears to be a good candidate for a agb - manqu star . seven stars are detected on the uit image of the old ( @xmath58 gyr ) solar - metallicity open cluster ngc 188 , but four of these are certain non - members ( table [ tbl-6 ] ) . the early photometry of sandage ( 1962 ) included the hot star ii-91 ( v = 13.82 , = 0.22 ) , but only recently has this star been shown to be a probable proper - motion cluster member ( @xcite ) , and to have an sdb spectrum ( @xcite ) . green et al . also report that ii-91 is a spectroscopic binary . as with the sdb stars in ngc 6791 , the effective temperature and luminosity of ii-91 can be estimated from its uit magnitude and m@xmath9 v or m@xmath10 v color , for the adopted cluster distance ( 1700 pc ) and reddening ( e ( ) = 0.12 ) . use of either the m@xmath10 or m@xmath9 magnitude yields a temperature of about 30,000 k and luminosity of @xmath48 l / l@xmath2 = 1.14 . thus , the temperature and luminosity of ii-91 falls into the same range found for the sdb stars in ngc 6791 . the brightest source in ngc 188 at 1520 is the star d702 , from the catalog of dinescu et al . ( 1996 ) , who give v = 13.70 and = 0.26 from their photographic photometry , and also assign a 80% probability of cluster membership on the basis of its proper motion . dinescu et al . also remark , however , that d702 is part of a blended image , so that it has large proper motion errors , and possibly erroneous photometry . d702 is also e43 in the catalog of caputo et al . ( 1990 ) , who give v = 14.04 and = 0.28 from their ccd photometry . here we adopt the photometry ( v = 14.18 , = 0.26 ) quoted by green et al . ( 1998a ) . the ultraviolet photometry shows that d702 must have a composite spectrum , because its relatively red color is inconsistent with its blue ultraviolet colors ( m@xmath9 v = 1.89 ; m@xmath9 m@xmath10 = 0.11 ) . a minimum value of t@xmath17 = 15,000 k can be derived for d702 from the observed m@xmath9 v color by neglecting any contribution of a red companion to the v magnitude ; the t@xmath17could be much higher , if the v magnitude includes significant contamination from the companion . one possible decomposition of d702 consistent with the visible and ultraviolet photometry is that it consists of a turnoff star with v = 14.9 and = 0.69 , and a hot companion with v = 14.9 and = 0.04 and t@xmath17 = 15,000 k. alternatively , the hot companion could be similar to ii-91 with v = 16.2 and t@xmath17 = 30,000 k , in which case the primary would be a blue straggler with v = 14.36 and = 0.36 . unlike the case for m67 , none of the blue stragglers in ngc 188 or ngc 6791 are detected with uit ( with the possible exception of d702 ) , nor would any detections be expected . this is partly because these clusters are more distant and more reddened , and their uit exposures shallower than that of m67 , but also because these older clusters have a blue straggler population with redder colors . a final uit source , located @xmath59 from the cluster center , and designated here as uit-1 , could not be identified with any sources in the ngc 188 star catalogs . however , it is located @xmath60 from the source u1725_00042710 in the usno - sa1.0 sky survey of monet et al . ( 1996 ) . in their photometric system defined by the iiia - j and iiia - f emulsions , u1725_00042710 has the very blue colors , r=18.7 and b = 18.0 . however , a 4000 5000 spectrum of this star obtained at calar alto in 1996 august ( s. moehler , personal communication ) showed a very red continuum with no evidence of broad balmer lines . whether this discrepancy indicates an eclipsing binary , or is due to a misidentification is currently unknown . the star is robustly detected in both uit filters , and the ultraviolet color m@xmath9 m@xmath10 = 0.17 indicates a hot ( @xmath31 20,000 k ) spectrum . the cmd of ngc 188 shown by dinescu et al . ( 1995 ) has five red clump stars , so the two sdb candidates detected by uit provide 28% of the hb population . the cmd of ngc 6791 given by kaluzny & rucinski ( 1995 ) shows about 23 red clump stars , so the eight hot stars detected by uit , plus the two cooler blue hb stars discussed by green et al . ( 1998a ) , form about 30% of the hb population . by comparison , dorman et al . ( 1995 ) found that they could fit the ultraviolet spectra of even the strongest uv - upturn galaxies if only 25% of the core helium - burning stars followed ehb tracks , while brown et al . ( 1997 ) needed an ehb fraction of at most 10% to model the hut spectra of six ellipticals . thus , the integrated spectra of ngc 6791 and ngc 188 would likely show a strong uv - upturn , although the present incompleteness of the cluster membership information prevents us from computing semi - empirical integrated spectra for the two clusters , as we did for m67 in section 3.1 . the similarity of the ehb fraction for ngc 188 and ngc 6791 might suggest that the super - solar metallicity of ngc 6791 is not an essential ingredient in the origin of its hot subdwarf population . however , there are several reasons why this conclusion might be premature . first , high percentage of sdb stars in ngc 188 could simply be an artifact of small number statistics . second , apart from ngc 188 , no sdb stars have yet been found in any other open cluster with a metallicity less than or equal to solar . finally , both sdb candidates in ngc 188 are in binary systems , and , as pointed out by green et al . ( 1998a ) , ngc 188 is known to have a rich binary population , whereas ngc 6791 appears to have a relatively low binary fraction . there are at least three ways in which galactic open clusters are an inadequate model for understanding the stellar populations of old galaxies . first , the open clusters have experienced dynamical interactions which can alter the stellar population mix ; for example , by leading to evaporation of low - mass stars while concentrating the blue stragglers toward the cluster center . second , the higher mean stellar density of open clusters might result in enhanced stellar interactions and possibly increased blue straggler or hot subdwarf formation . finally , due to the relatively small population of open clusters , the rapid evolutionary phases may not be adequately sampled . for example , hot post - agb stars are believed to be a significant ( although not the dominant ) contributor to the observed uv - upturn in elliptical galaxies , but consistent with their short ( @xmath61 yr ) lifetime , there are no post - agb stars present in m67 , ngc 188 , or ngc 6791 . and while hot subdwarfs have now been detected in ngc 188 and ngc 6791 , their total number ( @xmath62 ) remains uncomfortably low to draw any robust conclusions . keeping this limitations in mind , we summarize below the main results derived from the uit observations of m67 , ngc 188 , and ngc 6791 , and their implications for the study of the integrated ultraviolet spectra of old stellar populations . 1 . the uit image of m67 is dominated by the eleven detected blue stragglers ; in particular , the blue straggler f81 contributes 90% of the integrated flux of m67 at 1520 . the ultraviolet flux of the two blue stragglers , f90 and f131 , is significantly higher than that predicted on the basis of optical photometry , and probably indicates the presence of hot , subluminous companions . a semi - empirical calculation of the integrated ultraviolet spectrum of m67 shows that , even when neglecting the possibly anomalous star f81 , the blue stragglers dominate at wavelengths shorter than @xmath4 2600 . as pointed out by spinrad et al . ( 1997 ) , neglect of blue stragglers in population synthesis model fits of the rest - frame near - ultraviolet spectra of a galaxy will result in an underestimate of the galaxy age . 2 . eight white dwarf candidates are identified in the uit image of m67 , including the core - helium white dwarf companion of the yellow giant s1040 . optical spectroscopy of two of these sources has been obtained by fleming et al . ( 1997 ) : g152 is a hot ( t@xmath17@xmath4 68,000 k ) da white dwarf , and mmj 5973 is cooler ( t@xmath17 @xmath4 18,000 k ) db white dwarf . the number of white dwarf candidates is in reasonable agreement with that expected from theoretical white dwarf cooling models and a cluster age of 4 gyr . the integrated ultraviolet flux at 1500 along a white dwarf cooling model is @xmath48 e@xmath49 l@xmath43 gyr @xmath14 , and the contribution of white dwarfs to the integrated spectra of old galaxies is roughly 10% of that expected from hot post - agb stars . eight probable cluster members are detected on the uit image of ngc 6791 , including the five hot subdwarfs studied spectroscopically by liebert et al.(1994 ) , and the two additional sdb / o candidates , b9 and b10 , reported by kaluzny & rucinski ( 1995 ) . three probable cluster members are detected on the uit image of ngc 188 , including the sdb spectroscopic binary ii-91 . the star d702 in ngc 188 is probably a composite including a hot subdwarf , since it has a hot ultraviolet color ( m@xmath9 m@xmath10 = 0.11 ) , but a relatively cool optical color ( = 0.26 ) . the derived luminosity range , @xmath1 l / l@xmath2 @xmath53 , of the five sdb stars in ngc 6791 , and ii-91 in ngc 188 , is consistent with that expected for metal - rich , hot hb stars . the fraction of hot hb stars in both clusters is about 30% , implying that the integrated spectra of both clusters should show a pronounced uv - upturn , as strong as that observed in any elliptical galaxy . we thank the many people involved with the _ astro-2 _ mission who made these observations possible . we thank p. bergeron for his white dwarf atmosphere models , a. milone for providing the latest status of the m67 radial velocity monitoring , and d. dinescu for providing an electronic version of her ngc 188 catalog . milone , a.a.e . , latham , d.w . , mathieu , r.d . , morse , j.a . , & davis , r.j . 1992 , in evolutionary processes in interacting binary stars , edited by y. kondo , r.f . sister , and r.s . polidan ( kluwer , dordrecht ) , p. 473 81 & 977 & 10.03 & 0.09 & & 12750 & 4.26 & 8.17 & 280 & 1434 & 10.70 & 0.10 & 12.06 & 8600 & 3.89 & 12.25 & 156 & 1066 & 10.99 & 0.11 & 12.10 & 8680 & 4.13 & 12.39 & 153 & 968 & 11.28 & 0.13 & 13.15 & 8490 & 4.31 & 13.04 & am 55 & 752 & 11.32 & 0.17 & 15.61 & 7620 & 4.00 & 14.90 & p = 1003d 238 & 1267 & 10.94 & 0.22 & 14.20 & 8010 & 3.85 & 13.72 & p = 846d 190 & 1284 & 10.94 & 0.25 & 14.47 & 7750 & 3.78 & 14.16 & p = 4.2d , @xmath22 sct 185 & 1263 & 11.06 & 0.26 & 14.31 & 8290 & 4.14 & 13.18 & 184 & 1280 & 12.26 & 0.29 & 15.28 & 8090 & 4.12 & 15.33 & @xmath22 sct 261 & 1466 & 10.60 & 0.34 & 14.32 & & & & nonmember ? 131 & 1082 & 11.25 & 0.41 & 15.16 & 6930 & 4.09 & 16.47 & 90 & 975 & 11.08 & 0.43 & 16.18 : & 6490 & 3.44 & 18.04 & p = 975d mmj 5670 , g152 & 18.61 & 0.11 & 13.92 & 1.7 & da1 mmj 5973 & 19.66 & 0.10 & 15.61 & 6.7 & db s1040 & 11.52 & 0.82 & 15.82 & 1.0 & g8iii + wd batc 4672 & 17.66 & 0.46 & 15.84 & 10.1 & batc 3009 & 17.02 & 0.24 & 16.09 & 14.2 & mmj 6061 & 20.53 & 0.14 & 16.50 & 4.3 & batc 2776 & ... & ... & 16.58 & 7.5 & @xmath63batc 3337 & 18.44 & 0.09 & 16.66 & 12.0 & b1 & 16.97 & 0.07 & 16.20 : & & & 15,800 & bhb ? b2 & 17.43 & 0.12 & 14.47 & & & & sdo b3 & 17.77 & 0.12 & 14.96 & 24,170 & 1.21 & 22,900 & sdb b4 & 17.87 & 0.13 & 14.85 & 26,790 & 1.27 & 25,200 & sdb b5 & 17.90 & 0.10 & 15.21 & 22,770 & 1.10 & 21,800 & sdb b6 & 17.97 & 0.10 & 15.07 & 25,290 & 1.17 & 23,800 & sdb b9 & 18.18 & 0.15 & 14.91 & 30,050 & 1.25 & & b10 & 16.28 & 0.01 & 13.60 & @xmath64 & @xmath65 & & hd 4041 & 8.87 & 0.17 & 9.96 & ... & a2v , nonmember hd 4816 & 8.26 & 0.20 & 12.45 & 11.65 & a5v , nonmember hipp 3354 & 9.58 & 0.50 & 16.13 & ... & f3/f5v , nonmember tyc 4619 & 11.28 & 0.08 & 15.13 & 14.38 & nonmember ii-91 & 16.31 & 0.17 & 12.55 & 12.78 & sdb d-702 & 14.18 & 0.26 & 12.29 & 12.40 & dinescu et al . ( 1996 ) uit-1 & ... & ... & 13.95 & 14.12 & usno 1725 - 00042710 ?

we study the possibility that the very broad ( 1500 km s ) and luminous ( ) [ oiii ] line emission observed in the globular cluster rz 2109 might be explained with the photoionization of nova ejecta by the bright ( ) x - ray source hosted in the same globular cluster . we find that such scenario is plausible and explains most of the features of the rz 2109 spectrum ( line luminosity , absence of h emission lines , peculiar asymmetry of the line profile ) ; on the other hand , it requires the nova ejecta to be relatively massive ( ) , and the nova to be located at a distance from the x - ray source . we also predict the time evolution of the rz 2109 line emission , so that future observations can be used to test this scenario . novae , cataclysmic variables - galaxies : individual ( ngc 4472 ) - galaxies : star clusters : individual ( rz 2109 ) - x - rays : binaries - ism : jets and outflows

1203.4233

many x - ray sources ( xrss ) are inside globular clusters ( gcs ; e.g. heinke 2011 ) , but only five of them unambiguosly host a black hole ( bh ; maccarone et al . 2007 , hereafter m07 ; brassington et al . 2010 ; shih et al . 2010 ; maccarone et al . 2011 ; irwin et al . 2010 , hereafter i10 ; see king 2011 for an alternative explanation ) . these five objects were observed as luminous xrss ( @xmath6 ) with strong variability ( so that blending is excluded ; kalogera et al . the spectra of at least two of their host gcs are quite peculiar . one of them , rz 2109 ( in the elliptical ngc 4472 ; zepf et al . 2007 , 2008 , and steele et al . 2011 ; hereafter z07 , z08 , s11 ) displays prominent and broad [ oiii]@xmath75007,4959 emission ( the fwhm of [ oiii]@xmath8 is @xmath9@xmath10 ; its flux is @xmath11 ; at @xmath12 it corresponds to @xmath13 ; see z08 ) , but no h line emission@xmath14 , [ nii]@xmath15 , [ sii]@xmath16 ) is detected , except perhaps [ oiii]@xmath17 . ] . the spectrum of the gc hosting the xrs cxoj033831.8 - 352604 ( in the elliptical ngc 1399 ; see i10 ) is somewhat similar in the absence of h and the presence of [ oiii ] lines ; but widths and luminosities are lower ( fwhm @xmath18 , @xmath19 ) , and [ nii ] @xmath76583,6548 emission is comparable to [ oiii ] . these observations are hard to explain , because of the high [ oiii]/h@xmath20 ratio ( z08 estimate @xmath21 ; i10 finds a 3@xmath22 lower limit of 5 ; both are very uncertain ) and of the large broadening , implying motions much faster than gc velocity dispersions ( @xmath23 ) . optical lines are too wide for a planetary nebula , and too narrow for a supernova remnant . z08 suggest that the [ oiii ] emission from rz 2109 might come from a strong wind originating around the bh , photoionized by the xrs . such scenario strongly favours a bh with mass @xmath24 , rather than an intermediate - mass bh ( imbh ) with mass in the 10@xmath25 - 10@xmath26m@xmath27 range . this was preferred to the hypothesis that the [ oiii]-emitting gas closely orbits ( @xmath28 ) an imbh , on the basis of considerations about the maximum possible emission from such a small volume . such argument is severely weakened if the orbiting gas is very metal - rich , e.g. as the result of the tidal disruption of a white dwarf ( wd ) by an imbh ( see i10 ) : according to porter ( 2010 ) , the cxoj033831.8 - 352604 data are compatible with this scenario , whereas in the case of rz 2109 the conflict remains ; however , clausen & eracleous ( 2011 ) showed that simulations of wd disruption can reproduce the [ oiii ] line profiles observed in rz 2109 . on the other hand , maccarone & warner ( 2011 ) argued that the line emission of cxoj033831.8 - 352604 might be better explained with the photoionization ( by the xrs ) of the wind from an r coronae borealis star in the same gc . in this paper , we examine a further possible scenario , and apply it to the case of rz 2109 : [ oiii ] lines might be associated with a recent nova eruption in the core of the gc . in particular , we discuss the case ( first mentioned by s11 ) where the [ oiii ] emission comes from the photoionization of the nova ejecta by the xrs . x - ray emission from novae is typically fainter ( @xmath29 , even though at least one nova was reported to reach @xmath30 - see henze et al . 2010 ) and shorter ( decay times @xmath31 ) than what is observed for these sources . thus , we assume that the nova eruption is _ not _ associated to the xrs , i.e. that the co - existence of the two phenomena in the same gc is serendipitous . nova eruptions occur in binary systems where a wd accretes from a normal star ( e.g. gehrz et al . 1998 ) . nuclear reactions ignite explosively in the material accumulated on the wd surface , leading to a brief super - eddington phase during which the wd outer layers are ejected ( e.g. starrfield et al . 1972 , prialnik 1986 ) . ejecta velocities are in the 1002500@xmath32 range , and their metal mass fractions can be as high as @xmath33 ( yaron et al . 2005 , shara et al . 2010 ; hereafter y05 , s10 ) . typical ejecta masses are in the @xmath34@xmath35 range ; but y05 and s10 showed that nova eruptions on slowly - accreting , low - temperature wds can eject up to @xmath36 of gas . ejecta mass estimates in x - ray detected novae ( pietsch et al . 2007 ) somewhat confirm theoretical expectations : ejecta masses of @xmath37 are likely in at least 2 out of 18 novae . wd - hosting binaries are common in gcs ( sigurdsson & phinney 1995 ) ; thus , novae should be frequent in gcs . this disagrees with the low number of novae observed in gcs ; but the discrepancy might be explained by observational biases . we parametrize the nova rate in a gc as @xmath38 , where @xmath39 is the nova rate per unit stellar mass in elliptical galaxies ( della valle et al . 1994 ; mannucci et al . 2005 ; henze et al . 2009 ) , @xmath40 is the gc mass , and @xmath41 is the enhancement of the nova rate in gcs ( compared to ellipticals ) . we expect @xmath42 because of the over - abundance of wd - hosting binaries in gcs . on the other hand , observations of the m31 and m87 gc systems can put upper limits on nova occurrence rates : for m87 , shara et al . ( 2004 ) infer @xmath43 , whereas for m31 we have an upper limit @xmath44 ( henze et al . 2009 ) : assuming @xmath45 , these rates imply @xmath46 . we use the photoionization code cloudy ( version 08.00 , see ferland et . al . 1998 ) to model the optical spectrum resulting from the photoionization of the nova ejecta by the xrs . we represent the nova ejecta as a slab of material with thickness @xmath47 , number density @xmath48 , filling factor @xmath49 , and metallicity @xmath50 , at a distance @xmath51 from the xrs , covering a fraction @xmath52 of the solid angle around it . we note that @xmath52 determines two different geometrical configurations : 1 . if @xmath52 is large ( @xmath53 ) , the xrs must be _ within _ the nova ejecta . in such case , we neglect the offset between the xrs and the centre of the ejecta shell , so that the radius of the ejecta shell is @xmath54 ; then , the time elapsed after the nova explosion is @xmath55 , where @xmath56 , and @xmath57 . 2 . if @xmath52 is small , we assume the xrs to be _ outside _ the ejecta , and estimate @xmath58 from simple geometry : @xmath59 , so that @xmath60 . the xrs spectrum was taken from m07 : we use a disk blackbody spectrum with @xmath61 and 0.210 kev luminosity @xmath62 . we note that m07 reported a large variation ( a drop by a factor of @xmath63 ) in the count rate during a single _ xmm - newton _ observation . such variation is consistent with an increase in the intervening column density , from @xmath64 ( consistent with the galactic value ) to @xmath65 @xmath66 . thus , we assume that the intrinsic x - ray spectrum does not change ; but we experiment with different @xmath65 values . lc parameter & range of values + @xmath67 [ cm ] & 1618.5 , with 0.25 steps + @xmath68 [ cm ] & 15.0 , 15.5 , 16 , 16.5 + @xmath52 & 0.003 , 0.01 , 0.03 , 0.1 , 2/3 + @xmath69 [ @xmath70 & -2 to 10 , with 0.1 steps + @xmath49 & 2/3 + @xmath50 & 0.14 , 0.29 , 0.50 + @xmath71 [ kev ] & 0.22 + @xmath72 [ erg s@xmath1 ] & @xmath74 + @xmath75 [ @xmath76 & 19 , 20 , 21 , 21.5 + @xmath77 [ m@xmath27 ] & @xmath79 + table 1 describes the extent of the explored parameter space . we point out that : 1 . we use the cloudy pre - defined nova abundance set . it represents typical nova ejecta , and corresponds to an o / h number ratio of @xmath80 ( @xmath81 times larger than in the solar mixture ) , and to a metal mass fraction @xmath82 . we obtain mixtures with different metallicities by rescaling the metal abundances of this set by a factor 0.4 ( @xmath83 ) or 2.5 ( @xmath84 ) . the slab mass is @xmath85 f_{\rm cov } f_{\rm fill } [ n/(10 ^ 4\,{\rm cm^{-3}})]$ ] , where @xmath86 is the mass of the h atom . since we expect that @xmath87 ( y05 , s10 ) , ionization models were calculated only when @xmath88 . models were required to agree with two basic observational constraints from the spectrum shown in z081.3 years ) . instead the two spectra described by z07 ( taken in 20042005 , i.e. in the previous @xmath03.9 years ) appear quite different ( fwhm @xmath89@xmath90 , @xmath91 ) . z07 warn that uncertainties amount to a factor of @xmath92 , and the discrepancy might be explained by the ( very different ) observational setups . ] 1 . the flux of the [ oiii]@xmath8 line , @xmath93 . z08 do not quote an error on this measurement : the equivalent width comparison performed by s11 suggests an uncertainty of 2030 per cent ; but the discrepancy with the z07 spectrum suggests a larger amount . furthermore , we account for the approximations of the ionization models by increasing such uncertainty . thus , we consider viable models with @xmath94 within a factor of 3 from the z08 estimate . the absence of any other emission lines in all the spectra : we exclude all the models where the luminosity in any optical line , h@xmath20 , [ oi]@xmath95 , [ oii]@xmath14 , [ nii]@xmath15 , and [ sii]@xmath96 . ] exceeds @xmath97 . about 4 per cent of the models we calculated models , of whom 2144 are viable ( 655 if we consider only models with @xmath98 ) . if we require @xmath99 the number of viable models is 1278 ( 262 if we further impose @xmath98 ) , and requiring @xmath100 further decreases this number to 652 ( 100 of whom have @xmath98 ) . ] satisfy the above criteria . in fig . 1 , we show the distribution of the various parameters ( including the ionization parameter , defined as the number ratio of ionizing photons to h atoms , @xmath101 , where @xmath102 is the number of ionizing photons - i.e. with energy @xmath103 - emitted by the xrs in the unit time ) for these viable models , and how these distributions are influenced by the choice of priors about the models ( e.g. the effects of using different thresholds on the ejecta mass , or imposing a minimum for the nova - xrs distance ) . such distributions show us where a large amount of parameter space is available . it can be seen that : 1 . it is almost impossible to have viable models if @xmath104 ; 2 . thin shells ( @xmath105 ) are favoured , especially when high-@xmath106 models are excluded ; 3 . models where the xrs is inside the nova shell ( i.e. with @xmath107 ) are moderately unlikely , especially when high-@xmath108 models are excluded ; 4 . shell densities are in the @xmath109 range ; but models with high values of @xmath51 show a narrower distribution , favouring the low-@xmath48 part ; 5 . @xmath110 models are favoured ; but it is possible to get reasonable results if @xmath111 or even @xmath83 . the ionization parameter is mostly close to @xmath112 , a fairly typical value for hii regions ; this is somewhat remarkable , as the ionizing spectrum is not produced by hot stars , and most of the radiation energy is in the form of quite energetic ( 0.11 kev ) photons ( rather than photons just above the lyman limit ) . lcccc parameter & model a & model b & model c & model d + @xmath67 [ cm ] & 17.25 & 17.00 & 16.75 & 16.50 + @xmath68 [ cm ] & 15.5 & 15.0 & 15.0 & 15.5 + @xmath52 & 0.03 & 0.03 & 2/3 & 0.10 + @xmath69 [ @xmath70 & 4.7 & 5.5 & 5.0 & 5.5 + @xmath113 & -2.81 & -2.68 & -2.58 & -2.58 + @xmath50 & 0.50 & 0.29 & 0.50 & 0.29 + @xmath114 [ @xmath76 & 21 & 20 & 21.5 & 21.5 + @xmath106 [ 10@xmath115 m@xmath27 ] & 10.6 & 6.7 & 14.9 & 7.0 + @xmath58 [ 10@xmath116 cm ] & 6.2 & 3.5 & 5.6 & 2.0 + @xmath117 [ yr ] & 13.1 & 7.4 & 11.9 & 4.2 + [ oiii ] emission should evolve on a timescale @xmath117 . thus , our scenario can be tested by monitoring the rz 2109 spectrum . we arbitrarily select four viable models ( see table 2 ) for a study of the time evolution of line emission . this was done by assuming that the nova shell keeps expanding with @xmath118 ( cfr . z08 , s11 ) , and properly scaling the relevant parameters ( @xmath119 , @xmath47 and @xmath48 if xrs is outside the expanding shell ; @xmath51 , @xmath47 and @xmath48 otherwise ) . as the evolution of the thickness @xmath47 of the nova shell is unclear , we consider two different cases : ( a ) @xmath120constant ( appropriate for small velocity dispersions within the ejecta ) , and ( b ) @xmath121 ( appropriate for a significant ejecta velocity dispersion ) . we calculated the properties of each of these four models at five times : three past epochs ( april 2005 , december 2007 , and march 2009 , i.e. the times of the last z07 observation , of the z08 observation , and of the last s11 observation , respectively ) , and two future ones ( december 2012 and december 2017 , i.e. 5 and 10 years after the z08 observation ) . the behaviour of the results are summarized in fig . 2 , where we show the time evolution of the ionization parameter , of the ionization state of oxygen , and of the intensities of five lines ( or groups of lines ) . such lines are representative of the main ionization stages and of the most abundant heavy elements in the ejecta , but are _ not _ always the most intense : an exhaustive list of possibly detectable optical and uv lines can be found in the appendix . in 7 out of 8 models the ionization parameter increases with time , so that heavy elements ( such as oxygen ) move to higher ionization states . therefore , high - ionization lines ( such as the ovi lines around @xmath1221035 ) tend to grow stronger , whereas low - ionization lines ( e.g. [ nii]@xmath76548,6583 ) tend to weaken with time , and intermediate - ionization lines ( e.g. the civ doublet at @xmath71549,1551 ) grow to a peak , and then start to decline . the case ( a ) of model c is the main exception : since it assumes that ( i ) the xrs is within the expanding shell , and ( ii ) the shell thickness is constant , the density evolution in the shell ( @xmath123 ) compensates the geometrical dilution of the ionizing flux , resulting in a constant ionization parameter . as a consequence , the ionization state of oxygen ( and of all the other heavy elements ) remains practically constant , and line fluxes tend to decrease very slowly . the four models were required to agree with the december 2007 observations , so their spectra for this epoch are dominated by the [ oiii ] doublet , with @xmath94 within 20 per cent from the z08 value . in addition , [ neiii]@xmath124 and [ oiii]@xmath17 might be strong enough to be detectable : the latter was marginally detected by z08 ; as to [ neiii ] , we note that the z08 spectrum is very noisy at @xmath1254000 , and this line might be confused with other features in that region . all the models ( except model d - see next subsection ) are in quite satisfactory agreement with the 2009 observation , too : this is simply because the 1.3 years interval from december 2007 to march 2009 is a relatively small fraction of @xmath126 . instead we find larger ( and more interesting ) differences if we look at the april 2005 spectrum : models a and c predict that the [ oiii ] emission was as strong ( a ) or stronger ( c ) than in 2007 , and require that the discrepancy among the fluxes reported by z07 and z08 derives from observational factors only . instead , models b and d provide a different explanation : their quite short @xmath117 imply large differences in @xmath58 : in 2005 @xmath58 was only 2/3 ( model b ) or 1/3 ( model d ) of its value in 2007 ; such a difference not only reduces the amount of energy intercepted by the ejecta , but at the same time lowers the ionization parameter , favouring low - ionization states and moving the emission from [ oiii ] to [ oii ] ( or even [ oi ] ) lines . the predictions for 2012 and 2017 offer an opportunity to test these models , with model d predicting that all anomalous emission lines should disappear before 2012 , model b predicting large declines , and models a and c predicting a slower , steady decline . as we already mentioned , such decline is mostly understood in terms of an increasing ionization parameter , so that the o@xmath127 abundance drops . it should be noted that all the above predictions suffer from a number of approximations : 1 . the structure of a nova shell is not uniform . there are large density fluctuations both at early times ( @xmath128 years after the eruption ; see e.g. the images of nova cygni 1992 in chochol et al . 1997 ) and at late times ( @xmath129 years after the eruption ; see e.g. the images of t pyx in shara et al . 2 . the residual nuclear burning on the wd responsible for the nova eruption might produce a significant amount of uv / x - ray photons for a few years ( see henze et al . 2010 ) : furthermore , when @xmath130 , @xmath131 , and even moderate wd luminosities ( say @xmath132 ) are enough for the wd emission to dominate over the one from the xrs . therefore , the predictions for model b in 2005 , for model d in 2007 , and especially for model d in 2005 might suffer from the fact that our models do not consider the wd emission . the cloudy calculations assume steady state , but models with a long ( @xmath1331 yr ) recombination timescale would require a time - dependent approach . for h recombination , @xmath134 ( where @xmath135 is the h recombination coefficient ) , so that models with @xmath136 might suffer from this problem ; however , the recombination timescales of highly ionized metals are shorter ( by factors of 210 ) than @xmath137 , and predictions about their lines should be reliable for @xmath138 . our geometry is extremely simple , and might suffer from assuming that the whole nova shell is at the same distance from the xrs : this is reasonable if the xrs is within the shell ( model c ) , or if @xmath139 , but there are a number of cases ( models a and b at the 2012 and 2017 epochs ; model d at all epochs except 2005 ) where @xmath140 ( so that the ionization parameter in the xrs - approaching part of the shell might be @xmath141 times higher than in the xrs - receding part ) , and the xrs is engulfed by the shell ; we try to adapt to the new situation by switching to a model with the xrs at the centre of the shell ( @xmath142 , @xmath143 ) , but these two set of predictions should be considered extremely uncertain . ] . the timing of the evolution depends upon the exact value of @xmath144 : the adopted 1.5 value is reasonable ( s11 estimates @xmath145 for the broad component of the line profile - see the next subsection ) , but not unique : a lower(higher ) choice of @xmath146 would delay(fasten ) the predicted evolution ( for example , if @xmath147 , our predictions for december 2012 and december 2017 would apply to june 2015 and december 2022 , respectively ) . this is particularly relevant for model d : the [ oiii ] flux if fig . 2 and table a4 evolves too rapidly ( the 2007 flux agrees with the observations , but both the 2005 and the 2009 fluxes are too low ) ; however , the model should not be discarded ( yet ) because these discrepancies can still be reconciled using a value of @xmath144 ( @xmath148 ) that is still in agreement with the s11 measurements . some of these issues might be resolved by adopting more complicated modelling ( e.g. assume a spectrum / temporal evolution for the wd emission , use some density distribution , adopt a more realistic geometry , etc . ) ; however , a more detailed modelling , while providing more reliable predictions , would also require us to introduce a number of additional parameters . this is not only a problem in terms of resources ( such as computing power ) , but would further complicate the interpretation of the results . s11 model the spectral profile of the [ oiii]@xmath149 complex as the superposition of the identical profiles of the two lines of the doublet , each one made of a central , relatively narrow ( fwhm@xmath150 ) gaussian peak , plus a much broader ( fwhm@xmath151 ) ` flat ' component ( possibly the result of a bipolar outflow ) ; notably , the latter component is not really flat : the red side is about 1.4 times brighter than the blue side . as noted by s11 , this is consistent with a nova shell between the xrs and the observer ) isotropic component , plus a fast bipolar one ( @xmath152 ) . nova observations ( e.g. iijima & naito 2011 ) and theoretical models ( see e.g. the ejecta velocity plots in s10 ) support these hypotheses by finding / predicting complex profiles for nova emission lines . ] . if this is the case , the flux difference among the two sides suggests that in 2007 - 2008 @xmath58 was of the order of 10 per cent of the distance between the nova and the xrs : this would favor models with @xmath153 0.01 . furthermore , since @xmath58 increases with time , the asymmetry between the red and blue sides should increase as well ( at least as long as the xrs is outside the nova shell ) . if the nova lies between us and the xrs , the ejecta might even intercept the line of sight to the xrs , increasing the value of @xmath65 . we checked whether such an increase can explain the fading of the xrs seen in the 2004 _ xmm _ observation ( see m2007 ) , and in the subsequent _ swift _ and _ chandra _ observations ( see maccarone et al . the effective x - ray extinction due to isotropic nova ejecta is @xmath154 where @xmath155 is a geometrical factor accounting for the number of times that the photons cross the shell ( 1 if the xrs is within the shell ; 2 if the xrs is outside the shell ) , and for the inclination of the photon direction with respect to the normal to the shell ( note that @xmath155 can be quite larger than 2 if the line of sight to the xrs is almost tangential to the ejecta shell ) . maccarone et al ( 2010 ) suggest that the weakness of the xrs in rz 2109 during the _ chandra _ 2010 observation might be explained by a large increase in the effective @xmath65 in front of the xrs . we used the web version of w3pimms ( http://heasarc.nasa.gov/tools/w3pimms.html ) to estimate that @xmath156 is necessary to explain the _ chandra _ 2010 observations . this is marginally consistent with our scenario : if we set @xmath157 , @xmath84 , @xmath158 , we can obtain such a high @xmath65 for @xmath159 . in turn , this implies that the nova eruption occurred in 20002004 ( for @xmath160 ) , which might be consistent with the timing of the z07 observations ( 20042005 ) . however , the available parameter space is very small : a slight change in the choice of @xmath155 , @xmath50 and @xmath106 can reduce the upper limit on @xmath161 , moving the nova occurrence to an epoch _ after _ the first z07 observation . therefore , we deem quite unlikely that the weakening of the xrs is ( entirely ) due to the additional absorption from the nova shell ; however , if this is the case , the xrs should re - brighten within a few years . we estimate the probability of having a nova shell with properties suitable to explain the rz 2109 observations within a distance @xmath162 from the xrs as @xmath163 where @xmath164 and @xmath41 were introduced in sec . 2 , @xmath165 is the number of stars in a gc , @xmath166 is the average stellar mass in a gc , @xmath167 is the number density of stars at the centre of a gc ( @xmath168@xmath169 ; see e.g. pryor & meylan 1993 ) , @xmath170 is the fraction of novae whose ejecta can produce the observed line emission ( a highly uncertain quantity ; since relatively massive ejecta are needed , we use 0.01 as a reference ) , and @xmath171 v_{\rm ej,3}^{-1}$ ] acts as the visibility timescale for the [ oiii ] line emission . such a probability is of the same order of magnitude as that predicted in other scenarios such as the disruption of a wd by an imbh ( clausen & eracleous 2011 quote a disruption rate of @xmath172 from sigurdsson & rees 1997 ) , and it must be noted that the large uncertainty about several parameters in the above formula might lead to a substantial increase of @xmath173 . we note that , if the xrs is outside the nova shell , a further factor @xmath174 accounting for the possible beaming of the xrs radiation should be introduced into eq . ( 2 ) : therefore , an observational confirmation of this scenario might disfavour the models of the xrs ( e.g. king 2011 ) that assume a beamed emission ( appropriate for this geometry ) should be multiplied by a factor of @xmath175 : if @xmath176 , no viable model of this kind would survive . ] . we used the photoionization code cloudy to check whether the peculiar emission spectra from xrs - hosting gcs ( especially rz 2109 ) can be explained as the result of the photoionization of a nova shell by the radiation emitted by the xrs . we found that this is definitely possible , but requires that the nova ejecta are relatively massive ( @xmath177 ) , and that the xrs - nova distance is quite small ( @xmath5 ) . the combination of these two requirements is quite unlikely ; on the other hand , it can explain observations in a straightforward manner ( although the line profile needs further investigation ) . while not entirely satisfactory , this compares well with the two other proposed scenarios for rz 2109 : ( i ) the disruption of a wd by an imbh ( an event as unlikely as the one we discuss ) explains the line profile and the absence of h lines ( i10 , clausen & eracleous 2011 ) , but appears to be unable to produce the high [ oiii ] luminosity ( porter 2010 , s11 ) ; ( ii ) the ionization of material ejected from the xrs itself , while not suffering from the `` unlikely coincidence '' problem , does not explain either the absence of h emission lines , nor some properties of the line profile ( asymmetry in the flat component , origin of the gaussian component ) . finally , we predict the future evolution of the line emission in the xrs+nova scenario , by looking at both optical and uv lines . we find that in most cases the nova shell tends to move towards a higher ionization state , so that the line flux should move from moderate - ionization lines such as [ oiii]@xmath8 to high - ionization lines ( e.g. the ovi lines in the 10321038wavelength range ) . such predictions might provide a crucial test for discriminating among the various hypotheses described above . we thank the anonymous referee for a careful reading of the manuscript , and for several very valuable suggestions ( e.g. about uv lines ) , and thank a. wolter , l. zampieri , a. bressan , d. fiacconi , s. hachinger , r. porter , and t. maccarone for useful discussions and comments . this research has made use of nasa s astrophysics data system bibliographic services , and of tools provided by the high energy astrophysics science archive research center ( heasarc ; which is provided by nasa s goddard space flight center ) . allende prieto c. , lambert d.l . , asplund m. , 2001 , apj , 556 , l63 allende prieto c. , lambert d.l . , asplund m. , 2002 , apj , 573 , l137 brassington n.j . , et al . , 2010 , apj , 725 , 1805 chochol d. , grygar j. , pribulla t. , komzik r. , hric l. , elkin v. , 1997 , a&a , 318 , 908 clausen d. , eracleous m. , 2011 , apj , 726 , 34 della valle m. , rosino l. , bianchini a. , livio m. , 1994 , a&a , 287 , 403 gehrz r.d . , truran j.w . , williams r.e . , starrfield s. , 1998 , pasp , 110 , 3 ferland g.j . , shields g.a . , 1978 , apj , 226 , 172 ferland g.j . , korista k.t . , verner d.a . , ferguson j.w . , kingdon j.b . , verner e.m . , 1998 , pasp , 110 , 761 grevesse n. , sauval a.j . , 1998 , space science review , 85 , 161 heinke c.o . , 2011 , to appear in the proceedings of the conference `` binary star evolution : mass loss , accretion and mergers '' ( eds . v. kalogera and m. van der sluys , aip conf . ser . ) , mykonos , greece , june 22 - 25 , 2010 , ( arxiv.org/1101.5356 ) holweger h. , 2001 , joint soho / ace workshop `` solar and galactic composition '' ( bern , switzerland , march 69 2001 ) , edited by r.f . wimmer - schweingruber , aipc proceedings , vol . 598 , p. 23 henze m. , et al . , 2009 , a&a , 500 , 769 henze m. , et al . , 2010 , a&a , 523 , 89 iijima t. , naito h. , 2011 , a&a , 526 , a73 irwin j.a . , brink t.g . , bregman j.n . , roberts t.p . , 2010 , apj , 712 , l1 [ i10 ] kalogera v. , henninger m. , ivanova n. , king a.r . , 2004 , apj , 603 , l41 king a. , 2011 , apj , 732 , l28 maccarone t.j . , kundu a. , zepf s.e . , rhode k.l . , 2007 , nature , 445 , 183 [ m07 ] maccarone t.j . , kundu a. , zepf s.e . , rhode k.l . , 2010 , mnras , 409 , l84 maccarone t.j , kundu a. , zepf s.e . , rhode k.l . , 2011 , mnras , 410 , 1655 maccarone t.j . , warner b. , 2011 , mnras , 410 , l32 mannucci f. , della valle m. , panagia n. , cappellaro e. , cresci g. , maiolino r. , petrosian a. , turatto m. , 2005 , a&a , 433 , 807 moore k. , bildsten l. , 2011 , apj , 728 , 81 peacock m.b . , maccarone t.j . , knigge c. , kundu a. , waters c.z . , zepf s.e . , zurek d.r . , mnras , 402 , 803 pietsch w. , et al . , 2007 , a&a , 465 , 375 porter r.l . , 2010 , mnras , 407 , l59 prialnik d. , 1986 , apj , 310 , 222 pryor c. , meylan g. , 1993 , in asp con ference series 50 , structure and dynamics of globular clusters , eds djorgovski s.g . & meylan g. ( san francisco , ca : asp ) , 357 shafter a.w , quimby r.m . , 2007 , apj , 671 , l121 shara m.m . , zurek d.r . , williams r.e . , prialnik d. , gilmozzi r. , moffat a.f.j . , 1997 , aj , 114 , 258 shara m.m . , zurek d.r . , baltz e.a . , lauer t.r . , silk j. , 2004 , apj , 605 , l117 shara m.m . , yaron o. , prialnik d. , kovetz a. , zurek d. , 2010 , apj , 725 , 831 [ s10 ] shih i.c . , kundu a. , maccarone t.j . , zepf s.e . , joseph t.d . , 2010 , apj , 721 , 323 sigurdsson s. , phinney e.s . , 1995 , apjs , 99 , 609 sigurdsson s. , rees m.j . , 1997 , mnras , 284 , 318 starrfield s. , truran j.w . , sparks w.m . , kutter g.s . , 1972 , apj , 176 , 169 steele m.m . , zepf s.e . , kundu a. , maccarone t.j . , rhode k.l . , salzer j.j . , 2011 , apj , in press ( arxiv:1107.4051 ) [ s11 ] yaron o. , prialnik d. , shara m.m . , kovetz a. , 2005 , apj , 623 , 398 [ y05 ] zepf s.e . , maccarone t.j . , bergond g. , kundu a. , rhode k.l . , salzer j.j . , 2007 , apj , 669 , l69 [ z07 ] zepf s.e . , et al . , 2008 , apj , 683 , l139 [ z08 ] in this appendix , we list the detailed predictions about line fluxes at earth , and about the properties of the nova ejecta ( e.g. the ionization state of c , n , o , and ne ) from the four models described in the text , evolved to five past or future epochs : april 2005 , december 2007 , march 2009 ( chosen to coincide with the times of past optical observations ) , december 2012 , and december 2017 . we report all the emission lines in the uv - optical range ( 913@xmath17810000 ) reaching a flux @xmath179 in at least one of the five considered epoch , plus lyman@xmath180 and h@xmath180 . for well - separated multiplets we list only the strongest line ( for example , in the case of the [ oiii]@xmath74959,5007 doublet we only list [ oiii]@xmath1225007 ) , whereas close multiplets are listed as a single line ( for example the [ oii]@xmath73726,3729 doublet is listed as [ oii]@xmath1223727 ) . all the fluxes are in units of @xmath181 , and were obtained assuming a distance @xmath182 . no reddening correction was applied . all the data columns - except the one referring to the z08 observation , which is taken as a reference point - report two set of data : the first was obtained assuming that the shell thickness @xmath183 remains constant at the value used for december 2007 ( implying that the density @xmath48 scales as @xmath184;assumption a ) , whereas the second was obtained assuming that @xmath47 increases linearly with time ( i.e. that @xmath185 remains constant at the value estimated for december 2007 ; in such case the density @xmath48 scales as @xmath186 ; assumption b ) . the two cases are generally separated by a `` / '' , but in the case of ionization states it was necessary to use two separate lines . lccccc line & april 2005 & december 2007 & march 2009 & december 2012 & december 2017 + + niii@xmath187 & 0.020/0.014 & 0.043 & 0.063/0.074 & 0.129/0.094 & 0.077/0.026 + ovi@xmath188@xmath189 & 0/0 & 0.005 & 0.020/0.035 & 0.232/0.446 & 0.654/0.900 + lyman @xmath180 @xmath190 & 0.168/0.182 & 0.103 & 0.081/0.072 & 0.045/0.033 & 0.029/0.018 + ov]@xmath191 & 0/0 & 0.004 & 0.011/0.016 & 0.076/0.131 & 0.170/0.178 + nv@xmath192 & 0.005/0.002 & 0.055 & 0.131/0.189 & 0.667/0.878 & 0.977/0.704 + oiv]@xmath193 & 0.004/0.002 & 0.038 & 0.090/0.130 & 0.460/0.548 & 0.516/0.279 + niv]@xmath194 & 0.012/0.006 & 0.081 & 0.172/0.235 & 0.714/0.770 & 0.647/0.267 + civ@xmath195 & 0.068/0.004 & 0.031 & 0.057/0.071 & 0.162/0.151 & 0.119/0.043 + oiii]@xmath196 & 0.030/0.020 & 0.092 & 0.135/0.147 & 0.168/0.088 & 0.073/0.025 + niii@xmath197 & 0.067/0.046 & 0.196 & 0.286/0.314 & 0.383/0.203 & 0.146/0.037 + ciii]@xmath198 & 0.026/0.022 & 0.059 & 0.078/0.082 & 0.092/0.049 & 0.033/0.007 + @xmath199neiv]@xmath200 & 0.007/0.003 & 0.041 & 0.085/0.114 & 0.326/0.348 & 0.310/0.156 + @xmath199oii]@xmath201 & 0.050/0.100 & 0.014 & 0.011/0.008 & 0.004/0.002 & 0.002/0.001 + @xmath199nev]@xmath202 & 0.003/0.002 & 0.012 & 0.023/0.031 & 0.103/0.153 & 0.178/0.171 + @xmath199oii]@xmath203 & 0.055/0.098 & 0.012 & 0.010/0.008 & 0.005/0.002 & 0.003/0.001 + @xmath199neiii]@xmath204 & 0.166/0.148 & 0.232 & 0.242/0.229 & 0.139/0.058 & 0.041/0.010 + @xmath199oiii]@xmath205 & 0.030/0.023 & 0.058 & 0.071/0.071 & 0.056/0.025 & 0.019/0.006 + @xmath199oiii]@xmath206 & 2.77/2.24 & 3.6 & 3.43/3.14 & 1.48/0.555 & 0.385/0.097 + @xmath199nii]@xmath207 & 0.023/0.053 & 0.005 & 0.004/0.003 & 0.001/0.001 & 0.001/0 + h@xmath180 @xmath208 & 0.014/0.015 & 0.008 & 0.006/0.004 & 0.003/0.002 & 0.002/0.001 + @xmath199nii]@xmath209 & 0.433/0.893 & 0.055 & 0.037/0.029 & 0.012/0.006 & 0.006/0.001 + @xmath199oii]@xmath210 & 0.068/0.140 & 0.019 & 0.015/0.012 & 0.006/0.003 & 0.003/0.001 + + @xmath69 [ @xmath70 & 4.87/4.96 & 4.70 & 4.63/4.59 & 4.45/4.33 & 4.26/4.04 + @xmath211}$ ] & 10.4 & 13.1 & 14.4 & 18.1 & 23.1 + @xmath212}$ ] & 4.9 & 6.2 & 6.8 & 8.6 & 10.9 + @xmath52 & 0.020 & 0.030 & 0.0354 & 0.533 & 0.0831 + @xmath113 & -2.98/-3.07 & -2.81 & -2.74/-2.70 & -2.56/-2.44 & -2.37/-2.15 + c(i@xmath213vii ) & -,11,69,6,13,1,- & -,3,55,11,27,4,0 & -,2,45,13,33,7,0 & -,0,17,12,45,23,2 & -,0,5,6,40,40,9 + & 0,31,56,5,8,0,- & & -,1,40,13,36,9,1 & -,0,8,9,44,33,5 & -,0,1,2,26,50,21 + n(i@xmath213viii ) & 0,25,63,7,4,1,-,- & -,2,70,15,9,4,0,- & -,1,61,20,11,7,1,- & -,0,24,30,15,25,5,0 & -,0,7,21,14,40,16,1 + & 3,48,42,4,2,0,-,- & & -,1,56,22,12,9,1,- & -,0,11,28,15,35,10,1 & -,0,2,9,9,43,32,5 + o(i@xmath213ix ) & 0,31,58,9,2,0,0,-,- & -,2,69,23,4,2,0,-,- & -,1,58,31,6,3,1,-,- & -,0,19,53,13,10,5,0,- & -,0,6,41,20,15,15,2,0 + & 3,52,39,6,1,0,-,-,- & & -,1,53,35,7,4,1,0,- & -,0,8,50,18,13,10,1,- & -,-,2,22,20,17,30,8,0 + ne(i@xmath213xi ) & 0,25,69,5,1,0,-,-,-,-,- & -,1,79,15,3,0,-,-,-,-,- & -,1,70,23,5,1,-,-,-,-,- & -,0,29,49,19,3,0,-,-,-,- & -,0,10,44,36,9,1,0,0,-,- + & 2,39,55,3,1,-,-,-,-,-,- & & -,1,65,27,7,1,0,-,-,-,- & -,0,13,50,30,6,1,0,-,-,- & -,-,3,26,47,18,4,1,1,0,- + lccccc line & april 2005 & december 2007 & march 2009 & december 2012 & december 2017 + + niii@xmath187 & 0.010/0.002 & 0.132 & 0.216/0.220 & 0.115/0.042 & 0.038/0.005 + ovi@xmath188@xmath189 & 0/0 & 0.032 & 0.167/0.285 & 1.06/1.39 & 1.47/0.961 + lyman @xmath180 @xmath190 & 0.523/0.872 & 0.342 & 0.245/0.208 & 0.136/0.094 & 0.095/0.057 + ov]@xmath191 & 0/0 & 0.034 & 0.118/0.173 & 0.371/0.287 & 0.234/0.073 + nv@xmath192 & 0.009/0.002 & 0.252 & 0.744/1.01 & 1.76/1.30 & 1.09/0.301 + oiv]@xmath193 & 0.007/0.001 & 0.181 & 0.495/0.637 & 0.607/0.268 & 0.200/0.030 + niv]@xmath194 & 0.020/0.002 & 0.341 & 0.850/1.07 & 0.982/0.420 & 0.303/0.040 + civ@xmath195 & 0.016/0.002 & 0.127 & 0.228/0.250 & 0.175/0.068 & 0.045/0.005 + oiii]@xmath196 & 0.067/0.003 & 0.306 & 0.347/0.285 & 0.077/0.024 & 0.022/0.002 + niii@xmath197 & 0.140/0.006 & 0.655 & 0.809/0.710 & 0.221/0.065 & 0.056/0.004 + ciii]@xmath198 & 0.074/0.053 & 0.144 & 0.151/0.128 & 0.035/0.009 & 0.006/0 + nii]@xmath214 & 0.205/0.316 & 0.014 & 0.009/0.006 & 0.003/0.001 & 0/0 + cii]@xmath215 & 0.067/0.114 & 0.007 & 0.004/0.003 & 0.001/0 & 0/0 + @xmath199neiv]@xmath200 & 0.005/0 & 0.087 & 0.180/0.206 & 0.132/0.046 & 0.028/0.002 + @xmath199oii]@xmath201 & 0.546/0.798 & 0.024 & 0.011/0.006 & 0.002/0 & 0/0 + @xmath199nev]@xmath202 & 0.004/0 & 0.083 & 0.204/0.256 & 0.303/0.168 & 0.110/0.011 + @xmath199oii]@xmath203 & 0.070/0.076 & 0.004 & 0.002/0.001 & 0.001/0 & 0/0 + @xmath199neiii]@xmath204 & 0.316/0.305 & 0.358 & 0.231/0.156 & 0.017/0.003 & 0.002/0 + @xmath199oiii]@xmath205 & 0.071/0.003 & 0.206 & 0.176/0.129 & 0.024/0.006 & 0.005/0 + @xmath199oiii]@xmath206 & 1.37/0.050 & 4.84 & 3.36/2.33 & 0.368/0.095 & 0.080/0.005 + @xmath199nii]@xmath207 & 0.333/0.500 & 0.013 & 0.006/0.004 & 0.001/0 & 0/0 + @xmath199oi]@xmath216 & 0.336/0.581 & 0 & 0/0 & 0/0 & 0/0 + h@xmath180 @xmath208 & 0.027/0.030 & 0.025 & 0.017/0.014 & 0.008/0.005 & 0.004/0.002 + @xmath199nii]@xmath209 & 0.748/0.812 & 0.027 & 0.013/0.008 & 0.004/0.001 & 0.001/0 + @xmath199oii]@xmath210 & 0.726/1.06 & 0.033 & 0.015/0.009 & 0.002/0 & 0/0 + + @xmath69 [ @xmath70 & 5.83/6.00 & 5.50 & 5.38/5.32 & 5.10/4.90 & 4.82/4.48 + @xmath211}$ ] & 4.7 & 7.4 & 8.7 & 12.4 & 17.4 + @xmath212}$ ] & 2.2 & 3.5 & 4.1 & 5.9 & 8.2 + @xmath52 & 0.0139 & 0.030 & 0.0399 & 0.0759 & 0.1429 + @xmath113 & -3.01/-3.18 & -2.68 & -2.56/-2.50 & -2.28/-2.08 & -2.00/-1.66 + c(i@xmath213vii ) & 1,60,33,2,3,0,- & -,2,41,14,36,6,0 & 0,1,24,14,47,14,1 & -,0,4,7,44,39,6 & -,0,1,2,26,52,19 + & 2,74,24,0,0,-,- & & -,0,18,13,49,19,1 & -,0,1,3,32,50,14 & -,-,0,0,9,46,46 + n(i@xmath213viii ) & 20,63,14,2,1,0,-,- & -,1,65,19,9,5,0,- & -,1,43,29,14,13,1,- & -,0,7,23,18,40,11,1 & -,0,2,8,11,46,29,3 + & 33,66,1,0,-,-,-,- & & -,0,32,31,15,18,2,0 & -,0,3,11,14,48,23,2 & -,-,0,1,3,30,50,15 + o(i@xmath213ix ) & 32,52,13,2,0,-,-,-,- & -,1,62,28,6,2,1,-,- & -,0,36,44,11,6,2,0,- & -,0,4,34,25,18,17,1,- & -,-,1,12,18,22,40,8,0 + & 43,56,0,-,-,-,-,-,- & & -,0,25,49,14,8,4,0,- & -,-,1,16,22,23,33,5,0 & -,-,0,2,6,11,52,26,2 + ne(i@xmath213xi ) & 5,39,52,3,0,-,-,-,-,-,- & -,1,60,28,10,2,0,-,-,-,- & -,0,33,38,23,6,1,0,-,-,- & -,-,3,20,42,27,6,2,1,0,- & -,-,0,5,24,36,19,10,6,0,- + & 8,47,45,0,-,-,-,-,-,-,- & & -,0,22,39,29,9,1,0,-,-,- & -,-,1,8,32,36,14,6,3,0,- & -,-,-,0,4,15,20,24,35,1,0 + lccccc line & april 2005 & december 2007 & march 2009 & december 2012 & december 2017 + + ciii@xmath217 & 0.349/0.308 & 0.284 & 0.247/0.223 & 0.175/0.125 & 0.116/0.069 + niii@xmath187 & 1.140/1.130 & 0.878 & 0.751/0.656 & 0.529/0.348 & 0.359/0.180 + ovi@xmath188@xmath189 & 2.360/1.180 & 2.090 & 1.920/2.270 & 1.380/2.090 & 0.880/1.620 + lyman @xmath180 @xmath190 & 0.332/0.448 & 0.212 & 0.178/0.161 & 0.121/0.088 & 0.078/0.044 + ov]@xmath191 & 0.890/0.505 & 0.717 & 0.631/0.722 & 0.413/0.550 & 0.240/0.331 + nv@xmath192 & 4.930/3.100 & 4.240 & 3.810/4.160 & 2.670/3.070 & 1.690/1.790 + oiv]@xmath193 & 3.850/2.590 & 3.160 & 2.730/2.740 & 1.760/1.570 & 1.030/0.736 + niv]@xmath194 & 6.590/4.400 & 5.400 & 4.640/4.640 & 2.960/2.510 & 1.690/1.060 + civ@xmath195 & 1.940/1.300 & 1.570 & 1.340/1.340 & 0.842/0.694 & 0.467/0.282 + oiii]@xmath196 & 1.150/1.580 & 0.733 & 0.605/0.473 & 0.439/0.230 & 0.321/0.115 + niii@xmath197 & 3.210/4.040 & 2.200 & 1.820/1.450 & 1.280/0.659 & 0.889/0.295 + ciii]@xmath198 & 1.230/1.470 & 0.867 & 0.722/0.581 & 0.497/0.261 & 0.329/0.114 + @xmath199oiii]@xmath218 & 0.113/0.184 & 0.059 & 0.047/0.034 & 0.032/0.015 & 0.022/0.006 + @xmath199neiv]@xmath200 & 1.520/1.000 & 1.460 & 1.340/1.340 & 0.982/0.855 & 0.637/0.447 + @xmath199nev]@xmath202 & 0.865/0.546 & 0.713 & 0.631/0.705 & 0.423/0.536 & 0.248/0.322 + @xmath199neiii]@xmath204 & 0.964/1.620 & 0.528 & 0.425/0.315 & 0.301/0.130 & 0.218/0.055 + @xmath199oiii]@xmath205 & 0.485/0.789 & 0.254 & 0.200/0.149 & 0.136/0.063 & 0.095/0.028 + @xmath199oiii]@xmath206 & 9.350/16.90 & 4.630 & 3.720/2.680 & 2.730/1.150 & 2.050/0.498 + h@xmath180 @xmath208 & 0.023/0.033 & 0.014 & 0.012/0.010 & 0.008/0.006 & 0.005/0.003 + @xmath199nii]@xmath209 & 0.032/0.054 & 0.036 & 0.038/0.032 & 0.048/0.026 & 0.056/0.017 + + @xmath69 [ @xmath70 & 5.19/5.29 & 5.00 & 4.92/4.88 & 4.73/4.60 & 4.52/4.28 + @xmath211}$ ] & 9.2 & 11.9 & 13.2 & 16.9 & 21.9 + @xmath212}$ ] & 4.3 & 5.6 & 6.2 & 8.0 & 10.3 + @xmath52 & 2/3 & 2/3 & 2/3 & 2/3 & 2/3 + @xmath113 & -2.58/-2.68 & -2.58 & -2.57/-2.53 & -2.58/-2.45 & -2.57/-2.33 + c(i@xmath213vii ) & -,1,34,17,37,11,1 & -,1,26,17,41,14,1 & -,1,24,17,42,15,2 & -,2,25,15,42,15,2 & -,2,26,14,41,15,2 + & -,1,50,14,28,6,0 & & -,1,20,16,44,18,2 & -,1,13,12,45,25,4 & -,1,10,8,41,33,8 + n(i@xmath213viii ) & -,0,30,34,12,20,4,0 & -,0,21,35,13,25,5,0 & -,0,20,34,14,26,6,0 & -,1,20,32,14,27,6,1 & -,1,23,29,14,26,7,1 + & -,0,49,27,10,12,2,0 & & -,0,15,33,14,29,8,1 & -,0,10,26,14,36,13,1 & -,0,8,17,12,39,21,3 + o(i@xmath213ix ) & -,0,21,51,12,9,6,1,- & -,0,12,53,15,12,8,1,- & -,0,11,51,15,12,8,1,- & -,0,12,48,16,13,9,1,- & -,1,14,45,16,13,10,2,0 + & -,0,39,42,9,6,3,0,- & & -,0,8,49,17,13,10,2,0 & -,0,6,39,19,15,17,4,0 & -,0,5,28,18,16,24,8,1 + ne(i@xmath213xi ) & -,0,29,48,19,4,1,0,-,-,- & -,0,20,52,23,5,1,0,-,-,- & -,0,18,52,24,5,1,0,0,-,- & -,0,20,49,24,5,1,0,0,-,- & 0,0,23,47,23,5,1,0,0,-,- + & -,0,49,37,12,2,0,0,-,-,- & & -,0,14,51,28,6,1,0,0,-,- & 0,0,10,44,35,9,2,1,0,-,- & -,0,8,36,40,12,3,1,1,0,- + lccccc line & april 2005 & december 2007 & march 2009 & december 2012@xmath219 & december 2017@xmath219 + + ciii@xmath217 & 0/0 & 0.061 & 0.063/0.039 & 0.003/0 & 0.002/0 + niii@xmath187 & 0/0 & 0.213 & 0.208/0.138 & 0.009/0 & 0.005/0 + ovi@xmath188@xmath189 & 0/0 & 0.371 & 1.41/1.83 & 0.714/0.093 & 0.309/0 + lyman @xmath180 @xmath190 & 0.391/0.411 & 0.403 & 0.191/0.145 & 0.040/0.015 & 0.023/0 + ov]@xmath191 & 0/0 & 0.149 & 0.492/0.599 & 0.050/0.003 & 0.021/0 + nv@xmath192 & 0/0 & 0.746 & 2.01/2.27 & 0.155/0.013 & 0.050/0 + oiv]@xmath193 & 0/0 & 0.493 & 1.11/1.09 & 0.028/0.001 & 0.014/0 + niv]@xmath194 & 0/0 & 0.784 & 1.72/1.65 & 0.018/0.001 & 0.009/0 + civ@xmath195 & 0/0 & 0.216 & 0.456/0.391 & 0.006/0.001 & 0.003/0 + @xmath199neiv]@xmath220 & 0/0 & 0.086 & 0.161/0.142 & 0.006/0.001 & 0.003/0 + oiii]@xmath196 & 0/0 & 0.296 & 0.198/0.088 & 0.003/0 & 0.002/0 + niii@xmath197 & 0.001/0 & 0.697 & 0.496/0.251 & 0.004/0 & 0.003/0 + ciii]@xmath198 & 0.008/0.004 & 0.242 & 0.173/0.086 & 0.001/0 & 0.001/0 + nii]@xmath214 & 0.118/0.092 & 0.005 & 0.002/0.001 & 0/0 & 0/0 + cii]@xmath215 & 0.062/0.063 & 0.005 & 0.002/0.001 & 0/0 & 0/0 + @xmath199neiv]@xmath200 & 0/0 & 0.137 & 0.341/0.348 & 0.023/0.003 & 0.013/0 + @xmath199oii]@xmath201 & 0.283/0.200 & 0.007 & 0.002/0.001 & 0/0 & 0/0 + mgii@xmath221 & 0.065/0.031 & 0.004 & 0.001/0 & 0/0 & 0/0 + @xmath199nev]@xmath202 & 0/0 & 0.127 & 0.394/0.473 & 0.059/0.005 & 0.026/0 + @xmath199nei]@xmath222 & 0.056/0.084 & 0 & 0/0 & 0/0 & 0/0 + @xmath199neiii]@xmath204 & 0.079/0.062 & 0.355 & 0.151/0.051 & 0.001/0 & 0.001/0 + @xmath199oiii]@xmath205 & 0/0 & 0.173 & 0.088/0.031 & 0.001/0 & 0/0 + @xmath199oiii]@xmath206 & 0.007/0.002 & 4.30 & 1.73/0.480 & 0.005/0 & 0.003/0 + @xmath199nii]@xmath207 & 0.185/0.168 & 0.005 & 0.001/0.001 & 0/0 & 0/0 + @xmath199oi]@xmath216 & 0.348/.454 & 0 & 0/0 & 0/0 & 0/0 + h@xmath180 @xmath208 & 0.022/0.030 & 0.031 & 0.014/0.010 & 0.002/0.001 & 0.001/0 + @xmath199nii]@xmath209 & 0.244/0.159 & 0.013 & 0.003/0.001 & 0/0 & 0/0 + @xmath199oii]@xmath210 & 0.375/0.264 & 0.010 & 0.003/0.001 & 0/0 & 0/0 + + @xmath69 [ @xmath70 & 6.20/6.55 & 5.50 & 5.30/5.20 & 4.46/4.15 & 4.11/3.63 + @xmath211}$ ] & 1.5 & 4.2 & 5.5 & 9.2 & 14.2 + @xmath212}$ ] & 0.7 & 2.0 & 2.6 & 4.3@xmath219 & 6.7@xmath219 + @xmath52 & 0.02 & 0.10 & 0.160 & 2/3@xmath219 & 2/3@xmath219 + @xmath113 & -3.28/-3.63 & -2.58 & -2.38/-2.28 & -1.76/-1.44 & -1.76/-1.28 + c(i@xmath213vii ) & 41,57,2,-,-,-,- & -,2,65,8,20,5,0 & -,0,20,12,40,22,4 & -,-,0,0,4,30,66 & -,-,0,0,5,31,63 + & 59,40,1,-,-,-,- & & -,0,7,10,42,33,8 & -,-,-,-,1,15,85 & -,-,-,-,-,3,97 + n(i@xmath213viii ) & 74,26,0,-,-,-,-,- & -,1,68,15,7,8,2,0 & -,0,21,27,12,28,10,1 & -,-,0,0,0,9,42,48 & -,-,0,0,0,9,41,48 + & 80,20,-,-,-,-,-,- & & -,-,7,25,12,36,17,2 & -,-,-,-,-,2,23,75 & -,-,-,-,-,0,5,95 + o(i@xmath213ix ) & 70,30,0,-,-,-,-,-,- & -,1,64,23,6,4,3,0,- & -,0,18,40,16,11,12,3,0 & -,-,0,1,1,2,21,49,26 & -,-,0,1,1,2,20,49,26 + & 76,24,-,-,-,-,-,-,- & & -,-,4,39,19,13,18,6,0 & -,-,-,-,0,0,6,38,56 & -,-,-,-,-,-,0,10,90 + ne(i@xmath213xi ) & 61,30,10,-,-,-,-,-,-,-,- & -,1,67,21,10,2,0,0,-,-,- & -,0,20,40,31,8,1,1,0,-,- & -,-,0,3,15,21,14,11,27,9,1 & -,-,1,3,15,20,13,11,28,9,1 + & 71,23,6,-,-,-,-,-,-,-,- & & -,-,6,38,40,12,3,1,1,0,- & -,-,0,0,2,5,5,6,39,35,6 & -,-,-,-,-,-,-,0,8,46,46 +

a connection relating tamari lattices on symmetric groups regarded as lattices under the weak bruhat order to the positive monoid of thompson group is presented . tamari congruence classes correspond to classes of equivalent elements in . the two well known normal forms in correspond to endpoints of intervals in the weak bruhat order that determine the tamari classes . in the monoid these correspond to lexicographically largest and lexicographically smallest form , while on the level of permutations they correspond to-avoiding and-avoiding permutations . forests appear naturally in both contexts as they are used to model both permutations and elements of the thompson monoid . the connection is then extended to tamari orders on partitions of-gons into-gons and thompson monoids , .

math0604015

the purpose of this note is to present a connection between the positive thompson monoid ( of type @xmath1 ) and tamari lattices ( of type @xmath8 ) . the fact that thompson groups and monoids are related to trees is certainly well known and established among the people interested in these groups . ever since @xcite trees are heavily used as helpful tools in representing elements in order to aid both calculations and conceptual understanding . on the other hand , researchers in combinatorics have usually , with rare exceptions , heard very little about thompson groups , and even when they have it is mostly in the context of providing examples of infinite simple groups . we start with some very well known and understood classes of objects in combinatorics that are related to tamari lattices on symmetric groups and then naturally arrive at thompson monoids , which essentially capture all instances of these combinatorial objects along with their inter - relations . the connection in question relates tamari lattices on finite symmetric groups ( tamari lattices of type @xmath8 ) to the positive thompson monoid @xmath9 the connection is obtained in a natural way as follows . first some well known connections between permutations , inversion sequences and linearized labeled binary rooted trees are recalled . the simple observation that concatenation is closed in the set @xmath10 of inversion sequences leads to a definition of a graded product on the set of all finite permutations @xmath11 . the corresponding product on the set of linearized labeled binary rooted trees @xmath12 is just stacking of trees . at this stage we have three isomorphic monoids @xmath10 , @xmath11 and @xmath12 . tamari congruence on @xmath12 is the congruence obtained when trees that have the same shape but different linearization are identified . this leads to a corresponding congruence on the level of permutations and also on the level of inversion sequences and we get three monoids @xmath13 , @xmath14 and @xmath15 . it turns out that these three monoids are free . we then extend our considerations to the set of all sequences of non - negative integers @xmath16 , all linearized labeled binary rooted forests @xmath17 and the corresponding set of @xmath18-permutations @xmath19 . we extend the notion of tamari congruence and identify two forests of the same shape regardless of the linearization . the corresponding factor monoids @xmath20 are isomorphic to the thompson monoid @xmath21 . after going through the details of the connection between tamari lattices on symmetric groups and thompson monoid @xmath21 in section [ delinearization ] , a similar connection is established between thompson monoids @xmath6 , @xmath22 , and tamari orders ( they do not form lattices for @xmath23 ) corresponding to partitions of @xmath4-gons into @xmath5-gons in section [ partitions ] . tamari lattices of type @xmath8 are homomorphic images of the weak bruhat order lattices over finite coxeter groups of type @xmath8 , i.e. , finite symmetric groups . recall that the ( left ) weak bruhat lattice on @xmath24 as coxeter group of type @xmath25 is just the ( left ) cayley graph of @xmath24 with respect to the standard generating set of reflections @xmath26 ordered by declaring that @xmath27 if there exists a geodesic path from @xmath28 to @xmath29 that passes through @xmath30 . alternatively , we may say that @xmath30 is covered by @xmath29 if @xmath31 , for some standard reflection @xmath32 , and the length of @xmath29 ( in terms of the standard reflections ) is larger than the length of @xmath30 . then define the weak bruhat order as closure of this cover relation . for a fixed @xmath33 , there are many ways of thinking of the congruence classes on @xmath24 defining the corresponding tamari lattice @xmath34 . we recall some of them here , along with some additional notions . we consider linearized labeled binary rooted trees on @xmath33 interior vertices . when @xmath35 there is only one such tree and it has a single vertex which is simultaneously the root and a leaf labeled by 0 . if @xmath36 the root of such tree has degree 2 and the other @xmath37 interior vertices have degree 3 . the @xmath38 leafs are labeled bijectively by @xmath39 . in addition , the interior vertices are labeled bijectively by @xmath40 in such a way that the labels on the paths from the root to the leafs are decreasing ( this is the linearization part of the tree - we can use it to extend the partial order on the interior vertices induced by the tree structure to a linear order ) . we depict such trees as in figure [ tree - lin ] . a rooted binary tree often admits more than one linearization and some standard choices are well established . we mention here two such choices ( which are relevant to our discussion ) . the post - order linearization labels the interior vertices by @xmath41 exactly in the order they are visited by using the left - right - root rule . the inverse post - order ( or the right - left post order ) labels the interior vertices in the order they are visited by using the right - left - root rule . the in - order labeling ( using the left - root - right ) does not necessarily produce a proper linearization . in the rest of the text we often say linearized tree and tree when we mean linearized labeled binary rooted tree and labeled binary rooted tree ( the latter lack linearization labels , i.e. they lack labels on the interior vertices ) . we recall the interpretation of tamari lattice @xmath34 as given by huang and tamari in @xcite . it is defined by the product order on the set of integer sequences @xmath42 such that @xmath43 , for all @xmath44 , and the condition that whenever @xmath45 and @xmath46 then @xmath47 . while this is not directly apparent in @xcite , one can easily interpret these sequences as encodings of labeled binary rooted trees as follows . let @xmath48 be a tree with @xmath33 interior vertices . for @xmath49 , let @xmath50 be the largest leaf label of the subtree of @xmath48 hanging below the interior vertex @xmath44 in the in - order labeling . for example , the encoding of the tree in figure [ tree - lin ] is @xmath51 . the tamari lattice @xmath34 is defined by bjrner and wachs in @xcite as the product order on the set of integer sequences @xmath52 satisfying @xmath53 , @xmath49 , and @xmath54 , for @xmath55 and @xmath56 . a correspondence is established between permutations and labeled trees ( the labeling on the interior vertices does not respect the partial order imposed by the tree , but it is related to it in a different way ) . all permutations are encoded by integer sequences of the above type as follows . given a permutation @xmath30 , for each @xmath44 , count the number of consecutive terms in @xmath57 following @xmath58 that are smaller than @xmath58 . for example , for @xmath59 we have @xmath60 and the encoding sequence is @xmath61 . in the corresponding tree this sequence records , for each @xmath44 , the number of interior vertices in the right subtree below the vertex visited at position @xmath44 using the in - order . for example , the tree in figure [ tree - lin ] ( ignore the labels on interior vertices ) is encoded by the sequence @xmath61 . this same tree encodes the permutation @xmath62 . the tamari classes are then classes of permutations encoded by the same integer sequence . the top permutation in each of these classes is 312-avoiding and the bottom one is 132-avoiding . a correspondence between linearized binary trees on @xmath33 interior vertices and permutations in @xmath24 is given by loday and ronco in @xcite . the tamari classes correspond to classes of permutations that are associated to the same tree ( ignoring the linearization ) . this is exactly the way in which we will think of tamari congruence classes on @xmath24 . there is a way to define triangulations of a @xmath63-gon corresponding to permutations in @xmath24 . the tamari congruence classes then consists of permutations that produce the same triangulation ( see @xcite ) . the partial order on triangulations inherited from the weak bruhat order is actually rather natural and can be expressed in its own right , with no reference to the weak bruhat order ( the cover relation expresses a local change in the triangulation due to a single `` diagonal flip '' ) . this is precisely defined in a more general setting in section [ partitions ] . purely in terms of the weak bruhat order one can define the tamari congruence on @xmath24 as the coarsest congruence @xmath64 that collapses the edges @xmath65 for @xmath66 , in the weak bruhat order on @xmath24 . this and many other lattices on @xmath24 are described in this manner by n. reading in @xcite . the collapsing of edges is encoded in the coxeter diagram @xmath25 by directing the edges . the tamari congruence corresponds to orienting all the edges the same way , as in @xmath67 there are many other ways to arrive at an ordered lattice isomorphic to the tamari lattice corresponding to @xmath24 , with or without referring to permutations . the author has stumbled upon yet another way in @xcite in which fixed points of a certain endomorphism of an infinite rooted tree are studied . each vertex stabilizes after finitely many applications of the endomorphism and reaches a , so called , self - describing sequence . each class of points at level @xmath33 that eventually stabilizes to the same self - describing sequence corresponds in a rather natural way ( through site inversion counting ) to a congruence class in the tamari lattice on @xmath24 . note that there are certainly different congruences on @xmath24 producing the same lattice quotient and thus deserving of the title tamari congruence . the point is that there are always some choices involved and there is often more than one natural choice . one could work with the right bruhat order instead of the left one , or define slightly different way to associate triangulations to permutations , or apply some obvious automorphisms to the weak bruhat order lattice , and so on . there is just too much symmetry involved to claim any canonical choices ( in our discussion so far we already mentioned a few different choices existing in the literature ) . we now fix a particular tamari congruence on @xmath24 . the congruence will be denoted by @xmath64 . it is the one we already defined above in terms of collapsing edges in the weak bruhat order . we will make our definition of a triangulation corresponding to a permutation consistent with this choice . we will also make all our subsequent choices in accordance to this choice . this makes all the connections we display possible at the price of not always choosing the most standard way of representing some objects . it is all matter of left versus right , bottom versus top , etc . , and it seems a standard choice in one aspect leads to non - standard choices in another aspect , so some degree of `` oddness '' is unavoidable . lattice congruence classes in finite lattices always form intervals , so the tamari congruence classes are intervals in the weak bruhat order . in agreement with our choices tamari congruences correspond to intervals in the weak bruhat order in which the upper bound is a @xmath2-avoiding permutation and the lower bound is a @xmath3-avoiding permutation . we will later say more on pattern avoidance . here we list some basic facts on thompson group @xmath68 and the related positive monoid @xmath21 . other than the presentation , already given in , nothing in this section is needed to follow the text . the monoid @xmath21 satisfies the ore condition and embeds into its group of left fractions @xmath69 . the group @xmath68 is the celebrated thompson group @xmath1 , given by the group presentation @xmath70 which looks exactly the same as the monoid presentation for @xmath21 . the monoid @xmath21 is just the positive submonoid of @xmath68 , i.e. the submonoid generated by the set @xmath71 . the element @xmath72 , @xmath36 , in this presentation can be written it terms of @xmath73 and @xmath74 as @xmath75 . thus @xmath68 is a finitely generated group . it is also finitely presented ( with only two relations ) , but is it often more convenient to use the above infinite presentation . the group @xmath68 has many fascinating properties and has been studied and re - discovered many times in the last 40 years . it has been a steady source of highly non - trivial and important examples or counter - examples , especially in topology . the group @xmath68 is infinite and torsion free . m. brin and c. squier showed that @xmath68 has no subgroups isomorphic to the free group of rank 2 and satisfies no group laws @xcite . all normal subgroups of @xmath68 contain the commutator , which is a simple infinite group . the abelianization @xmath76 $ ] is @xmath77 ( obvious from the presentation above ) . k. brown and r. geoghegan singled out thompson group @xmath68 as the first example of a finitely presented torsion free group of type @xmath78 but not of type @xmath79 @xcite . thompson group @xmath68 has a universal property with respect to homotopy idempotents @xcite . it is the group of order preserving automorphisms of the free finitely generated algebra in the variety of binary cantor algebras ( all finitely generated free algebras in this variety are isomorphic , thus there is no notion of a rank ; see @xcite ) . v. guba and m. sapir showed that @xmath68 is the diagram group of the monoid presentation @xmath80 @xcite . v. guba recently showed that the dehn function of @xmath68 is quadratic @xcite ( this is exactly on the boundary between hyperbolic and non - hyperbolic groups ) . on a very concrete level , the group @xmath68 may be realized as the group of piecewise linear and order preserving homeomorphisms of the unit interval @xmath81 $ ] such that all the slope breaks occur at dyadic rational numbers and the slopes away from the finitely many breaks are integer powers of 2 ( this interpretation has been attributed to thurston ) . a closely related concrete realization is as the group generated by the two piecewise linear homeomorphisms given in figure [ x0x1 ] acting ( on the left ) on the interval @xmath82 . and @xmath74 ] the group @xmath68 was constructed by thompson in 1965 in relation to his study of questions in logic . the simplicity of the commutator @xmath84 $ ] and the simplicity of two related finitely presented groups , @xmath85 and @xmath86 , were established by thompson in his famous unpublished notes @xcite . the groups @xmath85 and @xmath86 were the first examples of finitely presented simple groups . a survey of some properties of @xmath1 and the related simple groups @xmath85 and @xmath86 is given in @xcite . it is known that @xmath68 is not elementary amenable , but it is not known if it is amenable . j. belk and k. brown @xcite showed recently that the isoperimetric constant of @xmath68 is no greater than @xmath87 . the question of amenability of @xmath68 was raised by r. geoghegan in 1979 and is one of the most interesting open questions related to thompson group @xmath68 . the question of amenability of @xmath68 can be related to the question of amenability of the positive monoid @xmath21 . it is shown by r. grigorchuk and a. stepin in @xcite that the positive monoid @xmath21 is not left amenable ( following the side convention we use in this text ) and that the group @xmath68 is amenable if and only if the monoid @xmath21 is right amenable ( the notion of amenability in monoids requires attention to be paid to the side ; left and right amenability are the same in groups ) . as seen from section [ fact - tamari ] there is a long and fruitful history of encoding permutations , trees ( linearized or not ) and various integer sequences in terms of each other . we use this section to establish a particular relation between trees , permutations and integer sequences that is relevant to our discussion and leads to a particular choice of a lattice congruence @xmath64 on @xmath24 defining the tamari lattice @xmath88 . consider a permutation @xmath30 in @xmath24 . let @xmath89 be the number of inversions of @xmath30 that involve @xmath44 and a term to the left of @xmath44 in @xmath30 . in other words , @xmath90 counts the number of terms in @xmath30 that are larger than @xmath44 and are positioned to the left of the term @xmath44 . the sequence @xmath91 is called the inversion sequence of @xmath30 . for , example , the inversion sequence for the permutation @xmath92 is @xmath93 . we will be thinking of sequences of natural numbers as elements of the free monoid with basis @xmath94 . let @xmath95 and let @xmath16 be the free monoid on @xmath96 . the inversion sequence @xmath97 is then identified with the element @xmath98 in @xmath16 . thus we have a map @xmath99 that associates an @xmath96-word of length @xmath33 to any permutation in @xmath24 . denote the image @xmath100 by @xmath101 . then @xmath102 and @xmath103 is bijective ( see @xcite ) . let @xmath104 be the inverse map of @xmath103 . one can try to write down explicit formulae for @xmath105 , but it is more important for our purposes to think of the following constructive way of calculating the permutation @xmath105 in @xmath24 corresponding to an inversion sequence @xmath106 in @xmath101 . start with @xmath33 empty slots @xmath107 . place 1 in such a way that @xmath108 open slots are left to the left of it . then place @xmath109 in such a way that @xmath110 open slots are left to the left of it . continue this procedure until a permutation in @xmath24 is obtained . in other words , @xmath105 is obtained in @xmath33 steps by placing , at step @xmath111 , the number @xmath111 in such a way that @xmath112 open slots are left to the left of @xmath111 , @xmath113 . for example to calculate @xmath114 one starts with @xmath115 then gets @xmath116 and finally @xmath117 denote the set of linearized labeled rooted binary trees on @xmath33 interior vertices by @xmath118 . observe that the leafs @xmath39 are usually drawn in line ordered from left to right by their labels . each par of consecutive leafs @xmath119 , @xmath49 , is called a gap . associate to each gap @xmath119 the last common vertex on the unique paths from the root to leaf @xmath120 and leaf @xmath44 . we say that the associated interior vertex covers the gap ( in order theoretic terms this is just the join of the two leafs ) . in the example in figure [ tree - lin ] the correspondence between the gaps and the interior vertices that cover them is given by @xmath121 the correspondence between gaps and interior vertices that cover them is bijective . the map @xmath122 given by @xmath123 for @xmath49 and a tree @xmath48 in @xmath118 , is bijective . as already observed , for our example from figure [ tree - lin ] we have @xmath124 . we could express @xmath125 without referring to the gaps as follows . for @xmath49 , @xmath126 is the linearization label of the interior vertex visited at position @xmath44 in the in - order . the reason we do not do this is that gaps will be relevant in section [ partitions ] when the discussion moves to trees of higher degree . a rooted binary tree with single interior vertex ( and two leafs ) is called a caret . each interior vertex determines a caret consisting of the vertex itself and its two children . each tree with @xmath33 interior vertices is composed of @xmath33 carets . thus we can bijectively associate gaps and carets in a labeled rooted binary tree . we also say that the associated caret covers the corresponding gap . we describe now the map @xmath127 which is inverse to @xmath128 , that associates a linearized tree @xmath129 with @xmath33 interior vertices to a permutation @xmath30 in @xmath24 . start with @xmath38 leafs placed on a line and labeled ( from left to right ) by @xmath39 . in step @xmath111 , @xmath113 , add a caret labeled by @xmath111 covering the gap @xmath130 . for example , for the permutation @xmath92 in the first step we add a caret labeled by 1 covering the gap ( 4,5 ) , after two steps we have two carets as depicted in the top half of figure [ tree - lin2 ] , after 4 steps we have 4 carets as in the bottom half of figure [ tree - lin2 ] ] and in the last step we obtain the linearized tree in figure [ tree - lin ] . of course , the compositions @xmath131 and its inverse @xmath132 provide bijection between @xmath101 and @xmath118 for every @xmath33 . it is actually possible to write down the correspondence more directly without referring to @xmath24 as an intermediate step , but a natural way to do this is to leave the world of trees and inversion sequences and extend all considerations to forests and arbitrary elements in the free monoid @xmath16 . the reason for this is already obvious when one notes that the intermediate steps in figure [ tree - lin2 ] consists of forests rather than trees . similarly , the intermediate steps in are not permutations in @xmath24 . we want to develop a language that will work with such intermediate steps . additional benefit of this extension is that , on the level of @xmath96-words , we will work in the more natural environment of the full monoid @xmath16 rather than its submonoid @xmath133 . concatenation of inversion sequences of length @xmath134 and @xmath33 is an inversion sequence of length @xmath135 . thus @xmath136 is indeed a submonoid of @xmath16 . we can define an operation on @xmath137 that agrees with the concatenation operation in @xmath10 and has the natural extensions @xmath138 and @xmath139 as mutually inverse monoid isomorphisms . the operation is denoted by @xmath140 and defined as follows . if @xmath141 and @xmath142 then @xmath143 is given by @xmath144 in other words , @xmath145 is produced by first increasing all the terms of @xmath30 by @xmath134 and then concatenating them to the right of the terms of @xmath29 . this operation on permutations will be called interlacing ( the same operation is used in @xcite in the definition of product on the hopf algebra @xmath146 $ ] ) . before we move on to forests and @xmath16 let us provide a definition of the operation , also denoted by @xmath140 , on linearized trees in @xmath147 compatible with the concatenation operation on @xmath10 and the interlacing operation @xmath140 defined in on @xmath11 . the operation is performed by stacking the second linearized tree on top of the first ( hence the notation @xmath140 ) . more precisely , for trees @xmath148 in @xmath149 and @xmath150 in @xmath118 the product @xmath151 is the tree @xmath48 in @xmath152 obtained by deleting the leaf label 0 in @xmath150 , increasing all other labels ( both in the interior and on the leafs ) in @xmath150 by @xmath134 , identifying the leaf with deleted label in @xmath150 with the root of @xmath148 and declaring the root od @xmath150 to be the root of @xmath48 . an example is given in figure [ product - trees ] . thus right now we have three canonically isomorphic monoids , namely @xmath10 , @xmath11 and @xmath12 , with the operations concatenation , permutation interlacing and tree stacking , respectively . since the free monoid @xmath16 does not need a special introduction , we start by introducing the notion of linearized labeled binary rooted forests . such forests consists of countably many rooted binary trees , only finitely many of which are non - trivial ( the forest has only finitely many carets ) . furthermore , the roots are labeled in bijective fashion by the numbers in @xmath153 , the leafs are also labeled in bijective fashion by the numbers in @xmath154 and a leaf on one tree is labeled by a smaller number than a leaf on another tree if and only if the same is true for their corresponding roots . note that vertices that are both a root and a leaf have two labels , one as a root and one as a leaf , and these labels may be different . finally , if the number of carets is @xmath33 then they are labeled bijectively by @xmath41 in such a way that the labels on all paths from a root to a leaf are decreasing ( thus , again , the labeling of the interior vertices is compatible with the order structure imposed by the forest structure ) . denote the set of linearized forests by @xmath17 . we represent forests by diagrams of the type depicted in figure [ forest - lin ] in which it is assumed that the labeling of both the roots and the leafs is done from left to right and all trees that are not drawn are singletons labeled by higher numbers . the labeling of the roots is usually left out , since it is determined uniquely by the labeling of the leafs . note that leaf 6 is also labeled as root 1 , while leaf 10 is also labeled as root 3 . we define now a bijective map @xmath155 the forest corresponding to the word @xmath156 in @xmath16 can be constructed inductively as follows . start with the trivial forest in which all trees are singletons . throughout the whole construction the leafs and their labels are left unchanged . all that happens in the process is that we add labeled carets and relabel the roots . for a letter @xmath157 the corresponding linearized forest is given in figure [ xi - lin ] . ] note that the root labels to the right of leaf @xmath158 do not agree any longer with the leaf labels ( the root labels are smaller by 1 ) . if @xmath159 is the forest representing @xmath160 construct the forest @xmath161 representing @xmath156 by adding a caret , labeled by @xmath33 , connecting root @xmath162 and root @xmath163 . the newly added root gets root label @xmath162 , old roots @xmath164 and higher get their labels decreased by 1 and the old roots @xmath162 and @xmath163 loose their root labels ( they are not roots any longer ) . for example , the forest in figure [ forest - lin ] corresponds to the word @xmath165 . conversely , the inverse map @xmath166 can be understood as follows . the linearization part of the labeling of the forest @xmath167 gives a recipe for constructing @xmath167 starting from the trivial forest . namely , first add the caret labeled by 1 , then the one labeled by 2 , and so on until the caret labeled by @xmath33 is added . in each step @xmath111 we record the label @xmath112 of the root that the newly introduced caret labeled by @xmath111 uses as the left leaf . the corresponding element of @xmath16 is then @xmath168 . one can check that this procedure applied to the tree in figure [ forest - lin ] does indeed yield the word @xmath165 . forests can be multiplied in a way compatible with the concatenation multiplication in @xmath16 as follows . the operation , still denoted by @xmath140 , is performed by stacking the second forest on top of the first . namely , the product of the forests @xmath167 and @xmath169 is the forest @xmath170 obtained by , first increasing all labels on the carets of @xmath169 by @xmath134 , where @xmath134 is the number of carets in @xmath167 , then identifying root @xmath44 in forest @xmath167 with leaf @xmath44 in forest @xmath169 and then deleting their root / leaf labels , correspondingly . the leafs of @xmath171 are the leafs of @xmath167 and the roots of @xmath171 are the roots of @xmath169 . for example , the product of the forest in the bottom half of figure [ tree - lin2 ] and the forest in the top half of the same figure is the linearized forest in figure [ product - forest ] . we now turn to the world of permutations . the intermediate steps in provide motivation for the following definition . a @xmath18-permutation of length @xmath33 is a map @xmath172 , where @xmath173 , such that the inverse image of each element in @xmath174 is a singleton . in other words , @xmath30 is an infinite sequence such that each of @xmath40 appears exactly once as a term , and the rest of the terms are @xmath18 s . denote the set of @xmath18-permutations by @xmath19 . a bijective correspondence @xmath175 between linearized forests and @xmath18-permutations is defined as follows . each gap @xmath119 , @xmath176 in a linearized forest @xmath167 is either covered by a caret or is not covered by a caret ( the latter happens in case the two vertices defining the gap come from different trees ) . define @xmath177 for example , the @xmath18-permutation corresponding to the forests in figure [ product - forest ] and figure [ forest - lin ] are @xmath178 and @xmath179 , respectively . note that we agree ( sometimes ) not to write ( some of ) the stars to the right of the last non-@xmath18 symbol . the non-@xmath18 symbols will be called concrete symbols in the rest of the text . the inverse map @xmath180 is , just as its restriction to @xmath11 , simply defined by drawing the forest caret by caret . for a @xmath18-permutation of length @xmath33 , in step @xmath111 , @xmath113 , add a caret labeled by @xmath111 covering the gap @xmath130 and relabel the roots accordingly . at the end of the procedure only @xmath33 carets will be drawn and the rest of the gaps are uncovered and correspond to @xmath18 s . the mutually inverse maps @xmath181 are defined in the same fashion as before . if @xmath30 has @xmath33 concrete symbols the corresponding word @xmath182 in @xmath16 is @xmath98 , with added stipulation that any occurrence of a @xmath18 to the left of @xmath111 in @xmath30 is counted in @xmath183 . in other words , for all our purposes , @xmath18 is considered to be larger than any concrete symbol ( symbol from @xmath184 ) . in the other direction , given a word @xmath185 of length @xmath33 over @xmath96 , once the symbols @xmath41 are placed by leaving an appropriate number of open slots as prescribed by @xmath185 , the rest of the open slots are filled with @xmath18 s . for example , @xmath178 and @xmath186 correspond to the @xmath96-words @xmath187 and @xmath165 , respectively . to complete the picture we define the operation , still denoted by @xmath140 , on @xmath19 that agrees with concatenation on @xmath16 and stacking of linearized trees in @xmath17 . the operation is performed by interlacing @xmath18-permutations . one can write down definite formulae , but the operation is more easily understood as a process . for @xmath18-permutation @xmath29 with @xmath134 concrete symbols and @xmath18-permutation @xmath30 with @xmath33 concrete symbols the product @xmath188 is obtained as follows . first increase the concrete symbols in @xmath30 by @xmath134 . then interlace @xmath30 into @xmath29 by placing the @xmath44-th symbol of @xmath30 in the position of the @xmath44-th star in @xmath29 . for example @xmath189 . the following diagram may be helpful for imagining the process [ cols="^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^,^ , < " , ] at this moment we have three monoids @xmath16 , @xmath17 and @xmath19 related by canonical isomorphisms . since @xmath16 is free so are @xmath17 and @xmath19 and we may be disappointed that all that happened so far is that we obtained two strange copies of the free monoid @xmath16 of countable rank namely @xmath17 with a free basis consisting of linearized forests @xmath190 as in figure [ xi - lin ] , @xmath191 , and @xmath19 with a free basis consisting of @xmath192 , @xmath193 . there are at least two ways to motivate what comes next . one is to observe that the multiplication rule @xmath140 on @xmath17 does not essentially depend on the linearization part of the labeling of the involved forests . this labeling is just carried around and adjusted here and there by increasing labels accordingly , but nothing in the definition depends on it . this means that the equivalence relation @xmath64 on @xmath17 obtained by dropping the labels on interior vertices is not only equivalence on @xmath17 but it is also a monoid congruence . the equivalence @xmath64 is a congruence on the monoid @xmath194 . another way to motivate the introduction of @xmath64 is as an extension of a well known connection between permutations and their linearized trees obtained when the linearization is striped away . in that case , several permutations correspond to the same labeled rooted binary tree . it is known that there are catalan number @xmath195 labeled rooted binary trees on @xmath33 interior vertices . thus the @xmath196 permutations in @xmath24 are split into @xmath197 classes of equivalent permutations . we fix the equivalence classes obtained in this process as classes defining the tamari congruence on @xmath24 . formally , for any forest @xmath48 in @xmath17 define @xmath198 to be the forest obtained when the labeling on all interior vertices is deleted . define an equivalence on @xmath17 by @xmath199 and , by use of the corresponding bijections , define the induced equivalences on @xmath19 by @xmath200 and on @xmath16 by @xmath201 thus we have a monoid congruence @xmath64 on @xmath16 , @xmath19 and @xmath17 and we want to understand the corresponding factor monoid . the diagram in figure [ s3all ] depicts the situation for @xmath202 . the hexagon in the middle is the ( left ) cayley graph of @xmath202 as coxeter group of type @xmath203 generated by the standard generating set @xmath204 . it is drawn in a way that represents the hasse diagram of the ( left ) weak bruhat order on @xmath202 . and its associated words and trees ] the only edge drawn as a full line in the right half of the diagram indicates that the corresponding permutations ( or words ) are to be identified in @xmath205 ( or in @xmath206 ) , since they correspond to the same tree . therefore @xmath207 , @xmath208 and @xmath205 and @xmath206 have 5 elements ( which is the catalan number @xmath209 ) . observe that the restriction of @xmath64 to @xmath11 is a congruence on the monoid @xmath11 ( again , this is because when the labeling is stripped in @xmath12 the operation @xmath140 is not affected ) . [ free ] the monoid @xmath210 is free . indeed , any non - trivial tree for which the left subtree is trivial is indecomposable in @xmath13 ( it can not be written as a product of two or more nontrivial trees ) . moreover , every tree in @xmath13 has a unique decomposition as a product of such indecomposable trees . thus @xmath13 is free . this is perhaps a reason why the connection to the thompson monoid is not immediately obvious . for a researcher in combinatorics there seem to be no particular gain in constructing free monoids using strange definitions of products of permutations . on the other hand , looking from thompson monoid @xmath21 point of view , the connection to tamari lattices on @xmath24 is not immediately obvious since on its basic level , working only with ordinary permutations before the expansion to @xmath19 , the information is encoded in a certain ( not particularly distinguished ) free submonoid of @xmath21 that seemingly does not demand any particular attention ( there are plenty of free monoids inside @xmath21 ) . however , we want to understand @xmath211 and this is where the interesting things happen . our understanding of the weak bruhat order and tamari lattice congruences on @xmath24 can still be used in this extended situation . note that every element in @xmath19 has finite congruence class . this is because there are only finitely many ways to linearize a forest with finitely many carets . let @xmath212 be the set of @xmath196 different @xmath18-permutations with @xmath33 concrete ( non-@xmath18 ) terms and @xmath18 s in some fixed positions . more precisely , let us say that there are exactly @xmath213 blocks of consecutive positions in which concrete symbols appear ( any two blocks are separated by some finite number of @xmath18 s ) . let the sizes of the concrete blocks , from left to right , be @xmath214 and let the last concrete symbol appears at position @xmath215 . clearly , all the elements related by @xmath64 to a @xmath18-permutation in @xmath212 are also in @xmath212 . we will describe the congruence classes on @xmath212 in terms of the congruence classes on @xmath216 . there is a canonical correspondence @xmath217 obtained by removal / insertion of @xmath18 s in appropriate positions . this enables us to induce the weak bruhat order on @xmath212 . we write @xmath218 for the @xmath18-permutation obtained from @xmath30 when @xmath44 and @xmath158 exchange their positions in @xmath30 . extending to @xmath24 this defines a ( left ) action of @xmath24 on @xmath212 . technically speaking , the hasse diagram of the weak bruhat order induced on @xmath212 is not the cayley graph of @xmath24 but rather the schreier graph of the action of @xmath24 on @xmath212 with respect to the standard generating set @xmath219 , but these two graphs are canonically isomorphic and we borrow the terminology from @xmath24 and use it on @xmath212 . in particular , we keep the notation @xmath220 for the weak bruhat order in the extended sense . if @xmath221 in @xmath212 is obtained from @xmath30 by interchanging @xmath44 and @xmath158 it is still valid to say that the lengths of @xmath29 and @xmath30 differ by 1 . moreover , @xmath29 covers @xmath30 in the weak bruhat order if and only if @xmath44 is to the left of @xmath158 in @xmath30 and it is directly below @xmath30 in the other case ( this says that the weak bruhat order is compatible with the lexicographic order on @xmath212 ) . we note that there is a very important difference . namely the @xmath64 classes on @xmath24 and on @xmath212 are not the same . for example , @xmath202 has five classes , while @xmath222 corresponding to the block pattern @xmath223 has only two . the two classes in @xmath222 are indicated in figure [ s3prime ] as the components connected by edges drawn as full lines . corresponding to the block pattern @xmath223 ] the first thing we want to know is under what conditions two neighbors in the weak bruhat order correspond to the same forest . we recall the explicit agreement that @xmath18 is larger than any concrete symbol . [ move - up ] let @xmath225 in @xmath212 . then the forests @xmath226 and @xmath129 are equal if and only if there exists a term between @xmath44 and @xmath158 in @xmath30 ( and therefore in @xmath29 as well ) that is larger than @xmath158 . exchanging the labels @xmath44 and @xmath158 in a linearized forest @xmath167 leads to another linearized forest if and only if the interior vertices @xmath44 and @xmath158 are not comparable with respect to the order induced by the forest @xmath167 . this happens exactly when @xmath44 and @xmath158 belong to different trees in the forest , in which case there is a @xmath18 between @xmath44 and @xmath158 in @xmath227 , or @xmath44 and @xmath158 belong to the same tree , but are both descendants of a vertex @xmath111 , in which case @xmath228 and @xmath111 is between @xmath44 and @xmath158 in @xmath227 . this means that for every @xmath30 in @xmath212 we may go up step by step in the weak bruhat order until we reach a @xmath18-permutation that has no occurrence of @xmath229 , with @xmath228 . this leads us to pattern avoiding considerations . we say that @xmath230 occurs in @xmath30 in @xmath212 if there are three indices @xmath231 such that @xmath232 ( recall that @xmath215 is the index of the last concrete symbol in the @xmath18-permutations in @xmath212 ) . this allows the possibility that the middle symbol @xmath233 is a @xmath18 , but the other two symbols must be concrete . if @xmath230 does not occur in @xmath30 the @xmath18-permutation is @xmath230-avoiding . similarly , we say that 132 occurs in a @xmath18-permutation @xmath30 in @xmath212 ( or in any sequence over a linearly ordered set ) if there are three indices @xmath231 such that @xmath234 . if 132 does not occur in @xmath30 the permutation is 132-avoiding . once again , the definition implies that the middle symbol @xmath233 may be a @xmath18 but the other two symbols involved must be concrete . it is easy to observe that @xmath30 is a 132-avoiding @xmath18-permutation if and only if all concrete terms to the left of any occurrence of @xmath18 in @xmath30 are larger than all concrete terms to the right of the same occurrence of @xmath18 and each concrete block in @xmath30 satisfies the 132-avoiding constraint by itself . [ 132=132 ] a @xmath18-permutation @xmath30 in @xmath212 is @xmath2-avoiding if and only if it is @xmath230-avoiding . any occurrence of @xmath230 in @xmath30 is also an occurrence of @xmath2 . for the converse , assume @xmath2 appears in @xmath30 . let @xmath235 be an occurrence of @xmath2 with minimal difference @xmath236 . if @xmath158 is to the right of @xmath213 we have an appearance of @xmath230 . but @xmath158 can not appear to the left of @xmath213 because of the minimality in the choice of @xmath44 and @xmath111 . the above observations lead to the following proposition . [ unique - up ] for every @xmath18-permutation @xmath30 in @xmath212 there exists a unique @xmath2-avoiding @xmath18-permutation @xmath237 in @xmath212 such that @xmath238 . moreover @xmath239 and @xmath240 if and only if @xmath241 . the existence is clear ( just go up step by step as long as possible without changing the forest ) . everything else follows from counting arguments . there are exactly @xmath242 ordered @xmath213-tuples of labeled rooted binary trees with @xmath214 carets , respectively . as for @xmath2-avoiding @xmath18-permutations , the symbols in each concrete block are uniquely determined by the condition that all symbols in a block to the left of some other block must be larger than the symbols in the other block . since there are exactly @xmath243 @xmath2-avoiding arrangements of @xmath244 symbols ( see the catalan addendum in @xcite ) we get that there are @xmath242 @xmath2-avoiding @xmath18-permutations in @xmath212 . thus every equivalence class in @xmath212 must have exactly one @xmath2-avoiding representative . in a completely analogous way , starting from any @xmath18-permutation @xmath30 in @xmath212 we may move step by step down in the weak bruhat order by preserving the corresponding forest as long as we see an occurrence of @xmath245 , with @xmath246 . define @xmath3-avoiding @xmath18-permutations as @xmath18-permutations in which there are no three indices @xmath231 such that @xmath247 . [ unique - down ] for every @xmath18-permutation @xmath30 in @xmath212 there exists a unique @xmath3-avoiding @xmath18-permutation @xmath248 in @xmath212 such that @xmath249 . moreover @xmath250 and @xmath240 if and only if @xmath251 . it is known that @xmath88 is a lattice , known as tamari lattice . in other words , it is known that @xmath64 is a lattice congruence on @xmath24 . we claim that @xmath252 is also a lattice , i.e. @xmath64 is a lattice congruence on the weak bruhat order lattice on @xmath212 . this essentially follows from the fact that certain maps @xmath253 , where @xmath254 is parabolic subgroup of @xmath24 ( as coxeter groups ) are lattice homomorphisms on the corresponding weak bruhat order lattices . define the flattening of a sequence @xmath255 of distinct integers ( or members of any linear order ) as the unique permutation @xmath30 in @xmath254 such that @xmath256 if and only if @xmath257 . [ left - right ] the map @xmath253 that maps a permutation @xmath30 to the flattening of @xmath258 is a surjective lattice homomorphism . the statement is equivalent to the following . consider @xmath24 and @xmath254 under their right weak bruhat order . define a map @xmath259 as follows . for @xmath30 in @xmath24 , let @xmath260 be the permutation in @xmath254 obtained when all terms in @xmath30 except for the terms @xmath261 are deleted and then flattened ( decreased by @xmath262 ) . the equivalence comes by applying the inverse to the elements in @xmath24 . in the left weak bruhat order we are concerned with the positions @xmath261 , while after inversion takes place and we land in the right weak bruhat order we are concerned with the terms @xmath261 . with respect to the right bruhat order , the map @xmath263 is the surjective lattice homomorphism @xmath253 corresponding to the parabolic subgroup @xmath254 generated by the @xmath264 reflections @xmath265 ( see @xcite for example ) . the equivalence @xmath64 is a lattice congruence on @xmath212 . moreover @xmath266 as lattices . patching together @xmath213 parabolic homomorphisms as in lemma [ left - right ] we get a surjective lattice homomorphism @xmath267 , which can then be composed further to get a lattice homomorphism @xmath268 . we claim that @xmath64 is the kernel of this homomorphism . recall that the concrete terms of a @xmath18-permutation @xmath30 in @xmath212 just indicate in what order the carets are added in the forest @xmath129 , which consists of @xmath213 trees with @xmath214 carets , respectively . in particular , it is clear that the @xmath44-th de - linearized tree that corresponds to the @xmath44-th concrete block depends only on the @xmath64 class of the flattening of the corresponding block ( the gaps in the numbers before the flattening correspond to carets added in the other trees of the forest ) . thus two @xmath18-permutations correspond to the same forest if and only if the corresponding flattenings in each block are @xmath64 related and the relation @xmath64 on @xmath212 is indeed the kernel of the surjective homomorphism @xmath269 . [ intervals ] each @xmath64 conjugacy class of @xmath18-permutations in @xmath212 is an interval in the weak bruhat order and is a union of several tamari congruence classes of @xmath24 ( after identification of @xmath212 and @xmath24 ) . for each class , the top bound of the interval is a @xmath2-avoiding @xmath18-permutation and the bottom bound is a @xmath3-avoiding @xmath18-permutation . congruence classes in finite lattices are always intervals . proposition [ unique - up ] and proposition [ unique - down ] show that the top and the bottom must be @xmath2-avoiding and @xmath3-avoiding @xmath18-permutations , respectively . finally , if @xmath270 and @xmath240 in @xmath24 , then there exists a term between @xmath44 and @xmath158 larger than @xmath158 . after @xmath18 s are placed in appropriate places to land in @xmath212 it is still correct that there is a larger term between @xmath44 and @xmath158 . thus the corresponding @xmath18-permutations are also related . this shows that each tamari class of @xmath24 is included in a @xmath64 class of @xmath212 . thus we have a thorough understanding of the equivalence classes in @xmath19 . we translate now this understanding to @xmath16 . we know that we can connect any two equivalent @xmath18-permutations by several steps involving transpositions , with the extra constraint that when we apply @xmath32 some term between @xmath44 and @xmath158 must be larger than @xmath158 . here is the corresponding statement in the @xmath16 world . [ move - upx ] let @xmath271 , @xmath27 and @xmath240 in @xmath19 . further let @xmath272 . then @xmath273 and @xmath274 can be obtained from @xmath182 by applying the substitution @xmath275 at positions @xmath111 and @xmath276 in @xmath182 . by proposition [ move - up ] , @xmath277 , where @xmath213 is a term larger than @xmath276 ( possibly a @xmath18 ) . this immediately means that there are more inversions to the left of @xmath276 than to the left of @xmath111 , i.e. , @xmath273 . the exchange of @xmath111 and @xmath276 causes move upwards in the weak bruhat order ccccccccc & = & & j+1 & & k & & j & + & & & & & ( j j+1 ) + & = & & j & & k & & j+1 & after @xmath111 and @xmath276 exchange their positions , no inversion numbers other than those at position @xmath111 and @xmath276 in @xmath182 can possibly be affected . we have @xmath278 and @xmath279 . the extra 1 in @xmath280 comes from the fact that now @xmath276 is to the left of @xmath111 and should be counted as extra inversion . thus cccl x ( ) & = & x_n+1 x_m + & & & ( j , j+1 ) + x ( ) & = & x_m x_n a converse to the previous proposition holds . let @xmath281 , with @xmath273 , and let @xmath282 be obtained from @xmath185 by applying the substitution @xmath283 . then @xmath284 , @xmath285 and @xmath286 . [ xsimy ] let @xmath281 , with @xmath273 , and let @xmath282 be obtained from @xmath185 by applying the substitution @xmath283 . then @xmath287 . we can now prove that the monoid @xmath288 is thompson s monoid @xmath21 . the congruence @xmath64 on @xmath16 is generated by @xmath289 for all pars of non - negative integers @xmath44 and @xmath111 with @xmath290 . in other words , @xmath288 is equal to thompson s monoid @xmath21 , given by the presentation @xmath291 it is clear that @xmath292 does hold , for @xmath290 , in @xmath16 ( this is a special case of corollary [ xsimy ] ) . another way to see this is to realize that the two words @xmath293 and @xmath294 correspond to the two ways to linearize ( i.e. to draw ) the forest in figure [ xixj ] ] on the other hand , the relations @xmath295 , for @xmath45 , are sufficient . this is because any two equivalent @xmath18-permutations can be related by a sequence of applications of appropriate transpositions ( say by going up in the weak bruhat order on @xmath212 and reaching the same @xmath2 avoiding @xmath18-permutation ) . by proposition [ move - upx ] this translates to a sequence of applications of relations of the type @xmath292 , for @xmath45 . every class of equivalent elements in @xmath21 corresponds to an interval in the weak bruhat order on @xmath18-permutations . moreover , the top always corresponds to a @xmath2-avoiding and the bottom to a @xmath3-avoiding @xmath18-permutation . the following proposition characterizes the words over @xmath96 that correspond to the endpoints of weak bruhat order intervals that determine the classes of equivalent elements in thompson s monoid @xmath21 . a @xmath18-permutation is @xmath2-avoiding if and only if the corresponding word @xmath182 has non - increasing indices . a @xmath18-permutation is @xmath3-avoiding if and only if the corresponding word @xmath182 has no decrease of an index larger than 1 . as a corollary we obtain the two well known normal forms on @xmath21 . every class of equivalent elements in @xmath21 has two normal forms . one is a word with non - increasing indices and the other is a word whose indices never decrease by more than 1 ( and can possibly increase ) . the normal form with non - increasing indices is obtained when we move up in the weak bruhat order by moving letters @xmath157 with `` small '' indices to the right of letters @xmath296 with `` large '' indices ( we assume @xmath45 ) by applying the substitution @xmath297 . the normal form with unit decrease is obtained when we move down in the weak bruhat order by moving letters @xmath296 with `` very large '' indices to the right of letters @xmath157 with `` small '' indices ( we assume @xmath298 ) by applying the substitution @xmath299 . the set of rules @xmath300 for @xmath301 , represents a confluent rewriting system on @xmath21 . the same is true for the reversed set of rules @xmath302 for @xmath45 . we note here that both normal forms are well known , but the top one is used more often in the literature on thompson monoids and groups . however , j. belk and k. brown use the bottom one quite efficiently in @xcite to get length functions for the elements in @xmath21 and @xmath68 and then use these length functions in further applications . taking a different approach , j. belk shows independently in @xcite that that the two normal forms of an element @xmath169 of @xmath96-length @xmath33 in @xmath21 bound the class of @xmath96 words representing @xmath169 in the so called word graph of @xmath169 based on the rewriting rules above ( without describing these classes as unions of tamari lattice congruence classes ) . we observe that an inversion sequence word @xmath185 in @xmath10 represents a basis element of the free monoid @xmath15 from proposition [ free ] if and only if @xmath185 has a single occurrence of @xmath73 ( necessarily at the very end ) . the corresponding basis elements in @xmath14 are represented by those permutations that start with their largest term . one can easily pick normal representatives ( either top or bottom ) for basis elements either in @xmath15 or in @xmath14 . finally , we mention that the linearizations corresponding to the bottoms of tamari classes on forests in @xmath17 are the ones obtained by post - order linearizations , while the ones corresponding to the tops are the inverse post - order linearizations . in this section we briefly indicate how , for @xmath22 , tamari orders on partitions of @xmath4-gons into @xmath5-gons lead to thompson monoid @xmath303 we fix @xmath213 and @xmath33 to discuss the general case , but provide concrete examples with @xmath304 and @xmath305 . first we define tamari order on the partitions of a fixed @xmath4-gon into @xmath5-gons . label the vertices of the @xmath4-gon by @xmath306 in the positive direction . any diagonal @xmath307 used in the partition is common to two uniquely determined @xmath5-gons in the partition that form a @xmath308-gon using @xmath307 as a diagonal ( connecting opposite vertices in the @xmath308-gon ) . the labels of the @xmath308-gon are still ( cyclically ) ordered from the smallest to the largest in positive direction . let the @xmath213 smallest labels on the vertices of the @xmath308-gon be @xmath309 . the diagonal @xmath307 has size @xmath44 if it uses the vertex labeled @xmath310 . a partition @xmath311 covers a partition @xmath312 if it is obtained from @xmath312 by removing a diagonal of size @xmath44 from a @xmath308-gon in @xmath312 and replacing it by the diagonal of size @xmath158 in the same @xmath308-gon . the tamari partial order on partitions is then just the closure of the cover relation . an example of partition ( with @xmath304 , @xmath305 ) is given in the left half of figure [ partition ] . the diagonal @xmath313 has size @xmath28 , while the diagonals @xmath314 and @xmath315 have size @xmath109 . next we define linearized labeled @xmath213-ary rooted trees . a linearized labeled rooted @xmath213-ary tree on @xmath33-interior vertices is a rooted tree in which the root has degree @xmath213 ( unless the tree has only the root as a vertex , in which case its degree is 0 ) , all interior vertices have degree @xmath316 , the leafs are labeled bijectively by @xmath317 and the interior vertices are labeled bijectively by @xmath318 in such a way that the labels on each path from the root to a leaf are decreasing . an example is given in the right half of figure [ partition ] . denote the set of such trees by @xmath319 . we can define the notion of an interior vertex covering a gap just as in the binary tree case . however , in this case every interior vertex covers exactly @xmath320 gaps . for every tree @xmath48 in @xmath319 define a sequence @xmath125 of length @xmath321 by . in our example in figure [ partition ] we have @xmath322 . denote the image of @xmath319 by @xmath323 . it consists of all sequences of length @xmath321 such that each term @xmath324 appears exactly @xmath320 times and all terms between two appearances of a term @xmath111 are smaller than @xmath111 . the map @xmath325 is a bijection . the inverse map @xmath326 can be defined in a way analogous to the binary case . namely , for @xmath30 in @xmath323 the @xmath320 element set @xmath58 will be called the set of locations of the term @xmath44 . starting from @xmath327 appropriately labeled leafs we add , in step @xmath44 , a @xmath213-ary tree with single interior vertex ( @xmath213-caret ) labeled by @xmath44 in such a way that for each location @xmath328 of @xmath44 the gap @xmath329 is covered by the interior vertex @xmath44 . from @xmath323 we can go by counting inversions to @xmath96-words of length @xmath33 . the only important remark is that the number of inversions @xmath90 does not depend on the particular occurrence of @xmath44 in @xmath30 . thus we have a bijective map @xmath330 , where @xmath331 is the set of words @xmath332 going back from @xmath331 to @xmath323 is accomplished by starting from @xmath321 open slots and then , in step @xmath111 , placing @xmath320 copies of the term @xmath111 in consecutive available open slots after leaving the first @xmath112 leftmost slots open . the operation of concatenation of @xmath96-words still makes perfect sense and leads to corresponding interlacing operation @xmath140 on sequences such as those in @xmath323 and stacking operation @xmath140 on @xmath213-ary trees . every sequence in @xmath323 provides a way to build a polygon partition . start with the polygon with no edges or diagonals drawn . draw the edge @xmath333 ) . in a sense that will be clear later this is the root edge . then , for @xmath334 ( in that order ! ) , in step @xmath44 add the @xmath320 locations of @xmath44 in @xmath30 to the path ( keep the vertices in the path always in increasing order ) . the union of all the obtained paths is the desired partition . in our running example @xmath335 and we start with the edge ( 0,9 ) , then for @xmath336 , we add the vertices 5 and 8 , since these are the locations of 4 in @xmath30 and we obtain the path @xmath337 . then for @xmath338 , we add the vertices 1 and 4 ( locations of 3 ) to get the path @xmath339 . for @xmath340 we add vertices 2 and 3 to get the path @xmath341 and finally for @xmath342 we add 6 and 7 to get the path @xmath343 . for each @xmath344 , each time we add @xmath320 new vertices to the path we add a new @xmath5-gon @xmath345 to the partition . if we keep the label @xmath44 on @xmath345 we obtain linearized partitions . dropping the labels on @xmath345 amounts to de - linearization in the corresponding trees and equivalence relation @xmath64 on @xmath323 . there is a direct way to relate trees and partitions . essentially , the leafs @xmath346 represent the edges @xmath347 , the root represent the edge @xmath348 and the interior vertices @xmath349 represent the diagonals . the interior vertex @xmath33 , being the root , represents , the edge @xmath348 . the diagram in figure [ treepartition ] depicts the correspondence in our model case for the tree and the partition from figure [ partition ] . the leafs are labeled by @xmath310 and the root is labeled by @xmath350 . the edges of the three are dashed , while the partition edges are in full line . the vertices of the tree are emphasized by representing them by small black disks . note that the operation @xmath140 on the level of polygon partitions amounts to gluing partitioned polygons . namely , if @xmath312 is a partition of an @xmath351-gon and @xmath311 is a partition of an @xmath352-gon into @xmath5-gons , then @xmath353 is obtained by lifting all nonzero vertex labels in the second polygon by @xmath354 , and then gluing the two polygons along the edge @xmath355 in both polygons . once again the situation can be lifted to arbitrary words in the monoid @xmath16 , which correspond to linearized @xmath213-ary forests in @xmath17 with stacking operation , @xmath18-sequences with interlacing operation , and partitions of finite sequences of polygons with gluing operation . the operation on the level of polygon partitions involves sequences of partitioned polygons and amounts to gluing the root edges in the first sequence to the leaf edges with matching label in the second partition . an example is given in figure [ product - partition ] . the leaf edges in the first partition are labeled by @xmath310 and the root edges by @xmath356 , while capital letters are used in the second partition . note that trivial trees ( single vertex , no edges ) in forests correspond to @xmath109-gons in polygons ( represented as singe edges ) . the dashed double arrows indicate which edges are to be identified in the gluing process . one choice to describe the multiplication of the depicted partitions by @xmath18-sequences is as @xmath357 . on the level of @xmath96-words the above product corresponds to @xmath358 . the tamari equivalence @xmath64 is again a monoid congruence , the top element in each class is always @xmath2-avoiding , the bottom one is @xmath3-avoiding , and the obtained factor monoid is thompson monoid @xmath6 , given by the presentation . this is just the positive monoid in the corresponding thompson group @xmath359 given by the same presentation but as a group , which is the group @xmath360 of fractions of @xmath6 . the groups @xmath359 , @xmath361 , share many properties with @xmath68 ( see @xcite ) . note that if @xmath44 is a concrete symbol that occurs on the left ( right ) of some larger symbol @xmath111 ( concrete or a @xmath18 ) , then all occurrences of @xmath44 are on the left ( right ) of @xmath111 . further , the corresponding @xmath213-ary forest does not change when we exchange all occurrences of @xmath44 and @xmath158 if and only if they are separated by some larger symbol @xmath111 ( thus @xmath230 or @xmath362 occurs - in the former case we go up and in the latter we go down in the order ) . the defining relations in @xmath6 have the form given in precisely because after all the occurrences of @xmath44 and @xmath158 switch their places in a @xmath18-sequence in which all @xmath320 occurrences of @xmath44 were to the left of all @xmath320 occurrences of @xmath158 ( separated by some larger symbol ) , the number of inversions for the term @xmath44 increases by @xmath320 . it seems that thompson monoid @xmath21 naturally codifies many instances of catalan - like objects in the sense that provides `` recipes '' for their construction as well as relations to indicate which recipes lead to the same catalan - like object . the ubiquity of catalan - like objects ( enumerated by catalan numbers ) is well known in combinatorics . on the other hand , the ubiquity of thompson s monoid ( in fact the group ) is equally well known in infinite group theory . so it is fitting that these objects are closely related . the fact that the finite coxeter groups of type @xmath8 play a role in the mix is also not extremely surprising in the light of their own relevance in many situations . it would be interesting to explore / establish connections between thompson monoids ( not necessarily of type @xmath1 ) and tamari lattices of type @xmath363 and @xmath364 @xcite ( corresponding to factor lattices of finite coxeter groups of type @xmath363 and @xmath364 ) . the connection between the higher thompson monoids @xmath6 , @xmath361 , and the higher catalan objects ( @xmath213-ary forests ) leads to a natural question of exploring the sequences in @xmath365 as a kind of higher coxeter objects of type @xmath8 ( ordinary permutations play this role when @xmath366 ) . thanks to marcelo aguiar , james belk and matt brin for sharing their thoughts . also , thanks to nathan reading for very careful reading of the text and numerous valuable suggestions and references .

we address here the issue of congestion in the modeling of crowd motion , in the non - smooth framework : contacts between people are not anticipated and avoided , they actually occur , and they are explicitly taken into account in the model . we limit our approach to very basic principles in terms of behavior , to focus on the particular problems raised by the non - smooth character of the models . we consider that individuals tend to move according to a desired , or spontanous , velocity . we account for congestion by assuming that the evolution realizes at each time an instantaneous balance between individual tendencies and global constraints ( overlapping is forbidden ) : the actual velocity is defined as the closest to the desired velocity among all admissible ones , in a least square sense . we develop those principles in the microscopic and macroscopic settings , and we present how the framework of wasserstein distance between measures allows to recover the _ sweeping process _ nature of the problem on the macroscopic level , which makes it possible to obtain existence results in spite of the non - smooth character of the evolution process . micro and macro approaches are compared , and we investigate the similarities together with deep differences of those two levels of description . b. maury , a. roudneff - chupin , f. santambrogio j. venel

1101.4102

congestion phenomena in population dynamics cover a wide range of mechanisms , which can be classified according to their stiffness : 1 . soft congestion : as the distance between individuals becomes smaller , the behavior of a single person is affected by the presence of others ; 2 . hard congestion : actual contacts between individuals occur , and the overall motion is perturbed by the fact that two persons may not occupy the same place at the same time . soft congestion models can be included in both microscopic and macroscopic models . in the first class there are discrete - space models like cellular automata - based models which constrain pedestrians to be located at squares of a fixed grid @xcite . we can also mention models based on networks as route choice models @xcite or queuing models @xcite . a lot of evacuation softwares rest on such models as for example buildingexodus in @xcite . some microscopic models are space continous as the social force model introduced in @xcite and its forerunner which has been proposed by @xcite . in @xcite , people are identified to particles submitted to the laws of newtonian mechanics . as for macroscopic models , soft congestion is also usually favored . some of these models ( see for example @xcite , @xcite and @xcite ) own their origins in vehicular traffic models , and deal with the one - dimensional case . in higher dimension , similarly to the microscopic case , many models rely on social forces . one solution , used e.g. by bellomo and dogbe in @xcite and @xcite , or degond in @xcite , is to add to the equation satisfied by the velocity a repulsive term that forces people to avoid high density areas . another possibility is to modify directly the velocity by adding a term that make people deviate from their preferred path as soon as they approach a crowded area . the modeling of the velocity is the key point of these strategies , see for example the work of hughes @xcite , @xcite , coscia @xcite , or piccoli @xcite , @xcite . we aim here at addressing the particular issues pertaining to _ hard _ congestion . in this spirit , we shall make very crude assumptions on the social aspects triggering crowd dynamics ( see section [ sec : ext ] for possible improvements on these aspects ) , and simply consider that , at any time , a spontaneous ( or desired ) velocity field is given , and that it is purely _ reptilian _ : everyone s priority is to achieve his own goals , without accounting for others . actual behavior shall result from some sort of compromise between individual tendencies and congestion constraints . more precisely , we shall consider that the actual velocity is defined as the projection of the spontaneous one onto the set of feasible velocities ( i.e. which do not lead to a violation of the non - overlapping constraint ) . those basic principles can be applied to microscopic and macroscopic descriptions of the crowd . in the microscopic setting , the degrees of freedom are the positions of individuals ( identified to rigid disks ) , and the non - overlapping constraints can be written straightforwardly by prescribing a minimal value for the distance between centers . the problem takes the form of a differential inclusion which fits into the general framework of sweeping processes introduced by moreau in the 70 s ( see @xcite ) . this approach has been extended more recently to non convex cases , namely the case of uniformly prox - regular sets in @xcite , in @xcite and later in @xcite . the perturbed ( i.e. inhomogeneous ) problem has been studied in @xcite . as we shall see , it gives a natural framework for the microscopic model we propose . in the macroscopic setting , the crowd is seen as a population density and the non - overlapping constraint consists in prescribing a maximal value for this density . although both microscopic and macroscopic models express the same type of modeling assumptions , the mathematical structure of the macroscopic model is less obvious . expressed in the eulerian framework as a transport equation by the actual velocity field , which is defined as the projection of the spontaneous one onto the set of feasible velocities , it involves a conservative transport equation by a field whose regularity can not be controlled a priori , so that classical results can not be used . a first attempt to address those issues was proposed in @xcite , in the case where the spontaneous velocity is the gradient of a given dissatisfaction function ( typically the distance to the exit for the evacuation of a building ) : the framework of optimal transportation allows to reformulate the problem as a gradient flow in the wasserstein space of measures ( see e.g. @xcite for an overview of the theory of optimal transport , and @xcite for the notion of gradient flow in this setting ) . we show here that the sweeping process framework , which does not rely on any assumption on the gradient nature of the forcing term , can be extended in the macroscopic situation . stemming from similar principles , both microscopic and macroscopic models present nevertheless deep differences . for instance we shall point out the fact that the notion of maximal density is not properly defined in the microscopic setting . thus the macroscopic one can not be expected to be obtained from the microscopic one by any homogenization process . we shall focus here on the particular issues related to hard congestion . in order to highlight the very problems it raises , we will favor a very basic form of the crowd motion model . in particular we shall disregard in the main part of this paper any social or strategical aspects of human behavior . we are aware of the very crude character of these assumptions , and we gathered in section [ sec : ext ] other aspects of pedestrian behavior that could be included in our approach . besides , we hope that the framework presented here , and the issues we raise in attempting to establish links between microscopic and macroscopic settings , may be fruitful in other contexts , e.g. micro - macro issues in granular flows or modeling of chemotactic motion of large populations of cells . we consider a population of @xmath0 individuals , identified with rigid disks centered at @xmath1 , @xmath2 , , @xmath3 , with common radius @xmath4 ( it can be extended straighfowardly to the so - called _ polydisperse _ case , i.e. with different radii ) . we denote by @xmath5 the generalized spontaneous velocity of the crowd ( @xmath6 is the velocity which the individual @xmath7 would like to have in the absence of others ) . we assume here that the desired velocity of @xmath7 depends on his location only , and that this dependence is the same for all individuals ( see section [ sec : ext ] for more general behavioral models ) : @xmath8 where @xmath9 is a given field . the set of feasible configurations is defined as @xmath10 we disregard here obstacles ( like walls , furniture ) to alleviate notations , but they can be included in the definition of @xmath11 straightforwardly . non - overlapping is preserved by prescribing that @xmath12 may not decrease if it is @xmath13 , i.e.@xmath14 where @xmath15 . it leads to define the set of feasible velocities as [ eq : cqdef ] c_= . the model simply writes [ eq : pbmic ] = p_c _ ( ) where the projection is performed according to the euclidean norm over @xmath16 . + problem ( [ eq : pbmic ] ) is self consistent and drives the evolution process , as will be shown in section [ eq : pbmic ] . yet , the saddle - point formulation of the projection problem sheds light on the underlying darcy - like structure of the problem , and introduces a pressure field which will have a straightforward interpretation in terms of modeling , and which will have a continuous counterpart in the macroscopic setting . at some given configuration @xmath17 , let us denote by @xmath18 the set of couples @xmath19 , @xmath20 , such that @xmath7 and @xmath21 are in contact . constraints on the velocity can be written in a matrix form ( we drop the dependence of matrix @xmath22 upon the configuration ) @xmath23 where the rows of @xmath22 correspond to active constraints : @xmath24 the saddle - point formulation of the projection problem consists in finding @xmath25 ( where @xmath26 is the number of active constraints ) , such that [ eq : spmic ] \ { lcl + b^p & = & + b & & 0 . . supplemented by the complementarity condition @xmath27 pressure @xmath28 can be seen as the interaction force between individuals @xmath7 and @xmath21 . in the macroscopic setting the crowd is represented by a density @xmath29 , which we shall assume to be supported within some domain ( the room ) @xmath30 . in order to alleviate the notations , we shall present the model in the case of a closed room ( see the end of section [ sec : theomac ] for some details on the way an exit door can be accounted for in this framework ) . if we set at @xmath31 the saturation value , the set of feasible densities is @xmath32 the density is advected by the actual velocity @xmath33 ( macroscopic counterpart of the microscopic one @xmath34 ) @xmath35 and @xmath33 is defined as @xmath36 where @xmath37 is the set of feasible velocities . it can be defined unformally as the set of all those velocities which have a non - negative divergence on the saturated zone @xmath38 $ ] ( where the density may not increase ) . more precisely , it is defined by duality as @xmath39 \right \},\ ] ] @xmath40 the dual expression of @xmath37 induces a natural saddle - point formulation for the projection problem , based on a pressure field @xmath41 : find @xmath42 , where @xmath43\}$ ] , such that [ eq : sppmac ] \ { lcl + p & = & + - _ q & & 0 q h^1 _ ( ) , . with the complementarity condition : @xmath44 we address here the question of existence and uniqueness of solutions to @xmath45 where @xmath46 describes the motion of the crowd , @xmath47 is the cone of feasible velocities defined by ( [ eq : cqdef ] ) , and @xmath48 is the desired velocity . before stating the main results of this section , let us describe the general framework into which it fits , namely the sweeping processes , introduced by moreau in the late 70 s ( see @xcite ) . the original setting was the following : consider @xmath49 a hilbert space and @xmath50 a path of convex closed sets in @xmath49 , with some regularity in time ( e.g. @xmath51 is continuous for the hausdorff distance ) . he was interested in the evolution of a point @xmath52 subjected to remain in @xmath51 . assuming that the evolution process tends to minimize the norm of @xmath53 , one ends up with the following process @xmath54 where @xmath55 is the outward normal cone to @xmath11 at @xmath56 : [ eq : onc ] n_k(q ) = this normal cone has been introduced in @xcite and is called more precisely the proximal normal cone . note that , as @xmath11 is convex , this normal cone identifies to @xmath57 , the subdifferential of the indicatrix function of @xmath51 . the following discrete process ( so called _ catching up _ algorithm ) , which is used by moreau to establish well - posedness of the problem , gives a clear idea of the evolution mechanism . consider a time step @xmath58 , an initial condition @xmath59 , successive positions are built according to @xmath60 as @xmath61 is closed and convex , the projection is well - defined and contractant , which allows to establish convergence results for the sequence of piecewise constant solutions @xmath62 . it is clear that the recursive process extends straightforwardly to the case where the projection onto @xmath11 is properly defined ( i.e. single - valued ) into its neighborhood . the finite - dimensional sets satisfying this property were introduced by federer in @xcite under the name of _ positively reached sets_. then , they were called _ p - convex sets _ by canino in @xcite and later _ proximally smooth sets _ by clarke , stern and wolenski in @xcite . the final name `` uniformly prox - regular set '' will be given by rockafellar et al . in @xcite ( see definition [ def : pr ] below ) . the crowd motion model differs slightly from the sweeping process , as the feasible set is fixed whereas the point @xmath17 ( which represents the whole crowd ) tends to evolve according to some given velocity @xmath63 . the same principles can be extended straightforwardly , in particular : assuming @xmath63 is not too large , and @xmath11 is uniformly prox - regular , then for a sufficiently small time step @xmath64 , the catching up algorithm can be used to build discrete solutions . in the same manner , one may write the evolution process as [ eq : dqnmic ] - n_k ( ( t ) ) ( ( t ) ) . note that this new formulation is a direct consequence of ( [ eq : pbmic ] ) : as @xmath65 and @xmath47 are mutually polar , the identity operator can be written as @xmath66 ( see @xcite ) . equivalence between both formulations is not obvious , as an identity has been replaced by an inclusion , yet we shall see that it holds true under some conditions . the catching up algorithm reads as follows @xmath67 @xmath68 let us now express in a rigorous manner that this is indeed an algorithm for @xmath64 sufficiently small , and that this strategy allows to build a solution to our problem . let us first give a proper definition of prox - regularity ( see theorem 1.3 in @xcite ) . [ def : pr ] a closed set @xmath11 is said to be @xmath69-prox - regular at @xmath70 if there exists a neighborhood @xmath71 of @xmath70 such that for all @xmath72 and @xmath73 , with @xmath74 , we have @xmath75 where @xmath76 is the proximal normal cone to @xmath11 at @xmath56 ( defined by ( [ eq : onc ] ) ) . it is said to be uniformly prox - regular with a constant @xmath77 if it is @xmath69 prox - regular at every point of its boundary . as for our microscopic model , it holds k is uniformly prox - regular with @xmath78 proof . first it can be shown that @xmath79 is uniformly prox - regular . in this case , tools of differential geometry can be used to compute the constant of prox - regularity which corresponds to the smallest radius of curvature ( i.e. the largest eigenvalue of weingarten operator ) . after calculation , we show that @xmath80 is uniformly prox - regular with constant @xmath81 . one may wonder whether the intersection of such sets ( which is the case for @xmath11 ) is uniformly prox - regular with a constant depending only on the constants of prox - regularity of the smooth sets . from a general point of view , this is wrong as illustrated in figure [ fig : idfausse ] . indeed , we have plotted in solid line the boundary of a set @xmath82 which is the intersection of two identical disks complements . this set is uniformly prox - regular but its constant of prox - regularity ( equal to the radius of the disk plotted in dashed line ) tends to zero when the disks centers move away from each other . in this situation , the scalar product between the normal vectors @xmath83 and @xmath84 ( see fig . [ fig : psvectnorm ] ) tends to -1 , which suggests that the prox - regularity constant also depends on the angle between normal vectors . [ l]@xmath83 [ l]@xmath84 and @xmath84.,title="fig:",scaledwidth=20.0% ] consequently the proof of the uniform prox - regularity of @xmath11 rests on a good estimate of the angles between gradients of active constraints . the existence of such a constant @xmath85 relies on the positive linearly independence of gradients of active constraints . more precisely it can be shown that there exists @xmath86 such that for all @xmath87 , [ eq : revtriang ] _ ( i , j ) ( ) _ ij |_ij ( ) | | _ ( i , j ) ( ) _ ij _ ij()| . it can be deduced from this inequality that @xmath11 is @xmath85-prox - regular with @xmath88 ( a detailed proof of this result can be found in @xcite ) . furthermore , the given upper bound is obtained by considering a configuration @xmath17 of aligned disks and by computing the constant of local prox - regularity at @xmath17 ( see proposition 3.18 in @xcite for more details ) . note that the prox - regularity coefficient of @xmath11 degenerates ( i.e. goes to @xmath13 ) as the size of the disks goes to zero , for a constant total mass . furthermore the prox - regularity coefficient of @xmath11 depends on @xmath0 with @xmath89 for @xmath0 large enough ( see @xcite ) . now the following result can be established : let @xmath63 be lipschitz and bounded , for all @xmath90 and all @xmath91 , there exists a unique absolutely continuous solution @xmath92 to @xmath93 \vspace{6pt } \\ \qq(0)=\qq_0 . \end{array } \right.\ ] ] moreover , this solution satisfies the differential equation @xmath94 \vspace{6pt } \\ \qq(0)=\qq_0 . \end{array } \right.\ ] ] [ th : wpmicro ] the well - posedness of the differential inclusion can be proved thanks to results in @xcite . then proposition 3.3 in @xcite claims that the solution satisfies the following differential equation @xmath95 which is equivalent to ( [ eq : pbmic ] ) since the cones @xmath96 and @xmath47 are mutually polar . we are concerned here with the question of existence of solution to the macroscopic model : given a desired velocity field @xmath97 defined in @xmath98 , given an initial density @xmath99 , find @xmath100 such that [ eq : pbmactransp ] _ t + ( ) = 0 , where the actual velocity field @xmath33 verifies [ eq : pbmacproj ] = p_c _ , and @xmath37 is the set of feasible velocities defined by . . [ eq : pbmactransp ] is meant in a weak sense : @xmath101 the overall problem can be written as a non - local and nonlinear transport equation @xmath102 where the notation @xmath103 expresses a dependence of @xmath33 upon the whole density field @xmath29 , through the projection of @xmath63 onto @xmath37 . the nonsmooth character of this projection and the fact that the regularity of @xmath33 is not controlled ( the projection is performed in the @xmath104 sense ) , rule out the possibility to use standard tools to define solutions to this equation . one may wonder whether the catching - up approach , which proved successfull in the microscopic setting , can be followed in the present context . the prediction step is straightforward : one transports @xmath29 according to the desired velocity field @xmath63 during a time step @xmath58 . as for the correction ( or projection ) step , it appears immediately that the standard eulerian way to measure distances between densities ( by estimating a given norm of their difference ) is not suitable . in the microscopic setting , the catching up approach proved successful because the difference between two configurations identifies with a displacement field , and this is due to the very lagrangian description of individuals . recovering this feature calls for a new way to measure differences between densities , more respectful of the lagrangian description of the motion , and this way is the wasserstein setting . we assume here that @xmath98 is convex . let @xmath105 be a measurable mapping , and @xmath106 a probability measure . we say that @xmath107 pushes forward @xmath108 onto @xmath109 ( written @xmath110 ) if @xmath111 for any measurable set @xmath112 . for given measures @xmath113 and @xmath114 , we denote by @xmath115 the set of transport maps between @xmath113 and @xmath116 . the quadratic wasserstein distance @xmath117 is defined by @xmath118 notice that this is a sloppy definition that holds for atomless measures , which is anyway the case we are interested in ; for general measures one should pass through the so - called _ transport plans _ , which we do not want to introduce here , and we address the interested reader to @xcite . however , this quantity @xmath119 can be proven to be a distance on the space @xmath120 and it makes the space of probability measures a _ geodesic space _ , i.e. every pair of points is linked by a curve ( which is in this case a curve of measures ) , such that the length of this curve exactly equals the distance between the points . in case in the minimization defining @xmath119 there is an optimal transport map @xmath121 ( which is the case if @xmath113 is absolutely continuous with respect to the lebesgue measure @xmath122 ) , then this geodesic curve is known explicitly and it is given by @xmath123 . in the spirit of the structure that we are imposing on @xmath124 , we define the outward normal cone to a subset @xmath125 ( we actually apply the definition of the strong frchet subdifferential in @xcite to the indicatrix function of @xmath11 ) by @xmath126 for all @xmath121 such that @xmath127 . it makes it possible to express the problem like we did in the microscopic setting ( eq . ( [ eq : dqnmic ] ) ) : [ eq : dqnmac ] - n_k ( ) , and @xmath33 transports @xmath29 according to ( [ eq : pbmactransp ] ) . we may now write the catching up algorithm ( which we consider for the time being as a theoretical tool ) : given an initial density @xmath99 and a time step @xmath128 , build @xmath129 , , @xmath130 according to [ eq : cumac ] \ { rcll ^n+1 & = & ( + ) _ # ^n & + ^n+1 & = & p_k ( ^n+1 ) & . where the projection is performed in the wasserstein sense and the set @xmath11 is the constrained set introduced in . as for the microscopic model , we must check that both steps are well defined ( possibly under some restriction on the time step @xmath64 ) , and that the obtained discrete solutions converge to a limit which can be identified as a solution to problem ( [ eq : pbmactransp])([eq : pbmacproj ] ) . we shall assume here two properties of @xmath63 . first we require that it tends to keep people within the room @xmath98 , i.e. that for @xmath131 sufficiently small @xmath132 maps @xmath98 onto @xmath98 . obviously , in the case of an emergency evacuation , with an open exit door , we will of course alleviate this assumption by allowing @xmath63 to cross the exit line ( we will discuss later on how to `` catch up '' the possible part of the mass that exits @xmath98 because of this ) . secondly , we need it regular enough ( lipschitz continuous is sufficient ) , so that , still for @xmath131 sufficiently small @xmath132 preserves the absolute continuity of the measure , i.e. @xmath133 implies @xmath134 . this assumption being made , the prediction step is well defined for any measurable field @xmath63 . the key point is the projection step , which differs significantly from the microscopic situation . it is a variational problem in the space of measures , which consists in minimizing the distance to a fixed measure among elements of @xmath11 . as for existence , standard compactness arguments may be applied , but the question of uniqueness is more delicate . first of all , let us rule out the possibility to establish uniform prox - regularity in the spirit of definition [ def : pr ] . consider the one - dimensional situation . a dirac mass at zero projects onto the characteristic function of @xmath135 . more generally , a combination of dirac masses @xmath136 ( we assume that the distance between supports of any two of them is always larger than half the total mass carried by the couple , so that they do not interact ) projects onto the sum of characteristic function of the flattened dirac masses : @xmath137 the wasserstein cost can be computed straightforwardly , it is @xmath138 on the other way around , consider a density @xmath139 ( i.e. @xmath140 almost everywhere ) and @xmath141 the largest open set such that @xmath142 a.e . in @xmath141 . this open set @xmath141 is a countable union of open intervals of lengths @xmath143 ( with possibly @xmath144 ) . among all densities which projects onto @xmath11 at @xmath29 , the farthest is the combination of dirac masses with weights @xmath145 , supported at the middles of intervals , and it is at distance @xmath146 . as a consequence , @xmath69 prox - regularity can be expected at some points , but @xmath69 is not bounded away from @xmath13 . note that some measures are the projection of themselves only , even if they saturate the constraint at every point of their support : consider e.g. a dense open set of total measure @xmath31 in @xmath147 , and define @xmath29 as the characteristic function of its complement . as the saturated zone contains no interval , it can not be the projection of a measure which violates the constraint . note that this non prox - regularity can be seen as a natural consequence of the fact that the property in the microscopic setting degenerates as the granularity gets finer ( i.e. when the radius goes to @xmath13 ) . on the other hand , the set @xmath11 enjoys some kind of convexity property : it is geodesically convex ( see @xcite ) : @xmath148 is said to be geodesically convex if for any @xmath113 , @xmath114 @xmath149 , the geodesic curve @xmath150 joining @xmath113 and @xmath114 belongs to @xmath11 for all @xmath151 $ ] . this property might suggest that uniqueness of the projection can be obtained by standard arguments : in a so - called aleksandrov non - positively curved ( npc ) length space ( see @xcite ) , considering @xmath152 minimizers and the measure halfway along the geodesic , convexity properties of the distance function @xmath153 lead to a contradiction . unfortunately , as soon as @xmath154 , @xmath120 is not npc and the previous distance function even turns out to have _ concavity _ properties . yet , despite the previous assertions ( non prox - regularity and the concavity of the squared distance ) , other kind of interpolation curves ( called _ generalized geodesics _ ) can be used to assert well - posedness of the projection problem , without any restriction on the distance from @xmath11 . ( generalized geodesics , see @xcite , p. 207 ) a generalised geodesic joining @xmath113 to @xmath114 with base @xmath29 is a curve defined by @xmath155 where @xmath156 ( resp . @xmath157 ) is an optimal transport map from @xmath29 to @xmath113 ( resp . @xmath114 ) . notice that when @xmath158 this gives again the expression of a geodesic in @xmath124 according to the distance @xmath119 . this generalized notion can be used to establish the following proposition for any @xmath159 , the wasserstein distance to the set @xmath11 of admissible densities defined in is attained at a unique point @xmath160 . the projection operator @xmath161 is continuous . it is easy to check ( but one can refer to lemma 9.2.1 p.206 in @xcite ) , that the square wasserstein distance enjoys strict convexity properties along generalized geodesics , i.e. @xmath162 . this gives the uniqueness once we know that the set @xmath11 is still convex along these generalized geodesics , i.e. that @xmath163 implies @xmath164 . this is obtained by a computation of the jacobian factor of the map @xmath165 , see @xcite . now that the catching - up algorithm is defined properly , the obtained sequence of discrete solutions allows to obtain an existence result : let @xmath166 be the piecewise constant interpolation of the discrete densities given by the catching - up algorithm ( [ eq : cumac ] ) . if @xmath167 is a @xmath168 velocity field , and @xmath169 , then @xmath170 converges as @xmath64 tends to @xmath13 to a solution of the macroscopic problem @xmath171 let us first describe the discrete quantities we obtain thanks to the catching up algorithm . we denote by @xmath172 the optimal transport between the projected density @xmath173 and @xmath174 , and we write @xmath175 ( see figure [ fig : transports ] ) . these transport maps allow to define a discrete velocity as follows : @xmath176 [ l]@xmath177 [ l]@xmath172 [ l]@xmath178 [ l]@xmath174 [ l]@xmath179 [ l]@xmath180 $ ] we then define two different interpolations of these quantities . first , the piecewise constant interpolation is given by : @xmath181n \tau , ( n+1 ) \tau].\ ] ] using @xcite for the projection part , it is possible to prove that @xmath182 satisfies the following discrete decomposition @xmath183 where @xmath184 converges uniformly to @xmath13 as @xmath64 tends to @xmath13 . we also define a continuous interpolation of @xmath185 as follows @xmath186n \tau , ( n+1 ) \tau],\ ] ] where @xmath187 . this second interpolation satisfies the transport equation at velocity @xmath188 . it is possible to prove a priori estimates on these interpolated curves , and therefore prove that they both converge to the same limit . the continuous interpolation of @xmath185 gives that the limit density satisfies a transport equation , and the discrete decomposition of the piecewise constant interpolation proves that the limit velocity is indeed the projection of the desired velocity onto @xmath37 . we finish this section by considering the case where , due to modeling reasons , we allow the vector field @xmath63 to let the mass exit through a part of the boundary . this means that @xmath189 is no longer supposed to map @xmath98 into @xmath98 but the segment connecting @xmath190 to @xmath191 is allowed to cross @xmath192 in a prescribed subset @xmath193 which stands for the exit . in such a case the transport step of the algorithm does not need to be changed , but we have to face , during the projection step , a density @xmath194 which is no longer supported in @xmath195 . to bring back the mass to the original domain we want to consider a modified set @xmath11 , that we call @xmath196 : @xmath197 the reason for this choice is the following : we consider that as soon as a particle reaches @xmath198 , instead of following its movement after @xmath198 , we leave it on it . this is done for simplicity , but it only means that we are no longer concerned with what happens to the particles that have reached @xmath198 , not that they are really blocked on the exit . obviously , we needed to withdraw the density constraint on @xmath198 , so as to let particles stay on it , and also to represent the fact that @xmath198 stands actually for everything that happens at the door _ and beyond_. once the set @xmath196 is introduced , the projection step is done by defining @xmath199 which means that we take a measure whose support may go beyond @xmath200 and we project it onto the measures over @xmath98 satisfying some extra density constraint . we stress that the same kind of argument ( projecting on a set of measures concentrated on @xmath200 ) could also be used , in the case with no exit , to withdraw the ( yet , natural ) assumption that @xmath189 maps @xmath98 into @xmath98 . the mathematical problem with this modified @xmath196 is much trickier . one of the difficulties , that prevent the usual theory to be applied , is the fact that the set @xmath196 loses some of the properties that @xmath11 had previously ( in particular it is no more geodesically convex : the geodesic - for the @xmath119 distance - between two points of @xmath196 could go out of @xmath196 ) . in particular we have no clue about the uniqueness of the projection , but the algorithm works in the same way if we accept to take any minimizer of the distance . in case the spontaneous velocity is the gradient of some dissatisfaction function ( e.g. distance to the exit in case of an emergency evacuation ) , the microscopic model can be put into a gradient flow form : @xmath201 where @xmath202 is the frchet subdifferential of @xmath203 defined by @xmath204 the function @xmath205 that we consider is the total dissatisfaction , i.e. @xmath206 , where @xmath207 may be , as we said , the distance to the exit . indeed , as @xmath11 is uniformly prox - regular , the proximal normal cone identifies with the frchet subdifferential of @xmath208 hence @xmath209 ( see @xcite for more details ) . in the macroscopic case , under the same assumption that @xmath63 is a gradient @xmath210 , then one defines the global dissatisfaction function in a continuous setting : @xmath211 and the overall process can be seen as a gradient flow in the wasserstein space , i.e. @xmath29 is advected by @xmath33 , @xmath35 with @xmath212 where @xmath213 , and @xmath214 is defined in the following sense ( see @xcite ) : the strong frchet subdifferential of a function @xmath203 at @xmath29 is the set of fields @xmath215 such that for all transport maps @xmath216 , the following inequality holds true @xmath217 notice that the definition we gave before of @xmath218 is nothing but a particular case of this one , when @xmath219 . the main advantage of this gradient - flow formulation is the fact that it allows to handle vector fields @xmath63 less regular than what we need in the general case ( where it is natural to require @xmath63 to be lipschitz continuous , i.e. @xmath220 ) . for the whole theory on gradient flows , first in a finite - dimensional setting , then in hilbert spaces , and finally in metric spaces and particularly in @xmath120 , we refer again to @xcite . in particular , the catching up algorithms made by the coupling of a prediction and a correction step , may be replaced in the case of a gradient structure by a single - step procedure , called _ proximal algorithm _ , where @xmath221 this algorithm has also been formalized in a metric setting , where it is known as _ minimizing movements _ ( see @xcite ) : it has first been applied to the case of @xmath120 with the wasserstein metric by jordan , kinderlehrer and otto in @xcite . the details about a gradient flow approach to the macroscopic framework of crowd motion are contained in @xcite , where both the case with no exit ( i.e. with the constraint @xmath222 ) and with exit ( @xmath223 ) are dealt with , the second one being much trickier than the first . moreau s catching up algorithm suggests a strategy to build approximate solutions to the evolution problem . yet , projecting a configuration @xmath17 onto @xmath11 is not straightforward . the scheme we propose extends ideas introduced in @xcite for granular flows . it consists in performing a catching up step with @xmath11 replaced by some kind of inner local convex approximation . more precisely , considering that the configuration @xmath224 at time @xmath225 is known , the predicted configuration is obtained by @xmath226 then @xmath227 is obtained by projecting @xmath228 onto @xmath229 @xmath230 where @xmath231 stands for @xmath232 [ cols="^ " , ] in figure [ fig : projs ] , we illustrate the set @xmath233 , intersection of sets @xmath234 whose boundaries are plotted in solid line . the set @xmath235 is delimited by the dashed line . the theoretical and numerical projections , respectively @xmath236 and @xmath237 are represented ( for two examples of @xmath238 ) . indeed , as @xmath239 is uniformly prox - regular , the projection onto @xmath11 of @xmath240 is well - defined for @xmath131 small enough . the replacement of @xmath239 by the convex set @xmath241 is convenient because it allows us to use classical numerical methods to compute this projection . indeed , this projection problem can be reformulated in a saddle - point form which can be solved by uzawa algorithm : find @xmath242 satisfying \ { l ^n+1= ^n+1 + _ ij _ ij(^n ) + i < j , d_ij(^n)+_ij(^n ) ( ^n+1-^n ) 0 + _ ij ( d_ij(^n)+_ij(^n ) ( ^n+1- ^n)=0 . . [ eq : projs ] note that the lagrange multipliers are not unique in general . indeed there exist configurations @xmath243 with strictly more than @xmath244 active constraints so that the associated gradients @xmath245 are linearly dependent . the matrix appearing in uzawa algorithm is @xmath246 where @xmath247 is the matrix whose rows are vectors @xmath248 . in @xcite , the authors quantify how the condition number of matrix @xmath249 varies with the parameter @xmath250 ( setting a lower bound of the local prox - regularity of @xmath11 at point @xmath251 ) when @xmath249 is non - singular . as a consequence we expect that the uzawa algorithm converges less quickly for configurations with low local prox - regularity . in numerical simulations , we noticed indeed that solving the saddle - point problem requires more iterations in case of a jam . in the proposed scheme ( [ eq : nummicrocorr ] ) , the replacement of @xmath239 by the convex set @xmath241 is computationally convenient but arises many difficulties in the numerical analysis . the convergence result is based on the fact that @xmath252 is a good approximation of @xmath239 near @xmath17 . let us denote by @xmath253 the continuous piecewise linear function satisfying @xmath254 defined by ( [ eq : nummicrocorr ] ) . let @xmath63 be lipschitz and bounded , for all @xmath90 , the sequence @xmath255 uniformly converges to @xmath17 defined in theorem [ th : wpmicro ] , when @xmath131 tends to 0 . here we just give a sketch of the proof . we refer the reader to @xcite for the details . it can be easily proved that @xmath256 is bounded and so a convergent subsequence can be extracted . thanks to the uniqueness result in theorem [ th : wpmicro ] , it suffices to check that the limit function ( which is absolutely continuous ) satisfies the differential inclusion . since @xmath257 , it comes for all @xmath258 , latexmath:[\[(\qq^n+ \tau \uu(\qq^n)-\qq^{n+1 } ) \cdot ( \tilde \qq - \qq^{n+1 } ) \leq dividing by @xmath131 , we obtain for all @xmath260 @xmath261 represents the discrete actual velocity . we aim at passing to the limit in the previous inequality and the crucial term is the last one . thus the convergence result rests on the local continuity of the map @xmath262 in a neighborhood of the set @xmath263 . more precisely it can be checked that for all @xmath264 and all @xmath265 , @xmath266 { } p_{k_{\qq } } ( \tilde \qq)\ ] ] by using the saddle point form of these projections ( [ eq : projs ] ) . thanks to the reverse triangle inequality , the lagrange multipliers ( which are not unique in general ) are bounded . then compactness arguments allow us to obtain the convergence of the projections . we deduce that the limit function @xmath17 satisfies the following differential inclusion : @xmath267.\ ] ] as for all @xmath268 , @xmath269 , the required result is proved . the convergence order of this scheme will be specified in a forthcoming paper @xcite . this more accurate result is based on metric qualification conditions between sets @xmath270 . more precisely there exists a constant @xmath271 such for all @xmath264 and all @xmath265 , @xmath272 this result rests on the reverse triangle inequality ( [ eq : revtriang ] ) . the strategy we propose relies again on the catching up philosophy which already made it possible to establish an existence result . the density is first advected by the spontaneous velocity field ( with no consideration of the congestion constraint ) , and then projected ( in the wasserstein sense ) onto the set of feasible densities . the time discretization scheme writes as follows : given a time step @xmath128 , an initial configuration @xmath99 , approximate configurations @xmath129 , , @xmath130 are built recursively according to @xmath273 @xmath274 the first step consists in transporting @xmath130 according to the given transport map @xmath275 . we simply transport the center of each cell at velocity @xmath167 during a time @xmath64 , and then distribute the mass of the transported cell on the different cells of the mesh it intersects , as illustrated in fig . [ fig : lagrangien ] . the transported density writes @xmath276 where @xmath277 represents the area of the set @xmath22 . + [ l]@xmath278 [ l]@xmath279 [ l]@xmath280 the second step is less standard . the scheme we propose to approximate the projection onto @xmath11 with respect to the wasserstein distance is based on the following considerations : consider a domain @xmath141 and a non - zero density @xmath108 supported by @xmath281 . for @xmath282 , @xmath283 , and its projection onto @xmath11 is identically @xmath31 in @xmath141 . let us denote by @xmath284 the displacement field which corresponds to the optimal transport between @xmath285 and @xmath286 . one has @xmath287 as the latter is identically @xmath31 in @xmath141 , one has @xmath288 so that , at the first order in @xmath289 , @xmath290 as @xmath291 is an optimal map , @xmath33 is a gradient : @xmath292 , where @xmath41 verifies the poisson problem @xmath293 since we are considering a violation of the density constraint on a set @xmath141 and assuming that , after the projection , the density will be saturated on @xmath141 itself , we are only interested in finding the mass that will exit @xmath141 when the displacement is given by @xmath284 . this means , at a first order approximation , we only need to estimate the flux of density through the boundary @xmath294 , i.e. @xmath295 . now consider the stochastic interpretation of the poisson equation @xmath293 considering that @xmath108 is a probability measure , we consider a random variable @xmath296 which follows the law of probability with density @xmath108 . we now consider a brownian motion stemming from @xmath297 . its path crosses @xmath294 for the first time at @xmath298 . the random variable @xmath299 is known to follow the probability law with density @xmath300 on @xmath301 ( see @xcite ) . the idea is therefore to redistribute the exceeding mass of the saturated zone in the following way ( see fig . [ fig : proj_stoch ] ) : for each saturated cell , a random walk is started , which transports the exceedind mass @xmath302 , where @xmath303 is the measure of the cell . when this random walk encounters a non - saturated cell , it gets rid of as much mass as it can , and continues as long as the transported mass is not fully distributed . when all the saturated cells have been treated , the obtained density @xmath179 is admissible . + let us notice that the gradient flow structure when @xmath304 would also allow for another time - discretized algorithm , namely the proximal ( or minimizing movement , or jordan - kinderlehrer - otto ) one instead of the prediction - correction one . this would lead to a sequence of minimization problems involving the computation of the distance @xmath119 . the formulation by means of the transport plans , that we did not develop here for the sake of readability , transforms these problems into linear programming problems , that could a priori be solved through a simplex algorithm . yet , it turns out to be so slow that 2d problems could not be efficiently solved in this way . [ sec : num_an_macro ] the numerical analysis of the transport part of the algorithm is standard , but the projection part is more delicate , and we are not able to provide a rigorous error estimate for the stochastic scheme we propose . let us simply say here that , beyond the unformal justification of the approach given previously , the asymptotic behavior of the so - called diffusion limited aggregation ( dla ) process ( see @xcite ) , and the link between the distance based on the energy norm between measures ( see @xcite or @xcite ) and the wasserstein distance give some arguments to support the chosen strategy . both microscopic and macroscopic model express the same assumptions at their respective levels : a spontaneous velocity is given , and the system evolves according to a velocity which is the closest to the spontaneous one among all feasible velocities , in a least square sense . yet , the macroscopic model was not built from the microscopic one by a rigorous homogenization process . we describe here some obstacles to such a process , to shed light on the deep differences between both settings , in spite of formal similarities . + first of all , let us point out that the notion of maximal density is somewhat ambiguous as far as the microscopic model is concerned . let us consider the case of identical radii ( monodisperse situation ) . the maximal packing density for identical disks is known to be @xmath305 , and corresponds to the triangular lattice ( see fig . [ fig : triang ] ) . + yet the actual density of moving collections of rigid disks is generally strictly less than this maximal value , which is only attained for this very particular configuration . as an example , in the evacuation situation represented in fig . [ fig : densites_micro ] , the mean density upstream the exit door ranges typically between @xmath306 and @xmath307 . the fact that the actual density is strictly less than the maximal one does not mean that the flow is unconstrained ( as the macroscopic setting would suggest ) . those considerations call for a clear identification of configurations which saturate the constraint . such a notion is proposed for the three - dimensional situation in @xcite , in a nash equilibrium spirit : _ we say that a particle ( or a set of contacting particles ) is jammed if it can not be translated while fixing the positions of all of the other particles in the system . _ note that this property is defined as _ local jamming _ in it is tempting to consider as maximal in some sense any density corresponding to such configurations , for which there are no free disks , so that constraints are activated everywhere . the triangular lattice is clearly jammed , but so is the cartesian lattice ( @xmath308 ) , and it is possible to build looser jammed configuration ( see fig . [ fig : triangloose ] , with @xmath309 ) . + beyond this fuzzy definition of maximal packing , the set of feasible velocities in the microscopic setting is strongly dependent on the local structure and not only on the density , a feature which is lost in the macroscopic approach . let us first consider the triangular lattice represented in fig . [ fig : triang ] . having the number of disks going to infinity and the radius going to zero , the density ( characteristic function of the solid phase ) converges to the uniform density @xmath310 . microscopic feasible velocities can be defined in the solid phase , and one may wonder what are the corresponding feasible velocities in the macroscopic limit . such velocities are clearly constrained in 3 directions . using self - evident notations @xmath311 , @xmath312 and @xmath313 to design the principal directions of the lattice , one obtains at the limit some sort of unidirectional expansion constraints in those directions , which can be written @xmath314 it still allows for non - rigid motions at the limit : fig . [ fig : triangmove ] represents a feasible velocity field at the microscopic level , whose macroscopic counterpart is @xmath315 note that , although the microscopic velocity field is tangent to the boundary of @xmath11 , the corresponding macroscopic field is strictly expansive ( @xmath316 ) . + + the considerations which we presented above have a crucial consequence on the behavior of both approaches : the microscopic model has the ability to reproduce blocked jams ( which are observed in practice ) , whereas the macroscopic one does not . the latter property is a consequence of the maximum principle . let us consider the situation represented in fig . [ fig : jammac ] , with a saturated zone ( @xmath317 ) upstream the exit door . the desired velocity @xmath63 is assumed to point to the exit door , so that @xmath318 . the actual velocity is @xmath319 , where @xmath41 solves a poisson problem in the saturated zone , with homogeneous neumann condition on the walls , and homogeneous dirichlet boundary conditions at the exit and on the interface with the non - saturated zone . by virtue of the maximum principle , @xmath41 is nonnegative over the saturated zone , so that the velocity correction @xmath320 is non - negative on the door : people exit _ quicker _ than they would if there were no congestion ( they are always pushed out by other people behind them ) . as a consequence , if the desired velocity field tends to move people out of the room , the evacuation process will never stop before the room is empty . [ l]@xmath321 [ l]@xmath322 [ l]@xmath323 [ l]saturated zone the microscopic situation is quite different . the pressures are still nonnegative by definition of the saddle - point formulation ( [ eq : spmic ] ) , but they are likely to act in the `` wrong '' direction for the individuals which are the closest to the door , as soon as they form an arch ( see figure [ fig : jammic ] ) . + the precise way for setting convergence questions when passing from the microscopic to the macroscopic model needs to associate a density to every microscopic configuration . an easy way to do that is to associate to every non - overlapping disks configuration the uniform measure ( with unit density ) on the union of those disks . in this way we obtain a density @xmath29 obviously satisfying @xmath324 but we also know that , when we let the size of the disks go to zero ( with the number of disks increasing consequently to infinity ) we can not obtain _ any _ density @xmath324 in the macroscopic limit . in particular in dimension 2 it is impossible to go beyond @xmath325 , since we know that the maximal density is realized by the triangular lattice . this suggests that it is better to rescale the approximating @xmath29 by the maximal density , which is @xmath31 in dimension one , @xmath326 in dimension two now the question is the following : take a sequence of initial data @xmath327 , i.e. the densities associated ( in the way we described above ) to a sequence of microscopic configurations with @xmath0 non overlapping disks of radius @xmath328 . let @xmath329 and @xmath330 and consider the limit density @xmath99 . can we say that , for fixed velocity field @xmath331 , the constrained evolution stemming in the microscopic model from @xmath327 converges to that obtained in the macroscopic one from @xmath99 ? the answer can be expected to be positive if @xmath332 , but the situation is much more complicated for @xmath333 . actually , the unidimensional case gives no ambiguity between jammed and maximal density configuration , which is not the case in higher dimension . for instance one can consider a sequence of jammed cartesian lattices in a two - dimensional domain @xmath98 ( a square , for instance ) , with a concentrating vector field @xmath331 ( for instance @xmath334 ) . in this situation , the configuration is completely blocked , and the evolution gives , for any time @xmath335 , the constant density @xmath327 . on the other hand , the limit as @xmath329 is a constant density , but it does not activate the constraints . this is due to the fact that the density realized by the cartesian grid is strictly smaller than that of the triangular lattice , which was taken as a reference . hence , in the evolution , @xmath336 would differ from @xmath99 and move towards a more concentrated configuration . but this ( jammed configuration which are not of maximal density ) is not the only source of problems in the micro - macro limit . actually , one can consider a similar example with a sequence of triangular grids . in such a case , both @xmath327 and the limit @xmath99 activate and saturate the constraints . this induces some constraints on the admissible velocities but , as we saw in the paragraph devoted to these constraints , they act differently in the microscopic and in the macroscopic models . in particular , the limits of the admissible velocities for the microscopic problem are those vector fields which are `` unilaterally incompressible '' in the three main direction of the lattice , which is not at all the same as imposing only a positive divergence . the actual limit constrained space at the macroscopic level necessitates obviously to account for the microscopic structure of the contact network . beyond packing fraction , different types of _ order metrics _ have been introduced in @xcite to give a better description of the local structure ( see also @xcite for a recent review on this matter ) , but the use of those concepts for evolution problem is still widely open . this shows some deep differences , for @xmath333 , between the limit of the microscopic case and the macroscopic one . in the case of a gradient vector field @xmath63 we could express this issue , thanks to the gradient - flow formulation , in terms of the associated dissatisfaction functionals . some general results give conditions to translate a variation form of convergence on these functionals ( called @xmath337convergence ) into a convergence of the associated flows , but we do not want to insight this question in details because of its complexity ; it would go out of the scope of this paper and it is moreover matter of an ongoing work . we specify here how the model we presented can be improved in order to account for more elaborate individual behaviors than the purely reptilian one on which we built our approach . + we aim here at integrating the fact that individuals are likely to elaborate complex strategies to escape a building . for example , in case of congestion , they may decelerate or try to avoid the jam , instead of keeping pushing inefficiently . the velocity of a person becomes then dependent upon the position of people he can see in front of him . such a strategy can be defined as follows : we define the set @xmath338 ( see fig . [ fig : ni ] ) containing persons who are near and visible to the individual @xmath7 : @xmath339 where @xmath340 and @xmath341 . the constant @xmath342 is taken equal to the half angle of view ( i.e. @xmath343 ) . two choices are possible for the individual @xmath7 if his neighbours ( belonging to @xmath338 ) walk slower than him . first , he can decelerate instead of going through the crowd . in this case , his speed @xmath344 ( i.e. his velocity s norm ) at time @xmath345 is made dependent upon his neighbours behavior at time @xmath346 . more precisely , it is computed as a barycenter of his neighbours speeds , weighted by their relative positions . otherwise , he can also be in a hurry and prefer changing his way instead of slowing down . in that case , if there exists another clear way through the set @xmath347 , he follows it while keeping his desired speed , or else he will go round the group @xmath338 . among all directions allowing him to do so , he chooses @xmath348 the closest to the one he wanted ( see fig . [ fig : bypass ] ) . numerical simulations with these strategies are presented in @xcite . if every pedestrian prefers decelerating , there is typically no contact between people and rarefaction waves are observed . on the contrary , the avoiding strategy leads to situation where the constraints are activated . integrating strategies in the macroscopic setting is more delicate , as it is less natural to follow the motion of a single individual in the crowd . however , it is possible to model the influence of local density on the behavior of the crowd by changing the desired velocity . more precisely , it is natural to assume that when people arrive upstream a crowded area , they tend to decrease their desired velocity . in the spirit of @xcite , a two dimensional generalization of road traffic models can be integrated in the model : at time @xmath349 , people located at @xmath350 have a desired velocity @xmath351 with the same direction as the initial desired velocity @xmath352 and a norm that decreases when @xmath29 increases : @xmath353 where @xmath354 goes monotonically from @xmath31 to @xmath13 as @xmath29 goes from @xmath13 to @xmath31 ( saturation value ) . the evolution of the crowd then obeys the same principles as before , namely : @xmath355 it can be proved that the density transported by @xmath356 stays in @xmath11 : the projection step in the evolution equation becomes useless . yet , if @xmath357 ( people continue to push even at saturation density ) , the congestion constraint is likely to be activated , and the approach we propose in this paper can be adapted to this situation . note that , from a modeling standpoint , it is natural to use _ downstream _ information to determine the desired velocity ( people adapt their velocity according to what they see , i.e. to the density in front of them , which is downstream their desired direction ) , whereas , once the velocity is determined , the transport can be performed using an standard upwind scheme . we investigate here the possibility to account for different types of individuals , who might have different , and possibly antagonist , strategies . in the microscopic setting , because of the lagrangian character of the approach , this can be done without any change in the mathematical structure : it is sufficient to alleviate the assumption that the desired velocity depends on the position only , but can be defined differently depending of individuals . in the macroscopic setting , this is less straightforward , but it can be formulated in the following way : consider two populations @xmath31 and @xmath152 , with associated densities @xmath358 and @xmath359 and desired velocity fields @xmath360 and @xmath361 , we consider that in the saturated zone @xmath362 $ ] , desired velocities are perturbed by a common correction velocity @xmath363 which ensures that the constraint @xmath364 remains satisfied , and we define this velocity as the one that minimizes the @xmath104 norm . the model can be written @xmath365 where @xmath41 vanishes in the non - saturated zone @xmath366 $ ] , and verifies the complementarity condition @xmath367 note that , in case @xmath360 and @xmath361 are both gradients of dissatisfaction functions ( one for each population ) , the system presents a gradient - flow structure in the wasserstein setting ( see @xcite for a similar problem in the context of cell migration ) . we give here some hints on another possibility to define the actual velocity , in a nash equilibrium spirit . note that this approach is likely to change the mathematical nature of the evolution problem . in particular , the instantaneous velocity is no longer defined in a unique way , but as any ( among infinitely many ) equilibrium in the following sense : the actual generalized velocity @xmath368 is such that , for any @xmath7 , @xmath33 considered as a function of @xmath369 only minimizes the distance to @xmath63 among feasible velocities , i.e. @xmath370 where @xmath371 is the set @xmath372 as soon as two disks are in contact and push against each other , there are infinitely many solutions to this problem , as illustrated by the situation of fig . [ fig : mashmic ] : the desired velocity of person @xmath31 is @xmath31 ( in the horizontal direction ) , whereas person @xmath152 tends to stay still . in the previous approach ( @xmath373 projection onto the cone of feasible velocity ) , the actual velocity is @xmath374 for both . in the present approach , any diagonal couple @xmath375 , with @xmath376 $ ] , realizes an instantaneous nash equilibrium . [ c]@xmath31 [ c]@xmath152 in the macroscopic framework , the problem can be described as follows : for any @xmath377 , one prescribes that @xmath378 with @xmath379 as indicated previously , this approach changes the mathematical nature of the evolution problem , which calls for further developments . let us simply say that the solutions we built in the previous sections are particular solutions to this new problem , among infinitely many others . the first problem of the comparison between these models lies in the initial configuration : given an initial configuration for the disks in the microscopic setting , it is difficult to choose the macroscopic density that best fits this configuration . as long as uniform densities are considered , we adopt the following method : we first estimate the mean density in the initial microscopic configuration , and then normalize it with the maximal density of disks we encounter throughout the microscopic evolution . in fig . [ fig : init_max_density ] , we present an example of such a computation . ( on the left ) is normalized by the maximal density encountered @xmath380 ( on the right ) to obtain a macroscopic density of @xmath381 . ] ( on the left ) is normalized by the maximal density encountered @xmath380 ( on the right ) to obtain a macroscopic density of @xmath381 . ] we then let evolve both systems and compare the configurations at equivalent time steps . the formation of the saturated zones is quite the same in microscopic and macroscopic settings ( see fig . [ fig:5obs_01 ] ) . however , as pointed out in section [ sec : micro_macro ] , the behavior of the two models at the exit is very different , in particular the evacuation is faster in the macroscopic model ( see fig . [ fig:5obs_15 ] ) . in the microscopic model , the pressure exerted between people appears naturally as a lagrange multiplier , and is computed using usawa algorithm . the macroscopic pressure field does not appear explicitly in the numerical scheme , but it can be recovered by estimating during the projection step a discrete equivalent of the odometer function ( see section [ sec : num_an_macro ] ) : on each cell , we define the pressure as the total mass emitted by this cell during the stochastic projection . in fig . [ fig : pressions ] , we represented the microscopic pressure and the isovalues of the macroscopic pressure field upstream an exit door . in both settings , the pressure field is maximal in the middle of the saturated zone , and decreases at the entrance and the exit of this zone . let us finally present some examples which correspond to more realistic evacuations situations . the microscopic setting , as suggested in section [ sec : micro_macro ] , describes more properly situations where people leave through narrow exits . [ fig : bat_maths ] shows the evacuation of the first floor of a large building ( departement of mathematics at orsay ) . + the macroscopic setting , on the other hand , would best fit situations where many people have to evacuate a large domain . we present in fig . [ fig : stade ] the evacuation of the stade de france . even if the benches of the stadium are quite narrow passages , the macroscopic model gives results that correspond to real evacuation configurations . we would like to thank e. maurel - segala and l. thibault for fruitful suggestions and comments .

the main features of the cross sections of low - lying orbital excitations with in heavy deformed nuclei are studied in rpa on the example ofgd . the dependence of the dwba e2 and m1 cross sections on the scattering angle and incident electron energy mev is analyzed in pwba . the cross section is larger for m1 than for e2 transitions at any angle if mev . the longitudinal ( coulomb ) c2 excitation dominates the e2 response for . only transverse m1 and e2 excitations compete for and the former one is dominant for . the m1 response is almost purely orbital up to even in backward scattering . qualitative pwba estimates based on the-dependence of the form factors alone are not able to predict some important features of the cross sections stemming from the strong magnetic and orbital character of the studied 1 excitations . the expectation for m1 over e2 dominance in backward scattering should not be extended to higher momentum transfers and incident energies . 1.0 cm 1.0 cm -1.5 cm 16.0 cm 1.0 cm 1.0 cm 5.0 mm = cmbx10 scaled2

nucl-th9604001

the backward inelastic electron scattering has established itself in the last 30 years as one of the main tools for the experimental study of nuclear magnetic excitations . the attention was focused in the past mainly on spherical nuclei , as it can be seen , e.g. , from the review articles @xcite . the detailed study of low - lying magnetic dipole ( m1 ) excitations in heavy deformed nuclei started with @xmath0 experiments at the linear accelerator in darmstadt @xcite , reviewed recently in @xcite . most of the data were collected at a backward scattering angle @xmath13 , where one expects a strong damping of electric excitations . this can be understood from qualitative considerations @xcite in the plane wave born approximation ( pwba ) . the @xmath0 cross section can be decomposed in this case into longitudinal ( coulomb ) and transverse ( electric and magnetic ) terms , multiplied by corresponding kinematic factors . the longitudinal kinematic factor @xmath14 vanishes for @xmath15 and only transverse multipoles are excited in backward scattering @xcite . among those with the same multipolarity , the magnetic excitation is dominant over the electric transverse one @xcite . this qualitative estimate is obtained in the long - wave limit @xmath16 , or small momentum transfer @xmath11 in comparison with the radius @xmath17 of the target nucleus . magnetic dominance was found at backward angle also in @xcite by assuming a purely spin - flip transition . the more realistic distorted wave born approximation ( dwba ) @xcite leads to important corrections especially in heavy nuclei with a strong coulomb field ( a large charge number @xmath18 ) . but this approach is not suitable for simple qualitative predictions , since the longitudinal and transverse electric contributions to the @xmath0 cross section interfere with each other in dwba . moreover , their separation would be meaningless when the longitudinal and transverse electric contributions are related in the dwba formalism @xcite through the continuity equation . in contrast to the above discussed common expectations based on the pwba , we have found recently @xcite e2 contributions to the cross sections of low - lying orbital m1 excitations measured at @xmath13 . such admixtures take place in heavy deformed nuclei where the m1 transition with @xmath19 is accompanied by a closely lying e2 transition with @xmath20 to the first member of the rotational band built on the intrinsic @xmath1 excitation . the large moment of inertia of these nuclei leads to a small separation between the two excitation energies , comparable with the energy resolution of the @xmath0 experiments . the accompanying e2 transition provides an important contribution to the measured cross sections ( @xmath13 ) for an effective momentum transfer @xmath21 @xmath9 corresponding to incident electron energies in the range @xmath22 mev @xcite . the m1 excitations in spherical nuclei are mainly of spin - flip type , which is expected to be favoured in backward electron scattering @xcite . in contrast , we are considering here a qualitatively different type of predominantly orbital m1 excitations . they appear only in deformed nuclei @xcite and their experimental study through backward @xmath0 scattering started much later @xcite . we are going to examine here in more detail some of the well - known qualitative pwba predictions for the interplay of magnetic and electric excitations , paying attention also to the approximations involved . these predictions ignore the nuclear dynamics because they are made for the general case of unknown transition densities . the latter could exhibit however some important peculiarities , typical for the considered nuclear excitations . the qualitative pwba estimates will be compared here to dwba and pwba cross sections obtained from realistic rpa transition densities in order to understand to what extent one could use the general pwba predictions for qualitative interpretation of realistic microscopic results . we shall study to this purpose the dependence of our theoretical dwba @xmath0 cross sections on the electron incident energy and momentum transfer in the whole possible range of scattering angles @xmath23 . cross sections of 1@xmath12 excitations in deformed nuclei have been presented and discussed until now almost exclusively for @xmath13 in order to compare them with the corresponding experimental data measured at this angle . only the cross section of the strongest ( low - lying orbital ) m1 excitation is known experimentally in each one of the several heavy deformed nuclei studied until now . we choose here @xmath2gd as an example , because this is the only nucleus , where the accompanying e2 transition was identified experimentally @xcite . we shall present and discuss in the next section the dependence of e2 and m1 cross sections of this typical 1@xmath12 excitation on the incident electron energy and momentum transfer in the whole range of scattering angles . the results are analyzed qualitatively in pwba for small and large momentum transfer in sections 3 and 4 , respectively . the conclusions are summarized in sect . 5 and expressions for e2 transition densities are given in the appendix . we describe intrinsic excitations with @xmath25 in deformed nuclei within the quasiparticle random phase approximation ( qrpa , or shorter rpa ) using a model hamiltonian specified in @xcite . it contains a quasiparticle ( q.p . ) mean field , given by a deformed woods - saxon potential plus pairing . the separable residual interaction consists of a spin - spin force and a quadrupole - type interaction , derived self - consistently from the deformation of the mean field . the isoscalar coupling constant of the latter interaction , determined microscopically , ensures the rotational invariance of the model hamiltonian , which is violated by the mean field alone . the choice of the parameters of the model hamiltonian is explained in more detail elsewhere @xcite . in contrast to many other microscopic calculations , energies of single - particle levels are not shifted but taken exactly as provided by the deformed woods - saxon potential . expressions for m1 and e2 transition probabilities in terms of the rpa amplitudes are given in @xcite . the large single - particle basis allows us to work without effective charges @xmath26 when calculating @xmath27 values and e2 transition densities . the @xmath28 values and m1 transition densities are obtained with bare orbital and effective spin gyromagnetic factors . the latter account for the renormalization of the spin operator by a factor of 0.7 , i.e. , @xmath29 the strongest m1 transition found experimentally in @xmath2gd has an excitation energy @xmath30 mev and a reduced magnetic transition probability @xmath31(m1 ) = 1.30 @xmath32 @xcite . it is easily identified with the strongest theoretical m1 transition at 2.90 mev with @xmath31(m1 ) = 1.24 @xmath33 . our theoretical reduced probability for the accompanying e2 transition , @xmath31(e2 ) = 42 e@xmath34@xmath35 , agrees also well with the corresponding experimental value 40 @xmath36 e@xmath34@xmath35 @xcite . the @xmath31(e2 ) values of @xmath25 excitations are small because the main collectivity of the e2 strength of this mode is concentrated in the spurious rotational state , i.e. in the lowest e2 transition of the ground state rotational band . the @xmath0 cross sections are , however , much more sensitive to the details of the rpa wave functions than transition probabilities or excitation energies . we obtain the transition densities , necessary for the calculation of dwba cross sections , as reduced matrix elements of the corresponding multipole density operators between the nuclear ground state and the @xmath37-th rpa excitation with @xmath38 : @xmath39 where @xmath40 and @xmath41 are longitudinal charge ( c2 ) and transverse current ( t@xmath42 ) density operators , respectively . the latter reduce to @xmath43 for m1 excitations ( @xmath44 ) and @xmath45 for e2 excitations ( @xmath46 ) . the transverse ( current ) transition densities @xmath47 are written in ( [ jden ] ) as a sum of two terms originating from the proton convection current density @xmath48 and the magnetization ( or spin ) current density @xmath49 of protons and neutrons . the current density operators and expressions for the m1 transition densities in rpa can be found in @xcite . expressions for the e2 transition densities are given in the appendix . the @xmath0 cross sections are calculated numerically from the transition densities ( [ rhoden ] ) , ( [ jden ] ) , using the dwba formalism described in @xcite . the theoretical dwba cross sections of the above described low - lying orbital 1@xmath12 excitation in @xmath2gd are displayed in figs . [ fig1 ] and [ fig2 ] . the m1 and e2 @xmath0 cross sections are plotted in fig . [ fig1 ] versus the incident electron energy in the range @xmath50 mev and scattering angles @xmath51 . since we are interested here mainly in qualitative conclusions , the agreement with experiment is relevant to the extent it could promote the confidence in the theoretical cross sections beyond the experimentally studied scattering angle of 165@xmath52 . two - dimensional sections of the three - dimensional plots in fig . [ fig1 ] , corresponding to a fixed angle @xmath53 and @xmath54 mev , were plotted together in @xcite and compared with experimental data @xcite , taken at this angle . the e2 excitation provides important contributions to the measured m1 cross section in the region @xmath22 mev and leads to a very good agreement with experiment after being taken into account . these features can be seen also by comparing the shaded areas ( @xmath55 ) of the m1 and e2 cross sections in fig . [ fig1 ] . for small incident energies @xmath56 the e2 cross section is two orders of magnitude smaller than the m1 cross section but both are comparable at higher @xmath56 . at 175@xmath52 the two cross sections are comparable for @xmath57 mev and the e2 response is dominant for @xmath58 mev . we are going to analyze qualitatively our microscopic dwba cross sections using pwba , where the cross section can be decomposed into a sum of longitudinal and transverse terms @xcite : @xmath59 \bigr \ } , \label{pwba } \\ v_{\ell } = { { \rm cos}^2 ( \theta /2 ) \over 4 { \rm sin}^4 ( \theta /2 ) } , \ ; \ ; v_t = { 1 + { \rm sin}^2 ( \theta /2 ) \over 8 { \rm sin}^4 ( \theta /2 ) } , \quad v_t = v_t / v_{\ell } = { 1 \over 2 } + \tan ^2 \left ( { \theta /2 } \right ) . \label{vlt}\end{aligned}\ ] ] @xmath18 and @xmath60 are the nuclear proton number and the unit charge , respectively . the factor @xmath61 from @xcite is incorporated in our definitions @xcite of the form factors @xmath62 ( [ pwba ] ) , coinciding with those of @xcite . only the term @xmath63 survives in ( [ pwba ] ) in the case of @xmath64 transitions , while the @xmath65 excitations are specified by longitudinal ( coulomb ) @xmath66 and transverse electric @xmath67 contributions . the transferred momentum has in our notations the inverse dimension of length , @xmath11 ( @xmath9 ) . the pwba formalism is used in this paper for two different purposes . first , in addition to the dwba results , cross sections are calculated in some cases also in pwba with the same rpa transition densities in order to separate the longitudinal c2 from the transverse e2 response and compare them with the transverse m1 cross sections . this separation is not possible in dwba . hence , the theoretical pwba cross sections are used only as a tool for studying different constituents of the cross sections and their origin . secondly , dwba and pwba cross sections , both obtained from realistic rpa transition densities , are compared to qualitative pwba predictions which ignore completely the nuclear dymanics . these estimates are based on general considerations of the scattering kinematics and the momentum dependence of the unknown form factors . on the other hand , the latter are fully specified in our formalism by the microscopic transition densities ( [ rhoden ] ) , ( [ jden ] ) . it is seen from fig . [ fig1 ] that for small incident energies the @xmath64 cross section remains almost constant when going from backward to forward angles . it starts to increase for @xmath68 but , contrary to what should be expected from the diverging transverse kinematic factor @xmath69 ( [ vlt ] ) , the @xmath64 cross section does not diverge in the limit @xmath70 . the expressions ( [ vlt ] ) , derived in the relativistic limit ( neglected electron rest mass ) , must be corrected for extreme forward or backward angles and acquire finite values , as we shall see in the next section . the dwba cross section of the accompanying @xmath65 transition , displayed in fig . [ fig1 ] , decreases substantially in the backward direction ( large @xmath71 ) for small incident energies . this is an indication for a dominant coulomb response which is damped by the longitudinal factor @xmath72 at backward angles . dwba cross sections are plotted usually @xcite versus the effective momentum transfer @xmath73 , @xmath74 , \quad q = { 2 \over \hbar c } \sqrt{e_i e_f } { \rm sin } ( \theta /2 ) = { 2 e_i \over \hbar c } { \rm sin } ( \theta /2 ) \sqrt{1 - ( e_x /e_i ) } , \label{qeff}\ ] ] in order to be compared with pwba cross sections plotted versus @xmath11 . in the above expression @xmath75 is the energy of the outgoing electron . the definition ( [ qeff ] ) coincides with the formula for @xmath73 in @xcite used to plot the experimental cross sections when a radius constant @xmath76 fm @xcite is assumed in ( [ qeff ] ) for the equivalent nuclear radius @xmath77 . the form factors @xmath62 ( [ pwba ] ) are fourier transforms of the transition densities ( [ rhoden ] ) , ( [ jden ] ) with bessel functions @xmath78 @xcite . therefore , the explicit dependence of the theoretical m1 and e2 cross sections on the momentum transfer is also of prime interest . it is plotted in fig . [ fig2 ] versus @xmath73 ( [ qeff ] ) and the scattering angle @xmath71 . the indented edges in fig . [ fig2 ] result simply from the adopted finite grid of mesh points . the graph cut - off at the edge corresponds to the maximal incident energy @xmath79 mev included in calculations . let us rewrite the kinematic factors in a form that is more appropriate for a constant @xmath11 . we neglect the small term @xmath80 in ( [ qeff ] ) and express @xmath81 in ( [ pwba ] ) through @xmath11 : @xmath82 , \quad { v_t \over e_i^2 } = { 1 \over 2 q^2 } \bigl [ \sin ^{-2 } ( \theta /2 ) + 1 \bigr ] , \nonumber \\ d \sigma_{\ell } ( \theta ) = d \sigma_{\ell } ( 90^{\circ } ) \cot ^2 ( \theta /2 ) , \quad d \sigma_t ( \theta ) = d \sigma_t ( 90^{\circ } ) { 1 \over 3 } \biggl [ { 1 \over \sin ^2 ( \theta /2 ) } + 1 \biggr ] . \label{vltq}\end{aligned}\ ] ] these expressions give the angular dependence of longitudinal and transverse cross sections for a fixed momentum transfer @xmath11 . it is seen from ( [ vltq ] ) that for a given @xmath11 the transverse cross section decreases only 33% between 90@xmath52 and 180@xmath52 . in contrast , the longitudinal cross section decreases in comparison with its value at 90@xmath52 : 58 times at 165@xmath52 , where most experiments were done , and 5835 times at the largest accessible backward angle of 178.5@xmath52 @xcite . thus , the almost constant m1 cross section in fig . [ fig2 ] for a given momentum transfer is a typical behaviour of transverse excitations , while the steep decrease of the e2 cross section towards backward angles is an indication for a predominant longitudinal ( coulomb ) component . the low-@xmath11 limit @xmath83 is reached in the considered heavy nuclei with @xmath84 fm for very small momentum transfer @xmath85 @xmath9 . the pwba cross section ( [ pwba ] ) is obtained in the relativistic limit , i.e. the electron rest mass @xmath86 mev is neglected in comparison with the incident electron energy @xmath56 and a small excitation energy @xmath87 is assumed in comparison with the momentum transfer . the former condition is always satisfied in our calculations because @xmath88 mev . the latter condition is violated at small angles , where one has to use the more accurate expressions provided by eqs . ( 4 - 16c ) and ( 6 - 38 ) of ref . @xcite : @xmath89 , \quad \delta ^2 = q^2 - \bigl ( e_x / \hbar c \bigr ) ^2 , \label{vtd } \\ \mbox{for } \theta = 0^{\circ } : \ ; \ ; v_{\ell } = 4 \biggl ( { e_i \over e_x } \biggr ) ^4 , \ ; \ ; v_t = { 2 e_i^6 \over e_x^4 ( m_ec^2)^2 } , \ ; \ ; v_t = { 1\over 2 } \biggl ( { e_i \over m_e c^2 } \biggr ) ^2 . \label{vtzero}\end{aligned}\ ] ] the electron rest mass is still neglected in ( [ vtd ] ) , which can not be used for @xmath90 since the four - momentum is @xmath91 and the transverse factor diverges . after taking the electron mass into account , one obtains @xmath92 and the finite values ( [ vtzero ] ) in the forward direction . the transverse cross section is strongly dominant for any realistic incident energy @xmath93 , as seen from ( [ vtzero ] ) . the limit ( [ vtzero ] ) is however not necessary even for a very small but different from zero angle ( a fraction of the degree ) because the expression ( [ vtd ] ) provides already a very good numerical accuracy @xcite . the smallest angle displayed in the three - dimentional graphs is @xmath94 . it is seen from ( [ vtd ] ) that @xmath95 is a very large number for small angles where @xmath11 is close to the photon point @xmath96 and @xmath97 is small . one can understand in this way the behaviour of the m1 and e2 cross sections at small scattering angles , exhibited in fig . the m1 cross section is dominant in the forward direction , in line with the general expectations @xcite for suppression of the longitudinal response throught its small kinematic factor at small angles . but this is no more true for higher momentum transfer . the factor @xmath95 ( [ vtd ] ) that damps the longitudinal ( coulomb ) excitation is large at @xmath98 only for small momentum transfer . however , eq . ( [ vtd ] ) reduces already to the simpler estimate ( [ vlt ] ) @xmath99 for @xmath79 mev . the coulomb response is favoured in this case and the e2 cross section is comparable in magnitude with the m1 cross section . the interplay between longitudinal and transverse e2 components in the forward direction can be seen in more details in fig . [ fig3 ] , where the pwba e2 cross sections are plotted versus the incident electron energy for different small scattering angles . the pwba cross sections are obtained from the same realistic rpa transition densities of the considered 1@xmath12 excitation in @xmath2gd , which were used to obtain the dwba results presented in figs . [ fig1 ] and [ fig2 ] . it is seen from fig . [ fig3 ] that the transverse e2 cross section ( dotted lines ) is almost constant and independent of the incident energy @xmath56 , while the coulomb cross section ( dashed curves ) is increasing quadratically with @xmath56 . the e2 cross section is purely transverse for @xmath100 and 0.5@xmath52 in the whole range ( @xmath101 mev ) of incident energies studied . in contrast , the coulomb term dominates for @xmath102 mev at @xmath103 and for @xmath104 mev at @xmath94 . its strong increase with @xmath56 in the latter case is seen also in the dwba e2 cross section plotted in fig . this behaviour can be understood by rewriting the pwba cross section ( [ pwba ] ) with the help of ( [ qe ] ) and ( [ vltq ] ) : @xmath105 \bigr \ } , \nonumber \\ w_{\ell } = \cot ^2 ( \theta /2 ) , \quad w_t = { 1 + \sin ^2 ( \theta /2 ) \over 2 \sin ^2 ( \theta /2 ) } , \quad w_t / w_{\ell } = v_t . \label{pwbq}\end{aligned}\ ] ] the @xmath11-dependence of the form factors @xmath62 ( [ pwbq ] ) is estimated in pwba by taking the leading terms ( lowest @xmath42 ) of the corresponding multipole operators into account @xcite . the bessel functions are approximated afterwards in the long - wave limit ( small @xmath11 ) as @xmath106 , so that their contribution to the form factors is @xmath107 @xcite , where @xmath17 is the nuclear radius . the transverse operators have convection and spin parts ( [ jden ] ) . it turns out that both parts exhibit the same leading @xmath11-dependence for magnetic , but not for electric operators . the choice of willey @xcite for the leading electric term is objected by berall @xcite who notes that only the spin term was considered in @xcite . willey has , however , both terms in view when deciding to neglect @xmath108 in favour of @xmath109 , where @xmath110 is the nucleon mass . thus , the choice of willey will be correct for @xmath111 , where @xmath112 and the excitation energy @xmath113 must be supplied in mev . we obtain @xmath114 @xmath9 for the low - energy orbital m1 excitations with @xmath115 mev , studied here . therefore , in our case we agree with the choice of berall for the leading convection electric term in the low-@xmath11 limit , while the choice of willey remains valid for the spin current where the problem of two competing leading terms does not arise . after introducing the above discussed correction , we obtain in the low-@xmath11 limit from the expressions ( 2.64 ) for multipole operators in ref . @xcite the following estimates for the @xmath11-dependence of the pwba form factors ( [ pwbq ] ) : @xmath116 upon insertion of ( [ fflow ] ) into ( [ pwbq ] ) , we get the following estimates for the pwba cross sections in the low-@xmath11 limit : @xmath117 where @xmath118 is defined in ( [ qnot ] ) . the expression for the magnetic cross section applies to both convection and spin current contributions . although the above estimates ( [ sigl ] ) are in agreement with previous results @xcite , the @xmath11-dependence in ( [ sigl ] ) is qualitatively different . this can be seen by comparison , e.g. with table xxiv of @xcite . the difference arises from the @xmath11-dependence of @xmath81 ( [ pwba ] ) , expressed explicitly in ( [ pwbq ] ) through ( [ qe ] ) . the previous pwba estimates for the @xmath11-dependence of the cross sections are valid only for a fixed incident energy , i.e. they are meaningful when the angular dependence of the form factor is studied for a constant incident energy . the scattering angle is however fixed in modern experiments @xcite and the form factor is studied by measurements for different incident energies . one has to apply in this case the estimates ( [ sigl ] ) . they are more general since they are valid also for a fixed incident energy . the angular dependence is given in this case by the kinematic factors @xmath119 ( [ pwbq ] ) . we obtain from ( [ sigl ] ) the following estimates for the considered m1 and e2 pwba cross sections by choosing the low-@xmath11 limit of the convection e2 excitation : @xmath120 turning back to the inspection of fig . [ fig3 ] , we see from ( [ sig21 ] ) that the quadratic @xmath11-dependence of the coulomb e2 excitation ( dashed curves ) agrees with the pwba prediction ( [ sig21 ] ) . moreover , it becomes clear that the constant transverse e2 cross section ( dotted lines ) must originate from a dominant convection current , because the transverse spin current would exhibit a strong @xmath121-dependence . let us note that the previous estimates @xcite would not be able to explain the constant transverse cross section of fig . [ fig3 ] , because they predict @xmath122 and @xmath123 dependencies for the convection and spin e2 cross sections , respectively . the following qualitative estimates for the relative transverse contributions are obtained directly from ( [ sigl ] ) : @xmath124 where @xmath118 is defined in ( [ qnot ] ) and a nuclear radius @xmath125 fm was assumed for the rare - earth region to evaluate numerically the latter relationship in ( [ sme ] ) , valid when the excitation energy @xmath113 is given in mev . the ratios ( [ see])-([sme ] ) should hold for any scattering angle . they apply to both the convection and spin parts of the magnetic excitation @xmath126 . it is seen from ( [ see ] ) that the convection electric cross section dominates over the spin electric one since it is taken from ( [ sigl ] ) just in the low-@xmath11 limit @xmath127 , necessary to ensure its dominance . hence , the longitudinal cross section has to be compared only with the transverse magnetic and convection electric ones : @xmath128 after inserting @xmath129 from ( [ vtzero ] ) in the above relationships and taking into account that @xmath130 at the photon point , one can easily verify that the cross section is dominated by transverse excitations at @xmath131 and the longitudinal contribution is negligible : @xmath132 where the same numerical estimate was used in the last relationship as in ( [ sme ] ) . this particular case of extreme forward scattering was discussed already above . let us check now the predictions ( [ smespi])-([scm ] ) by comparison with realistic results . the pwba cross sections displayed in fig . [ fig4 ] are plotted versus the transferred momentum @xmath11 for @xmath133 . this angle ensures @xmath134 , as seen from eq . ( [ vlt ] ) , i.e. a kinematic condition which is equally favourable for both longitudinal and transverse excitations . the pwba plots represent therefore the form factors @xmath62 ( [ pwba ] ) , ( [ pwbq ] ) up to a coefficient common for all of them . the cross sections are calculated with our microscopic rpa transition densities for the strongest low - lying orbital 1@xmath12 excitation in @xmath2gd . the ratio between the orbital and spin transition matrix elements of the m1 operator @xcite is @xmath135 = 7.8 for this excitation , i.e. it has a predominant orbital nature . the pwba m1 cross section is displayed as a continuous curve in the top plot of fig . [ fig4 ] , together with the orbital contribution alone ( dashed curve ) . this corresponds to the decomposition of the current transition density @xmath136 ( [ jden ] ) into convection ( orbital ) and magnetization ( spin ) parts . it is seen that the m1 cross section is almost purely orbital in the considered @xmath11-range with negligible spin contributions . this is a typical property of the studied orbital 1@xmath12 excitations in heavy deformed nuclei @xcite . the predominant orbital nature at the photon point ( see @xmath135 above ) is preserved up to @xmath137 @xmath9 , as seen from fig . [ fig4 ] . this typical feature , determined by the nuclear dynamics , is not present in the qualitative pwba estimates ( [ fflow ] ) , ( [ sigl ] ) , where the nuclear orbital , spin , and charge transition matrix elements are assumed to be all of comparable magnitude . the transverse e2 cross section should be one order of magnitude larger than the m1 cross section , as predicted by ( [ sme ] ) for the considered 1@xmath12 excitations with @xmath115 mev . comparison of the upper two plots in fig . [ fig4 ] invalidates this pwba prediction because the m1 cross section is two orders of magnitude larger than the transverse e2 contribution . this is observed also in the plots of fig . [ fig5 ] , where the m1 cross section is larger than the e2 contribution for small scattering angles . the e2 cross section arises in this case mainly from the transverse e2 excitation , as one can check from fig . [ fig3 ] for small angles and small incident energies . let us verify now the predictions ( [ sce ] ) , ( [ scm ] ) . it is seen from the middle plot of fig . [ fig4 ] that for small @xmath11 the pwba coulomb ( c2 ) cross section ( continuous curve ) is more than one order of magnitude larger than the transverse e2 cross section ( dashed curve ) , as it should be expected from ( [ sce ] ) for @xmath115 mev . comparison of the upper two plots in fig . [ fig4 ] shows that the m1 response dominates over the c2 cross section only for small momentum transfer @xmath138 @xmath9 ( @xmath5 mev ) . this crossing point , where the m1 and c2 cross sections are equal , is confirmed also by more realistic dwba results , displayed in the bottom plot of fig . [ fig4 ] ( note that they are plotted versus @xmath73 ) . on the other hand , eq . ( [ scm ] ) predicts that the crossing should take place for @xmath139 @xmath9 or @xmath140 mev , which is impossible since the incident energy must be larger than the excitation energy @xmath115 mev . therefore , the pwba prediction ( [ scm ] ) overestimates the c2 with respect to the m1 cross section by three orders of magnitude . in order to study this problem in more detail , we compare in fig . [ fig5 ] m1 and e2 dwba cross sections plotted versus the scattering angle . they are sections of the two plots in fig . [ fig1 ] for different fixed values of the incident electron energy @xmath56 . it is seen from fig . [ fig5 ] that the m1 cross section ( continuous curves ) is larger than the e2 cross section ( dashed curves ) in the whole kinematical region for small incident energies @xmath5 mev . above this energy the e2 response is dominant for almost any angle , except for very small ( forward ) or very large ( backward ) angles . one can check that the crossing point of the two curves at small angles is characterized by @xmath141 for different incident energies . thus , the qualitative estimate ( [ scm ] ) is not confirmed and the m1 response is dominant up to a much higher momentum transfer than predicted by pwba . small scattering angles are not favourable for the study of nuclear excitations through inelastic electron scattering because of the large background originating from the radiative tail of the strong elastic peak @xcite . on the other hand , a technique was developed for precise measurement of the inelastic cross section relative to the elastic peak , which can not be applied for backward angles @xcite . there are many other sources of radiative background , contributing at 180@xmath52 as well . among them , the magnetic bremsstrahlung is particularily strong @xcite . we turn our attention to the backward direction where most of the experiments on magnetic excitations have been done . let us check the estimates ( [ smespi])-([scm ] ) in the case of backward scattering where the dominant longitudinal contribution is strongly damped by the kinematics . the expression ( [ vlt ] ) for the kinematic factor @xmath95 diverges at 180@xmath52 . its correct limiting behaviour @xcite can be obtained from eqs . ( 4 - 12b , c ) of @xcite by taking the electron rest mass in the expression for @xmath72 into account . the transverse factor @xmath69 ( [ vlt ] ) does not need such a correction at backward angles , so that , @xmath142 where @xmath143 is given by ( [ vtzero ] ) . the upper index @xmath144 is to remind that the electron mass was taken into account . the corrected factors ( [ vlpi ] ) differ from the rough estimate ( [ vlt ] ) only for small incident energies and angles close to 180@xmath52 . one obtains from ( [ sce ] ) , ( [ scm ] ) and ( [ vtpi ] ) for a full backward angle : @xmath145 it is seen from ( [ scepi ] ) that the c2 and m1 cross sections should be of comparable magnitude even in fully backward scattering . the dominant contribution arises however from the transverse convection e2 response . according to ( [ scepi ] ) , this contribution should be one order of magnitude larger than the longitudinal c2 cross section for the considered excitation energies @xmath115 mev . the above qualitative pwba estimates have been made without disposing of any information about the dynamics of the considered nuclear excitations . let us compare them with our pwba results obtained from the realistic rpa transition densities of the studied orbital 1@xmath12 excitations in heavy deformed nuclei . the pwba cross sections are plotted in fig . [ fig6 ] for @xmath146 . the m1 cross section arising from the convection current alone ( dashed curve ) is shown in the upper plot together with the total ( convection plus spin , continuous curve ) m1 cross section . it is seen that the m1 response is almost purely orbital for small momentum transfer , but the spin m1 contributions become larger for @xmath147 @xmath9 . the pwba e2 cross sections , displayed in the lower plot of fig . [ fig6 ] , demonstrate that the c2 response is already negligible at this large angle . the transverse e2 cross section is almost two orders of magnitude smaller than the m1 cross section for small momentum transfer @xmath11 . these results , obtained from a realistic rpa wave function , stand in contrast with the general pwba expectations ( [ scepi ] ) where the nuclear dynamics is not taken into account . according to ( [ scepi ] ) , the transverse e2 excitation arising from the convection current should dominate over the m1 and c2 responses for small @xmath11 in backward scattering . the displayed cross sections are calculated with the simple kinematic factors ( [ vlt ] ) but the above conclusions are not altered when the more precise longitudinal factor @xmath148 ( [ vlpi ] ) is used instead of @xmath72 to calculate the c2 cross section from the bottom plot of fig . the two expressions differ considerably only for small incident energies and angles very close to 180@xmath52 . but even for @xmath146 and realistic incident energies @xmath149 mev the largest discrepancy ( for @xmath150 mev ) is not very relevant . the correct longitudinal factor @xmath148 ( [ vlpi ] ) is about 40 times larger than the rough value ( [ vlt ] ) , but @xmath148 is still 1270 times smaller than the transverse factor @xmath69 . thus , also with the correct factor @xmath148 the situation characterized by a negligible longitudinal response is not altered : the c2 contribution is still one order of magnitude smaller than the e2 cross section for small @xmath11 ( @xmath151 mev ) . the intensity of the scattered electrons decreases several orders of magnitude in the backward direction but this kinematical region is more interesting because the maximal momentum is transferred there for a given incident energy , as seen from ( [ qe ] ) . one can investigate in this way the high-@xmath11 region , far from the photon point which could be studied also with more precise methods as photonuclear reactions @xcite . the contribution of the bessel functions to the form factors ( [ pwba ] ) are approximated in the high-@xmath11 limit @xcite by @xmath152 . this corresponds to @xmath153 @xmath9 for the considered rare - earth nuclei . let us assume additionally that @xmath154 @xmath9 , see eq . ( [ qnot ] ) , so that the leading convection and spin terms in the transverse e@xmath42 and m@xmath421 form factors have the same @xmath11-dependence . the estimates ( [ fflow ] ) acquire now the simpler form : @xmath155 one obtains from ( [ fflar ] ) and ( [ pwbq ] ) the following estimates for the pwba cross sections for large momentum transfer : @xmath156 where ( [ vtpi ] ) and ( [ qeff ] ) were used to get ( [ sclpi ] ) . if we choose an angle where @xmath157 , the ratio ( [ sclar ] ) is always a large number since @xmath158 = 4.75 @xmath9 . this is indeed the condition for validity of the non - relativistic treatment of the nucleus @xcite . one can conclude in this way from the estimates ( [ emla ] ) , ( [ sclar ] ) , that the transverse e2 and m1 cross sections should be of comparable magnitude , while the longitudinal c2 excitation should dominate over them for not very large backward angles . the prediction ( [ sclar ] ) is confirmed in fig . [ fig5 ] where the e2 response ( mainly longitudinal c2 ) dominates over m1 for large @xmath11 ( @xmath104 mev ) apart from very large backward angles . this qualitative conclusion is confirmed also in fig . [ fig4 ] , where the relative magnitudes of the form factors are clearly seen due to @xmath134 . the longitudinal c2 dominates over the transverse e2 and m1 responses in fig . [ fig4 ] for large transferred momentum . let us remind that the c2 dominance over the transverse e2 excitation , predicted by ( [ sce ] ) for low @xmath11 , was confirmed in the previous section , i.e. this relationship is valid for a wide region of transferred momenta . the longitudinal dominance over the transverse currents is usually considered as a signature for collective excitations @xcite . one should note , however , that this dominance is predicted by eqs . ( [ sce ] ) , ( [ scm ] ) and ( [ sclar ] ) , which are not restricted by any assumptions about the nature or extent of collectivity of the nuclear transition . on the other hand , collective excitations have been associated usually with the assumption for an irrotational , incompressible flow @xcite , whose signature is the vanishing transverse electric current @xmath159 ( [ mdop ] ) . our rpa transition densities for the transverse e2 currents with @xmath160 and @xmath161 are of comparable magnitude ( not in favour of collectivity according to the latter criterium ) but their amplitude is one order of magnitude smaller than the amplitude of the c2 transition density ( in favour of collectivity according to the former criterium ) . on the basis of different considerations we came to the conclusion @xcite that the low - lying orbital 1@xmath12 excitations represent a weakly collective scissor mode . in contrast to the validity of ( [ sclar ] ) , the relationship ( [ emla ] ) , which should hold for any angle , is not confirmed by realistic results . decomposition is made in figs . [ fig4 ] and [ fig6 ] for 70.53@xmath52 and 179.5@xmath52 , respectively . it is seen from these two figures that for large @xmath11 the transverse e2 excitation ( mainly convection ) is almost two orders of magnitude smaller than the m1 response ( also mainly convection ) . it was found in the previous section that the corresponding prediction ( [ sme ] ) for this ratio at low @xmath11 contradicts our realistic results in the same way , but the discrepancy is three orders of magnitude for small momentum transfer . the condition of m1 dominance over the c2 response for full backward scattering and for large @xmath11 is obtained from ( [ sclpi ] ) : @xmath162 or @xmath163 mev . this estimate is confirmed in fig . [ fig6 ] , where pwba cross sections for @xmath164 are displayed . they are obtained with realistic rpa transition densities . the lower plot shows that the transverse e2 contribution dominates over the c2 cross section . it is seen from the comparison of the two plots that also the m1 response dominates over the c2 excitation , as expected from ( [ sclpi ] ) . the upper plot demonstrates that the m1 response is almost purely orbital for @xmath165 @xmath9 : it originates mainly from the convection current ( dashed curve ) . the spin m1 current becomes more important above this momentum transfer . this is due to the fact that the convection transition density is peaked close to the nuclear surface ( at @xmath166 fm ) and almost vanishes for @xmath167 fm , while the spin transition density has an appreciable amplitude also deep inside the nucleus , a region reached at high momentum transfer . it has been argued @xcite that the spin - flip m1 response should dominate in backward scattering when the electron rest mass can be neglected in comparison with its incident energy and the transferred momentum is much larger than the nuclear excitation energy . these approximations are reasonable in backward scattering for not very small incident energy . in this case the electron can be viewed as a massless particle with a good helicity @xcite , i.e. it is longitudinally polarized and the electron spin is aligned along or opposite to its momentum . the scattering to 180@xmath52 is considered in ref . @xcite as an occurrence of spin - flip , so that the longitudinal and transverse convection parts of the interaction should be strongly damped and only transverse spin m1 and spin e2 interactions will compete . the ratio of the corresponding cross sections is given by the pwba estimates ( [ smespi ] ) for small @xmath11 and ( [ emla ] ) for large @xmath11 . hence , if only transverse spin interactions had to compete with each other , a dominant spin m1 response could be expected from the above helicity arguments only for small @xmath85 @xmath9 where the necessary high - momentum approximation is not justified very well . moreover , so small @xmath11-values are not reached in the @xmath0 experiments @xcite investigating m1 excitations in heavy deformed nuclei , where @xmath168 mev or @xmath169 @xmath9 in backward scattering . the spin m1 and e2 cross sections should be almost equal at high @xmath11 according to ( [ emla ] ) . inspection of fig . [ fig6 ] demonstrates that in backward scattering ( negligible longitudinal response ) the convection m1 cross section is much larger than the transverse e2 contribution even for large momentum transfer up to @xmath170 @xmath9 . in this way , both the qualitative pwba prediction ( [ emla ] ) and the above helicity arguments for a dominant spin m1 response contradict the realistic results for the considered orbital m1 transition . the equality of the transverse m1 and e2 cross sections is predicted in ( [ emla ] ) to take place for any @xmath153 @xmath9 while it is reached in reality for a much higher momentum transfer @xmath170 @xmath9 . it is true that the longitudinal interaction is strongly suppressed in backward scattering . this is easily seen from the small longitudinal kinematic factor @xmath148 ( [ vlpi ] ) . it is also true that high - energy electrons are longitudinally polarized . but even in the pwba treatment of backward scattering of longitudinally polarized electrons both convection and spin currents contribute on equal footing to the cross section @xcite . let us compare finally the m1 and e2 dwba cross sections plotted in fig . [ fig7 ] versus the effective momentum transfer @xmath73 ( [ qeff ] ) for different scattering angles . the m1 cross section is larger than the e2 contribution in the whole region of incident energies studied ( @xmath4 mev ) for small forward angles @xmath171 ( not shown in the figure , but seen in fig . [ fig1 ] ) . at larger scattering angles there is always a crossing point for a given momentum transfer , beyond which the e2 response is dominant . the @xmath11-value of the crossing point obeys the relationship discussed in the previous section . it is seen from fig . [ fig7 ] that the large @xmath11-values lie beyond the crossing point where the e2 excitation ( mainly longitudinal ) is dominant over m1 . however , the crossing point moves towards higher @xmath11-values when approaching the backward direction , so that the m1 response becomes more important even for moderately large momentum transfer . the two cross sections decouple only at 175@xmath52 . at this angle the m1 excitation dominates up to @xmath172 @xmath9 , while the c2 and the transverse e2 responses are of comparable magnitude . the c2 contributions are negligible beyond 175@xmath52 and the total cross section is determined only by the relative transverse m1 and e2 responses which are multiplied by the same transverse factor @xmath69 ( [ vlt ] ) . cross sections of lower multipolarity decrease faster for larger @xmath11 @xcite . according to ( [ sigl ] ) , the cross sections obey the following order of increasing multipolarity in their momentum dependence : @xmath173 , where the high-@xmath11 limit , common for convection and spin transverse electric responses , is chosen . in addition to this general trend , our rpa results show that the studied orbital excitations have typical convection m1 transition densities similar to the charge c2 transition density : they are characterized by a well - localized single bump . in contrast , the spin m1 and transverse e2 transition densities exhibit more oscillations . hence , the spin m1 response decreases less for larger @xmath11 in comparison with the convection m1 term , i.e. the spin response behaves effectively as a higher multipolarity with respect to @xmath11 . one should expect , therefore , that the transverse e2 response will dominate over the transverse m1 excitation for very large transferred momenta . this can be seen in the two bottom plots of fig . [ fig7 ] for @xmath174 @xmath9 . thus , the transverse e2 response will be dominant even in a fully backward scattering for very large momentum transfer . we have studied the @xmath0 cross sections of low - lying excitations with @xmath1 in heavy deformed nuclei . they have a predominantly orbital nature , in contrast to the more extensively explored spin - flip m1 excitations in spherical nuclei . moreover , we have found recently @xcite that the accompanying e2 transitions with @xmath20 provide appreciable contributions to the measured m1 cross sections in heavy deformed nuclei even in backward scattering where the m1 response is expected to dominate . we obtain the cross sections from realistic microscopic rpa transition densities . the low - lying 1@xmath12 excitation with the strongest orbital m1 transition in @xmath2gd is studied as a typical example for such excitations which have been interpreted as a weakly collective scissors mode @xcite . the already reported good agreement @xcite of the theoretical dwba cross sections with the available experimental data ( @xmath175 ) confirms the importance of the accompanying e2 transitions in heavy deformed nuclei and places some confidence in the validity of the theoretical cross sections for different scattering conditions . the dependence of the theoretical m1 and e2 dwba cross sections upon the scattering angle @xmath71 and the incident electron energy @xmath56 is studied here in the whole kinematical region @xmath176 for @xmath50 mev . in some special cases the cross sections are calculated also within pwba in order to separate longitudinal from transverse contributions ( not possible in dwba ) . the dwba and pwba cross sections , obtained from realistic rpa transition densities , are compared to qualitative pwba predictions which ignore completely the nuclear dynamics . they are based on general considerations of the scattering kinematics and the momentum dependence of the unknown form factors . the latter are fully specified by our rpa transition densities which is the main advantage of the microscopic approach . \i ) for a given momentum transfer and @xmath177 the ( purely transverse ) m1 cross section is almost constant with @xmath71 , while the e2 cross section decreases fast in the backward direction . this effect is simply due to the interplay between the longitudinal and transverse kinematic factors . \ii ) for small momentum transfer @xmath11 and a fixed scattering angle @xmath71 the transverse convection m1 and e2 cross sections are almost independent of @xmath11 , while the longitudinal c2 cross section increases as @xmath122 . the convection current is dominant in the transverse e2 response for small @xmath11 . \iii ) the longitudinal c2 ( coulomb ) form factor is much larger than the transverse m1 and e2 form factors . therefore , the c2 response is dominant if the longitudinal kinematic factor is not very small . this condition is met away from the extreme forward and backward directions . even for very small angles the coulomb suppresion is no more effective at high incident energies because of the momentum dependence in the kinematic factor . \iv ) for scattering angles close to 180@xmath52 , where the coulomb response is negligible , the transverse e2 excitation dominates over the m1 transition at high transferred momenta . this is due to the fact that cross sections of lower multipolarity decrease faster with the increase in @xmath11 . the longitudinal dominance over the transverse currents is usually considered as a signature for collective excitations . however , this dominance is predicted in pwba by eqs . ( [ sce ] ) , ( [ scm ] ) and ( [ sclar ] ) , whose derivation does not involve any assumptions about the nature or extent of collectivity of the nuclear transition . the pwba predictions are not able to provide reliable numerical estimates for the limiting values of angles , incident energies and transferred momenta which specify the validity range of different predicted relationships . the qualitative pwba estimates are either not able to predict or contradict the following realistic microscopic results for low - lying orbital 1@xmath12 excitations : \v ) the ratio between the c2 and m1 cross sections is three orders of magnitude smaller than estimated in pwba . the pwba predictions for a strong coulomb response could be more appropriate for rotational transitions but not for the orbital m1 excitations considered here . thus , a more pronounced m1 dominance is present in our results at small momentum transfer , especially in backward scattering . the m1 response is dominant at any angle for incident energies @xmath5 mev . it is dominant in the forward direction , e.g. for @xmath178 at @xmath103 and even up to larger energies at smaller angles . \vi ) the longitudinal c2 excitation is comparable with the transverse e2 response at @xmath179 but becomes negligible beyond this angle in the backward direction . for such large angles the total cross section is determined by the interplay between transverse m1 and e2 excitations . \vii ) in backward scattering ( @xmath180 ) the transverse m1 response is stronger than the transverse e2 excitation up to @xmath170 @xmath9 . this feature stands in contrast to the qualitative pwba predictions for a dominant transverse e2 transition in both low- and high-@xmath11 regions ( the latter refers to @xmath153 @xmath9 ) . the two cross sections are comparably large in the region @xmath181 @xmath9 and the transverse e2 cross section is expected to dominate for @xmath182 @xmath9 ( @xmath58 mev in backward scattering ) . \viii ) the m1 response originates mainly from the convection current up to @xmath137 @xmath9 . the spin m1 current becomes more important above this momentum transfer due to the volume character of the oscillating spin m1 transition density . the above properties apply to the low - lying 1@xmath12 excitations in heavy deformed nuclei studied here . they are mainly orbital with small spin contributions . such typical properties can not be predicted without information about the wave function of the considered nuclear excitation . though high - energy electrons can be viewed as longitudinally polarized massless particles with a good helicity , there is no suppression of the convection current interaction in backward scattering . this is confirmed by the studied orbital excitations where the convection current provides the main contribution to the m1 response up to a high momentum transfer . thanks are due to henk p. blok and jochen heisenberg for providing us with their dwba code and useful communications on the underlying fromalism . this work is supported by the deutsche forschungsgemeinschaft ( dfg ) under contracts fa 67/15 - 1 and fa 67/15 - 2 . the transition densities ( [ rhoden ] ) , ( [ jden ] ) are calculated with the rpa wave functions in the laboratory frame in the way described in @xcite for m1 transitions . in the case of e2 transitions they have the form : @xmath183 , \label{rhod } \\ { \cal j } _ { 2l^{\prime } } ^{\nu } ( r ) = { 1 \over 2 } \sum _ { 0 < i < f } \bigl [ f^{\nu } _ { -1 } ( fi ) { \cal j } _ { 2l^{\prime}1 } ( fi , r ) + f^{\nu } _ { -1 } ( f\tilde{i } ) { \cal j } _ { 2l^{\prime}1 } ( f\tilde{i},r ) \bigr ] , \label{jd } \end{aligned}\ ] ] where the summation runs over single - particle states of the woods - saxon potential with projections @xmath184 on the nuclear symmetry axis @xmath185 obeying @xmath186 and time - reversed states @xmath187 are treated explicitly . the nuclear dynamics is contained in the factors @xmath188 , which are linear combinations of the rpa forward- and backward - going amplitudes @xmath189 and @xmath190 @xcite , respectively : @xmath191 , \label{fup } \\ f^{\nu } _ { \sigma } ( fi , m ) = \psi _ { \nu } ( fi , m ) + \sigma m \ , \phi _ { \nu } ( fi , m ) , \label{fsig}\end{aligned}\ ] ] the signature index @xmath192 , corresponding to the indistinguishable @xmath193 and @xmath194 directions , is of technical interest only . it allows us to take symmetries into account and to avoid the problem of symmetrizing the momentum operator @xcite . final results do not depend on the signature , as seen from ( [ rhod ] ) , ( [ jd ] ) . in order to save space we skip in ( [ fup ] ) , ( [ fsig ] ) and below expressions involving time - reversed initial states @xmath187 , though we always take such terms in the numerical calculations into account . the notations @xmath195 in ( [ rhod ] ) and @xmath196 in ( [ jd ] ) stand for the q.p . matrix elements of the e2 charge and current density operators @xmath197 and @xmath198 , respectively : @xmath199 @xmath200 are the vector spherical harmonics @xcite . the nuclear charge and current density operators @xmath201 and @xmath202 can be found , e.g. in @xcite . the q.p . matrix elements are obtained with the help of the eigenfunctions @xmath203 of the deformed woods - saxon potential @xcite in cylindrical coordinates ( @xmath204 ) : @xmath205 where @xmath206 , are the coefficients of the expansion over the basis wave functions @xmath207 of the axially symmetric harmonic oscillator . the cylindrical quantum numbers @xmath208 , correspond to the directions along and perpendicular to the symmetry axis , while @xmath209 and @xmath210 are the projections of the orbital angular momentum and spin operators on this axis . the angular integration over @xmath211 in ( [ rhod ] ) , ( [ jd ] ) , reduces to integration over @xmath212 since the integration over the azymuth angle @xmath213 and the spin matrix elements can be performed analytically in the basis ( [ wswf ] ) . one obtains in this way the following expressions for the q.p . matrix elements of the charge transition density ( [ rhod ] ) : @xmath214 where the coefficients @xmath215 are linear combinations of the bcs occupation numbers . the functions @xmath216 contain only @xmath217 from ( [ wswf ] ) . thus , they depend only on @xmath185 and @xmath218 , i.e. they are products of hermite and associate laguerre polynomials @xcite : @xmath219 the q.p . matrix elements of the transverse e2 convection currents have the form : @xmath220 + { \rho \over r } { \partial \over \partial z_i } a^{fi } ( \lambda ^{\prime } + 1 , \lambda ^{\prime } ) \biggr \ } d \ > { \rm cos } \theta , \label{jc21}\end{aligned}\ ] ] @xmath221 a^{fi } ( \lambda ^{\prime } + 1 , \lambda ^{\prime } ) \nonumber \\ - { z \lambda^{\prime } \over r \rho } a^{fi } ( \lambda ^{\prime } + 1 , \lambda ^{\prime } ) \biggr \ } d \ > { \rm cos } \theta , \label{jc23}\end{aligned}\ ] ] where the factors @xmath222 are defined in ( [ afi ] ) . the notations @xmath223 and @xmath224 mean that only the wave functions of the initial state in @xmath216 must be differentiated . the transverse e2 magnetization currents give rise to the following q.p . matrix elements : @xmath225 \nonumber \\ + 2 { \partial \over \partial z } a^{fi } ( \lambda + 1 , \lambda + 1 ) \bigr ] \nonumber \\ + { \rho \over r } \bigl [ { \partial \over \partial \rho } \bigl [ a^{fi } ( \lambda + 1 , \lambda + 1 ) - a^{fi } ( \lambda + 2 , \lambda ) \bigr ] + { 2 \over \rho } a^{fi } ( \lambda + 2 , \lambda ) \bigr ] \biggr \ } d \ > { \rm cos } \theta , \label{js21}\end{aligned}\ ] ] @xmath226 \nonumber \\ + { ( z^2 - 4 \rho ^2 ) \over r^3 } \bigl [ \rho { \partial \over \partial \rho } \bigl [ a^{fi } ( \lambda + 2 , \lambda ) - a^{fi } ( \lambda + 1 , \lambda + 1 ) \bigr ] \nonumber \\ - 2 a^{fi } ( \lambda + 2 , \lambda ) + { z \over \rho } \bigl [ a^{fi } ( \lambda + 1 , \lambda ) - a^{fi } ( \lambda + 2 , \lambda + 1 ) \bigr ] \bigl ] \nonumber \\ - { z \over r^3 } \bigl [ \bigl ( 2z^2 - 3 \rho ^2 \bigr ) { \partial \over \partial z } a^{fi } ( \lambda + 1 , \lambda + 1 ) + 5 \rho ^2 { \partial \over \partial z } a^{fi } ( \lambda + 2 , \lambda ) \bigr ] \biggr \ } d \ > { \rm cos } \theta , \label{js23}\end{aligned}\ ] ] where @xmath227 are the spin gyromagnetic ratios ( [ grat ] ) . all differentiations in ( [ jc21])-([js23 ] ) are performed analytically @xcite using recurrency relations for hermite and laguerre polynomials . the numerical integration in ( [ rhome ] ) and ( [ jc21])-([js23 ] ) extends over @xmath228 . 99 l. w. fagg , rev . 47 ( 1975 ) 683 s. raman , l. w. fagg and r. s. hicks , internat . phys . 7 ( 1991 ) 355 d. bohle , a. richter , w. steffen , a. e. l. dieperink , n. lo iudice , f. palumbo and o. scholten , phys . b137 ( 1984 ) 27 a. richter , nucl . a522 ( 1991 ) 139c ; a553 ( 1993 ) 417c ; progr . nucl . phys . 34 ( 1995 ) 261 w. c. barber , ann . 12 ( 1962 ) 1 w. c. barber , j. goldemberg , g. a. peterson and y. torizuka , nucl . 41 ( 1963 ) 461 h. berall , electron scattering from complex nuclei ( academic press , new york , london , 1971 ) j. heisenberg and h. p. blok , ann . nucl . part . 33 ( 1983 ) 569 j. heisenberg , adv . 12 ( 1981 ) 61 r. nojarov , a. faessler and m. dingfelder , j. of phys . g20 ( 1994 ) l111 m. dingfelder , r. nojarov and a. faessler , progr . 34 ( 1995 ) 319 r. nojarov , progr . nucl . phys . 34 ( 1995 ) 297 d. bohle , a. richter , k. heyde , p. van isacker , j. moreau and a. sevrin , phys . 55 ( 1985 ) 1661 r. nojarov , a. faessler and m. dingfelder , phys . c51 ( 1995 ) 2449 r. nojarov and a. faessler , nucl . a484 ( 1988 ) 1 d. bohle , a. richter , u. e. p. berg , j. drexler , r. d. heil , u. kneissl , h. metzger , r. stock , b. fischer , h. hollick and d. kollewe , nucl . a458 ( 1986 ) 205 a. faessler , r. nojarov and t. taigel , nucl . phys . a492 ( 1989 ) 105 m. petraitis , j. p. connelly , hall crannell , l. w. fagg , j. t. obrien , d. i. sober , j. r. deininger , s. e. williamson , r. lindgren and s. raman , phys . c49 ( 1994 ) 3000 r. s. willey , nucl . 40 ( 1963 ) 529 r. raphael and h. berall , phys . rev . 143 ( 1966 ) 671 h. theissen , springer tracts in mod . phys . 65 ( 1972 ) 1 u. kneissl , j. margraf , h. h. pitz , p. von brentano , r .- d . herzberg and a. zilges , progr . part . nucl . phys . 34 ( 1995 ) 285 a. i. ahieser , a. g. sitenko and v. k. tartakovskii , elektrodinamika yader ( naukova dumka , kiev , 1989 ) p. 177 l. j. weigert and m. e. rose , nucl . 51 ( 1964 ) 529 d. a. varshalovich , a. n. moskalev and v. k. khersonskii , quantum theory of angular momentum ( world scientific , singapore , 1988 ) j. damgaard , h .- c . pauli , v. v. pashkevich and v. m. strutinski , nucl . a135 ( 1969 ) 432 r. nojarov , z. bochnacki and a. faessler , z. phys . a324 ( 1986 ) 289

we present optical photometry of one type iin supernova ( 1994y ) and nine type ia supernovae ( 1993y , 1993z , 1993ae , 1994b , 1994c , 1994 m , 1994q , 1994ae , and 1995d ) . sn 1993y and sn 1993z appear to be normal sn ia events with similar rates of decline , but we do not have data near maximum brightness . the colors of sn 1994c suggest that it suffers from significant reddening or is intrinsically red . the light curves of sn 1994y are complicated ; they show a slow rise and gradual decline near maximum brightness in and numerous changes in the decline rates at later times . sn 1994y also demonstrates color evolution similar to that of the sn iin 1988z , but it is slightly more luminous and declines more rapidly than sn 1988z . the behavior of sn 1994y indicates a small ejecta mass and a gradual strengthening of the h emission relative to the continuum .

astro-ph0108299

supernova ( sn ) light curves reveal the structure of the stellar progenitors and reflect the underlying energy sources created in the explosion . they have also been the primary tool for intercomparison of individual events and selection of subsamples of supernovae ( sne ) for use as distance indicators . direct comparisons of peak luminosities and multiband light curves with the predictions of theoretical models lead to a better understanding of supernova explosions . useful reviews of the optical light curves of sne are given by doggett & branch ( 1985 ) , kirshner ( 1990 ) , wheeler & harkness ( 1990 ) , patat et al . ( 1994 ) , leibundgut ( 1996 ) , and suntzeff ( 1996 ) , among others . we have been engaged in systematic optical monitoring of bright , relatively nearby sne . these include hydrogen - rich objects ( type ii ) , as well as the various kinds of hydrogen - deficient sne ( types ia , ib , and ic ) . ( for a summary of the spectroscopic properties of sne , see filippenko 1997 . ) here we report our results for nine sne ia and one peculiar sn ii , from data obtained in 19931995 . section 2 discusses the discoveries of the sne , and section 3 contains a description of the observations . the `` template subtraction '' method of photometry used to extract the light of sne from their environs is presented in section 4 . section 5 describes the calibration of the measurements onto the standard johnson - kron - cousins system . the sn light curves and color curves appear in sections 6 and 7 , respectively , while estimates of the absolute magnitudes are in section 8 . section 9 summarizes our results . ut dates are used in all cases . the discoveries of the supernovae presented here were made by several groups . sn 1993y ( fig . 1 ) in ugc 2771 was discovered on 1993 september 18 , about @xmath2 east and @xmath3 north of the galaxy s nucleus , with the 1.2-m oschin telescope as part of the second palomar sky survey ( mueller 1993 ) . accurate position measurements of the object resulted in , ( balam 1993 ) . a spectrum taken on september 22 showed sn 1993y to be of type ia about three weeks past maximum brightness ( dressler & sargent 1993 ) . another spectrum taken on september 25 showed that it was close to one month past maximum ( filippenko & matheson 1993 ) . sn 1993z ( fig . 2 ) in the sab galaxy ngc 2775 was found on 1993 september 23 , about @xmath4 west and @xmath5 south of the nucleus , as part of the leuschner observatory supernova search ( loss ) using a 0.76-m telescope ( treffers et al . the accurate position measurement of the sn is , ( balam 1993 ) . a spectrum taken on september 25 indicated sn 1993z to be of type ia about four weeks past maximum ( treffers et al . 1993 ) . sn 1993ae ( fig . 3 ) in ugc 1071 was found with the 0.9-m schmidt telescope at the observatoire de la cote dazur ( oca ) on 1993 november 7.92 ; the sn was located at , ( pollas 1993 ) . spectroscopy on november 11.2 showed that sn 1993ae was a normal type ia about 10 d after maximum with an expansion velocity of about 10000 km s@xmath6 from si ii @xmath76355 ( cappellaro & bragaglia 1993 ) . the supernova was located about @xmath8 east and @xmath9 north of the nucleus . sn 1994b ( fig . 4 ) was discovered on 1994 january 16.98 in an anonymous galaxy , again using the oca 0.9-m schmidt telescope ; the supernova position was at , , or @xmath10 east and @xmath11 north of the center of the host face - on spiral galaxy ( pollas 1994 ) . a spectrum obtained on january 19 suggested that sn 1994b was of type ia near maximum ; the broad si ii @xmath76355 absorption trough was visible , and a redshift @xmath12 of the parent galaxy was also determined from the h ii region on which sn 1994b was superposed ( filippenko & matheson 1994 ) . sn 1994c ( fig . 5 ) was discovered on 1994 march 5 at , during the second palomar sky survey ( mueller & mendenhall 1994 ) . it was confirmed as a sn on march 9 ( djorgovski , thompson , & smith 1994 ) . sn 1994c was classified on march 12 as a type ia about 20 d past maximum , and the host galaxy s redshift was determined to be @xmath13 ( riess , challis , & kirshner 1994 ) . it was located about @xmath14 west and @xmath15 north of the anonymous galaxy s nucleus . sn 1994 m ( fig . 6 ) in ngc 4493 was found by wild ( 1994 ) at zimmerwald , university of berne , on 1994 april 29.9 ; the sn position was , , about @xmath14 west and @xmath16 south of the nucleus . a spectrum taken on may 4.14 showed it to be of type ia very close to maximum light ( schmidt , kirshner , & peters 1994 ) . sn 1994q ( fig . 7 ) was found with the oca 0.9-m schmidt telescope on 1994 june 2.92 at , , or @xmath17 west and @xmath18 south of an anonymous galaxy s nucleus ( pollas & albanese 1994 ) . a spectrum taken on june 3 suggested that sn 1994q was of type ia about one week past maximum ( filippenko , matheson , & barth 1994 ) . sn 1994ae ( fig . 8) in the sc galaxy ngc 3370 was discovered on 1994 november 14 by loss ( van dyk et al . 1994 ) long before maximum light ; various estimates of the date of maximum visual brightness were november 30 ( nakano et al . 1994 ) , december 1 ( patat , vician , & szentasko 1994 ) , and prior to december 6 ( vanmunster , villi , & cortini 1994 ) . the accurate position is , ( nakano , kushida , & kushida 1994 ) , which is @xmath19 west and @xmath20 north of the galaxy nucleus . a spectrum taken on november 23.16 indicated sn 1994ae to be of type ia ; the measured expansion velocity for the si ii @xmath76355 line was 11000 km s@xmath6 ( iijima , cappellaro , & turatto 1994 ) . sn 1995d ( fig 9 ) in the s0 galaxy ngc 2962 was found apparently on the rise on 1995 february 10.76 , using a 0.25-m telescope at the yatsugatake south base observatory ( kushida , nakano , & kushida 1995 ) at , , about @xmath21 east and @xmath22 south of the galactic nucleus . a spectrum obtained on february 13.18 confirmed it as a type ia , about one week before maximum ; the spectrum was dominated by p - cygni lines of intermediate - mass elements on a very blue continuum , and the deduced expansion velocity for si ii @xmath76355 was 10900 km s@xmath6 ( benetti , mendes de oliveira , & manchado 1995 ) . sn 1994y ( fig . 10 ) in the sbc galaxy ngc 5371 was discovered on 1994 august 19.15 with a 0.9-m telescope at mcdonald observatory ( wren 1994 ) . images obtained as part of the loss showed it to be about @xmath23 west and @xmath21 north of the nucleus , on the rise from august 12 to august 16 , and not visible on august 3 , 6 , and 9 ( paik et al . a spectrum taken on august 25 identified it as a type ii ( jiang , liu , & hu 1994 ) . furthermore , clocchiatti et al . ( 1994 ) took a spectrum on august 27.15 , which showed the balmer emission lines without the characteristic p - cygni profiles ; rather , there were narrow lines on top of broad bases , thereby resulting in the classification of sn 1994y as a type iin ( see filippenko 1997 ) . also evident in a spectrum taken on 1995 february 8 was a narrow absorption feature at 6622 superimposed on the strong h@xmath1 line , signifying a large amount of circumstellar material ( wang et al . accurate position measurements yielded , ( boattini & tombelli 1994 ) . observations were made at lick observatory with the nickel 1-m reflector and standard johnson - kron - cousins @xmath24 filters . the telescope utilizes a loral @xmath25 pixel ccd ( 15 @xmath26 m pixel@xmath6 ) , which we binned @xmath27 to yield a plate scale of @xmath28 pixel@xmath6 . the full - width at half - maximum ( fwhm ) of stellar objects generally varied from about @xmath29 to @xmath30 . the ccd images were bias corrected using an average of several bias images . they were also flattened by the average of several flatfield images of the twilight sky . exposure times were as short as 120 s for sn 1994ae near maximum light to 1800 s for deeper late - time and template images . however , some images were combined to give a longer overall exposure time . figures 110 illustrate lick images of the ten sne , their host galaxies , and the stars in the field that were used as comparisons ( henceforth referred to as field stars ) . sn 1993y , sn 1993z , and sn 1994y were also observed at leuschner observatory as part of the berkeley automatic imaging telescope ( bait ) project , which employed 50-cm and 76-cm telescopes ( see filippenko 1992 ; richmond , treffers , & filippenko 1993 ; richmond et al . 1996 ; and references therein ) . each telescope was equipped with a @xmath31 pixel ccd with a plate scale of about @xmath32 pixel@xmath6 , and stellar objects generally had a fwhm of about @xmath14 to @xmath33 . the filters were made according to the characteristics described by bessell ( 1990 ) . the images were bias corrected with the median of five bias images taken each afternoon . five flat - field images of the twilight sky were used to determine a median image , which was then used to flatten the data . exposure times were between 120 and 900 s. sn 1993y was observed from 1993 september 26 through december 20 , sn 1993z from 1993 september 23 through 1994 february 1 , and sn 1994y from 1994 august 22 through december 7 . in addition , three images of sn 1994y were taken at kitt peak national observatory on 1995 january 28 , using a 0.9-m telescope and tektronix @xmath34 pixel ccd ( 24 @xmath26 m pixel@xmath6 ) . they were each 300-s exposures at a plate scale of @xmath35 pixel@xmath6 and fwhm from @xmath30 to @xmath14 . to determine the instrumental magnitudes of most of these various sne , we use the technique of template subtraction described in richmond et al . ( 1995 ; see also filippenko et al . 1986 ) to minimize the light contamination by the background galaxy . this involves taking a `` template '' image of the galaxy long after the supernova has faded . the template image was matched to each early - time image by registering the template using several field stars in common . the template was obtained under superior seeing conditions and was then convolved with a gaussian to match the widths of the point - spread function ( psf ) of each early - time image . the magnitudes of the field stars for both the template and early - time images were measured using a circular aperture of 15 pixels radius and a sky annulus of width 10 pixels . comparing the magnitudes , an average scale factor was obtained and used to subtract the template image from the early - time image . the flux scale factor varied by @xmath36 10% typically from star to star within a given frame . the resulting template - subtracted image showed only the sn and , occasionally , the small residuals of the field stars and the host galaxy nucleus . the sn magnitude was then measured with negligible contamination from the galaxy light . for sne 1993ae , 1994 m , and 1995d , template subtraction was not needed , because the sn was sufficiently far from the galactic nucleus and in a region of relatively uniform background . simple aperture photometry was performed in these cases , with a circular aperture of 10 or 15 pixels radius and a 10-pixel - width sky annulus . psf - fitting photometry , via iraf / daophot ( stetson 1987 ) , was performed for sn 1994y and for the leuschner observations of sn 1993y . in both cases , the model psf was constructed using bright , isolated field stars on each of the images . sn 1993y was far enough from the host galaxy nucleus that the background was negligible in all bands . on particular nights when the conditions were photometric , standard star fields from landolt ( 1992 ) were observed in @xmath24 , along with images of the sn . the instrumental magnitudes of these stars were determined using either a 10- or 15-pixel - radius circular aperture and a 10-pixel - width sky annulus . using iraf , the photometric instrumental magnitudes were converted to apparent magnitudes with the transformation equations @xmath37 where @xmath38 are instrumental magnitudes , @xmath24 are standard apparent magnitudes , and @xmath39 are the airmass values . the second - order extinction terms were set to zero . the results , along with the number of landolt fields observed and used on each night to determine the photometric solution , are shown in table 1 . in solving these equations , the standard deviations in the residuals were on the order of a few hundredths of a magnitude . thus , the net uncertainty of the fit can be taken to be @xmath40 mag . in each of figures 110 , the labeled field stars were the comparisons used to determine the apparent magnitude of the respective sn . each field star s instrumental magnitude was measured , again using the same aperture and sky annulus as above , and transformed with equations ( 1)(4 ) into apparent magnitudes . table 2 lists the results and their respective uncertainties . the images were fairly clean , so that cosmic ray hits near the stars of interest occurred only rarely . in such cases where the cosmic - ray pixels could be easily identified , they were smoothed by linear interpolation using nearby pixels . in cases when there were no photometric observations made , the apparent magnitudes of the field stars were obtained from riess ( 1996 ) . also , observations of sne 1993z , 1994ae , 1995d , and 1994y occurred on multiple nights under photometric conditions ; the night with the smallest photometric uncertainties was adopted as the standard . comparisons of our field star calibrations with riess ( 1996 ) and riess et al . ( 1999 ) were made , whenever possible , and the results were generally quite consistent . each sn instrumental magnitude was then converted into an apparent magnitude on the standard johnson - kron - cousins @xmath24 system using a weighted mean of the difference between the instrumental and apparent magnitudes of the field stars . tables 36 present these magnitudes along with their uncertainties . in addition , a comparison of the sn 1994y data using psf fitting versus template - subtraction yielded negligible differences when the sn was bright . comparisons with the results of riess ( 1996 ) and riess et al . ( 1999 ) generally show good agreement for the objects and epochs in common : the mean difference for 34 points is @xmath41 0.01 mag , and the dispersion is @xmath41 0.07 mag . however , there may be a slight systematic trend , namely our average values are @xmath41 0.04 , 0.02 , 0.01 mag brighter in @xmath42 , @xmath43 , and @xmath44 , respectively , and @xmath41 0.03 mag fainter in @xmath45 than those of riess ( 1996 ) and riess et al . ( 1999 ) . finally , we note that the transformation equations ( [ eq : b])-([eq : i ] ) , which were used on photometric nights at lick to derive apparent magnitudes of the field stars from their instrumental magnitudes , are implicitly assumed to be valid for the leuschner and kpno data . from the apparent magnitudes of the field stars , we find the scaling relation for the nights that were not photometric and scale the sn instrumental magnitude to an apparent magnitude . although the loral chips were changed in the middle of the lick program , the two ccds were very similar ( both loral ) and the filters remained the same , so nearly the same transformation should apply . the lack of independent transformations for the leuschner and kpno data increases the uncertainties of these data but by an amount that is difficult to quantify . unfortunately , the leuschner system that was used is no longer available for calibration ; however , we still obtain consistent results between lick and leuschner . there were only three kpno measurements , and again , they are consistent with the other results . figure 11 shows the light curves of sn 1993y , and figure 12 shows the light curves of sn 1993z . the light curves of sne 1993ae , 1994b , 1994c , 1994 m , 1994q , 1994ae , and 1995d are shown in figure 13 . the line segments connecting the points have no significance other than to aid the reader in distinguishing between different sne and to indicate general trends . the time of maximum light for sn 1993y is uncertain . based on its spectra , it was estimated to be either around august 26 ( filippenko & matheson 1993 ) or september 1 ( dressler & sargent 1993 ) . for this paper , we adopt august 26 as the date of peak brightness . the observations of sn 1993y are therefore from 31 d to 137 d past maximum . normal type ia sne show a decrease in the decline rate to a slow linear magnitude decline at around 40 d past maximum in @xmath43 and @xmath44 and at @xmath41 60 d in @xmath45 . this is evident in our light curves for sn 1993y , with the change in slope occurring some time between october 1 ( @xmath41 36 d after maximum ) and october 18 ( @xmath41 53 d after maximum ) in @xmath43 and @xmath44 , and around october 27 ( @xmath41 62 d after maximum ) in @xmath45 . the data after the onset of slow decline was fit to a line using a weighted least - squares fit algorithm ( press et al . the corresponding rates of slow decline for sn 1993y are shown in table 7 , along with a comparison to the rates of slow decline for two normal sne ia , sn 1980n ( decline rates starting from 42 d past maximum ; hamuy et al . 1991 ) and sn 1989b ( decline rates from 42 d to 131 d past maximum ; wells et al . as is evident , the results for sn 1993y are in agreement with both sne 1980n and 1989b . we adopt august 28 as the date sn 1993z reached maximum brightness ; thus , the observations presented here are from 26 d to 258 d past maximum . by fitting to different segments of our data in @xmath44 , we determine that a change in the rate of decline occurred between september 26 ( @xmath41 29 d after maximum ) and october 22 ( @xmath41 55 d after maximum ) , which is the case for normal type ia sne . using the same method described above and fitting only to the data after the onset of the slow linear magnitude decline , the rates for sn 1993z are shown in table 7 . as in the case of sn 1993y , the rates of slow decline for sn 1993z are in agreement with the respective values of the two normal sne ia , sne 1980n and 1989b . the dates of maximum @xmath42-band brightness of sne 1993ae , 1994 m , 1994q , and 1994ae were determined from their light curves to be 1993 november 2.1 , 1994 may 3.6 , 1994 may 28.9 , and 1994 november 29 , respectively ( riess 1996 ) ; hence , our first observation of sn 1994ae is just 2 d after @xmath42 maximum . those of sne 1994b and 1994c were determined from their spectra to be 1994 january 27 and 1994 february 28 , respectively ( riess et al . 1999 ) . sadakane et al . ( 1996 ) present extensive light curves in @xmath0 during the first 100 d for sn 1995d , as well as several spectra near maximum brightness . they determine @xmath43 maximum to be on february 21.5 . it is also seen from their light curves that @xmath45 maximum occurs prior to the date of @xmath43 maximum . comparison of our data with their photometry at about the same epoch shows very good agreement . at the first epoch in common , our data in @xmath43 are fainter by 0.01 mag , which is well within the uncertainties . sadakane et al . do not have @xmath44 and @xmath45 data at this epoch . however , they have data 2 d prior and 3 d after this common epoch , and our @xmath44 and @xmath45 data do lie between these points . at the second epoch in common , our @xmath43 and @xmath44 data are brighter by @xmath36 0.07 mag , and our @xmath45 data is fainter by 0.21 mag , all within a few times the uncertainties . riess ( 1996 ) gives the date of @xmath42 maximum to be february 21.3 . early - time spectra of sn 1994y reveal balmer emission lines on a wide base , while at later times the balmer lines broaden ; at all times there is an absence of narrow absorption lines seen in some other type iin supernovae ( filippenko 1997 ) . figure 14 shows the light curves for sn 1994y . from the paucity of data in @xmath42 , it is difficult to say how long sn 1994y stayed near maximum , but it is apparent that sn 1994y had a `` plateau''-like phase near maximum with a duration of about 30 d in @xmath43 , 40 d in @xmath44 , and 20 d in @xmath45 . this plateau s appearance , however , differs from that seen in conventional sne ii - p ( doggett & branch 1985 ) . an estimate of the times and magnitudes of peak brightness was made by eye using the data between jd 2,449,580 and 2,449,700 ; table 8 lists the results . the uncertainty was estimated from the breadth of the maximum and the abundance of data points near the peak . in @xmath42 , the light curve seems to indicate a rapid rise and fall . in @xmath43 and @xmath44 , it reaches the `` plateau '' around ( or slightly before ) jd 2,449,600 and only gradually rises and falls . in particular , there is no clear indication of the time of @xmath43 maximum , which may have occurred in the gap between our observations . @xmath44 maximum seems to have occurred late in the plateau . in @xmath45 , there is a more distinct peak . it is evident from the @xmath44 and @xmath45 light curves that there is a change in the post - maximum rate of decline around jd 2,449,800 the decline rates become steeper . there is a similar change in @xmath43 at about jd 2,449,860 . we fit a linear decline to the data in @xmath43 after jd 2,449,700 but prior to 2,449,860 , and in @xmath44 for all data after jd 2,449,800 . the calculated rates of decline are shown in table 7 , along with the decline rates for another type iin , sn 1988z ( rate from 110 d to 450 d past maximum ; turatto et al . 1993 ) , and a peculiar type ii , sn 1993j ( rate from about 200 d to 400 d after maximum ; richmond et al . sn 1994y demonstrates a much more rapid decline than sn 1988z , but one that is on the same order of magnitude as sn 1993j . two indications imply that the mass of the ejecta in sn 1994y was small . the brightness of sn 1994y remained at a plateau for only several weeks ( instead of several months , as do most sne ii - p ) , which suggests a small ejecta mass , since the duration of the plateau phase depends on how long the effective photosphere takes to work its way back through the hydrogen envelope of the star into the inner regions . a short phase indicates that the hydrogen envelope is not massive ( or that the sn was expanding much more quickly than usual ) . also , barbon et al . ( 1995 ) and richmond et al . ( 1996 ) indicate in the case of sn 1993j that the small amount of mass ejected , implied from its double peak , leads to a decrease in the efficiency of @xmath46-ray trapping ; this generates a faster rate of decline than the 0.98 mag per 100 d resulting from the radioactive decay of @xmath47co , which powers the late - time light curves of normal type ii - p sne . the similarity in the decline rates of sn 1994y and sn 1993j , again , suggests that the ejecta of sn 1994y were also small in mass . finally , the @xmath42 and @xmath43 light curves exhibit another change in their rates of decline beginning around jd 2,449,925 , leveling off to about 0.140.15 mag per 100 d. this is similar to the very late - time decline rate of 0.150.16 mag per 100 d seen in sn 1988z between 730 d and 1150 d ( turatto et al . 1993 ) , which is a period well after our observations of sn 1994y . we suspect that interaction of the ejecta with circumstellar gas is providing additional light , as in the case for sn 1988z . however , the flattening of the light curves suggests that this process becomes more pronounced at an earlier time in sn 1994y , which again may be a result of the small ejecta mass . figure 15 shows the color curves of the nine type ia sne . note that these values have not been corrected for reddening . there are large variations in the late - time @xmath48 template color curve given by riess , press , & kirshner ( 1996 ) ; we have chosen a typical one , but there is a substantial difference between our observations and this template . sn 1993y may be more reddened than most of the other sne ia presented here . the proximity of sn 1994c to the host galaxy s nucleus and the sn s very red color at early times suggest that it suffered significant reddening or was intrinsically red . comparisons of our colors for sn 1995d with the color curves from sadakane et al . ( 1996 ) show good agreement . figure 16 presents the color curves of sn 1994y . sn 1988z had an initial @xmath49 @xmath50 0.4 mag , followed by a leveling off at later times to @xmath51 @xmath50 0.64 mag ( turatto et al . 1993 ) . due to the ambiguity of the time of maximum for sn 1994y , it is difficult to indicate the colors at peak brightness . however , at the assumed time of @xmath42 maximum , @xmath51 is between 0.12 and 0.25 mag . furthermore , @xmath51 at later times also seems to be approaching a plateau around 0.5 mag . @xmath52 appears to increase monotonically , almost linearly , with time . @xmath48 , on the other hand , is @xmath530.15 mag about two weeks prior to @xmath42 maximum , reddens to a peak at or above 0.27 mag some time after two months , then turns blueward for the duration of our observations , ending a year from @xmath42 maximum at @xmath530.78 mag . the @xmath52 and @xmath48 color curves at late times show the growing strength of @xmath44 relative to @xmath43 and @xmath45 , thus suggesting that the h@xmath1 line is getting stronger relative to the continuum , and optical spectra confirm this conclusion ( filippenko 1997 ; matheson & filippenko 1999 , private communication ) . equivalent width is 1200 on 1995 january 25 , @xmath54 on april 22 , and even larger at later times . ] this explains the lack of a change in the rate of decline at late times of the @xmath44 light curve , since most of the light is concentrated in the h@xmath1 emission . we can calculate the absolute magnitude of sn 1994y , for which we have data near maximum light . burstein & heiles ( 1984 ) determine the galactic extinction to ngc 5371 ( the host of sn 1994y ) to be @xmath55 = 0.00 mag . using the distance modulus to ngc 5371 , @xmath56 mag ( tully 1988 ) , the resulting absolute magnitudes are shown in table 8 . peak @xmath42 brightness is @xmath57 mag . average type ii supernovae have @xmath58 mag , but the dispersion is quite large , @xmath59 mag ( miller & branch 1990 ) . like sn 1988z , which has a @xmath60 mag ( turatto et al . 1993 ) , sn 1994y is more luminous than the typical sn ii but not abnormally so . we presented @xmath24 photometry of the type iin sn 1994y and the type ia sne 1993y , 1993z , 1993ae , 1994b , 1994c , 1994 m , 1994q , 1994ae , and 1995d . all observations were done using telescopes of diameter 1 m or less . light curves of sn 1993y and sn 1993z exhibit late - time rates of decline very similar to those of typical sne ia . the color curves of sn 1993y suggest that it suffers more reddening than most of the other sne ia presented here . sn 1994c may also have been significantly reddened or was intrinsically red . the light curves of sn 1994y show that it remained at a plateau near maximum light for several weeks , with no definitive date of peak brightness . the light curves also suggest that the ejecta mass was small . the estimated absolute magnitude of sn 1994y shows that it , like the sn iin 1988z , was slightly more luminous than normal sne ii . we are grateful to a. j. barth , b. leibundgut , and d. schlegel for their assistance in obtaining some of the data . we would also like to express our appreciation to a. g. riess for determining the ages of some of the sne ia and for providing helpful comments . the thoughtful and extensive suggestions of the anonymous referee led to substantial improvements in this paper . the research of a.v.f.s group was supported by nsf grants ast-8957063 , ast-9115174 , ast-9417213 , and ast-9987438 , as well as by nasa through grants go-6043 and go-6584 from the space telescope science institute ( which is operated by aura , inc . , under nasa contract nas 5 - 26555 ) . a.v.f . also thanks the guggenheim foundation for a fellowship . we are grateful to sun microsystems , inc . ( academic equipment grant program ) and photometrics , ltd . for equipment donations to the telescopes ( bait ) at leuschner observatory . the supernova search and follow - up efforts with bait were the precursors to ( and testbeds of ) the lick observatory supernova search with the katzman automatic imaging telescope ( li et al . 2000 ; filippenko et al . 2001 ) ; continued funding from the sylvia and jim katzman foundation is greatly appreciated . note added in proof . p. garnavich , a. noriega - crespo , & a. moro - martin ( 1996 , iau circ . , 6314 ) point out that sn 1994y exhibited a strong infrared excess in data obtained about 17 months after discovery . this excess could be emission from dust formed in the ejecta , or it may be an infrared echo from an existing dust shell . balam , d. d. 1993 , iau circ . 5883 barbon , r. , benetti , s. , cappellaro , e. , patat , f. , turatto , m. , & iijima , t. 1995 , a&as , 110 , 513 benetti , s. , mendes de oliveira , c. , & manchado , a. 1995 , iau circ . 6135 bessell , m. s. 1990 , , 102 , 1181 boattini , a. , & tombelli , m. 1994 , iau circ . 6065 burstein , d. , & heiles , c. 1984 , apjs , 54 , 33 cappellaro , e. , & bragaglia , a. 1993 , iau circ . 5888 clocchiatti , a. , garcia - lopez , r. , barker , e. s. , wren , w. , & wheeler , j. c. 1994 , iau circ . 6065 djorgovski , s. , thompson , d. , & smith , j. 1994 , iau circ . 5946 doggett , j. b. , & branch , d. 1985 , , 90 , 2303 dressler , a. , & sargent , w. 1993 , iau circ . 5870 filippenko , a. v. 1992 , in robotic telescopes in the 1990s , ed . a. v. filippenko ( san francisco : asp , conf . 34 ) , 55 filippenko , a. v. 1997 , , 35 , 309 filippenko , a. v. , & matheson , t. 1993 , iau circ . 5870 filippenko , a. v. , & matheson , t. 1994 , iau circ . 5923 filippenko , a. v. , matheson , t. , & barth , a. j. 1994 , iau circ . 6001 filippenko , a. v. , porter , a. c. , sargent , w. l. w. , & schneider , d. p. 1986 , aj , 92 , 1341 filippenko , a. v. , li , w. d. , treffers , r. r. , & modjaz , m. 2001 , in iau conf . 246 , small - telescope astronomy on global scales , ed . w. p. chen , c. lemme , & b. paczynski ( san francisco : asp ) , 121 hamuy , m. , phillips , m. m. , maza , m. , wischnjewsky , m. , uomoto , a. , landolt , a. u. , & khatwani , r. 1991 , , 102 , 208 iijima , t. , cappellaro , e. , & turatto , m. 1994 , iau circ . 6108 jiang , x. j. , liu , w. , & hu , j. y. 1994 , iau circ . 6065 kirshner , r. p. 1990 , in supernovae , ed . a. g. petschek ( new york : springer - verlag ) kushida , r. , nakano , s. , & kushida , y. 1995 , iau circ . 6134 landolt , a. u. 1992 , , 104 , 340 leibundgut , b. 1988 , ph.d . thesis , university of basel leibundgut , b. 1996 , in supernovae and supernova remnants , ed . r. mccray & z. wang ( cambridge : cambridge university press ) , 11 li , w. d. , et al . 2000 , in cosmic explosions , ed . s. s. holt & w. w. zhang ( new york : aip ) , 103 miller , d. l. , & branch , d. 1990 , , 100 , 530 mueller , j. 1993 , iau circ . 5870 mueller , j. , & mendenhall , j. d. 1994 , iau circ . 5946 nakano , s. , kushida , y. , & kushida , r. 1994 , iau circ . 6110 nakano , s. , takamizawa , t. , kushida , r. , & ikeya , k. 1994 , iau circ . 6109 paik , y. , filippenko , a. v. , treffers , r. r. , van dyk , s. d. , & richmond , m. w. 1994 , iau circ . 6058 patat , f. , barbon , r. , cappellaro , e. , & turatto , m. 1994 , a&a , 282 , 731 patat , f. , vician , z. , & szentasko , l. 1994 , iau circ . 6111 pollas , c. 1993 , iau circ . 5888 pollas , c. 1994 , iau circ . 5923 pollas , c. , & albanese , d. 1994 , iau circ . 6001 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , numerical recipes in c , 2nd ed . ( cambridge : cambridge university press ) richmond , m. w. , treffers , r. r. , & filippenko , a. v. 1993 , , 105 , 1164 richmond , m. w. , treffers , r. r. , filippenko , a. v. , & paik , y. 1996 , , 112 , 732 richmond , m. w. , et al . 1995 , , 109 , 2121 riess , a. g. 1996 , ph.d . thesis , harvard university riess , a. g. , challis , p. , & kirshner , r. 1994 , iau circ . 5948 riess , a. g. , press , w. h. , & kirshner , r. p. 1996 , , 473 , 88 riess , a. g. , et al . 1999 , , 117 , 707 sadakane , k. , et al . 1996 , pasj , 48 , 51 schmidt , b. , kirshner , r. , & peters , j. 1994 , iau circ . 5984 stetson , p. b. 1987 , , 99 , 191 suntzeff , n. b. 1996 , in supernovae and supernova remnants , ed . r. mccray & z. wang ( cambridge : cambridge university press ) , 41 treffers , r. r. , filippenko , a. v. , leibundgut , b. , paik , y. , lee , l. f. m. , & richmond , m. w. 1993 , iau circ . 5870 tully , r. b. 1988 , nearby galaxies catalog ( cambridge : cambridge university press ) turatto , m. , cappellaro , e. , danziger , i. j. , benetti , s. , gouiffes , c. , & della valle , m. 1993 , , 262 , 128 van dyk , s. d. , treffers , r. r. , richmond , m. w. , filippenko , a. v. , & paik , y. b. 1994 , iau circ . 6105 vanmunster , t. , villi , m. , & cortini , g. 1994 , iau circ . 6115 wang , l. , li , z. -w . , clocchiatti , a. , & wheeler , j. c. 1995 , iau circ . 6135 wells , l. a. , et al . 1994 , , 108 , 2233 wheeler , j. c. , & harkness , r. p. 1990 , 53 , 1467 wild , p. 1994 , iau circ . 5982 wren , w. 1994 , iau circ . 6058 cccccc + + + b & 4.49 ( 02 ) & 0.31 ( 01 ) & @xmath530.09 ( 01 ) & 97 sep 05 & 4 + v & 4.10 ( 02 ) & 0.20 ( 01 ) & 0.02 ( 01 ) & & + r & 3.95 ( 01 ) & 0.14 ( 01 ) & 0.05 ( 01 ) & & + i & 4.03 ( 03 ) & 0.10 ( 01 ) & @xmath530.03 ( 06 ) & & + + + + b & 4.09 ( 01 ) & 0.28 ( 01 ) & @xmath530.06 ( 00 ) & 94 dec 01 & 7 + v & 3.85 ( 01 ) & 0.14 ( 01 ) & 0.02 ( 01 ) & & + r & 3.74 ( 01 ) & 0.11 ( 01 ) & 0.04 ( 01 ) & & + i & 3.85 ( 01 ) & 0.08 ( 01 ) & @xmath530.02 ( 01 ) & & + + + + b & 4.55 ( 04 ) & 0.25 ( 03 ) & @xmath530.07 ( 02 ) & 97 apr 12 & 5 + v & 4.04 ( 03 ) & 0.25 ( 02 ) & 0.02 ( 01 ) & & + r & 3.92 ( 03 ) & 0.18 ( 02 ) & 0.05 ( 03 ) & & + i & 4.08 ( 03 ) & 0.09 ( 02 ) & @xmath530.00 ( 03 ) & & + + + + b & 4.12 ( 02 ) & 0.45 ( 01 ) & @xmath530.10 ( 01 ) & 95 mar 28 & 7 + v & 4.03 ( 01 ) & 0.18 ( 01 ) & 0.01 ( 01 ) & & + r & 3.94 ( 02 ) & 0.12 ( 01 ) & 0.02 ( 02 ) & & + i & 4.02 ( 02 ) & 0.08 ( 01 ) & @xmath530.07 ( 02 ) & & + + + + b & 4.29 ( 01 ) & 0.41 ( 01 ) & @xmath530.07 ( 01 ) & 95 may 23 & 5 + v & 3.98 ( 02 ) & 0.28 ( 01 ) & 0.01 ( 01 ) & & + r & 3.86 ( 01 ) & 0.24 ( 01 ) & 0.01 ( 02 ) & & + i & 3.98 ( 04 ) & 0.17 ( 01 ) & @xmath530.04 ( 07 ) & & ccccc + + + a & 20.17 ( 29 ) & 18.88 ( 10 ) & 18.90 ( 19 ) & 18.38 ( 24 ) + b & 18.83 ( 10 ) & 18.06 ( 05 ) & 17.82 ( 08 ) & 17.32 ( 10 ) + c & 18.86 ( 10 ) & 18.03 ( 05 ) & 17.51 ( 08 ) & 16.88 ( 09 ) + d & 18.77 ( 10 ) & 18.02 ( 05 ) & 17.57 ( 08 ) & 17.10 ( 09 ) + e & 19.65 ( 18 ) & 18.44 ( 07 ) & 17.73 ( 11 ) & 16.82 ( 11 ) + f & 17.35 ( 03 ) & 16.72 ( 01 ) & 16.34 ( 02 ) & 15.91 ( 03 ) + g & 18.65 ( 08 ) & 17.50 ( 03 ) & 16.74 ( 04 ) & 16.04 ( 05 ) + h & 16.66 ( 02 ) & 15.82 ( 01 ) & 15.35 ( 01 ) & 14.89 ( 01 ) + i & 16.79 ( 02 ) & 16.26 ( 01 ) & 15.92 ( 02 ) & 15.55 ( 02 ) + j & 17.28 ( 02 ) & 16.33 ( 01 ) & 15.76 ( 02 ) & 15.20 ( 02 ) + + + + a & 16.04 ( 00 ) & 14.87 ( 00 ) & 14.15 ( 00 ) & 13.49 ( 00 ) + b & 16.62 ( 01 ) & 15.77 ( 00 ) & 15.27 ( 01 ) & 14.82 ( 01 ) + c & 16.65 ( 01 ) & 15.79 ( 00 ) & 15.31 ( 01 ) & 14.86 ( 01 ) + d & 18.16 ( 03 ) & 17.46 ( 02 ) & 17.02 ( 03 ) & 16.62 ( 04 ) + e & 20.45 ( 17 ) & 18.97 ( 06 ) & 17.93 ( 10 ) & + f & 18.94 ( 05 ) & 17.83 ( 02 ) & 17.11 ( 03 ) & 16.57 ( 04 ) + g & 17.64 ( 02 ) & 17.05 ( 01 ) & 16.66 ( 02 ) & 16.36 ( 02 ) + + + + a & 18.25 ( 08 ) & 17.28 ( 03 ) & 16.78 ( 04 ) & 16.31 ( 05 ) + + + + a & 15.60 ( 02 ) & 14.96 ( 02 ) & 14.55 ( 02 ) & 14.22 ( 02 ) + b & 17.11 ( 02 ) & 16.53 ( 02 ) & 16.16 ( 02 ) & 15.84 ( 02 ) + c & 16.24 ( 02 ) & 15.45 ( 02 ) & 14.96 ( 02 ) & 14.55 ( 02 ) + d & 16.12 ( 02 ) & 15.58 ( 02 ) & 15.23 ( 02 ) & 14.94 ( 02 ) + + + + a & 18.82 ( 10 ) & 18.06 ( 05 ) & 17.69 ( 08 ) & 17.19 ( 10 ) + b & 19.70 ( 20 ) & 18.70 ( 09 ) & 17.95 ( 14 ) & 17.25 ( 15 ) + c & 20.30 ( 29 ) & 18.86 ( 10 ) & 17.92 ( 16 ) & 17.66 ( 17 ) + d & 15.78 ( 01 ) & 14.91 ( 00 ) & 14.40 ( 00 ) & 13.95 ( 01 ) + e & 16.07 ( 01 ) & 15.53 ( 00 ) & 15.20 ( 01 ) & 14.86 ( 01 ) + f & 13.86 ( 00 ) & 13.35 ( 00 ) & 13.06 ( 00 ) & 12.74 ( 00 ) + g & 16.30 ( 01 ) & 15.38 ( 00 ) & 14.88 ( 01 ) & 14.40 ( 01 ) + h & 17.93 ( 04 ) & 17.17 ( 02 ) & 16.68 ( 04 ) & 16.23 ( 04 ) + i & 17.97 ( 04 ) & 16.93 ( 02 ) & 16.24 ( 03 ) & 15.67 ( 03 ) + j & 17.71 ( 04 ) & 16.99 ( 02 ) & 16.60 ( 03 ) & 16.17 ( 04 ) + k & 14.32 ( 00 ) & 13.60 ( 00 ) & 13.21 ( 00 ) & 12.85 ( 00 ) + l & 14.73 ( 00 ) & 14.03 ( 00 ) & 13.64 ( 00 ) & 13.25 ( 00 ) + + + + a & 14.20 ( 02 ) & 13.19 ( 02 ) & 12.68 ( 02 ) & 12.32 ( 02 ) + b & 15.09 ( 02 ) & 14.46 ( 02 ) & 14.08 ( 02 ) & 13.69 ( 02 ) + + + + a & 15.63 ( 01 ) & 15.18 ( 00 ) & 14.89 ( 01 ) & 14.60 ( 01 ) + b & 14.77 ( 00 ) & 14.22 ( 00 ) & 13.87 ( 00 ) & 13.54 ( 00 ) + c & 18.04 ( 04 ) & 17.05 ( 02 ) & 16.48 ( 03 ) & 15.97 ( 03 ) + + + + a & 18.07 ( 11 ) & 17.58 ( 06 ) & 17.13 ( 09 ) & 16.58 ( 10 ) + b & 19.06 ( 26 ) & 18.57 ( 14 ) & 17.31 ( 20 ) & 15.84 ( 21 ) + c & 18.84 ( 22 ) & 18.34 ( 12 ) & 18.02 ( 19 ) & 17.93 ( 24 ) + d & 18.81 ( 95 ) & 20.32 ( 64 ) & 18.78 ( 94 ) & 18.40 ( 97 ) + e & 17.18 ( 05 ) & 16.53 ( 02 ) & 16.11 ( 04 ) & 15.66 ( 04 ) + f & 18.65 ( 24 ) & 18.67 ( 14 ) & 18.14 ( 23 ) & 17.59 ( 26 ) + g & 19.06 ( 18 ) & 17.66 ( 06 ) & 16.52 ( 09 ) & 14.95 ( 10 ) + + + + a & 18.51 ( 12 ) & 17.01 ( 03 ) & 16.00 ( 04 ) & 14.91 ( 04 ) + b & 18.22 ( 10 ) & 17.21 ( 04 ) & 16.61 ( 05 ) & 16.15 ( 06 ) + c & 14.97 ( 01 ) & 14.30 ( 00 ) & 13.93 ( 01 ) & 13.62 ( 01 ) + d & 15.00 ( 01 ) & 14.34 ( 00 ) & 13.97 ( 01 ) & 13.66 ( 01 ) + e & 16.64 ( 03 ) & 15.87 ( 01 ) & 15.44 ( 02 ) & 15.08 ( 02 ) + + + + a & 14.27 ( 00 ) & 13.93 ( 00 ) & 13.74 ( 00 ) & 13.62 ( 00 ) + b & 14.72 ( 00 ) & 14.27 ( 00 ) & 13.98 ( 00 ) & 13.78 ( 00 ) + c & 15.81 ( 01 ) & 15.35 ( 00 ) & 15.08 ( 01 ) & + d & 18.57 ( 08 ) & 17.66 ( 04 ) & 16.83 ( 06 ) & 16.24 ( 07 ) + e & 18.71 ( 11 ) & 18.16 ( 06 ) & 17.57 ( 10 ) & 17.06 ( 11 ) + f & 14.77 ( 00 ) & 14.19 ( 00 ) & 13.82 ( 00 ) & 13.53 ( 00 ) + g & 18.28 ( 08 ) & 17.84 ( 04 ) & 17.48 ( 08 ) & 17.14 ( 10 ) + h & 16.54 ( 02 ) & 15.92 ( 01 ) & 15.54 ( 01 ) & 15.27 ( 02 ) lccccc + + + 93 sep 29.34 & 2449259.84 & & 17.08 ( 04 ) & 16.67 ( 02 ) & 16.33 ( 03 ) + 93 oct 07.42 & 2449267.92 & & & 16.99 ( 08 ) & + 93 oct 08.41 & 2449268.91 & & & 17.22 ( 02 ) & 16.77 ( 02 ) + 93 oct 19.40 & 2449279.90 & & 18.13 ( 05 ) & & + 93 oct 30.30 & 2449290.80 & & 18.33 ( 10 ) & & 17.70 ( 10 ) + 93 nov 01.33 & 2449292.83 & & & & 17.95 ( 08 ) + 93 nov 02.25 & 2449293.75 & & 18.53 ( 12 ) & & 17.71 ( 09 ) + 93 nov 03.25 & 2449294.75 & & & & 18.01 ( 08 ) + 93 nov 04.32 & 2449295.82 & & & 18.24 ( 03 ) & 17.92 ( 05 ) + 93 nov 05.32 & 2449296.82 & & 18.42 ( 06 ) & 18.22 ( 04 ) & + 93 nov 06.34 & 2449297.84 & & 18.55 ( 06 ) & 18.27 ( 05 ) & 18.00 ( 05 ) + 93 nov 07.33 & 2449298.83 & & 18.62 ( 07 ) & 18.36 ( 04 ) & 18.13 ( 03 ) + 93 nov 08.35 & 2449299.85 & & & & 18.21 ( 05 ) + 93 nov 13.27 & 2449304.77 & & & 18.47 ( 04 ) & 18.27 ( 07 ) + 93 nov 15.30 & 2449306.80 & & & 18.47 ( 04 ) & + 93 nov 20.31 & 2449311.81 & & 18.79 ( 09 ) & & + 93 nov 24.27 & 2449315.77 & & & 18.73 ( 07 ) & 18.54 ( 09 ) + 93 nov 25.20 & 2449316.70 & & & 19.03 ( 09 ) & + 93 dec 03.17 & 2449324.67 & & & 19.01 ( 08 ) & 18.84 ( 10 ) + 93 dec 05.17 & 2449326.67 & & 19.11 ( 15 ) & 19.11 ( 08 ) & 18.85 ( 11 ) + 93 dec 06.17 & 2449327.67 & & & 19.00 ( 07 ) & + 93 dec 07.17 & 2449328.67 & & 18.92 ( 17 ) & & + + + + 93 oct 18.50 & 2449279.00 & 19.14 ( 02 ) & 18.06 ( 02 ) & 17.60 ( 02 ) & 17.29 ( 03 ) + 93 nov 16.40 & 2449307.90 & 19.57 ( 07 ) & 18.74 ( 05 ) & 18.54 ( 04 ) & 18.57 ( 07 ) + 93 dec 16.43 & 2449337.93 & & 19.49 ( 21 ) & & 19.28 ( 22 ) + 93 dec 17.34 & 2449338.84 & 19.75 ( 14 ) & 19.64 ( 18 ) & 19.18 ( 12 ) & 19.23 ( 18 ) + 94 jan 11.24 & 2449363.74 & 20.31 ( 11 ) & 20.20 ( 16 ) & 20.55 ( 25 ) & 20.29 ( 47 ) lccccc + + + 93 sep 23.52 & 2449254.02 & & & 13.87 ( 07 ) & + 93 sep 24.52 & 2449255.02 & & & 13.96 ( 08 ) & + 93 sep 26.53 & 2449257.03 & & & 13.94 ( 08 ) & + 93 oct 22.45 & 2449282.95 & & & 15.27 ( 17 ) & + 93 oct 25.44 & 2449285.94 & & & 15.31 ( 17 ) & + 93 oct 28.45 & 2449288.95 & & & 15.40 ( 18 ) & + 93 nov 04.42 & 2449295.92 & & & 15.68 ( 20 ) & + 93 nov 16.39 & 2449307.89 & & & 16.06 ( 24 ) & + 93 nov 19.48 & 2449310.98 & & & 16.34 ( 26 ) & + 93 nov 23.46 & 2449314.96 & & & 16.50 ( 28 ) & + 93 nov 26.38 & 2449317.88 & & & 16.31 ( 26 ) & + 93 dec 01.36 & 2449322.86 & & & 16.57 ( 28 ) & + 93 dec 19.37 & 2449340.87 & & & 16.84 ( 31 ) & + 93 jan 11.28 & 2449363.78 & & & 17.03 ( 33 ) & + 93 jan 17.29 & 2449369.79 & & & 17.57 ( 38 ) & + 93 feb 01.27 & 2449384.77 & & & 17.92 ( 42 ) & + + + + 93 nov 15.57 & 2449307.07 & & & & 15.79 ( 01 ) + 93 nov 16.57 & 2449308.07 & 17.02 ( 02 ) & & & + 93 dec 16.52 & 2449338.02 & 17.52 ( 02 ) & 16.81 ( 01 ) & 16.69 ( 02 ) & 16.91 ( 04 ) + & & & & 16.74 ( 03 ) & 16.75 ( 05 ) + 94 jan 10.32 & 2449362.82 & 17.99 ( 02 ) & 17.38 ( 02 ) & 17.40 ( 02 ) & 17.40 ( 02 ) + 94 feb 05.43 & 2449388.93 & & 17.88 ( 03 ) & & 18.07 ( 05 ) + 94 mar 14.29 & 2449425.79 & 19.02 ( 06 ) & 18.40 ( 05 ) & 18.88 ( 10 ) & + 94 apr 12.21 & 2449454.71 & & & & 19.37 ( 42 ) + 94 apr 13.22 & 2449455.72 & & 19.03 ( 09 ) & 19.45 ( 10 ) & 19.06 ( 12 ) + & & & 19.11 ( 08 ) & & 19.20 ( 19 ) + 94 may 14.22 & 2449486.72 & & 19.42 ( 16 ) & 19.68 ( 34 ) & lccccc + + + 93 nov 15.34 & 2449306.84 & 17.10 ( 01 ) & 16.25 ( 01 ) & & 16.17 ( 01 ) + 93 dec 17.21 & 2449338.71 & 18.92 ( 09 ) & 17.89 ( 05 ) & 17.75 ( 08 ) & 17.80 ( 10 ) + 94 jan 11.14 & 2449363.64 & 19.34 ( 06 ) & 18.69 ( 05 ) & 18.54 ( 05 ) & 18.82 ( 10 ) + + + + 94 feb 05.37 & 2449388.87 & & 19.23 ( 04 ) & & 19.25 ( 07 ) + 94 mar 15.23 & 2449426.73 & & 20.38 ( 17 ) & 19.28 ( 13 ) & 19.46 ( 17 ) + + + + 94 mar 12.27 & 2449423.77 & & 18.83 ( 07 ) & 18.40 ( 09 ) & + 94 mar 14.20 & 2449425.70 & 20.02 ( 11 ) & 18.88 ( 06 ) & 18.60 ( 06 ) & 18.08 ( 04 ) + + + + 94 may 14.33 & 2449486.83 & 17.23 ( 01 ) & 16.71 ( 01 ) & 16.70 ( 01 ) & 17.14 ( 02 ) + 94 jun 11.27 & 2449514.77 & & 18.35 ( 04 ) & 18.06 ( 04 ) & + & & & & 18.00 ( 04 ) & + + + + 94 jun 11.43 & 2449514.93 & 17.79 ( 01 ) & 17.13 ( 01 ) & 16.75 ( 01 ) & + 94 jul 12.35 & 2449545.85 & & 18.51 ( 02 ) & 17.92 ( 02 ) & 18.02 ( 02 ) + 94 jul 13.40 & 2449546.90 & 19.55 ( 07 ) & & & + 94 aug 06.28 & 2449570.78 & & 19.46 ( 07 ) & 18.10 ( 05 ) & + + + + 94 dec 01.58 & 2449688.08 & 13.15 ( 00 ) & 13.06 ( 00 ) & 13.07 ( 00 ) & 13.48 ( 00 ) + 95 mar 27.44 & 2449803.94 & & 16.95 ( 02 ) & 17.22 ( 03 ) & 17.52 ( 04 ) + 95 mar 30.36 & 2449806.86 & 17.35 ( 02 ) & 17.15 ( 03 ) & & 17.58 ( 03 ) + + + + + + 95 mar 28.33 & 2449804.83 & 16.12 ( 01 ) & 14.98 ( 00 ) & 14.61 ( 00 ) & 14.32 ( 00 ) + 95 may 23.23 & 2449860.73 & & 16.52 ( 01 ) & 16.53 ( 02 ) & 16.80 ( 02 ) + & & & & 16.55 ( 01 ) & lccccc + + + 94 aug 22.17 & 2449587.67 & 15.25 ( 07 ) & 15.00 ( 04 ) & 14.79 ( 03 ) & 14.95 ( 04 ) + 94 aug 31.16 & 2449596.66 & & 14.54 ( 02 ) & 14.29 ( 02 ) & 14.44 ( 02 ) + 94 sep 09.16 & 2449598.66 & & 14.55 ( 02 ) & & 14.44 ( 03 ) + 94 sep 03.16 & 2449599.66 & 14.59 ( 04 ) & 14.54 ( 02 ) & 14.30 ( 02 ) & 14.34 ( 03 ) + 94 sep 04.19 & 2449600.69 & 14.58 ( 04 ) & 14.46 ( 02 ) & 14.32 ( 01 ) & 14.41 ( 02 ) + 94 sep 06.15 & 2449602.65 & & 14.51 ( 02 ) & 14.35 ( 02 ) & 14.36 ( 02 ) + 94 sep 07.15 & 2449603.65 & & & 14.23 ( 02 ) & + 94 sep 08.20 & 2449604.70 & & 14.48 ( 02 ) & 14.28 ( 01 ) & 14.28 ( 02 ) + 94 sep 10.15 & 2449606.65 & 14.76 ( 04 ) & 14.49 ( 02 ) & 14.32 ( 02 ) & 14.28 ( 02 ) + 94 sep 11.19 & 2449607.69 & 14.84 ( 05 ) & 14.49 ( 03 ) & 14.31 ( 01 ) & 14.31 ( 02 ) + 94 sep 12.15 & 2449608.65 & 14.50 ( 04 ) & 14.62 ( 02 ) & 14.22 ( 02 ) & 14.33 ( 02 ) + 94 sep 14.19 & 2449610.69 & 14.90 ( 06 ) & 14.55 ( 03 ) & 14.24 ( 02 ) & 14.27 ( 02 ) + 94 sep 16.18 & 2449612.68 & & 14.44 ( 03 ) & 14.24 ( 02 ) & 14.32 ( 02 ) + 94 sep 17.14 & 2449613.64 & & 14.50 ( 03 ) & 14.29 ( 02 ) & 14.26 ( 03 ) + 94 sep 18.18 & 2449614.68 & & 14.46 ( 04 ) & 14.24 ( 02 ) & 14.25 ( 02 ) + 94 sep 27.13 & 2449623.63 & & & 14.26 ( 03 ) & 14.17 ( 03 ) + 94 sep 29.13 & 2449625.63 & & 14.41 ( 03 ) & & 14.13 ( 02 ) + 94 sep 30.13 & 2449626.63 & & 14.42 ( 05 ) & 14.28 ( 02 ) & 14.15 ( 02 ) + 94 oct 01.13 & 2449627.63 & & 14.62 ( 04 ) & 14.24 ( 02 ) & 14.11 ( 02 ) + 94 oct 03.12 & 2449629.62 & & 14.50 ( 04 ) & 14.20 ( 02 ) & 14.05 ( 03 ) + 94 oct 04.12 & 2449630.62 & & 14.71 ( 07 ) & & 14.12 ( 04 ) + 94 oct 07.12 & 2449633.62 & & 14.53 ( 04 ) & 14.21 ( 02 ) & 14.08 ( 02 ) + 94 oct 08.12 & 2449634.62 & & 14.49 ( 03 ) & 14.21 ( 02 ) & 14.08 ( 02 ) + 94 oct 10.12 & 2449636.62 & & 14.48 ( 13 ) & 14.21 ( 07 ) & 14.27 ( 14 ) + 94 oct 11.11 & 2449637.61 & & & 14.22 ( 02 ) & 14.15 ( 03 ) + 94 oct 12.11 & 2449638.61 & & & 14.15 ( 02 ) & 14.15 ( 03 ) + 94 oct 16.11 & 2449642.61 & & 14.63 ( 05 ) & 14.25 ( 03 ) & 14.15 ( 03 ) + 94 oct 17.10 & 2449643.60 & & & 14.33 ( 03 ) & + 94 nov 08.54 & 2449665.04 & & 15.18 ( 04 ) & 14.70 ( 02 ) & 14.43 ( 02 ) + 94 nov 14.55 & 2449671.05 & & & 14.82 ( 02 ) & 14.55 ( 02 ) + 94 nov 19.51 & 2449676.01 & & 15.46 ( 06 ) & 14.93 ( 02 ) & 14.69 ( 02 ) + 94 nov 29.56 & 2449686.06 & & & & 14.89 ( 02 ) + 94 dec 07.51 & 2449694.01 & & & 15.23 ( 03 ) & + + + + + 95 jan 28.51 & 2449746.01 & & 16.40 ( 03 ) & & 15.72 ( 02 ) + & & & 16.40 ( 05 ) & & + + + + 95 mar 30.53 & 2449807.03 & 17.45 ( 11 ) & 17.26 ( 01 ) & 16.37 ( 01 ) & 16.53 ( 01 ) + 95 apr 22.51 & 2449830.01 & & 17.55 ( 02 ) & 16.71 ( 02 ) & + 95 apr 26.37 & 2449833.87 & 17.98 ( 02 ) & 17.61 ( 04 ) & 16.85 ( 02 ) & 17.04 ( 01 ) + 95 may 23.45 & 2449860.95 & 18.56 ( 07 ) & 18.12 ( 03 ) & 17.23 ( 04 ) & 17.81 ( 04 ) + 95 may 30.40 & 2449867.90 & & 18.60 ( 04 ) & & 17.84 ( 02 ) + 95 jun 27.34 & 2449896.84 & & 19.34 ( 08 ) & & 18.68 ( 05 ) + 95 jun 28.36 & 2449897.86 & 19.49 ( 13 ) & & 17.84 ( 03 ) & + 95 jul 25.27 & 2449924.77 & & 19.84 ( 10 ) & & 18.99 ( 07 ) + 95 jul 26.28 & 2449925.78 & 19.74 ( 10 ) & & 18.23 ( 05 ) & + 95 aug 27.15 & 2449957.65 & 19.79 ( 23 ) & & & + 95 aug 28.17 & 2449958.67 & & 19.89 ( 13 ) & 18.71 ( 06 ) & 19.49 ( 10 ) lccccc + + + sn 1993y & @xmath61 & @xmath62 & @xmath63 & @xmath64 & + sn 1993z & @xmath65 & @xmath66 & @xmath67 & @xmath68 & + sn 1980n & @xmath69 & @xmath70 & & & hamuy et al . 1991 + sn 1989b & @xmath71 & @xmath72 & & & wells et al . 1994 + + + + sn 1994y & & @xmath73 & @xmath74 & & + sn 1988z & 0.47 & 0.44 & 0.18 & & turatto et al . 1993 + sn 1993j & @xmath75 & @xmath76 & @xmath77 & @xmath78 & richmond et al . 1996 ccccc jd & 2449603.7 @xmath79 0.5 & 2449617.7 @xmath79 3.0 & 2449618.5 @xmath79 4.0 & 2449629.0 @xmath79 2.0 + ut & sep 07.20 & sep 21.20 & sep 22.00 & oct 03.50 + @xmath80 & 14.52 @xmath79 0.04 & 14.36 @xmath79 0.05 & 14.14 @xmath79 0.02 & 14.04 @xmath79 0.04 + @xmath81 & @xmath5318.37 & @xmath5318.53 & @xmath5318.75 & @xmath5318.85

in this article we introduce a framization of the hecke algebra of type . for this framization we construct a faithful tensorial representation and two linear bases . we finally construct a markov trace on these algebras and from this trace we derive isotopy invariants for framed and classical knots and links in the solid torus .

1603.08487

the idea of framization of a knot algebra ( hecke algebras and bmw algebra among others ) was introduced by the two last authors in @xcite and consists in adding framing generators to the defining generators of the knot algebra with the aim of finding new invariants of classical links or , more generally , invariants of knot - like objects . the yokonuma hecke algebra is the prototype of framization ; indeed this algebra , introduced by t. yokonuma @xcite in the context of representations of chevalley groups , can be thought of as a framization of the hecke algebra of type @xmath1 . more precisely , the yokonuma hecke algebra supports a markov trace @xcite and then it becomes a peculiar knot algebra considering that , by using the jones recipe , one can construct invariants for : framed links @xcite , classical links @xcite and singular links @xcite . it is worth mentioning that recently it was proved that the invariants for classical links constructed in @xcite are not topologically equivalent either to the homflypt polynomial or to the kauffman polynomial , see @xcite . on the other hand , jones raised that his recipe for the construction of the homflypt polynomial might be used for hecke algebra not only of type @xmath2 , cf . then , the third author studied the jones recipe by using the hecke algebra of type @xmath0 ; namely , in @xcite she constructed the analogue of the homflypt polynomial for oriented knots and links inside the solid torus , see also @xcite . further , in @xcite lambropoulou constructed all possible analogues of the homflypt polynomial in the solid torus from the ariki - koike algebras and the affine hecke algebras of type @xmath2 . the purpose of this article is to introduce and to start a systematic study of a framization of the hecke algebras of type @xmath0 , denoted by @xmath3 , with the principal objective to explore their usefulness in knot theory . thus , having in mind both the role of the hecke algebra of type @xmath0 @xcite and the yokonuma hecke algebra of type @xmath2 @xcite in knot theory , it is natural to define , by using the jones recipe applied to @xmath3 , invariants in the solid torus for : classical knots , framed knots and singular knots . for these purposes a first key point is to prove that the algebra @xmath4 supports a markov trace . in fact , this is one of the main results proved in the present article . in @xcite , m. chlouveraki and l. poulain d andecy have introduced the affine and cyclotomic yokonuma hecke algebras . in the context of framization @xcite , the definition of the algebras introduced by chlouveraki and poulain d andecy can be understood by adding framing generators and making the framization according to the formula of framization of the generators of the yokonuma hecke algebra of type @xmath5 , that is , only of the braiding generators . now , the hecke algebra of type @xmath6 is a particular case of cyclotomic hecke algebra . the framization of the hecke algebra of type @xmath6 proposed here makes the framization of all generators of the hecke algebra of type @xmath6 ; in particular , also of the special loop generator of the algebra . before giving the organization of the article we note that , by taking into account the various articles generated recently from the algebra of yokonuma hecke of type @xmath2 ( view for example @xcite among others ) , the framization proposed here indicates that the algebra @xmath4 should be interesting in itself . the article is organized as follows . in section 1 we introduce our notation and explain the background notions . in section 2 we define our framizations for the coxeter group of type @xmath6 , for the artin braid group of type @xmath6 and for the hecke algebra of type @xmath6 , @xmath7 . in section 3 we construct a tensorial representation for the algebra @xmath8 . in section 4 we find linear bases for @xmath7 , one of which is used in section 5 for constructing a markov trace on the algebras @xmath7 . finally , in section 6 necessary and sufficient conditions are given for the trace parameters in order to proceed with the construction of topological invariants of framed and classical knots and links in the solid torus ( section 7 ) . in this section we review known results , necessary for this paper , and we also fix the following terminology and notations that will be used along the paper : 1 . the letters @xmath9 and @xmath10 denote two indeterminates . and we denote by @xmath11 , the field of rational functions @xmath12 . 2 . the term _ algebra _ means unital associative algebra over @xmath11 3 . for a finite group @xmath13 , @xmath14 $ ] denotes the group algebra of @xmath13 4 . the letters @xmath15 and @xmath16 denote two fixed positive integers 5 . we denote by @xmath17 a fixed primitive @xmath16th root of unity 6 . we denote by @xmath18 the group of integers modulo @xmath16 , @xmath19 , and by @xmath20 the cyclic group of order @xmath16 , @xmath21 . note that @xmath22 . 7 . as usual , we denote by @xmath23 the length function associated to the coxeter groups . the finite coxeter group of type @xmath24 ( @xmath25 ) can be realized as the symmetric group on the set @xmath26 . set @xmath27 the elementary transposition @xmath28 , so the coxeter presentation of @xmath29 is encoded in the following dynkin diagram : ( 350,40 ) ( 82,20)@xmath30 ( 120,20)@xmath31 ( 200,20)@xmath32 ( 240,20)@xmath33 ( 85,10 ) ( 87.5,9)(1,0)35 ( 125,10 ) ( 127.5,10)(1,0)10 ( 145,10 ) ( 165,10 ) ( 185,10 ) ( 205,10 ) ( 207.5,10)(1,0)35 ( 245,10 ) ( 192.5,10)(1,0)10 then , the artin braid group , @xmath34 , associated to @xmath29 is generated by the elementary braidings @xmath35 , which satisfy the following relations : @xmath36 the _ framed braid group _ , @xmath37 , is defined as the group presented by the braiding generators @xmath38 and the framing generators @xmath39 subject to the relations ( [ braidr ] ) together with the relations : @xmath40 notice that @xmath37 is isomorphic to the wreath product @xmath41 . the @xmath16_modular framed braid group _ @xmath42 is defined by adding to the above defining presentation of @xmath37 the relations @xmath43 . hence , @xmath44 . it is convenient to write the elements of @xmath45 in the split form @xmath46 , where the @xmath47 s are integers ( called the framings ) and @xmath48 . a framed braid can be represented as a usual geometric braid on @xmath15 strands by attaching the respective framing to each strand . for details and the above geometric interpretation of the framed braid group see @xcite . see also @xcite . set @xmath25 . let us denote by @xmath49 the coxeter group of type @xmath50 . this is the finite coxeter group associated to the following dynkin diagram define @xmath51 for @xmath52 . it is known , see @xcite , that every element @xmath53 can be written uniquely as @xmath54 with @xmath55 , @xmath56 , where @xmath57 furthermore , this expression for @xmath58 is reduced . hence , we have @xmath59 . the corresponding _ braid group of type _ @xmath60 associated to @xmath49 , is defined as the group @xmath61 generated by @xmath62 subject to the following relations @xmath63 geometrically , braids of type @xmath60 can be viewed as classical braids of type @xmath64 with @xmath65 strands , such that the first strand is identically fixed . this is called ` the fixed strand ' . the 2nd , , @xmath66st strands are renamed from 1 to @xmath15 and they are called ` the moving strands ' . the ` loop ' generator @xmath67 stands for the looping of the first moving strand around the fixed strand in the right - handed sense . in figure [ bbraid ] we illustrate a braid of type @xmath68 . -type . ] [ [ realized1 ] ] the group @xmath49 can be realized as a subgroup of the permutation group of the set @xmath69 . more precisely , the elements of @xmath49 are the permutations @xmath58 such that @xmath70 , for all @xmath71 . for example @xmath72 and @xmath73 further , the elements of @xmath49 can be parameterized by the elements of @xmath74 ( see ( * ? ? ? * lemma 1.2.1 ) ) . more precisely , the element @xmath53 corresponds to the element @xmath75 such that @xmath76 . then , we have that @xmath77 is parameterized by @xmath78 and @xmath79 is parameterized by @xmath80 . more generally , if @xmath53 is parameterized by @xmath81 , then @xmath82 [ coxeter ] ( * ? ? ? * lemma 1.2.2 ) let @xmath53 parameterized by @xmath81 . then @xmath83 if and only if @xmath84 and @xmath85 if and only if @xmath86 . set @xmath87 and @xmath88 . then we have that @xmath89 and @xmath90 , and @xmath58 is represented by @xmath91 . [ symbraids1 ] the above realization of the @xmath49 can be lifted also to the braid level . indeed in @xcite tom dieck defines the _ symmetric braids _ ( denoted by @xmath92 ) and proves that @xmath61 is isomorphic to this group of braids . specifically tom dieck considers the braids in @xmath93 $ ] with strands between @xmath94 and @xmath95 which are symmetric about the axis @xmath96 $ ] . moreover , the group of the symmetric braids is generated by the elements @xmath97 represented graphically as follows . an isomorphism between the groups @xmath61 and @xmath92 is induced by the mapping : @xmath98 , @xmath99 for @xmath100 . [ [ section ] ] the hecke algebra of type @xmath101 , denoted @xmath102 , is the algebra generated by @xmath103 subject to the following relations : @xmath104 it is well known that the dimension of @xmath102 is @xmath105 and clearly for @xmath106 it coincides with @xmath107 $ ] . in this section we introduce the main object studied in the paper , that is , a framization of the hecke algebra of type @xmath6 . to do that , previously we introduce a framed version of both , the braid group and the coxeter group of type @xmath108 . at the end of the section we include some useful relations derived directly from the defining relations of our framization algebra . [ [ section-1 ] ] we start with the definition of a @xmath16framed version of @xmath49 . @xmath16modular framed coxeter group of type @xmath109 _ , @xmath110 , is defined as the group generated by @xmath111 and @xmath112 satisfying the coxeter relations of type @xmath113 among @xmath79 and the @xmath77 s , the relations @xmath114 for all @xmath115 , the relations @xmath116 for all @xmath117 , together with the following relations : @xmath118 the analogous group defined by the same presentation , where only relations @xmath116 are omitted , shall be called _ framed coxeter group of type @xmath109 _ and will be denoted as @xmath119 . the _ framed braid group of type @xmath113 _ , denoted @xmath120 , is the group presented by generators @xmath121 , @xmath39 subject to the relations ( [ fbraida ] ) and ( [ braidb ] ) , together with the following relations : @xmath122 the _ @xmath16modular framed braid group _ , denoted @xmath123 , is defined as the group obtained by adding the relations @xmath124 , for all @xmath125 , to the above defining presentation of @xmath120 . in figure [ braidgenrs ] the generators of the groups @xmath120 and @xmath123 are illustrated . and @xmath123.[braidgenrs ] ] the mapping that acts as the identity on the generators @xmath79 and the @xmath77 s and maps the @xmath126 s to @xmath127 defines a morphism from @xmath110 onto @xmath49 . also , we have the natural epimorphism from @xmath123 onto @xmath110 defined as the identity on the @xmath126 s and mapping @xmath67 to @xmath79 and @xmath128 to @xmath77 , for all @xmath125 . thus , we have the following sequence of epimorphisms . @xmath129 where the first arrow is the natural projection of @xmath130 to @xmath123 . finally , we can lift trivially the length function @xmath23 on @xmath49 to @xmath110 . indeed , we deduce from relations ( [ wdn ] ) , that every @xmath131 can be written in the form @xmath132 , with @xmath53 . thus , we define the length of @xmath133 by @xmath134 . we denote again by @xmath23 the length function on @xmath110 . geometrically , elements in @xmath135 ( resp . @xmath136 ) are braids of type @xmath137 ( recall figure [ bbraid ] ) such that each one of the @xmath15 moving strands has a framing in @xmath138 ( resp . @xmath139 ) attached . [ symfrbraids ] following remark [ symbraids1 ] , one can define analogously the _ symmetric framed braids _ resp . _ @xmath16-modular symmetric framed braids _ ( denoted by @xmath140 resp . @xmath141 ) and prove that the liftings @xmath142 resp . @xmath143 are isomorphic to these groups of braids . in terms of geometric relizations , a symmetric framed braid is a braid in @xmath92 with an integer resp . @xmath16-modular framing attached on each strand , such that the strands @xmath125 and @xmath144 have the same framing , for all @xmath125 . note that @xmath145 can be regarded naturally as a subgroup of @xmath110 ; indeed , the elements of @xmath49 correspond to the elements of @xmath110 having all framings equal to @xmath146 . we will proceed now to lift the sets @xmath147 of @xmath49 , introduced in subsection [ coxeterb ] , to subsets @xmath148 of the group @xmath110 , with the aim to give a standard writing for the elements of @xmath110 which will be useful to parameterize later a basis of the framization of the hecke algebra of type @xmath101 defined here . we define inductively the subsets @xmath148 of @xmath110 as follows : @xmath149 and @xmath150 note that , for all @xmath151 and @xmath16 , we have @xmath152 and for @xmath153 the sets @xmath147 and @xmath154 coincide . also , every element @xmath155 can be written as @xmath156 , with @xmath157 . further , we have the following proposition . [ standardwdn ] every element of @xmath110 can be expressed in the standard form , that is , as a product @xmath158 , where @xmath159 . set @xmath160 . we will prove the claim by induction on @xmath15 . for @xmath161 , it is straightforward to check that @xmath133 can be written in standard form . e.g. if @xmath162 and @xmath163 , we have @xmath164 where @xmath165 and @xmath166 . suppose now that the proposition is true for all positive integers less than @xmath15 . then , if @xmath167 in @xmath133 is equal to @xmath127 or @xmath168 , we have : @xmath169 now , @xmath170 and applying the induction hypothesis on the word inside the first parenthesis above , we deduce that @xmath133 can be written in the standard form . if @xmath167 is equal to @xmath171 or @xmath172 , we have : @xmath173 again , noting that @xmath174 and applying the induction hypothesis on the word inside the first parenthesis above , it follows that @xmath133 can be written in the standard form . [ [ section-2 ] ] in order to define a framization of the hecke algebra of type @xmath6 , we need to introduce the following elements @xmath175 and @xmath176 , for @xmath177 , in @xmath178 $ ] , @xmath179 notice that the @xmath175 and the @xmath176 s are idempotents , cf @xcite for the @xmath176 s . let @xmath25 . the algebra @xmath180 is defined as the quotient of @xmath181 $ ] over the two sided ideal generated by the following elements : @xmath182 we shall denote the corresponding to @xmath128 ( respectively , to @xmath67 ) in @xmath183 by @xmath184 ( respectively , by @xmath185 ) and we shall keep the same notation for the @xmath126 s ( respectively , the @xmath176 s and @xmath175 ) in @xmath186 . hence , equivalently , the algebra @xmath7 can be defined by generators @xmath187 , @xmath188 and relations as follows . @xmath189 we extend the above definition for @xmath190 by defining @xmath7 as the algebra generated by @xmath191 and @xmath185 subject to the relations ( [ tb1 ] ) , ( [ modular1 ] ) and ( [ quadraticv ] ) . [ diff ] note that @xmath7 is different from the algebra @xmath192 , for @xmath193 and suitable parameters @xmath194 and @xmath195 , defined by m. chlouveraki and l. poulain dandecy in @xcite . indeed , they differ in the quadratic relation of the generator @xmath185 , since in @xmath7 the relation ( [ quadraticv ] ) involves framing generators , meanwhile the quadratic relation defined in @xmath196 does nt . from the above description by generators and relations of the algebras @xmath7 we have @xmath197 , for all @xmath198 . thus , by taking @xmath199 , we have that following tower of algebras . @xmath200 it is clear that @xmath175 commutes with @xmath185 and @xmath176 commutes with @xmath184 . these facts implies that the generators @xmath185 and @xmath184 s are invertible . moreover , we have : @xmath201 notice that , by taking @xmath153 , the algebra @xmath202 becomes @xmath203 . further , by mapping @xmath204 and @xmath205 , we obtain an epimorphism from @xmath7 to @xmath203 . moreover , if we map the @xmath206 s to a fixed non trivial @xmath16th root of the unity , we have an epimorphism from @xmath7 to @xmath207 . notice that the relations ( [ braid1])([tb1 ] ) are the defining relations of @xmath130 and the relations ( [ braid1])([modular1 ] ) are the defining relation of @xmath123 . then , @xmath186 can be obviously defined as a quotient of the group algebra @xmath208 $ ] or @xmath209 $ ] . alternatively , @xmath186 can be regarded as a @xmath210deformation of the group algebra @xmath211 $ ] of the @xmath16-modular framed braid group of type @xmath6 . [ [ section-3 ] ] we also have the following relations which are deduced easily and will be used frequently in the sequel . @xmath212 where @xmath213 is the natural generalization of @xmath175 , @xmath214 notice that the @xmath213 s are idempotents . finally , we finish the section by introducing certain elements @xmath215 and proving some algebraic identities which will be used along the paper . set @xmath216 figure [ bi ] illustrates the elements @xmath217 . . ] further , for all @xmath125 we have @xmath218 and a direct computation shows that , @xmath219 [ braidhbhb ] for @xmath25 and @xmath220 , the following relations hold in @xmath7 : 1 . @xmath221 , for all j 2 . @xmath222 , for @xmath223 or @xmath224 3 . @xmath225 4 . @xmath226 . the proof of relations ( i ) and ( ii ) follows directly by using the defining relations of @xmath7 . the proof of relations ( iii ) is by induction on @xmath151 . for @xmath227 , the relation in question is the defining relation ( [ braid4 ] ) . let us suppose now that relation ( iii ) holds for all positive integers less than @xmath228 . then for @xmath228 we have : @xmath229 we prove now relation ( iv ) . we have @xmath230 . then by using relation ( iii ) we obtain @xmath231 , so relation ( iv ) is true . we will define now a tensorial representation for the algebra @xmath7 . the definition of this representation is based on the tensorial representation constructed by green in @xcite for the hecke algebra of type @xmath0 and following the idea of an extension of the jimbo representation of the hecke algebra of type @xmath2 to the yokonuma - hecke algebra proposed by espinoza and ryom hansen in @xcite . the tensorial representation constructed here will be used to prove that a set of linear generators , denoted @xmath232 , is a basis for @xmath7 ( theorem [ basis ] ) . further , as a corollary , we obtain that this tensorial representation is faithful ( corollary [ faithful ] ) . [ [ section-4 ] ] let @xmath233 be a @xmath11vector space with basis @xmath234 . as usual we denote by @xmath235 the natural basis of @xmath236 associated to @xmath237 . that is , the elements of @xmath235 are of the form : @xmath238 where @xmath239 and @xmath240 . we define the endomorphism @xmath241 of @xmath233 by @xmath242 and the endomorphism @xmath13 of @xmath243 by @xmath244 for all @xmath100 , we extend these endomorphisms to the endomorphisms @xmath245 and @xmath246 of the @xmath15th tensor power @xmath247 of @xmath233 , as follows : @xmath248 where @xmath249 denotes the endomorphism identity of @xmath236 . further we define the endomorphism @xmath250 of @xmath247 by : @xmath251 the main goal of this section is to prove that these endomorphisms define a representation of the @xmath7 in the algebra of endomorphisms @xmath252 of @xmath247 . to do that , we will need certain endomorphisms @xmath253 of @xmath247 , introduced in @xcite , which are defined by , @xmath254 also , we will need the following element @xmath255 , @xmath256 [ f ] we have : 1 . @xmath257 2 . @xmath258 claim ( 1 ) is ( * ? ? ? * lemma 3 ) . to prove ( 2 ) , we note that @xmath259 . hence , @xmath260 now , from the fact that @xmath261 is 1 or 0 , depending if @xmath262 or not , claim ( 2 ) follows . [ tensorrepre ] the mapping @xmath263 , @xmath264 and @xmath265 defines a representation @xmath266 of @xmath7 in @xmath252 . to prove the theorem it is enough to verify that the operators @xmath246 , @xmath245 and @xmath250 satisfy the defining relations of @xmath7 whenever we replace @xmath184 by @xmath246 , @xmath206 by @xmath245 and @xmath185 by @xmath250 . having present ( * ? ? ? * theorem 4 ) follows that ( [ braid1 ] ) , ( [ braid2 ] ) , ( [ th])([modular1 ] ) and ( [ quadraticu ] ) it holds for the operators @xmath246 , @xmath245 , and @xmath250 , and it is easy to check that the identities ( [ braid3 ] ) and ( [ tb1 ] ) hold for these operators too . in order to finish the proof of the theorem we shall prove the relations ( [ quadraticv ] ) and ( [ braid4 ] ) . to do that , it is enough to assume @xmath161 and to prove the relations by evaluating in @xmath267 . we prove now that relation ( [ quadraticv ] ) is valid for @xmath250 . if in @xmath133 we have @xmath268 , we distinguish either @xmath269 or @xmath270 . for @xmath269 ; we have : @xmath271.\ ] ] for @xmath270 , we have : @xmath272.\ ] ] on the other hand , if @xmath273 it is clear that @xmath274 . then by lemma [ f ] relation ( [ tb1 ] ) holds in both cases . finally , we will prove that the identity ( [ braid4 ] ) holds if we replace @xmath185 by @xmath250 and @xmath275 by @xmath276 . to do that , we will distinguish first the cases according to the following exhaustive values of @xmath277 and @xmath278 : * case @xmath279 * case @xmath280 and @xmath281 * case @xmath268 and @xmath282 * case @xmath280 and @xmath283 . case ( a ) holds by @xcite , and case ( b ) is easy to check . we distinguish now according the values of @xmath125 and @xmath117 in the remaining cases ( c ) and ( d ) . for item ( c ) we have four cases and eight cases for item ( d ) . we will check only the most representative cases by evaluating in @xmath284 . * case : * @xmath268 , @xmath285 , @xmath269 and @xmath286 . we have : @xmath287 * case : * @xmath268 , @xmath282 , @xmath270 and @xmath288 . we have : @xmath289 * case : * @xmath280 , @xmath290 , @xmath291 and @xmath292 . we have : @xmath293 * case : * @xmath280 , @xmath290 , @xmath291 and @xmath294 . we have : @xmath295 * case : * @xmath280 , @xmath290 , @xmath296 and @xmath297 . we have : @xmath298 [ [ section-5 ] ] we shall finish the section by proving proposition [ aplicacion ] , which is an analogue of ( * ? ? ? * lemma 3.1.4 ) . this proposition will be used in the proof of theorem [ basis ] and describes , through @xmath266 , the action of @xmath49 on the basis @xmath299 . the defining generators @xmath185 and @xmath184 of the algebra @xmath7 satisfy the same braid relations as the coxeter generators @xmath300 and @xmath77 of the group @xmath49 . thus , the well known matsumoto lemma implies that if @xmath301 is a reduced expression of @xmath53 , with @xmath302 , then the following element @xmath303 is well defined : @xmath304 where @xmath305 , if @xmath306 and @xmath307 , if @xmath308 . the notation @xmath309 stands for the image by @xmath266 of @xmath310 . note that , for @xmath58 , @xmath311 such that @xmath312 , we have @xmath313 . [ aplicacion ] let @xmath53 parameterized by @xmath314 . then @xmath315 the proof follows by induction on the length of @xmath58 . for @xmath316 we have that @xmath317 , then the result is direct from the definition of @xmath266 . now suppose that the induction hypothesis holds for any @xmath318 with @xmath319 and let @xmath58 be an element with length @xmath15 . then we have two cases : @xmath320 or @xmath321 for some @xmath318 with @xmath319 . we only present the proof of the case @xmath321 , as the proof of the other case is analogous . suppose @xmath322 is parameterized by @xmath81 . then by the induction hypothesis we obtain : @xmath323 now , in lemma [ coxeter ] we have @xmath84 , therefore from the definition of @xmath246 s we obtain @xmath324 finally , from ( [ parametrized ] ) we have that @xmath58 is parameterized by @xmath325 . hence the claim follows . we introduce here two linear bases @xmath326 and @xmath327 for @xmath7 . the first one is used for defining in the next section a markov trace on @xmath7 , the second one plays a technical role for proving that @xmath326 is a linearly independent set . set @xmath328 , and @xmath329 for all @xmath56 , let us define inductively the set @xmath330 by @xmath331 and @xmath332 [ dn ] we define @xmath327 as the subset of @xmath7 formed by the following elements @xmath333 where @xmath334 . we will prove first that @xmath327 is a linearly spanning set for @xmath7 . to do that we need some formulas of multiplication among the defining generators of @xmath7 and the elements @xmath330 . these are given in lemmas [ relations1d ] and [ relations2d ] below . notice that every element of @xmath330 has the form @xmath335 or @xmath336 , with @xmath337 and @xmath338 , where @xmath339 and @xmath340 [ relations1d ] in @xmath7 the following relations holds : 1 . @xmath341 2 . @xmath342 , for all @xmath343 3 . @xmath344 4 . @xmath345 , if @xmath346 5 . @xmath347 , if @xmath227 6 . @xmath348 , for @xmath227 7 . @xmath349 all relations follow from direct computations . [ relations2d ] in @xmath7 we have : 1 . @xmath350 2 . @xmath351j = k@xmath352 in claim ( i ) we will check only the case @xmath353 , since the other cases are clear . we have @xmath354 the only non - trivial case in claim ( ii ) is whenever @xmath355 . we have @xmath356 [ spand ] the set @xmath327 is a spanning set for @xmath7 . the proof is by induction on @xmath15 . let @xmath357 be the linear subspace of @xmath7 spanned by @xmath327 . the assertion is true for @xmath190 , since @xmath358 and obviously @xmath359 is equal to the space spanned by @xmath360 . assume now that @xmath361 is spanned by @xmath362 . notice that @xmath363 . this fact and proving that @xmath357 is a right ideal , implies the proposition . now , we deduce that @xmath357 is a right ideal from the hypothesis induction and lemmas [ relations1d ] and [ relations2d ] . indeed , the multiplication of @xmath364 from the right by all defining generators of @xmath7 results in a linear combination of elements of the form @xmath365 , with @xmath366 and @xmath367 . in order to prove now that @xmath327 is a linearly independent set , we will firstly rewrite its elements in split form , that is , as the product between the braiding part and the framing part . more precisely , given an element in @xmath327 , then by using the relations ( [ th])([tb1 ] ) , the framing elements ( every power of the @xmath126 s ) that appears in this given element , can be moved to the right . thus , we deduce that the elements in @xmath327 can be written in the following form : @xmath368 with @xmath369 and @xmath370 , where the sets @xmath371 are defined inductively as follows : @xmath372 and @xmath373 recall now that @xmath374 and notice that @xmath375 is reduced , so @xmath376 ( see ( [ gw ] ) ) . these facts and noting that the elements of the sets @xmath147 ( see ( [ nwn ] ) ) are reduced , imply that @xmath377 then , the set @xmath232 can be described by : @xmath378 secondly , we shall use a certain basis @xmath379 of @xmath233 introduced by espinoza and ryom hansen in @xcite . more precisely , @xmath379 consist of the following elements : @xmath380 where @xmath151 is running @xmath381 and @xmath382 . notice that @xmath379 is a basis for @xmath233 , since for any fixed @xmath151 the base change matrix between @xmath383 and @xmath384 is non singular , see @xcite . further , it is easy to see that @xmath385 we are now in the position to prove that @xmath327 is a basis for @xmath386 [ basis ] @xmath327 is a linear basis for @xmath7 . hence the dimension of @xmath7 is @xmath387 . according to proposition [ spand ] we only need to prove that @xmath327 is a linearly independent set . indeed , suppose that we have a linear combination in the form : @xmath388 the proof follows by proving that @xmath389 for all @xmath390 . now , using the expression ( [ elementd ] ) for the elements of @xmath232 and applying @xmath266 to the above equation , we obtain the following equation : @xmath391 where @xmath58 runs in @xmath49 and @xmath392 runs in @xmath393 . now , set @xmath53 parameterized by @xmath394 . then from lemma [ aplicacion ] and the definition of the elements @xmath395 s , we get @xmath396 on the other hand , by using ( [ corre ] ) we have : @xmath397 where @xmath398 runs in @xmath393 . thus , evaluating equation ( [ principal ] ) in @xmath399 we obtain @xmath400 where @xmath401 runs in @xmath74 and @xmath402 runs in @xmath393 . therefore , @xmath403 for all @xmath125 and @xmath404 since the left side of the last equation is a linear combination of elements of the basis @xmath405 . in particular , the above theorem implies the following corollary . [ faithful ] the representation @xmath266 is faithful . for all @xmath56 , let us define inductively the sets @xmath406 by @xmath407 and @xmath408 [ cn]we define @xmath326 as the subset of @xmath7 formed by the following elements : @xmath409 where @xmath410 . to prove that @xmath326 is a linearly spanning set we will need some formulas of multiplication among the defining generators of @xmath7 and the elements @xmath406 . these are given in lemmas [ relations1c][relations3c ] below . now notice that every element of @xmath406 has the form @xmath411 or @xmath412 with @xmath337 and @xmath338 , where @xmath413 and @xmath414 [ relations1c ] the following hold : 1 . @xmath415 2 . @xmath416 the proof is straightforward . [ relations2c ] the following hold : @xmath417 the positive case follows directly from lemma [ relations2d ] ( ii ) , since @xmath418 . for the negative case we have : @xmath419 [ relations3c ] the following hold : 1 @xmath420 2 . @xmath421 + where @xmath422 $ ] . in particular we have : @xmath423 the claim of ( i ) is straightforward . to prove claim ( ii ) , we note first that @xmath424 . then , splitting @xmath425 according to ( [ inverse ] ) and invoking ( [ braid4 ] ) we deduce : @xmath426 by using again ( [ inverse ] ) , we write the second @xmath275 that appears inside the parenthesis above in terms of @xmath425 . so , we obtain : @xmath427 hence @xmath428\\ & & + b_1\mathfrak{m}_{n , k , m}^-.\end{aligned}\ ] ] now , by using the definition of @xmath176 , we obtain : @xmath429 in the same way we obtain : @xmath430 therefore , the proof follows . [ basisc ] the set @xmath326 is a basis for @xmath7 . we can prove that @xmath326 is a spanning set for @xmath7 analogously to the proof of proposition [ spand ] , but by using now lemmas [ relations1c][relations3c ] instead of lemmas [ relations1d ] and [ relations2d ] . now , the cardinal of @xmath326 is @xmath431 , hence the proof follows . we shall close the subsection with a lemma , which will be used in section 6 . [ tr1 ] for @xmath432 and @xmath433 the following identities hold : 1 . @xmath434 2 . @xmath435 3 . @xmath436 to prove claim ( i ) we use lemmas [ relations1c][relations3c ] . more precisely , we have : @xmath437 for claim ( ii ) we have by ( [ inverse ] ) : @xmath438 then , by expanding the @xmath439 above , the result follows . finally , we can prove claim ( iii ) similarly , but by using now ( [ quadraticu ] ) and ( [ inverse ] ) . the section is devoted to proving that the tower of algebras ( [ tower ] ) associated to the algebras @xmath7 supports a markov trace ( theorem [ trace ] ) . this fact is proved by using the method of relative traces , cf . probably this method is due to a. p. isaev and o. v. ogievetsky , see for example @xcite . in few words , the method consists in constructing a certain family of linear maps @xmath440 , called _ relative traces _ , which builds step by step the desired markov properties . the markov trace on @xmath7 is defined by @xmath441 . [ [ section-6 ] ] let @xmath442 be an indeterminate and denote by @xmath443 the field of rational functions @xmath444 . we work now on the algebra @xmath445 which , for simplicity , we denote again by @xmath7 . notice that @xmath446 . consequently , @xmath447 is taken as @xmath448 . we set @xmath449 and from now on we fix non zero parameters @xmath450 in @xmath451 . [ transitionmaps ] for @xmath198 , we define the linear functions @xmath440 as follows . for @xmath190 , @xmath452 and @xmath453 . for @xmath25 , we define @xmath454 on the basis @xmath326 of @xmath7 by : @xmath455 where @xmath456 . note that ( [ tracedef ] ) also holds for @xmath367 , since @xmath457 is a basis for @xmath361 . [ traza1 ] for all @xmath458 and @xmath459 , we have : * @xmath460 * @xmath461 * @xmath462 . for proving claim ( i ) notice that , due to the linearity of @xmath454 , we can suppose that @xmath463 is a defining generator of @xmath464 and @xmath465 , with @xmath466 . further , to prove the claim we shall distinguish the @xmath467 s according to the possibilities of @xmath468 . @xmath469 for @xmath470 , we have @xmath471 , then @xmath472 since @xmath473 . hence , @xmath474 . @xmath469 for @xmath475 , we consider first @xmath476 . then using lemma [ relations1c ] and [ relations2c ] we have : @xmath477 . hence , @xmath478 suppose now @xmath479 . by definition , @xmath480 . then by lemma [ relations3c ] @xmath481 where @xmath482 but , we have @xmath483 therefore , @xmath484 @xmath469 for @xmath485 , with @xmath486 , we have : @xmath487 if @xmath488 with @xmath489 , the claim follows directly from ( i ) lemma [ relations1c ] . for example , for @xmath490 , we have @xmath491 we can proceed in similar way for the other cases for @xmath117 . @xmath487 if @xmath492 with @xmath493 . the claim follows by using the formulas of lemma [ relations2c ] . below , we show only the prove of the case @xmath355 , since that the other cases for @xmath117 follows easily . we have : @xmath494\right ) \\ & = & \left(zw\left [ \mathfrak{m}_{n-1,k+1,m}^{\pm}+\frac{1}{d}({\mathsf{u}}-{\mathsf{u}}^{-1})\sum_s t_{j}^{m - s}\mathfrak{m}_{n-1,k , s}^{\pm})\right]\right)\\ & = & zw\mathfrak{m}_{n-1,k , m}^{\pm}g_{j}={\rm tr}_n(y)z.\end{aligned}\ ] ] @xmath487 if @xmath479 , we deduce the claim directly from ( ii ) lemma [ relations1c ] . to prove ( ii ) , by using the linearity of @xmath454 , we can suppose again that @xmath495 stands for the defining generators of @xmath464 and @xmath465 , with @xmath466 . note now that @xmath496 . hence claim ( ii ) follows directly from the definition of @xmath454 . finally , claim ( iii ) is a combination of claims ( i ) and ( ii ) . [ traza 2 ] for every @xmath198 and @xmath433 , we have that @xmath497 as we know , from linearity of the trace is enough consider @xmath495 in @xmath326 , then we have @xmath498 , with @xmath466 . whenever @xmath475 or @xmath499 the result is clear , since @xmath500 commute with @xmath495 . so , suppose @xmath485 . then , from lemma [ relations1c ] , we obtain @xmath501 on the other hand , we have @xmath502 thus , the proof of the lemma follows . [ exg ] for @xmath25 , @xmath503 and @xmath459 , we have : * @xmath504 * @xmath505 we prove ( i ) . expanding the left side and using lemma [ traza1 ] , we have : @xmath506 similarly , we expand the right side obtaining : @xmath507 hence , claim ( i ) is true . to prove claim ( ii ) we use lemmas [ traza1 ] and [ traza 2 ] . indeed , we have : @xmath508 [ traza3 ] for @xmath25 and @xmath503 . we have @xmath509 as before , we can suppose @xmath510 with @xmath511 . we will check the first equality by distinguishing the possibilities for @xmath512 . for @xmath513 the claim is only a direct computation . for @xmath514 , we have @xmath515 finally , for @xmath516 , we have : @xmath517 thus , the proof of the first equality is done . from a combination of lemma [ tr1 ] ( ii ) and lemma [ exg ] ( i ) , we deduce @xmath518 hence , from ( i ) the second equality follows . [ trazaimp ] for all @xmath433 , we have @xmath519 again , from the linearity of @xmath454 it is enough to consider @xmath495 in the basis @xmath326 . set @xmath520 , with @xmath466 . we are going to prove the statement by distinguishing according to the possibilities for @xmath468 . @xmath469 for @xmath470 , the claim follows from lemma [ traza 2 ] . @xmath469 for @xmath485 , we note that , using the formula ( [ inverse ] ) for the inverse of @xmath521 and lemma [ exg ] ( ii ) , we obtain the following : @xmath522 now , for the left side of this equality , we have : @xmath523 in the same manner one obtains this last expression for @xmath524 . in consequence the claim holds . @xmath469 finally , for @xmath525 we separate the proof depending on the form of @xmath58 in @xmath526 . @xmath487 suppose that @xmath527 , with @xmath528 . then , we have : @xmath529 on other hand : @xmath530 then using lemma [ tr1 ] ( iii ) , we obtain @xmath531 where @xmath532 we will compute the values of @xmath1 , @xmath533 and @xmath534 . a direct computation shows that : @xmath535 expanding @xmath536 in @xmath533 , we get : @xmath537 then @xmath538 by expanding also @xmath536 in @xmath534 , we have : @xmath539 ( notice that the last equality is obtain by making @xmath540 ) . thus @xmath541 , this imply @xmath542 @xmath487 suppose @xmath543 , with @xmath528 . we have : @xmath544 then @xmath545 on the other hand : @xmath546 ( in the last equality we have used ( iv ) proposition [ braidhbhb ] ) . hence @xmath547 @xmath487 suppose @xmath548 . we have : @xmath549 then @xmath550 on the other hand , we note that @xmath551 then @xmath552 hence , @xmath553 . @xmath487 finally , let us suppose that @xmath554 . we have @xmath555 then @xmath556 we shall compute now @xmath557 . to do that , we note that splitting the square and recalling the definition of @xmath558 , we can write @xmath559 where @xmath560 hence @xmath561 on the other side , we have @xmath562 then @xmath563 as before , we split the square and then we deduce the following @xmath564 where @xmath565\mathfrak{m}_{n-2,j,\beta}^+ \\ c_1 & : = & { \rm tr}_{n-1}({\rm tr}_n(e_{n-1}b_n t_n^{m})\mathfrak{m}_{n-1,j,\beta}^+)=z\left[\sum_s y_{m - s}t_{n-2}^s\right]\mathfrak{m}_{n-2,j,\beta}^+ . \end{array}\ ] ] hence , comparing ( [ caseproof1 ] ) and ( [ caseproof2 ] ) , the claim follows . therefore the lemma is proved . [ [ section-7 ] ] in this subsection we prove that the family @xmath566 supports a markov trace . let @xmath567 be the linear map , from @xmath7 to @xmath448 , defined inductively by setting : @xmath568 and @xmath569 the definition of @xmath567 says that @xmath570 and that @xmath571 let us denote @xmath572 the family @xmath573 . the following theorem is one of our main results . [ trace ] @xmath572 is a markov trace on @xmath566 . that is , for every @xmath198 the linear map @xmath574 satisfies the following rules : 1 . @xmath570 2 . @xmath575 3 . @xmath576 4 . @xmath577 5 . @xmath578 where @xmath579 . rules ( ii)(iv ) are direct consequences of lemma [ traza1 ] ( ii ) . indeed , for example for ( ii ) , we have : @xmath580 we prove rule ( v ) by induction on @xmath15 . for @xmath190 , the rule holds since @xmath359 is commutative . suppose now that ( v ) is true for all @xmath151 less than @xmath15 . we prove it first for @xmath581 and @xmath433 . we have @xmath582 hence , @xmath583 for all @xmath433 and @xmath584 . now , we prove the rule for @xmath585 . by using lemma [ trazaimp ] , we get @xmath586 summarizing , we have @xmath587 for all @xmath433 and @xmath588 . clearly , having in mind the linearity of @xmath567 , this last equality implies that rule ( v ) holds . the rules of the trace on the topological level are illustrated in the next figure . in this section we establish the necessary and sufficient conditions by which the parameters trace @xmath591 satisfies the following equation . @xmath592 this equation plays a key role for defining knot and link invariants in the next section . in this section we will prove that if the parameters satisfy the so called @xmath589conditon , and a set of new conditions , called @xmath590condition , then the above equation holds ; see theorem [ factortr ] . finally , we will compute such trace parameters , by using the method due to p. grardin to solve the so called @xmath589system , see ( * ? ? ? * appendix ) . [ [ section-8 ] ] in @xcite certain elements @xmath593 were introduced , associated to the trace parameters of the trace on the yokonuma hecke algebra . with these @xmath593 the authors defined a non - linear system of equations called the _ e system_. we say that the solutions of this @xmath589system satisfy the @xmath589_condition_. notably , whenever the trace parameters of the markov trace on the yokonuma hecke algebra satisfy the @xmath589condition we have an invariant for framed and classical knots and links . we consider here the same formal expressions of elements @xmath593 associated now to the trace parameters @xmath594 of @xmath567 . more precisely , we define @xmath595 note that @xmath596 . also we need to introduce the following elements @xmath597 in the summations above the @xmath404 s are regarded modulo @xmath16 . the @xmath589system is defined as the non linear system of equations in @xmath598 formed by the following @xmath599 equations : @xmath600 where @xmath601 . any solution @xmath602 of the @xmath589system is referred to by saying that it satisfies the @xmath589condition . assume that @xmath602 satisfies the @xmath589condition . the @xmath590system is the homogeneous linear system of equations in @xmath603 , formed by the following @xmath16 equations : @xmath604 where @xmath338 , and @xmath605 and @xmath606 are the elements that result from replacing @xmath607 by @xmath608 in ( [ e^k ] ) and ( [ f^k ] ) respectively , that is : @xmath609 also we have that @xmath610 , see ( * ? ? ? * section 4.3 ) . thus the @xmath590system is formed by the following equations @xmath611 notice that the matrix associated to this linear system is given by : @xmath612 any solution @xmath613 of the @xmath590system is referred to saying that it satisfies the @xmath590condition . we have the following theorem . [ factortr ] we assume that the trace parameters are specialize to complex numbers @xmath602 and @xmath613 that satisfy the @xmath589condition and the @xmath590condition respectively . then @xmath614 we shall prove this theorem at the end of the subsection and using the lemmas [ h1][h3 ] below . we will introduce first the elements @xmath615 . [ h1 ] let @xmath616 , where @xmath617 . then @xmath618 hence , @xmath619 . splitting @xmath620 , we have : @xmath621 now , @xmath622 . then @xmath623 [ h2 ] let @xmath624 , where @xmath617 . then @xmath625 in particular , we have @xmath626 . we have : @xmath627 [ h3 ] let @xmath628 , with @xmath617 . then @xmath629 , where @xmath630 . we have : @xmath631 by the linearity of @xmath632 we can assume that @xmath58 is an element in the inductive basis @xmath326 . we proceed by induction on @xmath15 . for @xmath190 we have two possibilities : @xmath633 or @xmath634 . for @xmath633 , we have : @xmath635 for @xmath634 , we have : @xmath636 thus , for @xmath190 the theorem is proved . suppose now that the theorem is true for every positive integer less than @xmath65 . set @xmath58 be an element in @xmath326 . we shall prove the theorem by distinguishing the three types of form for @xmath58 . @xmath469 suppose @xmath616 , where @xmath617 . by using the lemma [ h1 ] and the fact that @xmath637 s satisfies the @xmath589condition , we have : @xmath638 @xmath469 suppose @xmath624 , where @xmath639 . then , by using now lemma [ h2 ] and the fact that the @xmath640 s satisfied the @xmath590condition , we have : @xmath641 @xmath469 finally , suppose @xmath628 , where @xmath642 . from lemma [ h3 ] , we have @xmath643 now , by using the induction hypothesis , we get @xmath644 . but , now @xmath645 and @xmath646 therefore @xmath647 the @xmath589system was solved by p. grardin , by using some tools from the complex harmonic analysis on finite groups , see ( * ? ? ? * appendix ) . however , his method works on any field having characteristic @xmath146 . we shall introduce now some notations and definitions , necessary to explain the method used by grardin , which will be used to solve the @xmath590system as well . for more details on the tools of harmonic analysis used here , see @xcite . we shall regard the group algebra @xmath648 $ ] , as the algebra formed by all complex functions on @xmath18 , where the product is the convolution product , that is : @xmath649 as usual , we denote by @xmath650 the function with support @xmath651 . recall that @xmath652 is the unity with respect to the convolution product and that @xmath653 is a linear basis for @xmath654 . the algebra @xmath654 is commutative and is the direct sum of the simple ideals @xmath655 , where @xmath656 and the @xmath657 s are the characters of @xmath658 , that is : @xmath659 in @xmath654 we have another product , the punctual product , that is : @xmath660 the algebra @xmath654 with the punctual product has unity @xmath661 and is the direct sum of its simple ideals @xmath662 , where @xmath663 . the fourier transform @xmath664 on @xmath654 is the automorphism defined by @xmath665 , where @xmath666 recall that @xmath667 , where @xmath668 . the following proposition collects the properties of the fourier transform used here . these properties are well - known and can be found , for example , in @xcite . [ fourier ] for every @xmath663 and @xmath669 . we have : 1 . @xmath670 2 . @xmath671 3 . @xmath672 4 . @xmath673 5 . @xmath674 . to solve the @xmath589system , grardin considered the elements @xmath675 , defined by @xmath676 . then , he interpreted the @xmath589system as the functional equation @xmath677 with the initial condition @xmath678 . now , by applying the fourier transform on this functional equation we obtain @xmath679 . this last equations implies that @xmath680 is constant on its support @xmath681 , where it takes the values @xmath682 . thus , we have @xmath683 by applying @xmath684 and the properties listed in the proposition above , grardin showed that the solutions of the @xmath589system are parameterized by the non empty subsets of @xmath18 . more precisely , for such a subset @xmath681 , the solution @xmath685 is given as follows . @xmath686 now , in order to solve the @xmath590system with respect to @xmath685 , we define @xmath687 by @xmath688 . then we have @xmath689 . so , to solve the @xmath590system is equivalent to solving the following functional equation : @xmath690 which , applying the fourier transform and proposition [ fourier ] ( iv ) , is equivalent to : @xmath691 this equation implies that the support of @xmath692 is contained in the support of @xmath680 . now , set @xmath681 the support of @xmath680 . then we can write @xmath693 . in this last equation , by applying @xmath684 and proposition [ fourier ] ( i ) and ( iv ) , we get : @xmath694 thus , we have proved the following proposition . the solution of the @xmath590-system with respect to the solution @xmath685 of the @xmath589system is in the form : @xmath695 where the @xmath696 s are complex numbers . in this section we define invariants for knots and links in the solid torus , by using the jones recipe applied to the pairs @xmath697 where @xmath198 . to do that , we fix from now on that the trace parameters @xmath698 satisfy the @xmath589condition and the trace parameters @xmath699 satisfy the @xmath590system , with respect to the @xmath698 s . the invariants constructed here will take values in @xmath448 . more precisely , the closure of a framed braid @xmath700 of type @xmath108 ( recall section [ framizations ] ) is defined by joining with simple ( unknotted and unlinked ) arcs its corresponding endpoints and is denoted by @xmath701 . the result of closure , @xmath701 , is a framed link in the solid torus , denoted @xmath702 . this can be understood by viewing the closure of the fixed strand as the complementary solid torus . for an example of a framed link in the solid torus see figure [ flink ] . by the analogue of the markov theorem for @xmath702 ( cf . for example @xcite ) , isotopy classes of oriented links in @xmath702 are in bijection with equivalence classes of braids of type @xmath137 and this bijection carries through to the class of framed links of type @xmath137 . for @xmath700 in @xmath120 , the markov trace @xmath572 with the trace parameters specialized to solutions of the @xmath589system and the @xmath590system , and @xmath705 the natural epimorphism of @xmath120 onto @xmath7 we define @xmath706 where @xmath707 is the exponent sum of the @xmath128 s that appear in @xmath700 . then @xmath708 is a laurent polynomial in @xmath709 and @xmath442 and it depends only on the isotopy class of the framed link @xmath701 , which represents an oriented framed link in @xmath702 . the invariants @xmath708 , when restricted to framed links with all framings equal to 0 , give rise to invariants of oriented classical links in @xmath702 . by the results in @xcite and since classical knot theory embeds in the knot theory of the solid torus , these invariants are distinguished from the lambropoulou invariants @xcite . more precisely , they are not topologically equivalent to these invariants on _ links_. as we have said previously the cyclotomic yokonuma hecke algebra @xmath192 provides a framization of the hecke algebra of type @xmath6 when @xmath193 . in @xcite where this algebra was introduced , it was also proved that @xmath192 supports a markov trace , which will be denoted here by @xmath710 , for details see ( * ? ? ? * section 5 ) . then using jones s recipe a new invariant for framed links in the solid torus is constructed , which is given by @xmath711 where @xmath712 is the natural algebra epimorphism given by @xmath713 see ( * ? ? ? * section 6.3 ) . as we see in the previous section , in order that this polynomial becomes an invariant , the trace parameters ( of @xmath710 ) have to satisfy a non - linear system of equations , which for @xmath193 is equivalent to the systems given here ( e and f system ) . + now , we would like to make some comparison between @xmath714 and @xmath715 . at first sight the invariants look similar , but the structural differences between the @xmath7 and @xmath196 make them differ ( see remark [ diff ] ) . for example , for the loop generator twice , we have the following chlouveraki m. , j. juyumaya , k. karvounis and s. lambropoulou , _ identifying the invariants for classical knots and links from the yokonuma - hecke algebras_. sumbitted for publication . see arxiv:1505.06666 . juyumaya j. and s. lambropoulou , _ an adelic extension of the jones polynomial _ , m. banagl , d. vogel ( eds . ) the mathematics of knots , contributions in the mathematical and computational sciences , vol . 1 , springer ( 2011 ) . see arxiv:0909.2545 .

two types of gaussian processes , namely the gaussian field with generalized cauchy covariance ( gfgcc ) and the gaussian sheet with generalized cauchy covariance ( gsgcc ) are considered . some of the basic properties and the asymptotic properties of the spectral densities of these random fields are studied . the associated self - similar random fields obtained by applying the lamperti transformation to gfgcc and gsgcc are studied . faculty of engineering , multimedia university , jalan multimedia , cyberjaya , 63100 , selangor darul ehsan , malaysia . faculty of information technology , multimedia university , jalan multimedia , cyberjaya , 63100 , selangor darul ehsan , malaysia .

0807.0022

generalization of some well - known stochastic processes indexed by a single parameter to processes indexed by two parameters has attracted considerable interest recently . for example , houdre and villa @xcite generalized fractional brownian motion parametrized by a single hurst index to the bifractional brownian motion characterized by two indices . another example is provided by the multidimensional stationary gaussian fields with generalized cauchy covariance indexed by two parameters introduced by gneiting and schlather @xcite . these processes can be regarded as extension of the gaussian processes with cauchy covariance used in geostatistics . for simplicity , we call such processes the gaussian field with generalized cauchy covariance ( gfgcc ) . here we would like to point out that one should not confuse such a process with a stable process with cauchy marginals . in general , processes parametrized by two indices can provide more flexibility in their applications in modeling physical phenomena . in particular , the gfgcc model has an additional nice and useful property as it allows separate characterization of fractal dimension and long range dependence ( lrd ) by two different indices . this is in contrast to models based on fractional brownian motion ( or fractional brownian noise ) which use a single index to characterize these two properties . models based on a stochastic process or field parametrized by a single index seem inadequate . a detailed analysis on network traffic carried by park et al @xcite shows that fractional brownian motion / fractional gaussian noise model is inadequate for description of network traffic for all scales since at very small time scales the traffic fluctuations are no longer statistically self - similar . the need to replace global scaling by local scaling is essential for processes such as multifractional brownian motion introduced independently by peltier and vehel @xcite and benassi , jaffard and roux @xcite . the main aim of this paper is to study gfgcc and its anisotropic counterpart , which we call gaussian sheet with generalized cauchy covariance ( gsgcc ) . in view of the fact that gfgcc is widely used in geostatistics and other applications @xcite , it will be useful to consider its properties in more detail . in these existing applications , usually only the covariance structure of gfgcc are used , and the sample path properties of gfgcc are rarely mentioned . however , a better understanding of the sample properties of gfgcc and gsgcc will render more versatility and flexibility to their applications . our approach to this subject is mainly from a physical viewpoint . basic sample properties such as the long range dependence and the local self - similarity properties of gfgcc and gsgcc are investigated . a simpler method is used to derive the asymptotic properties of the spectral densities of gfgcc and gsgcc . by generalizing the lamperti transformation to @xmath2-dimensional processes , new types of random field and random sheet with global self - similar property associated with gfgcc and gsgcc are obtained . properties of these random field and random sheet are also studied . in this section we consider gfgcc , which is a multidimensional isotropic gaussian random field in @xmath2-dimensional euclidean space . we first introduce some notations and state some basic definitions and properties of gfgcc . denote by @xmath3 , @xmath4 , @xmath5 and @xmath6 the sets of positive integers , integers , real numbers and positive real numbers respectively . let @xmath7 and @xmath8 be two vectors in @xmath9 , and @xmath10 , @xmath11 be its euclidean norm . by @xmath12 and @xmath13 , we mean @xmath14 and @xmath15 respectively for all @xmath16 . a random field @xmath17 on @xmath9 is called a gaussian field with generalized cauchy covariance ( or gfgcc ) if it is a stationary gaussian field with mean zero and covariance given by @xmath18 where @xmath19 $ ] and @xmath20 . note that has the same functional form as the characteristic function of the generalized multivariate linnik distribution first studied by anderson @xcite . @xmath21 is positive - definite for the above ranges of @xmath22 and @xmath23 , and it is completely monotone for @xmath24 , @xmath25 . @xmath17 becomes the gaussian field with usual cauchy covariance when @xmath26 . recall that a random field @xmath27 is @xmath28-self - similar ( @xmath28ss ) if @xmath29 , where @xmath30 denotes equality in the sense of finite - dimensional distributions of @xmath31 . self - similar property requires scale invariance to hold for all scales . this is rather too restrictive for many applications . we also know from samorodnitsky and taqqu @xcite that a stationary gaussian random field such as @xmath17 can not be a self - similar field . however , @xmath17 satisfies a weaker self - similar property known as local self - similarity considered by kent and wood @xcite . [ def2]let @xmath32 $ ] . a centered stationary gaussian field is locally self - similar ( lss ) of order @xmath33 if for @xmath34 , its covariance @xmath35 satisfies @xmath36\end{aligned}\]]for some positive constants @xmath37 , @xmath38 and @xmath39 . since as @xmath34 , @xmath40,\end{aligned}\ ] ] the gfgcc @xmath17 is @xmath33 lss , with @xmath41 , @xmath42 and @xmath43 . adler @xcite called the class of gaussian fields which satisfy the indexed-@xmath22 fields . these processes are also known as the adler processes according to some authors , for example lang and roueff @xcite . they form a very rich class of gaussian random fields , which include the centered gaussian field @xmath44 with powered exponential covariance @xmath45which have the same functional form as the characteristic function of the multivariate symmetric stable distribution as given in kotz , kozubowski and podgorski @xcite , and garoni and frankel @xcite . instead of using to characterize local self - similarity , one can also use the definition of locally asymptotically self - similar ( lass ) property first introduced by benassi , jaffard and roux @xcite for multifractional brownian motion @xmath46 , which is a generalization of fractional brownian motion with the hurst index replaced by the hurst function @xmath47 , @xmath48 . it can be shown that under some regularity conditions on @xmath47 , the multifractional brownian motion @xmath46 is lass . this property can be adapted to gfgcc if we take @xmath47 as constant with its value in ( 0,1 ) . [ def1 ] a stochastic process @xmath27 is lass at a point @xmath49 with order @xmath50 if @xmath51and @xmath52 is nontrivial . here the convergence and equality are in the sense of finite dimensional distributions , and @xmath52 is called the tangent field of @xmath27 at the point @xmath49 . [ p1 ] gfgcc is a lass random field of order @xmath33 ; and its tangent field is l@xmath53vy fractional brownian field of index @xmath33 . by using , the covariance of the increment field @xmath54 , for @xmath55 , @xmath56 is given by @xmath57 let @xmath58 and @xmath59 , then @xmath60 where @xmath61 is the index@xmath33 l@xmath53vy fractional brownian field with zero mean and covariance given by @xmath62 thus up to a multiplicative constant @xmath63 , the tangent field of gfgcc at any point @xmath64 is the l@xmath53vy fractional brownian field indexed by @xmath33 . the tangent field of @xmath17 at the point @xmath49 reflects the local structure of the random field at @xmath49 . in other words , gfgcc behaves locally like a l@xmath53vy fractional brownian field . this provides an example to the general results on tangent fields considered by falconer @xcite . the fractal dimension of the graph of a random field @xmath27 depends on the local property of the random field . the local irregularities of the graph are measured by the parameter @xmath22 , which can be regarded as the fractal index of the random field . thus the behavior of the covariance function at the origin to a great extent determines the roughness of the random field . the results on the fractal dimension of an lss field are treated by adler @xcite , kent and wood @xcite , davies and hall @xcite . a nice property of @xmath33lss fields is that their fractal dimension is determined by @xmath22 . let @xmath27 be a stationary gaussian field and let @xmath65 be the variance of the increment process @xmath66 . if there exists @xmath67 $ ] satisfying @xmath68 then @xmath33 is called the fractal index of the random field @xmath27 . equivalently , @xmath33 is the local hlder index of the random field . clearly , condition is fulfilled by random field which satisfies such as the gfgcc @xmath17 . adler had shown that @xmath33 is the upper bound of the indices for which , with probability one , the graph of @xmath27 satisfies a global regularity of the same order . thus , @xmath22 characterizes the roughness of the sample path . the fractal dimension of gfgcc can be obtained by using the following result for the fractal ( hausdorff ) dimension of an lss field as given in adler @xcite , chapter 8 . [ p1026_4 ] the fractal dimension @xmath69 of the graph of a locally self - similar field @xmath70 , of fractal index @xmath33 , over a hyperrectangle @xmath71 $ ] , is given by @xmath72 the estimation of @xmath22 for lss field has been studied extensively , see for example wood and chan @xcite , istas and lang @xcite , kent and wood @xcite , lang and roueff @xcite . the parameter @xmath23 is also known as topothesy in the studies of roughness of surfaces by wood and chan @xcite , and davies and hall @xcite . the topothesy of a cross section provides a measure of the roughness which is scale dependent in contrast to fractal dimension which is scale invariant . the gaussian random field @xmath17 can have srd ( short range dependence ) or lrd ( long range dependence ) , depending on the values of the parameters @xmath22 and @xmath23 . for this purpose we make use of the following definition which is a generalization of the one - dimensional case considered by flandrin et al @xcite , lim and muniandy @xcite : a stationary centered gaussian field with covariance @xmath35 is said to be a long range dependent process if @xmath73 otherwise it is short range dependent . [ p2]the gfgcc @xmath17 is a long range dependent random field if and only if @xmath74 . in order to obtain the condition for the gaussian random field with covariance to be lrd , we make use of the following integral identity given in gradshteyn and ryzhik @xcite , 3.251 , no.11 : @xmath75 where @xmath76 , @xmath77 and @xmath78 is the beta function . using polar coordinates , one gets @xmath79 for large @xmath80 , @xmath81 therefore , the integral is divergent for all @xmath20 , @xmath74 . for @xmath20 and @xmath82 , we have @xmath83 therefore the condition for @xmath17 to be a gaussian field with lrd is @xmath74 . the discussion above shows that it is possible to characterize the fractal dimension @xmath69 and the lrd property separately . if the covariance is re - expressed as @xmath84 , which behaves like @xmath85 in the large@xmath86 limit , then @xmath87 is lrd if and only if @xmath88 . thus @xmath22 ( @xmath89 ) and @xmath90 ( @xmath91 ) respectively provide separate characterization of fractal dimension and lrd / srd . the separate characterization of the fractal dimension ( local property ) and lrd ( global property ) for gfgcc appears to offer a more natural and flexible model than that based on a single parameter such as in l@xmath53vy fractional brownian field . we note that this feature of separate characterization of local self - similarity ( hence fractal dimension ) and long range dependence is present in any stationary gaussian field with covariance @xmath35 satisfying the asymptotic behaviors @xmath92 as @xmath93 , and @xmath94 as @xmath95 , with @xmath19 $ ] , @xmath91 . similarly , one can also have a gaussian stationary process which has separate parametrization of fractal dimension and short range dependence @xcite . the ability to have separate characterization of fractal dimension and hurst effect is a desirable property in the modeling of physical and geological phenomena . in this section , we consider the spectral density of @xmath17 and its asymptotic properties . though the covariance of gfgcc is given by a relatively simple expression , the analytic simplicity of the covariance function is not inherited by the corresponding spectral density . this is similar to the case of the stationary gaussian field @xmath44 with powered exponential covariance which have simple form , but its spectral density in general does not have closed analytic expression . in the case of gfgcc in @xmath5 , a detailed study of spectral densities ( in terms of probability distributions correspond to the characteristic functions of generalized linnik distributions ) have been carried out by kotz et al . @xcite for @xmath96 , @xmath97 , @xmath98 ; by ostrovskii @xcite for @xmath96 , @xmath97 , @xmath99 ; and by erdogan and ostrovskii @xcite for @xmath96 , @xmath20 , @xmath98 . they employed the contour integration representations and series expansions of the generalized linnik distributions . however the techniques used in these works are less accessible to practitioners . in this section , we derive the asymptotic properties of the spectral densities of gfgcc for @xmath89 and @xmath20 , which can be regarded as an extension to the results on generalized multivariate linnik distributions . the techniques used in our derivations are mathematically more tractable . recall that the spectral density @xmath100 of a stationary field @xmath27 is defined as the fourier transform of its covariance function @xmath101 : @xmath102if the integral is convergent . if the integral does not converge , we consider @xmath103 as a generalized function and define @xmath100 as the fourier transform of @xmath103 in the schwartz space of test functions @xcite . namely , for any test function @xmath104 in the schwartz class of @xmath9 , we require @xmath105where @xmath106 alternatively , the spectral density can also be defined to be the function satisfying @xmath107 for gfgcc , since @xmath108as @xmath109 ( @xcite , page 209 ) , we find that when @xmath110 , it s spectral density is @xmath111where @xmath112 is the bessel function . when @xmath113 , using no.4 of 6.565 in @xcite , we have the explicit formula @xmath114if @xmath115 . here @xmath116 is the modified bessel function . the formula no.7 of 6.576 in @xcite shows that if @xmath117 , then @xmath118therefore is still the spectral density when @xmath119 $ ] . for general @xmath120 , no explicit formula such as can be found for @xmath121 . when @xmath98 , the formula gives the spectral density of @xmath17 for all values of @xmath19 $ ] and @xmath20 . for @xmath122 , we would also like to find a formula for the spectral density that is valid for all @xmath123 and @xmath20 . for this purpose , it would be beneficial to investigate the case @xmath98 first . when @xmath98 , we can rewrite as @xmath124let @xmath125 and consider the region @xmath126 in the complex plane defined by @xmath127when @xmath128 , the function @xmath129 is an analytic function on the domain @xmath126 . therefore , by cauchy integral formula , @xmath130notice that the boundary of @xmath126 , @xmath131 , consists of three components : the line segment @xmath132 along the real axis from @xmath133 to @xmath80 , the arc @xmath134 of the circle @xmath135 from @xmath80 to @xmath136 , and the line segment @xmath137 along the imaginary axis from @xmath136 to @xmath133 . on the arc @xmath134 , if @xmath138 , then @xmath139therefore @xmath140and implies that @xmath141this gives us @xmath142 for @xmath122 , we can derive a formula similar to . recall that the hankel s function of the first kind @xmath143 is defined as @xmath144 where @xmath145 is the modified bessel function of the second kind or called the neumann function . using hankel s function , we can rewrite as @xmath146for @xmath109 , we have ( @xcite , no.3 of 8.451 ) @xmath147therefore , we can show as in the @xmath98 case that @xmath148using the formula ( @xcite , no.1 of 8.407 ) @xmath149 we have finally @xmath150this formula agrees with the formula for multivariate linnik distribution proved in @xcite for @xmath151 and @xmath99 . notice that the right hand side of is well - defined for all @xmath152 . using the formula ( no.2 of 6.521 in @xcite ) , @xmath153we have @xmath154 @xmath155 when @xmath123 , residue calculus implies that this last integral is equal to @xmath156this shows that is indeed the spectral density of gfgcc for all @xmath123 and @xmath20 . we would also like to remark that although the formula is derived under the assumption @xmath122 , but since @xmath157 therefore when @xmath98 , the formula reduces to the formula . we summarize the result as follows . if @xmath123 and @xmath20 , the spectral density of the gfgcc @xmath17 is given by @xmath158if @xmath113 and @xmath20 , the spectral density of the gfgcc @xmath159 is given by @xmath160 to find the high frequency behavior of the spectral density , we first consider the case where @xmath113 . using the fact that ( @xcite , no.6 of 8.451 ) as @xmath109 , @xmath161this implies that as @xmath162 , @xmath163with leading term @xmath164 for general @xmath123 , to find the high frequency behavior of @xmath121 , we make use of eq . . making a change of variable and using @xmath165 as @xmath166 , we find that @xmath167using the formula ( @xcite , no.16 of 6.561 ) @xmath168and the definition of asymptotic expansion ( @xcite , page 611 ) , we conclude that as @xmath169 , @xmath121 behaves asymptotically as @xmath170 notice that there is a drastic change of high frequency limit of @xmath121 when @xmath120 and @xmath113 . in fact , naively putting @xmath113 in give identically zero terms . this is a hint that as @xmath162 , @xmath171 does not have polynomial decay , instead it decays exponentially as is verified by . we summarize the results as follows : [ p1113_1 ] if @xmath123 and @xmath20 , the high frequency limit of the spectral density @xmath121 is given by the following asymptotic series @xmath172if @xmath113 and @xmath20 , the high frequency limit of the spectral density @xmath171 is given by the following asymptotic series @xmath173 when @xmath174 and @xmath175 , agrees with the results given in @xcite and @xcite respectively . in particular , we observe that when @xmath123 , the high frequency behavior of the spectral density of gfgcc is @xmath176which is independent of @xmath23 . kent and wood @xcite have shown that if a random field has spectral density satisfying , then its covariance satisfies with locally self similar property . however , the converse is not true . in this connection we remark that for the gaussian stationary field @xmath44 with powered exponential covariance which is lss with @xmath177 , \hspace{1cm}\vert\tau\vert\rightarrow 0^+,\end{aligned}\]]its small @xmath86 behavior has a similar form as that of gfgcc . thus it is not surprising that this two random fields have the same tail behavior for their spectral densities at high frequencies as given by . the detailed calculation carried out by garoni and frankel @xcite for the probability distribution of the multivariate l@xmath53vy stable distribution with characteristic function given by confirms this . we also note that can be used to verify that the tangent field at any point @xmath49 has spectral density which varies as @xmath178 for @xmath179 . if we let @xmath180 , then the tangent field is just the l@xmath53vy fractional brownian field in @xmath9 . such a relationship can be viewed as a consequence of the tauberian abelian theorem ( see e.g. @xcite ) . for the low frequency behavior of the spectral density @xmath121 , we first consider the case @xmath113 . using 8.485 , 8.445 , 8.446 of @xcite , we find that if @xmath181 , @xmath182whereas when @xmath183 , where @xmath184 is a nonnegative integer , @xmath185here @xmath186 is the logarithm derivative of the gamma function . therefore from , we find that : @xmath187 if @xmath188 , then as @xmath189 , @xmath190 @xmath187 if @xmath191 , then as @xmath189 , @xmath192where @xmath90 is the euler constant . @xmath187 if @xmath193 , then as @xmath189 , @xmath194 in fact , by considering the cases @xmath195 and @xmath196 separately and substituting the series and into , we can express the spectral density @xmath171 in terms of convergent power series in @xmath197 . = .8 = .8 = .8 = .8 = .8 = .8 for general @xmath123 , the low frequency behavior of @xmath121 depends on the arithmetic nature of @xmath22 and @xmath23 . the method we are going to employ does not allow the derivation of the whole asymptotic series as obtained by kotz et al @xcite , erdogan and ostrovskii @xcite . we will only derive the leading behavior of the spectral density @xmath121 , which only depends on the algebraic conditions @xmath82 , @xmath198 or @xmath199 . these conditions are less stringent than the arithmetic conditions considered in @xcite . however , the simpler method employed here provides the necessary @xmath200 , @xmath189 asymptotic behaviors which are sufficient for most practical purposes . when @xmath82 , since ( @xcite , eq . 8.402 ) @xmath201we find from that as @xmath202 , @xmath203(@xcite , no.11 of 3.251 ) . when @xmath198 , we make a change of variable on to get @xmath204 in view of the leading behavior of the bessel function @xmath112 as @xmath205 , we write @xmath121 as the sum of two terms @xmath206 and @xmath207 where @xmath208and @xmath209as @xmath202 , and show that @xmath206 has a finite limit . by lebesgue s dominated convergence theorem , the limit is given by @xmath210 . namely @xmath211this expression can be evaluated using regularization method . more precisely , using lebesgue s dominated convergence theorem again , we find that @xmath212the formula no.14 of 6.561 in @xcite then gives @xmath213for the term @xmath207 , we make a change of variable @xmath214 or equivalently @xmath215 , to get @xmath216we split @xmath207 again into a sum of two terms @xmath217 and @xmath218 , where@xmath219and@xmath220when @xmath189 , @xmath221the integral is a convergent integral with value given by ( @xcite , 3.219 page 316 ) @xmath222 putting everything together , we find that @xmath223therefore , as @xmath224 , @xmath225 when @xmath199 , care has to be taken since is defined only for @xmath110 . making a change of variable , one finds that @xmath226it is easy to verify that for any @xmath227 and @xmath20 , when @xmath189 , @xmath228now using no.1 of 6.621 in @xcite , we have @xmath229where @xmath230 is the hypergeometric function @xmath231using the formula no.2 of 9.131 in @xcite , we find that @xmath232since @xmath233therefore , @xmath234where we have used the formula @xmath235 . this gives for @xmath236 , @xmath237it is easy to verify that by putting @xmath113 in , and , we get back , and respectively . the low frequency behaviors of @xmath121 are summarized below . [ p1113_2]for all @xmath19 $ ] and @xmath20 , the low frequency limit of the spectral density @xmath121 is given by @xmath238 we see that if @xmath74 , @xmath121 is divergent at the origin and if @xmath236 , @xmath239 as @xmath189 . note that the condition @xmath74 agrees with the lrd condition . in fact , it is a basic fact that for a stationary field with positive covariance , it is lrd if and only if its spectral density diverges at the origin . we would also like to point out that the low frequency limit of the spectral density of the gaussian field with powered exponent covariance @xmath44 is finite as shown by garoni and frankel @xcite , in agreement with the fact that @xmath44 is srd . we remark that by considering @xmath240 we can find the next order term in the asymptotic expansion of @xmath121 when @xmath189 using the same methods we employed above . the results depend on more complicated conditions on @xmath22 and @xmath23 . we also briefly remark that the asymptotic behavior of the spectral density at low frequency is connected to the large time behavior of the covariance function . for the covariance function @xmath21 which satisfies @xmath241where @xmath242 is a slowly varying function for large @xmath86 , i.e. @xmath243 as @xmath244 for all positive constant @xmath245 , hardy littlewood karamata tauberian theorem ( see e.g. @xcite ) implies that the spectral density @xmath121 has the following asymptotic behavior @xmath246where @xmath247 we can also introduce another type of @xmath2-dimensional process @xmath248 , called gaussian sheet with generalized cauchy covariance ( gsgcc ) . the gsgcc is a centered gaussian random field @xmath248 indexed by two multidimensional parameters @xmath249 and @xmath250 , with @xmath251 $ ] , @xmath252 for all @xmath16 , and with covariance given by @xmath253where @xmath254 is the covariance of one - dimensional gfgcc indexed by @xmath255 and @xmath256 . heuristically , one can regard @xmath248 as the product of @xmath2 independent one - dimensional gfgcc processes @xmath257 . however , since the product of independent normal random variables is not a normal random variable , we shall not write @xmath258 . nevertheless , it is true that for any integer @xmath259 , @xmath260^m\right)=\prod_{i=1}^n e\left(\left[x_{\alpha_i,\beta_i}(t_i)\right]^m\right).\ ] ] the gsgcc @xmath248 is an anisotropic gaussian random field and it does not satisfy the definition of local self similarity given in definition [ def2 ] . however , we can show that it is lass and has tangent field according to definition [ def1 ] . assume that @xmath261 . namely , for some @xmath262 , @xmath263 , and for all @xmath264 , @xmath265 . here @xmath266 [ p1026_2]gsgcc is lass with tangent field @xmath267 , which is a stationary centered gaussian field with covariance @xmath268 from definition , it is easy to see that@xmath269\right\ } \hspace{0.5cm}\text{as}\hspace{0.5cm}|\tau_i|\rightarrow 0^+ , \;i=1,\ldots , n . \end{aligned}\ ] ] therefore , @xmath270 from this proposition we see that in general , the tangent field @xmath271 of the gsgcc @xmath248 defined by definition [ def1 ] are uncorrelated in some of the directions of @xmath9 . it does nt fully capture the locally self similar property of @xmath248 . there is another measure of local self similarity , called local multiple self - similarity that better describe this kind of process . recall that @xcite a stochastic process @xmath27 is called multi - self similar ( mss ) of index @xmath272 if and only if for any @xmath273 , @xmath274 it is obvious that a @xmath28-mss field is a @xmath275$]-ss field , but in general a self similar field is not multi self similar . the notion of lss can be generalized accordingly . a centered stationary gaussian field is @xmath2-ple locally self similar of order @xmath33 if its covariance function @xmath35 satisfies for @xmath276 , @xmath277\right),\end{aligned}\]]for some @xmath278 , @xmath279 . in view of , it is easy to see that @xmath248 is @xmath2-ple locally self similar with @xmath280 , @xmath281 and @xmath282 . we also note that the local multi - self similarity is equivalent to @xmath283\;\;\text{as}\;\ ; \tau_i\rightarrow 0^+,\end{aligned}\ ] ] for any @xmath284 . here @xmath285 is the unit vector in the @xmath286 direction . this can also be rephrased as @xmath2-ple lass . @xmath248 is @xmath2-ple lass . more precisely , for every @xmath16 , @xmath287where @xmath61 is the one dimensional fractional brownian motion of index @xmath33 . the proof is the same as proposition [ p1 ] . one should observe that the @xmath2-ple lass here is different from the lass defined in definition [ def1 ] . we take the limit process in each of the @xmath286direction separately and the limit process can be regarded as the partial derivative process of @xmath248 in the @xmath286-direction . a disadvantage of this definition is that the limit process of an @xmath2-dimensional field is a one - dimensional process . therefore , we will define another limit process that will better reflect local multi - self similarity , as were considered in @xcite and @xcite . given a stochastic process @xmath288 , we define its total increment by @xmath289 at the point @xmath290 as @xmath291when @xmath98 , this is the same as the increment , i.e. @xmath292 . when @xmath293 , we have @xmath294which is also known as the rectangular increment . for general @xmath2 , given @xmath295 , the set of points @xmath296 are the vertices of the hyperrectangle with @xmath297 and @xmath298 as one of the main diagonals . by giving the vertex @xmath298 a weight @xmath299 , the other vertices of the hyperrectangle can be given a weight @xmath299 and @xmath300 alternatingly so that adjacent vertices has different weights . @xmath301 is then the weighted sum of the field @xmath27 at the vertices of the hyperrectangle . recall that the fractional brownian sheet @xmath302 of index @xmath303 is a centered gaussian process with covariance @xmath304it is well known that @xmath305 is a self similar field of order @xmath306 , and multi self similar of order @xmath33 . however , unlike the l@xmath53vy fractional brownian field , the fractional brownian sheet is not a process with stationary increment . nevertheless , it is a process with stationary total increment , i.e. , @xmath307 now returning to the local asymptotic multi self similarity of gsgcc , we can show the following : [ p1026_5]@xmath308 b_{\alpha/2}^{\#}(u).\end{aligned}\]]here @xmath309 . if we heuristically write @xmath248 as @xmath310 , then heuristically the total increment @xmath311 can be written as the product of increments @xmath312 , and the result follows from proposition [ p1 ] . for a more rigorous proof , notice that @xmath313e_i\right)\\ = & \sum_{\delta\in \{0,1\}^n}\sum_{\eta\in\{0,1\}^n}(-1)^ { \sum_{i=1}^n \delta_i+\sum_{i=1}^n \eta_i } \prod_{i=1}^n c_{\alpha_i,\beta_i}\left({\varepsilon}_i \left[\delta_i u_i-\eta_i v_i\right ] \right)\\ = & \prod_{i=1}^n \bigl\{c_{\alpha_i,\beta_i}(0)-c_{\alpha_i,\beta_i}({\varepsilon}_i u_i)-c_{\alpha_i,\beta_i}({\varepsilon}_i v_i)+c_{\alpha_i,\beta_i}({\varepsilon}_i[u_i - v_i])\bigr\}\\ = & \prod_{i=1}^n \left\langle \delta_{{\varepsilon}_i u_i}x_{\alpha_i,\beta_i}(t_i)\delta_{{\varepsilon}_i v_i}x_{\alpha_i,\beta_i}(t_i)\right\rangle.\end{aligned}\]]the result follows . from this proposition , we see that the total increment process can capture the locally multi - self similar property of the gsgcc better . it also shows that locally , gsgcc behaves similarly as the fractional brownian sheet . one would tend to use proposition [ p1026_2 ] and proposition [ p1026_4 ] to conclude that the hausdorff dimension of the graph of gsgcc over a hyperrectangle is @xmath314 however , proposition [ p1026_4 ] can not be applied here since proposition [ p1026_2 ] does not imply that the fractal index of gsgcc is @xmath315 . for the fractional brownian sheet @xmath316 , kamont @xcite showed that the hausdorff dimension of its graph is bounded above by @xmath317 . in @xcite , ayache used wavelet method to show that the hausdorff dimension of the graph of the fractional brownian sheet @xmath316 is indeed equal to @xmath317 . this fact was proved again by ayache and xiao @xcite as a special case of a more general result on hausdorff dimension of fractional brownian sheets from @xmath9 to @xmath318 , using a different method . therefore , it is natural for us to conjecture that the hausdorff dimension of the graph of gsgcc is also @xmath317 . in fact , implies that @xmath319 ^ 2\right\rangle = o\left(\vert \tau\vert^{\min\alpha}\right)\hspace{1cm}\text{as}\;\ ; \vert\tau\vert\rightarrow 0^+.\end{aligned}\ ] ] by a well - known theorem ( see e.g. @xcite ) , this implies that the hausdorff dimension of the graph of gsgcc is bounded above by @xmath317 . on the other hand , it is easy to show that [ lemma1122_1 ] let @xmath320 $ ] be a hyperrectangle in @xmath9 . there exist constants @xmath321 and @xmath322 such that @xmath323 ^ 2\right\rangle \leq c_2\sum_{i=1}^n |t_i - s_i|^{\alpha_i},\end{aligned}\ ] ] for all @xmath324 . then by adapting the proof in @xcite for the lower bound on the hausdorff dimension of general anisotropic gaussian fields with stationary increments , it can be verified that the graph of gsgcc over a hyperrectangle has hausdorff dimension equal to @xmath317 . the condition for lrd can be generalized to random sheet and it becomes @xmath325 since @xmath326 and proposition [ p2 ] for @xmath98 gives @xmath327we have gsgcc is lrd if and only if for some @xmath279 , @xmath328 ; it is srd if and only if @xmath329 for all @xmath279 . in order to determine the asymptotic behavior for the spectral density of the generalized cauchy sheet , we observe that the spectral density of gsgcc , which we denote by @xmath330 can be expressed as products of spectral densities of one dimensional isotropic gfgcc considered in section [ sec3 ] . namely , @xmath331 therefore , the high frequency and low frequency behaviors of the spectral density @xmath330 can be obtained from proposition [ p1113_1 ] and proposition [ p1113_2 ] respectively . the high frequency limit of the spectral density @xmath330 is given by @xmath332where @xmath333and@xmath334 the low frequency limit of the spectral density @xmath330 is given by @xmath335where@xmath336 here we have used the fact that @xmath337 ( @xcite , no.1 of 8.335 ) , @xmath338 and @xmath339 ( @xcite , no.2 of 8.366 ) . in his seminal paper @xcite , lamperti introduced a transformation which provides a one to one correspondence between a self similar process and a stationary process . for a stationary process @xmath27 , @xmath340 , we let @xmath341 for @xmath342 , @xmath343 , and @xmath344 be its @xmath28-lamperti transfrom . then @xmath345 is an @xmath28-self similar ( @xmath28-ss ) process . conversely , if @xmath346 is @xmath28-ss with zero mean , then the inverse lamperti transformation of @xmath345 defined by @xmath347 is a stationary process . there are two ways to define extensions of lamperti transformation to @xmath9 linking stationary random field to a self - similar random field . given a stationary random field @xmath288 , the first way to define its lamperti transformation is , given @xmath348 , defined by @xmath349for @xmath350 . it is easy to show that [ p3 ] let @xmath351 be the @xmath28-lamperti transform of a stationary field @xmath70 defined by . then @xmath345 is a @xmath28-ss field . for any @xmath284 , we have @xmath352,\ldots , \ln [ ct_n]\right)\\ & = _ dc^h\vert t\vert^h x(\ln t_1 + \ln c,\ldots , \ln t_n+\ln c)\\ & = _ d c^h \vert t\vert^h x(\ln t_1,\ldots , \ln t_n)\\ & = _ d c^h y(t),\end{aligned}\]]i.e . @xmath345 is @xmath28-ss . the inverse to this lamperti transformation transforms a random field @xmath345 to @xmath353for @xmath354 . for a field @xmath355 to be stationary , it must satisfy @xmath356 for all @xmath357 . this is an @xmath2-parameter family of conditions . however , in proving that @xmath345 is @xmath28-ss , we only use one parameter family of these conditions , i.e. , we only look at those @xmath358 of the form @xmath359 . therefore we can not expect that the inverse lamperti transform of a @xmath28-ss field is a stationary field . as an example , consider the l@xmath53vy fractional brownian field with index @xmath28 , @xmath360 , which is a @xmath28-ss field . define its inverse @xmath28lamperti transform @xmath361 by . we find that its covariance is @xmath362 ^ 2+\ldots+[e^{t_n+\tau_n}-e^{t_n}]^2\right)^{h}\biggr\}.\end{aligned}\]]it is easy to verify that for @xmath122 , this expression is not independent of @xmath354 . therefore , @xmath361 is not a stationary process . to make full use of the @xmath2-parameter families of symmetry in a stationary process , there is another definition of lamperti transformation introduced by @xcite . given a stationary process @xmath70 , and a multi - index @xmath363 , the second @xmath28-lamperti transform of @xmath27 , denoted by @xmath364 , is defined as @xmath365for @xmath350 . it was shown in @xcite that [ p4 ] let @xmath366 be the @xmath28-lamperti transform of a stationary field @xmath70 defined by . then @xmath364 is a @xmath28mss field . the inverse of this second lamperti transformation transform a field @xmath367 to @xmath70 , where @xmath368 it has a nice property , namely , as was shown in @xcite : let @xmath288 be the inverse second @xmath28lamperti transform of a @xmath28-mss field @xmath367 defined by . then @xmath27 is a stationary field . as a side remark , it can be shown that is essentially the unique ( up to some multiplicative constants ) transformation that takes a stationary process to a @xmath28-mss . however , if for @xmath369 , @xmath370 is a function of the form @xmath371then if @xmath372,\ldots , c[m]\in { \mathbb{r}}_+^n$ ] , @xmath373 and @xmath374 are such that @xmath375 , it is easy to verify that the transform @xmath376,\beta_i}(t)^{\gamma_i}x(\ln t_1,\ldots , \ln t_n)\ ] ] takes a stationary process to a @xmath28ss process , but its inverse in general does not take a @xmath28ss process to a stationary process . the first lamperti transformation corresponds to @xmath377 , @xmath372_1=\ldots = c[1]_n=1 $ ] , @xmath378 and @xmath379 . returning to the gfgcc @xmath17 and the gsgcc @xmath248 , proposition [ p3 ] shows that their first lamperti transforms , @xmath380 and @xmath381 , are @xmath28ss fields with covariances @xmath382^{-\beta},\end{aligned}\]]and @xmath383^{-\beta_i}\end{aligned}\]]respectively . on the other hand , proposition [ p4 ] shows that their second lamperti transforms , @xmath384 and @xmath385 , are @xmath28mss fields with covariances @xmath386^{-\beta},\end{aligned}\]]and @xmath387^{-\beta_i}\end{aligned}\]]respectively . next we consider whether the lrd property of gfgcc and gsgcc is preserved under the lamperti transformations . first we note that the lrd condition can be extended to non - stationary field in the following way . let @xmath388 be the correlation function of a non - stationary gaussian field @xmath27 . then the condition of lrd for the non - stationary gaussian field @xmath27 is given by @xmath389 + a. the @xmath28-ss gaussian field @xmath380 is lrd for all @xmath19 $ ] and @xmath20 . + b. the @xmath28-ss gaussian field @xmath381 is lrd for all @xmath390 satisfying @xmath251 $ ] and @xmath252 , @xmath279 . + c. the @xmath28-mss gaussian field @xmath384 is lrd for all @xmath19 $ ] and @xmath20 . + d. the @xmath28-mss gaussian field @xmath385 is lrd for all @xmath390 satisfying @xmath391 $ ] and @xmath252 , @xmath279 . + it is easy to verify that both the processes @xmath380 and @xmath384 have the same correlation function @xmath392\right)^2\right)^{\alpha/2}\right]^{-\beta } , \hspace{0.5cm}\tau\in { \mathbb{r}}_+^n,\nonumber\end{aligned}\]]which is independent of @xmath28 . in order to show that these processes are lrd , we have by condition , @xmath393\right)^2\right)^{\alpha/2}\right]^{-\beta}d^n\tau\\ = & \left[\prod_{i=1}^n t_i\right]\int\limits_{{\mathbb{r}}_+^n}\left[1+\left(\sum_{i=1}^n\left(\ln \left[1+u_i\right]\right)^2\right)^{\alpha/2}\right]^{-\beta}d^nu\\ = & \left[\prod_{i=1}^n t_i\right]\int\limits_{{\mathbb{r}}_+^n}\left[1+\left(\sum_{i=1}^nv_i^2\right)^{\alpha/2}\right]^{-\beta}\left[\prod_{i=1}^n e^{v_i}\right]d^nv,\end{aligned}\ ] ] where we have used the substitutions @xmath394 and @xmath395 . clearly , when @xmath396 , the integrand also approaches @xmath397 . therefore , the last integral diverges for all @xmath398 , which is the condition for @xmath380 and @xmath384 to be lrd . the statements for @xmath399 and @xmath385 are proved analogously . note that the lrd property of gfgcc and gsgcc is preserved under the lamperti transformations , but the srd property is not preserved . here we have an example that the application of lamperti transformation to a lrd stationary process ( in this case the gfgcc and gsgcc ) gives a ( multi)self similar process with lrd . examples of lamperti transformation encountered so far relate either two short memory processes ( for example , ornstein - uhlenbeck process and brownian motion ) , or between a srd process and a lrd process ( in the case of fbm and its inverse lamperti transformed process ) . for examples , one can show that the inverse lamperti transformation of fractional brownian sheet with lrd property gives rise to a stationary random sheet with short range dependence , and the stationary field associated with the l@xmath53vy fractional brownian field is a stationary field with srd . recall that l@xmath53vy fractional brownian field is essentially the only self similar gaussian field with stationary increments @xcite . similarly , one can show that fractional brownian sheet is essentially the only multi similar gaussian random field that has stationary total increments . hence @xmath380 , @xmath399 , @xmath384 and @xmath385 , the self - similar and multi self similar fields associated with gfgcc and gsgcc , do not have stationary increments or total increments . there exists a weaker stationary property known as asymptotically locally stationarity , which requires the field to be stationary in the limit @xmath34 . a centered gaussian random field @xmath27 is said to have asymptotically locally stationary increment if and only if as @xmath34 , the variance of its increment @xmath400 ^ 2\right\rangle$ ] is independent of @xmath297 . more precisely , @xmath401where @xmath402 is independent of @xmath297 , and for any fixed @xmath297 , @xmath403 as functions of @xmath404 . similarly , a centered gaussian random field @xmath27 is said to have asymptotically locally stationary total increment if and only if as @xmath34 , the variance of its total increment @xmath405 ^ 2\right\rangle$ ] is independent of @xmath297 . for the random field @xmath380 , we have @xmath406 ^ 2\right\rangle = \left\langle \left[y_{\alpha,\beta}(t+\tau)\right]^2\right\rangle+ \left\langle \left[y_{\alpha,\beta}(t)\right]^2\right\rangle-2\bigl\langle y_{\alpha,\beta}(t+\tau ) y_{\alpha,\beta}(t)\bigr\rangle\\ & = \vert t+\tau\vert^{2h } + \vert t\vert^{2h}-2\vert t+\tau\vert^{h}\vert t\vert^h \left[1+\left(\sum_{i=1}^n \left(\ln \left[1+\frac{\tau_i}{t_i}\right]\right)^2\right)^{\alpha/2}\right]^{-\beta}.\end{aligned}\]]using @xmath407and @xmath408\right)^2\right)^{\alpha/2}\right]^{-\beta}= & \left[1+\left(\sum_{i=1}^n \left ( \frac{\tau_i}{t_i}+o(\tau_i^2)\right)^2\right)^{\alpha/2}\right]^{-\beta}\\ = & \left[1+\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2+o(\vert \tau\vert^3 ) \right)^{\alpha/2}\right]^{-\beta}\\ = & \left[1+\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2\right)^{\alpha/2}+o(\vert \tau\vert^{\alpha+1 } ) \right]^{-\beta}\\ = & 1-\beta\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2\right)^{\alpha/2}+o\left(\vert\tau\vert^{\min\{2\alpha , \alpha+1\}}\right),\end{aligned}\]]we find that as @xmath409 , @xmath410 ^ 2\right\rangle=&\vert t\vert^{2h}\left(1 + 2h\frac{\sum_{i=1}^n t_i\tau_i}{\vert t\vert^2}\right)+\vert t\vert^{2h } \\\nonumber & - 2\vert t\vert^{2h}\left(1+h\frac{\sum_{i=1}^n t_i\tau_i}{\vert t\vert^2}\right)\left(1-\beta\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2\right)^{\alpha/2}\right)\\&+o\left(\vert\tau\vert^{\min\{2,2\alpha , \alpha+1\}}\right)\nonumber\\ \nonumber= & 2\beta \vert t\vert^{2h}\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2\right)^{\alpha/2}+o\left(\vert\tau\vert^{\min\{2,2\alpha , \alpha+1\}}\right).\end{aligned}\]]it is obvious that the leading term is not independent of @xmath297 for any @xmath28 and @xmath22 , unless when @xmath98 and @xmath180 . therefore for @xmath122 , @xmath380 does not have asymptotically locally stationary increments . similarly , for the fields @xmath384 , @xmath399 and @xmath385 , one can show similarly that as @xmath409 , @xmath411 ^ 2\right\rangle= & 2\beta t_1^{2h_1}\ldots t_n^{2h_n}\left(\sum_{i=1}^n \left [ \frac{\tau_i}{t_i}\right]^2\right)^{\alpha/2}+o\left(\vert\tau\vert^{\min\{2,2\alpha , \alpha+1\}}\right),\end{aligned}\ ] ] @xmath412 ^ 2\right\rangle= & 2\vert t\vert^{2h}\sum_{i=1}^n \beta_i \left|\frac{\tau_i}{t_i}\right|^{\alpha_i}+o\left(\vert\tau\vert^{\min\{2,2\alpha_i , \alpha_i+1\}}\right),\end{aligned}\]]@xmath413 ^ 2\right\rangle= & 2t_1^{2h_1}\ldots t_n^{2h_n}\sum_{i=1}^n \beta_i \left|\frac{\tau_i}{t_i}\right|^{\alpha_i}+o\left(\vert\tau\vert^{\min\{2,2\alpha_i , \alpha_i+1\}}\right).\end{aligned}\]]therefore for @xmath414 , none of the fields @xmath384 , @xmath399 and @xmath385 have asymptotically locally stationary increments . when @xmath98 , all the fields @xmath380 , @xmath384 , @xmath399 and @xmath385 are actually the same , and they have asymptotically locally stationary increment if and only if @xmath180 , in which case the variance of the increment @xmath415 behaves like @xmath416 next we consider the total increments . since increment is the same as total increment when @xmath98 , we only need to consider @xmath122 . using similar computations as given above , one can verify that for @xmath122 , the fields @xmath380 , @xmath384 and @xmath399 do not have asymptotically locally stationary increments , but @xmath385 have if @xmath180 . we show the computation of the latter case here . by definition , @xmath417 ^ 2\right\rangle=\sum_{\delta\in\{0,1\}^n } \sum_{\eta\in\{0,1\}^n } ( -1)^{\sum_{i=1}^n \delta_i+\sum_{i=1}^n \eta_i}\\&\hspace{4cm}\left\langle \mathbb{y}^{\#}_{\alpha,\beta}\left(t+\sum_{i=1}^n\delta_i\tau_i e_i\right)\mathbb{y}^{\#}_{\alpha,\beta}\left(t+\sum_{i=1}^n\eta_i\tau_i e_i\right ) \right\rangle\\ = & \prod_{i=1}^n \sum_{(\delta_i,\eta_i)\in\{0,1\}^2}(-1)^{\delta_i+\eta_i}(t_i+\delta_i\tau_i)^{h_i } ( t_i+\eta_i\tau_i)^{h_i}c_{\alpha_i,\beta_i}\bigl(\ln[t_i+\delta_i\tau_i ] , \ln[t_i+\eta_i\tau_i]\bigr)\\ = & \prod_{i=1}^n \biggl\{t_i^{2h_i}-2(t_i+\tau_i)^{h_i}t_i^{h_i } \left(1+\left|\ln\left[1+\frac{\tau_i}{t_i}\right]\right|^{\alpha_i}\right)^{-\beta_i}+(t_i+\tau_i)^{2h_i}\biggr\}\\ = & \prod_{i=1}^n\left\{2\beta_i t_i^{2h_i}\left|\frac{\tau_i}{t_i}\right|^{\alpha_i}+o\left(|\tau_i|^{\min\{2,2\alpha_i,\alpha_i+1\}}\right)\right\}\\ = & 2^n\prod_{i=1}^n\left\{\beta_i t_i^{2h_i}\left|\frac{\tau_i}{t_i}\right|^{\alpha_i}\right\}+o\left(\vert \tau\vert^{\sum_{i=1}^n\alpha_i+\delta}\right),\end{aligned}\]]where @xmath418 . if @xmath180 , then as @xmath34 , @xmath419 ^ 2\right\rangle\sim 2^n\prod_{i=1}^n [ \beta_i|\tau_i|^{\alpha_i}].\end{aligned}\ ] ] therefore , we have verified the following : the total increments of @xmath385 are asymptotically locally stationary if @xmath180 . next we consider the tangent fields ( definition [ def1 ] ) of the lamperti transforms of gfgcc @xmath380 and @xmath384 . we have [ p1026_1 ] + a. the field @xmath380 is lass of order @xmath33 , with tangent field at @xmath420 being @xmath421 + b. the field @xmath384 is lass of order @xmath33 , with tangent field at @xmath420 being @xmath422 b_{\alpha/2}\left(\frac{u_1}{t_1},\ldots,\frac{u_n}{t_n}\right).\ ] ] + using the formula @xmath423 ^ 2\right\rangle + \left\langle\left[\delta_{{\varepsilon}v}y_{\alpha,\beta}(t)\right]^2\right\rangle-\left\langle\left[\delta_{{\varepsilon}(u - v)}y_{\alpha,\beta}(t+{\varepsilon}v)\right]^2\right\rangle\bigr\},\end{aligned}\]]we obtain immediately from that as @xmath424 , @xmath425 ^ 2\right)^{\alpha/2}+ \left(\sum_{i=1}^n\left[\frac{v_i}{t_i}\right]^2\right)^{\alpha/2}- \left(\sum_{i=1}^n\left[\frac{u_i - v_i}{t_i}\right]^2\right)^{\alpha/2}\right\}.\end{aligned}\]]notice that this is up to the factor @xmath426 , the covariance of the scaled l@xmath53vy brownian field @xmath427the statement for @xmath384 is proved similarly . in fact , one can also deduce from and that the local fractal index of the fields @xmath380 and @xmath384 are both equal to @xmath33 , and the hausdorff dimension of their graphs over a hyperrectangle is @xmath428 . for the tangent fields of the lamperti transforms of gsgcc @xmath399 and @xmath385 , we recall that given @xmath19^n$ ] , @xmath429 , and we assume wlog that @xmath430 for some @xmath431 . then as in proposition [ p1026_1 ] , we can use and to show that [ p1026_3 ] + a. the field @xmath399 is lass of order @xmath432 , with tangent field at @xmath420 being @xmath433where the field @xmath434 is defined in proposition [ p1026_2 ] . + b. the field @xmath385 is lass of order @xmath435 , with tangent field at @xmath420 being @xmath436 t_{\alpha,\beta}\left(\frac{u_1}{t_1},\ldots,\frac{u_n}{t_n}\right).\ ] ] + from propositions [ p1 ] and [ p1026_1 ] and propositions [ p1026_2 ] and [ p1026_3 ] , we find that the tangent fields of gfgcc and gsgcc are related to the tangent fields of their lamperti transforms by some change of variable formulas . namely , if @xmath27 has tangent field @xmath437 at @xmath354 , then its first lamperti transform @xmath438 has tangent field @xmath439 at @xmath440 , and its second lamperti transform @xmath441 has tangent field @xmath442 t\left(\frac{u_1}{s_1},\ldots,\frac{u_n}{s_n}\right)\ ] ] at @xmath440 . we also notice that the order of self - similarity of the fields @xmath380 and @xmath399 , @xmath28 , is in general different from their order of local asymptotic self similarity , being @xmath33 and @xmath435 respectively . since the stochastic process @xmath385 resembles the gsgcc @xmath248 in the sense that both their covariances can be written as products of covariance of one - dimensional processes , we can consider the limit of the total increments of @xmath385 . it is easy to obtain as in proposition [ p1026_5 ] , using the @xmath98 case of proposition [ p1026_1 ] that @xmath443b_{\alpha/2}^{\#}(u).\end{aligned}\ ] ] here @xmath309 . we have studied some of the basic properties of gfgcc and gsgcc , and their associated lamperti transforms . the asymptotic properties of the spectral densities of gfgcc and gsgcc are considered . one expects the separate characterization of fractal dimension and long range dependence for gfgcc and gsgcc will provide more flexibility in their applications to modeling various surfaces and images . in the one - dimensional case , gfgcc has been applied to model the havriliak - negami relaxation law @xcite . the estimations of parameters for stationary gaussian processes have been widely studied . some of these estimations can be adapted for gfgcc and gsgcc , and they are crucial to the applications . further applications in spatial - temporal processes are possible if gfgcc is extended and modified to a space - time field to include non - stationarity and anisotropy @xcite . we hope to apply results obtained in this paper to model various physical systems such as thin film surfaces in semiconductors @xcite , surface ocean waves @xcite , geological morphology @xcite , etc , in a future work . * acknowledgements * the authors would like to thank malaysian academy of sciences , ministry of science , technology and innovation for funding this project under the scientific advancement fund allocation ( saga ) ref . no p96c . anderson , _ a multivariate linnik distribution _ , stat . * 14 * , 333 336 ( 1992 ) . george e. andrews , richard askey , and ranjan roy , _ special functions _ , encyclopedia of mathematics and its applications , vol . 71 , cambridge university press , cambridge , 1999 . p. flandrin , p. borgnat and p.o . amblard , _ from stationarity to self - similarity , and back : variations on the lamperti transformation _ , in _ processes with long - range correlations , theory and applications _ , ( lecture notes in physics vol . g. rangarajan and m. ding , springer , new york , 2003 ) . kent and a.t.a . wood , _ estimating fractal dimension of a locally self - similar gaussian process by using increments _ , statistics research report srr 03495 , centre for mathematics and its applications , austalia national university , canberra ( 1995 ) ; j. r. statist . b * 59 * , 679699 ( 1997 ) . s. c. lim and l.p . teo , _ fractional oscillator process with two indices _ , preprint arxiv : 0804.3906 ( 2008 ) . j. mateu , e. porcu , g. christakos and m. bevilacqua , _ fitting negative spatial covariances to geothermal field temperatures in nea kessani ( greece ) _ , envoronmetrics * 18 * , 759773 ( 2007 ) . n.k . sahoo , s. thakur and r.b . tokas , _ fractals and superstructures in gadolinia thin film morphology : influence of process variables on their characteristic parameters _ , thin solid films * 503 * , 8595 ( 2006 ) .

we characterize lower bounds for the bakry - emery ricci tensor of nonsymmetric diffusion operators by convexity of entropy on the-wasserstein space , and define a curvature - dimension condition for general metric measure spaces together with a square integrable-form in the sense of . this extends the lott - sturm - villani approach for lower ricci curvature bounds of metric measure spaces . in generalized smooth context , consequences are new bishop - gromov estimates , pre - compactness under measured gromov - hausdorff convergence , and a bonnet - myers theorem that generalizes previous results by kuwada . we show that-warped products together with lifted vector fields satisfy the curvature - dimension condition . for smooth riemannian manifolds we derive an evolution variational inequality and contraction estimates for the dual semigroup of nonsymmetric diffusion operators . another theorem of kuwada yields bakry - emery gradient estimates .

1606.06837

in this article we present a lagrangian approach for studying possibly nonsymmetric diffusion operators . for instance , we consider operators of the form @xmath3 where @xmath4 is the laplace - beltrami operator of a compact smooth riemannian manifold @xmath5 and @xmath6 is a smooth @xmath1-form on @xmath7 . then the bakry - emery @xmath2-ricci tensor associated to @xmath8 is defined by @xmath9 for @xmath10 $ ] where @xmath11 denotes the symmetric derivation of @xmath12 with respect to the levi - civita connection @xmath13 of @xmath7 . note @xmath14 is also meaningful for @xmath15 but we will not consider these cases . if @xmath16 , @xmath8 is the diffusion operator of the canonical symmetric dirichlet form associated to the smooth metric measure space @xmath17 with @xmath18 , and @xmath19 is understood as the ricci curvature of @xmath17 . in celebrated articles by lott , sturm and villani @xcite - built on previous results in @xcite - a definition of lower ricci curvature bounds for general metric measure spaces in terms of convexity properties of entropy functionals on the @xmath0-wasserstein space was introduced . in smooth context these definitions are equivalent to lower bounds for the bakry - emery tensor provided @xmath6 is exact . for diffusion operators where @xmath6 is not necessarily exact such a geometric picture was missing . though the operator @xmath8 yields a bilinear form , in general this form is not symmetric and therefore can not arise as dirichlet form of a metric measure space . nevertheless , there are numerous results dealing with probabilistic , analytic and geometric properties of @xmath8 under lower bounds on @xmath14 . the results are very similar to properties that one derives for symmetric operators with lower bounded ricci curvature , e.g. @xcite . in this article we derive a geometric picture associated to the diffusion operator @xmath20 for general @xmath1-forms @xmath12 in the spirit of the work by lott , sturm and villani . we characterize lower bounds on @xmath14 in terms of convexity for line integrals along @xmath0-wasserstein geodesics . moreover , for generalized smooth metric measures spaces ( definition [ gsmms ] ) we impose the following definition . for simplicity , in this introduction we assume @xmath21 . we will say @xmath22 together with a @xmath1-form @xmath6 satisfies the curvature - dimension condition @xmath23 if and only if for every pair @xmath24 there exists an @xmath0-wasserstein geodesic @xmath25 such that @xmath26 \\ & \hspace{1 cm } -\frac{1}{2}kt(1-t)kw_2(\mu_0,\mu_1)^2,\end{aligned}\ ] ] where @xmath27 and @xmath28 denotes the line integral of @xmath6 along @xmath29 . for the case @xmath30 corresponding definitions are made in definition [ defa ] and definition [ defb ] . in particular , we emphasize that definition [ defb ] is also meaningful in the class of general metric measure spaces together with @xmath0-integrable @xmath1-forms in the sense of @xcite . however , in this article we will only study the generalized smooth case . we prove several geometric consequences : generalized bishop - gromov estimates , pre - compactness under gromov - hausdorff convergence , and a generalized bonnet - myers theorem . the latter generalizes a result of kuwada in @xcite - even for smooth ingredients . then , we show that the condition @xmath31 is stable under @xmath2-warped product constructions . this also includes so - called euclidean @xmath2-cones and @xmath2-suspensions . in the last section we introduce the notion of @xmath32-flows that arise naturally on generalized smooth metric measure spaces together with a @xmath1-form satisfying a curvature - dimension condition . more preciely , if @xmath33 is the semigroup associated to the operator @xmath3 on a smooth riemannian manifold @xmath7 such that @xmath34 satisfies @xmath23 , and if @xmath35 is the dual flow acting on probability measures , then it is an absolutely continuous curve in @xmath0-wasserstein space and for any probability measure @xmath36 , @xmath37 satisfies the following inequality @xmath38 provided standard regularity assumption for the corresponding heat kernel ( proposition [ propa ] ) . @xmath39 is any absolutely continuous probability measure , and @xmath40 is the @xmath0-wasserstein geodesic between @xmath41 and @xmath39 . this yields a contraction estimate for @xmath35 that is again equivalent to the bakry - emery conditon for @xmath33 by @xcite . therefore , we obtain the following theorem . let @xmath5 be a compact smooth riemannian manifold , and let @xmath6 be a smooth @xmath1-form . we denote with @xmath42 the corresponding metric measure space , and let @xmath33 and @xmath35 be as in the section 7 . then , the following statements are equivalent . * @xmath43 , * @xmath44 satisfies the condition @xmath23 , * for every @xmath45 @xmath37 is an @xmath46-flow curve starting in @xmath36 , * @xmath35 satisfies the contraction estimate in corollary [ contractionxxx ] , * @xmath33 satisfies the condition @xmath47 . this work was partly done while the author was in residence at the mathematical sciences research institute in berkeley , california during the spring 2016 semester , supported by the national science foundation . i want to thank the organizers of the differental geometry program and msri for providing great environment for research . [ [ metric - measure - spaces ] ] * metric measure spaces * + + + + + + + + + + + + + + + + + + + + + + + let @xmath48 be a complete and separable metric space , and let @xmath49 be a locally finite borel measure . we call @xmath22 a metric measure space . the case @xmath50 is excluded . the space of constant speed geodesics @xmath51\rightarrow x$ ] is denoted with @xmath52 , and it is equipped with the topology of uniform convergence . @xmath53 denotes the evaluation map at time @xmath54 that is continuous . + the @xmath0-wasserstein space of probability measures with finite second moment is denoted with @xmath55 , and @xmath56 is the @xmath0-wasserstein distance . @xmath57 and @xmath58 denote the subset of compactly supported probability measures and the family of @xmath49-absolutely continuous probability measures , respectively . a coupling or plan between probability measures @xmath59 and @xmath60 is a probability measure @xmath61 such that @xmath62 where @xmath63 are the projection maps . a coupling @xmath64 is optimal if @xmath65 optimal couplings exist , and if an optimal coupling @xmath64 is induced by a map @xmath66 via @xmath67 where @xmath68 is a measurable subset of @xmath69 , we say @xmath70 is an optimal map . + a probability measure @xmath71 is called an optimal dynamical coupling if @xmath72 is an optimal coupling between its marginal distributions . let @xmath73}$ ] be an @xmath0-wasserstein geodesic in @xmath55 . we say an optimal dynamical coupling @xmath25 is a lift of @xmath74 if @xmath75 for every @xmath76 $ ] . if @xmath25 is the lift of an @xmath0-wasserstein geodesic @xmath74 , we call @xmath25 itself an @xmath0-wasserstein geodesic . we say @xmath25 has bounded compression if there exists a constant @xmath77 such that @xmath78 for every @xmath76 $ ] . + we say that a metric measure space @xmath79 is _ essentially non - branching _ if for any optimal dynamical coupling @xmath25 there exists @xmath80 such that @xmath81 and for all @xmath82 we have that @xmath83 \mbox { and for some } \epsilon>0\ \mbox{implies } \ \gamma=\gamma ' . $ ] for instance a metric measure space satisfying a riemannian curvature - dimension condition @xmath84 in the sense of @xcite is essentially non - branching @xcite . + if we assume that for @xmath85-almost every pair @xmath86 there exists a unique geodesic @xmath87 between @xmath88 and @xmath89 then by measurable selection there exists a measurable map @xmath90 with @xmath91 . for @xmath45 and @xmath92 borel with @xmath93 we set @xmath94 , and for @xmath95 we set @xmath96 . in this case @xmath97 is the unique optimal dynamical plan between @xmath98 and @xmath99 . again , the family of @xmath100-spaces is a class that satisfies this property @xcite . [ gsmms ] we say @xmath22 is a _ generalized smooth metric measure space _ if there exists an open smooth manifold @xmath101 , and a riemannian metric @xmath102 on @xmath7 with induced distance function @xmath103 such that the metric completion of the metric space @xmath104 is isometric to @xmath48 , and for any optimal dynamical plan @xmath105 such that @xmath75 is a geodesic and @xmath106 we have that @xmath107 in particular , @xmath108 , and if we choose the constant geodesic @xmath25 with @xmath109 for all @xmath76 $ ] where @xmath110 is any measurable set of finite @xmath49-measure , one gets that @xmath111 is of @xmath49-measure @xmath112 . we call @xmath7 the set of regular points in @xmath69 . the condition ( [ transportcondition ] ) yields that @xmath79 is essentially non - branching and that for @xmath85-almost every pair @xmath86 there exists a unique geodesic @xmath87 between @xmath88 and @xmath89 . moreover , for each pair @xmath113 there is a unique dynamcial optimal coupling @xmath25 such that @xmath114 where @xmath115 is the space of geodesic in @xmath7 , @xmath116 , and @xmath25 is induced by a map . to see this note that @xmath117 and transport geodesics are contained in @xmath7 then , since one can choose an exhaustion of @xmath7 by compact sets , we can assume that @xmath59 and @xmath60 are compactly supported in @xmath7 . then , the claim follows from statements in @xcite and since geodesics are unique . examples of generalized smooth metric measure spaces in the sense of the previous definition are riemannian manifolds with boundary that are geodesically convex , cones , suspensions @xcite and warped products @xcite . moreover , in an upcoming paper of the author with ilaria mondello , it will be shown that stratified spaces are generalized smooth provided certain assumptions on tangent cones at singular points . this result will also show that orbifolds are generalized smooth . [ [ forms - and - vector - fields ] ] * @xmath1-forms and vector fields * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + assume @xmath79 is a generalized smooth metric measure space . a @xmath1-form @xmath6 is a measurable map @xmath118 with @xmath119 . we say @xmath120 for @xmath121 if @xmath122 is finite where @xmath123 and @xmath110 compact . + similar , we can consider measurable and @xmath124-integrable vector fields on @xmath69 . note that a vector field @xmath125 on @xmath7 yields a @xmath1-form @xmath6 via @xmath126 . in the context of generalized smooth metric measures space this is the natural isomorphism between vectorfields and @xmath1-forms , and we will often switch between these viewpoints . if @xmath25 is an optimal dynamical coupling with bounded compression , the line integral @xmath127 exists @xmath25-almost surely , and it does not depend on the parametrization of @xmath29 up to changes of orientation . moreover , for any @xmath0-wasserstein geodesic @xmath25 with bounded compression , we set @xmath128 . [ [ the - case - of - arbirtrary - metric - measure - spaces ] ] * the case of arbirtrary metric measure spaces * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath129 be any function . the hopf - lax semigroup @xmath130 is defined by @xmath131 @xmath132 is the @xmath133-transform of @xmath134 . we say a function @xmath134 is @xmath133-concave if there exist @xmath135 such that @xmath136 . if @xmath69 is compact , by kantorovich duality for any pair @xmath24 with bounded densities there exist a lipschitz function @xmath134 such that @xmath137 for any geodesic @xmath138 with bounded compression . for instance , see @xcite . + if we follow the approach of gigli in @xcite , there is also a well - defined notion of @xmath139-integrable @xmath1-form @xmath12 for general metric measure spaces @xmath79 , and one can define the dual coupling @xmath140 as measurable function on @xmath69 where @xmath141 is a sobolev function . note that in this context @xmath142 does not necessarily exist . the notion of line integral along a geodesic is more subtle , but if we consider a @xmath0-wasserstein geodesic @xmath105 that has bounded compression , then we can define @xmath143 where @xmath144 is a kantorovich potential for @xmath25 , and @xmath145 is the density of @xmath146 . since @xmath145 and @xmath147 are bounded , @xmath148 is well - defined if @xmath149 is @xmath49-integrable . note that in a smooth context @xmath150 , and therefore in smooth context we have @xmath151 in the following we just write @xmath152 for @xmath153 . also note , that in general there is no identification between @xmath1-forms and vectorfields . [ [ entropy - functionals ] ] * entropy functionals * + + + + + + + + + + + + + + + + + + + + + for @xmath45 we define the boltzmann - shanon entropy by @xmath154 and @xmath155 otherwise . given a number @xmath156 , we define the @xmath2-rny entropy functional @xmath157 $ ] with respect to @xmath49 by @xmath158 where @xmath159 denotes the density of the absolutely continuous part in the lebesgue decomposition of @xmath36 . in the case @xmath160 the @xmath1-rny entropy is @xmath161 . if @xmath49 is finite , then @xmath162 and @xmath163 . moreover , if @xmath49 is finite and @xmath164 , then @xmath165 is lower semi - continuous . if @xmath49 is @xmath166-finite one has to assume an exponential growth condition @xcite to guarantee lower semi continuity . + if there is a @xmath1-form @xmath12 , we also define @xmath167 $ ] by @xmath168 and @xmath112 otherwise . if @xmath169 , then @xmath170 . [ [ distortion - coefficients ] ] * distortion coefficients * + + + + + + + + + + + + + + + + + + + + + + + + + for two numbers @xmath171 and @xmath156 we define @xmath172\times ( 0,\infty)\mapsto\sigma_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(\theta)= \begin{cases } \frac{\sin_{{\scriptscriptstyle{k}}/{\scriptscriptstyle{n}}}(t\theta)}{\sin_{{\scriptscriptstyle{k}}/{\scriptscriptstyle{n}}}(\theta ) } \ & \ \mbox { if } \sin_{{\scriptscriptstyle{k}}/n}(x)>0 \mbox { for } x\in ( 0,\theta],\\ \infty \ & \ \mbox { otherwise}. \end{cases}\end{aligned}\ ] ] @xmath173 is the solution of the initial value problem @xmath174 the modified distortion coefficients for number @xmath171 and @xmath164 are given by @xmath172\times ( 0,\infty)\mapsto \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(\theta)=\begin{cases } \theta\cdot\infty \ & \ \mbox { if } k>0 \mbox { and } n=1,\\ t^{\frac{1}{n}}\left[\sigma_{{\scriptscriptstyle{k}}/({\scriptscriptstyle{n}}-1)}^{(t)}(\theta)\right]^{1-\frac{1}{n } } \ & \ \mbox { otherwise}. \end{cases}\end{aligned}\ ] ] [ defa ] let @xmath22 be a generalized smooth metric measure space , and let @xmath6 be an @xmath0-integrable @xmath1-form . we say @xmath175 satisfies the _ curvature - dimension condition @xmath31 _ for @xmath171 and @xmath156 if and only if for each pair @xmath176 there exists a dynamical optimal plan @xmath25 with @xmath177d\pi(\gamma)\end{aligned}\ ] ] where @xmath178 . we call any such @xmath1-form @xmath12 admissible . + if we replace @xmath179 by @xmath180 in the previous definition we say @xmath175 satisfies the _ reduced curvature - dimension condition _ @xmath181 . [ defb ] let @xmath22 be a metric measure space , and let @xmath6 be an @xmath0-integrable @xmath1-form in the sense of @xcite . + we say @xmath175 satisfies the _ curvature - dimension condition _ @xmath23 if and only if for each pair @xmath182 with bounded densities there exists a geodesic @xmath25 with bounded compression and a potential @xmath134 as in ( [ phi ] ) such that @xmath26 \\ & \hspace{1 cm } -\frac{1}{2}kt(1-t)kw_2(\mu_0,\mu_1)^2,\end{aligned}\ ] ] where @xmath183 and @xmath184 . equivalently , the map @xmath185 is @xmath186-convex . + @xmath175 satisfies the _ entropic curvature - dimension condition _ @xmath187 if and only if for each pair @xmath182 with bounded densities there exists a geodesic @xmath25 with bounded compression and a potential @xmath134 as in ( [ phi ] ) such that @xmath188 where @xmath189 and @xmath184 . that is the map @xmath185 is @xmath190-convex in the sense of @xcite . if we can choose @xmath169 as an admissible @xmath1-form , the previous definitions become the ones from @xcite . it is easy to prove that * @xmath191 , * @xmath192 + for @xmath193 and @xmath194 , * if @xmath49 is finite , then @xmath195 . for instance , compare with similar statements in @xcite . _ definition [ defb ] makes sense for any possibly non - smooth metric measure space . but for simplicity , for the rest of the article we always assume that @xmath79 is a generalized smooth metric measure space . some of the statements that we prove for generalized smooth metric measure spaces extend to arbitrary metric measure spaces but in general not without additional assumptions . _ + let @xmath79 and @xmath196 be generalized smooth metric measure spaces . a map @xmath197 is a smooth metric measure space isomorphism if @xmath198 is a metric measure space isomorphism , and if @xmath198 is a diffeomophism between the subsets of regular points @xmath199 and @xmath7 . let @xmath200 be a generalized smooth metric measure space , and let @xmath6 be an @xmath0-intergrable @xmath1-form . assume @xmath201 satisfies the condition @xmath31 . then the following properties hold . * for @xmath202 define the generalized smooth metric measures space @xmath203 . then @xmath204 satisfies @xmath205 . * for a convex subset @xmath206 define the generalized smooth metric measure space @xmath207 . then @xmath208 satisfies @xmath31 . * let @xmath209 be a generalized smooth metric measure space , and let @xmath210 be a smooth metric measure space isomorphism . then @xmath211 satisfies the condition @xmath31 . we check ( iii ) . we define @xmath212 on @xmath213 . if @xmath29 is a geodesic in @xmath69 , then @xmath214 is a geodesic in @xmath215 . the line integral of @xmath216 along @xmath217 is @xmath218 then , the statement follows like similar results for metric measure spaces that satisfy a curvature - dimension condition ( for instance see @xcite ) . [ nonbranching ] let @xmath200 be a generalized smooth metric measure space , @xmath6 an @xmath0-integrable @xmath1-form , and @xmath171 and @xmath219 . then the following statements are equivalent : * @xmath220 satisfies @xmath181 . * for each pair @xmath221 there exists an optimal dynamical plan @xmath25 with @xmath222 such that @xmath223^{-\frac{1}{n } } \geq \sigma_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}|)\rho_0(\gamma_0)^{-\frac{1}{n}}+\sigma_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}|)\left[e^{-\phi_1(\gamma)}\rho_1(\gamma_1)\right]^{-\frac{1}{n}}\end{aligned}\ ] ] for @xmath224 $ ] and @xmath25-a.e . @xmath145 is the density of @xmath74 w.r.t . @xmath49 . * @xmath201 satisfies @xmath187 . moreover , the condition @xmath31 is equivalent with ( ii ) if the coefficients @xmath226 are replaced by the coefficients @xmath227 . first , we observe that in the context of generalized smooth metric measure spaces up to a set measure zero optimal couplings between @xmath49-absolutly continuous measures @xmath228 are unique ( also compare with the remark after definition [ gsmms ] ) . + `` ( i)@xmath229(ii ) '' : let @xmath230 be with bounded support , and let @xmath231 be the optimal coupling between @xmath59 and @xmath60 . let @xmath232 be an @xmath233-stable generator of the borel @xmath166-field of @xmath48 . for each @xmath234 we define a disjoint covering of @xmath69 of @xmath235 sets by @xmath236 where @xmath237 and @xmath238 . + we define @xmath239 and set @xmath240 if @xmath241 . then we consider the marginal measures @xmath242 and @xmath243 that are @xmath49-absolutely continuous , and @xmath244 is the unique optimal coupling . since geodesics are @xmath85-almost surely unique , the dynamical optimal plan @xmath245 is the unique optimal dynamical coupling between its endpoints where @xmath246 . therefore , @xmath247 satisfies the @xmath248-inequality for every @xmath198 and @xmath249 . in particular , @xmath250 is @xmath49-absolutely continuous . then , we define a dynamical coupling between @xmath59 and @xmath60 by @xmath251 @xmath252 is optimal since @xmath253 is an optimal coupling . therefore , we can apply lemma 3.11 in @xcite : since the measures @xmath254 for @xmath255 are mutually singular , @xmath256 are mutually singular as well . + now , for @xmath257 we consider the measure @xmath258 . since it decomposes into mutually singular , absolutely continuous measures @xmath259 with densities @xmath260 , @xmath261 is absolutely continuous as well , and by mutual singularity of the measure @xmath259 its density is @xmath262 . again , since geodesics are @xmath85-almost surely unique we have that @xmath263 , and @xmath264 for every @xmath265 . from the @xmath248-inequality for @xmath247 we have @xmath266 this holds for every @xmath267 and @xmath268 . since @xmath267 and @xmath268 are mutually disjoint , by summing up the previous inequality holds for @xmath269 and @xmath270 as well . since the family @xmath271 is a generator for the @xmath166-field , we have for @xmath64-almost every @xmath272 @xmath273 and since @xmath274 , this is the claim . + for the following recall that in the context of generalized smooth metric measure spaces @xmath275 where @xmath25 is an @xmath0-wasserstein geodesic , @xmath134 is an kantorovich potential , and @xmath276 for every @xmath76 $ ] . + `` ( ii)@xmath229(iii ) '' : recall from @xcite that @xmath277 is convex . then , apply @xmath278 to ( [ something1 ] ) and use jensen s inequality on the right hand side to obtain the condition @xmath187 . + `` ( iii)@xmath229(ii ) '' : this works like in `` ( i)@xmath229(ii ) '' where one has to use the convexity of @xmath279 again . see also @xcite . + `` ( ii)@xmath229(i ) '' : integrate ( [ something1 ] ) w.r.t . the optimal dynamical plan @xmath25 . + the proof of the equivalence for the condition @xmath31 is similar . if we consider @xmath6 such that @xmath280 on @xmath7 for a smooth function @xmath281 , then @xmath282 and we can reformulate ( [ something1 ] ) as @xmath283^{-\frac{1}{n } } \geq \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}|)\left[\rho_0(\gamma_0)e^{f(\gamma_0)}\right]^{-\frac{1}{n}}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}|)\left[\rho_1(\gamma_1)e^{f(\gamma_1)}\right]^{-\frac{1}{n}}\end{aligned}\ ] ] for @xmath25-a.e . @xmath29 . that is the condition @xmath31 in the sense of @xcite for the metric measure space @xmath284 . in this section we consider a vector field @xmath125 rather than a @xmath1-form @xmath12 . recall that for smooth metric measure spaces we always can identifiy @xmath125 with a @xmath1-form @xmath6 . let @xmath22 be a smooth metric measure space with @xmath285 and @xmath286 . let @xmath13 be the levi - civita connection of @xmath102 and let @xmath287 be a smooth vector field . we define the bakry - emery @xmath2-ricci tensor for @xmath288 $ ] by @xmath289 where @xmath290 and @xmath291 . for @xmath292 we define @xmath293 for @xmath294 we define @xmath295 for all @xmath296 and @xmath112 otherwise . [ maintheorem ] let @xmath22 be a smooth metric measure space with @xmath285 and @xmath286 , @xmath171 and @xmath297 $ ] . let @xmath125 be an @xmath0-integrable smooth vector field . then @xmath298 satisfies the condition @xmath31 if and only if @xmath299 moreover , if @xmath2 is finite , then @xmath31 and @xmath181 are equivalent . assume @xmath300 . let @xmath30 . the case @xmath21 follows by obvious modifications . consider @xmath301 that are compactly supported . otherwise we can choose compact exhaustions of @xmath302 and consider the restriction of optimal couplings to these sets . there exists a @xmath133-concave function @xmath134 such that @xmath303 is the unique optimal map between @xmath59 and @xmath304 , and @xmath74 is the unique @xmath0-wasserstein geodesic between @xmath59 and @xmath60 in @xmath305 . @xmath74 is compactly supported and @xmath49-absolutely continuous @xcite . + the potential @xmath134 is semi - concave , and by the bangert - alexandrov theorem the hessian @xmath306 exists for @xmath49-almost every @xmath307 . moreover , for @xmath49-almost every @xmath307 the optimal map @xmath308 admits a jacobian @xmath309 for every @xmath224 $ ] . @xmath309 is non - singular for every @xmath76 $ ] , and the monge - ampere equation @xmath310holds @xmath49-almost everywhere . we pick a point @xmath307 where @xmath309 is non - singular and ( [ jacobi ] ) holds . for an orthonormal basis @xmath311 of @xmath312 @xmath313 is the jacobi field along @xmath314 with initial condition @xmath315 and @xmath316 . therefore @xmath317 we set @xmath318 . since @xmath309 is non - singular for every @xmath224 $ ] , riemannian jacobi field calculus ( for instance see @xcite ) yields for @xmath319 the differential inequality @xmath320 * 3 . * consider the vector field @xmath125 , the geodesic @xmath224\rightarrow \gamma_x(t)$ ] , and the corresponding line integral @xmath321 we set @xmath322 and compute @xmath323 in addition with ( [ firstone ] ) and ( [ assumption ] ) this yields @xmath324 hence @xmath325 if we set @xmath326 , then we obtain @xmath327 or equivalently @xmath328 note the dependence on @xmath307 . ( [ secondone ] ) holds for @xmath49-a.e . @xmath307 . + * 4 . * by the same computation as in @xcite we can improve ( [ secondone ] ) by taking out the direction of motion if @xmath329 . we know that @xmath330 where @xmath331 and @xmath332 . from lemma 3.1 in @xcite one sees that @xmath333 is symmetric . @xmath334 has the form @xmath335 for an @xmath336-matrix @xmath337 . hence , if @xmath338 , we have @xmath339 moreover , taking the trace in ( [ ricattii ] ) yields @xmath340 where @xmath341 . the jacobi determinant @xmath342 satisfies @xmath343 . therefore if we set @xmath344 , we have @xmath345ds}\end{aligned}\ ] ] and ( [ anotherequation ] ) becomes @xmath346 . @xmath347 describes volume distortion in direction of the transport geodesics . we remove this part and consider @xmath348 and if we set @xmath349 , then @xmath350 a computation yields @xmath351 and note that @xmath352 . setting @xmath353 as in * 3 . * we also obtain @xmath354 if @xmath355 . + * 5 . * set @xmath356 . note that ( [ kk ] ) implies @xmath357 . then @xmath358 . by construction @xmath359 . hence ( similar as in part ( c ) of the proof of theorem 1.7 in @xcite ) we obtain @xmath360 or @xmath361 consider @xmath59 , @xmath60 , @xmath308 and @xmath362 as in * 1.*. then for @xmath49-a.e . @xmath307 we have @xmath363^{-\frac{1}{n } } & = \rho_0(x)^{-\frac{1}{n}}\left[\mathcal{j}_t(x)^{}e^{\phi_t(\gamma_x)}\right]^{\frac{1}{n}}\\ & \geq \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}_x|)\rho_0(x)^{-\frac{1}{n}}\mathcal{j}_0(x)^{\frac{1}{n}}\\ & \hspace{2cm}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(|\dot{\gamma}_x|)\left[e^{\phi_1(\gamma_x)}\rho_0(x)^{-1}\mathcal{j}_1(x)\right]^{\frac{1}{n}}\\ & = \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}_x|)\rho_0(x)^{-\frac{1}{n}}\\ & \hspace{2cm}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(|\dot{\gamma}_x|)\left[e^{-\phi_1}\rho_1(t_1(x))\right]^{-\frac{1}{n}}.\end{aligned}\ ] ] recall that @xmath364 . @xmath25 is the unique dynamical optimal coupling between @xmath59 and @xmath60 . hence , integration with respect to @xmath59 yields @xmath365^{-\frac{1}{n}}d\pi(\gamma)=-\int\left[\rho_t(t_t(x))e^{-\phi_t(\gamma_x)}\right]^{-\frac{1}{n}}d\mu_0(x)\\ & \leq -\int \big[\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}_x|)\rho_0(x)^{-\frac{1}{n}}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(|\dot{\gamma}_x|)\left[e^{-\phi_1(\gamma_x)}\rho_1(t_1(x))\right]^{-\frac{1}{n}}\big]d\mu_0(x)\\ & = -\int \big[\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-t)}(|\dot{\gamma}|)\rho_0(\gamma_0)^{-\frac{1}{n}}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(t)}(|\dot{\gamma}|)e^{-\frac{1}{n}\phi_1(\gamma)}\rho_1(\gamma_1)^{-\frac{1}{n}}\big]d\pi(\gamma).\end{aligned}\ ] ] that is the claim . + * `` @xmath366 '' : we argue by contradiction . assume @xmath298 satisfies the curvature - dimension condition @xmath31 . we only consider the case @xmath367 . since @xmath368 , @xmath298 satisfies the condition @xmath181 as well . assume there is @xmath369 such that @xmath370 . choose a smooth function @xmath371 such that @xmath372 we set @xmath373 . it follows @xmath374 . we can assume that @xmath371 has compact support in @xmath375 , and @xmath371 and @xmath376 are kantorovich potentials by replacing @xmath371 by @xmath377 and @xmath378 by @xmath379 for a sufficiently small @xmath380 ( we apply theorem 13.5 in @xcite to @xmath371 and @xmath376 ) . then the map @xmath381 for @xmath382 $ ] induces a constant speed @xmath0-wasserstein geodesic @xmath383}$ ] . for this particular @xmath308 we repeat calculations of the previous steps and obtain for @xmath384 @xmath385 @xmath49-a.e . where @xmath386 . now , we consider @xmath387 and the induced wasserstein geodesic @xmath388 . we set @xmath389 , and note that @xmath390 . by continuity of @xmath391 we can choose @xmath392 and @xmath393 such that @xmath394.\end{aligned}\ ] ] moreover , by continuity we can choose @xmath392 and @xmath393 even smaller such that @xmath395 for @xmath396 and @xmath382 $ ] . and again by using just continuity of @xmath125 , @xmath371 and all its first and second order derivatives for @xmath397 we can choose @xmath398 and @xmath393 even more smaller such that @xmath399 for every @xmath400 and every @xmath396 . note that @xmath401 and @xmath402 . then we compute for @xmath403 - omitting the dependency on @xmath396 , and using @xmath404 for any @xmath396 and each @xmath382 $ ] - @xmath405\end{aligned}\ ] ] where we use @xmath406 in the third equality . recall @xmath407 . the previous differential inequality is equivalent to @xmath408''\geq - \frac{k-2\epsilon}{n}\left|\dot{\gamma}_t\right|^2e^{\frac{1}{n}(y_t+\phi_t ) } \ \ \mbox { on } [ -\delta,\delta].\end{aligned}\ ] ] a reparametrization @xmath409\rightarrow -s\delta+s\delta$ ] yields @xmath410''\geq - \frac{k-2\epsilon}{n}\left|\dot{\hat{\gamma}}_t\right|^2e^{\frac{1}{n}(\hat{y}_t+\hat{\phi}_t ) } \ \ \mbox { on } [ -\textstyle{\frac{1}{2}},\textstyle{\frac{1}{2}}].\end{aligned}\ ] ] where @xmath411 , @xmath412 and @xmath413 . @xmath414 is a kantorovich potential for the wasserstein geodesic @xmath415\mapsto \mu_{-\delta t+\delta t}$ ] , and by the previous computation ( [ lala ] ) is the differential inequality as in step ( 4 ) for the potential @xmath416 but with reversed inequalities . we obtain ( [ secondone ] ) with reverse inequalities and @xmath186 replaced by @xmath417 that contradicts the condition @xmath181 . consider a generalized smooth metric measure space @xmath22 with regular set @xmath7 . in this section we assume that @xmath418 is @xmath49-absolutely continuous . we consider again vector fields instead of @xmath1-forms . + recall the definition of the measure @xmath419 from section 2 . for any measurable set @xmath92 with @xmath93 and any point @xmath95 we define @xmath420 where @xmath421\rightarrow x\in\mathcal{g}(x)$ ] . since geodesics are @xmath85-almost everywhere unique , @xmath422 is the unique optimal dynamical plan between @xmath98 and @xmath423 . + we define for @xmath424 and @xmath425 @xmath426 if @xmath427 , then @xmath428 . + we say the vector field @xmath125 has _ locally finite flow _ if for any @xmath95 there is @xmath429 such that @xmath430 . if @xmath431 for some smooth @xmath141 , @xmath125 has locally finite flow if @xmath432 is a locally finite measure . + we define for @xmath424 @xmath433 if @xmath427 , then @xmath434 . [ bg ] let @xmath200 be a generalized smooth metric measure space , and @xmath125 an @xmath0-integrable vector field . assume @xmath435 satisfies the curvature - dimension @xmath31 for @xmath436 and @xmath156 . then each bounded set @xmath437 has finite @xmath49-measure , and either @xmath49 is supported by one point or all points and all spheres have mass @xmath112 . + moreover , if @xmath164 then for each fixed @xmath438\cap m$ ] and all @xmath439 ( with @xmath440 if @xmath441 ) , we have @xmath442 and @xmath443 in particular , if @xmath444 , then @xmath445 for all @xmath446 and @xmath447 , and the latter also holds for @xmath160 and @xmath448 . fix a point @xmath438\cap m$ ] , and assume first that @xmath449 . we set @xmath450 , and @xmath224\mapsto ( e_t)_{\star}\pi_{x_0,a_r}=\mu_t\in \mathcal{p}^2(x)$ ] . @xmath73}$ ] is the unique geodesic between between its endpoints . note that by definition of a generalized smooth metric measure space transport geodesics connecting @xmath451 and points in @xmath452 are @xmath453-almost surely contained in @xmath7 . therefore , since @xmath451 by riemannian calculus @xmath454 for any @xmath257 , and since @xmath418 is @xmath49-absolutely continuous , it follows that @xmath455 for any @xmath257 . in particular , @xmath456 is the unique wasserstein geodesic between @xmath457 and @xmath458 for any @xmath459 $ ] . the unique optimal dynamical plan @xmath460 is obtained by pushforward @xmath461 w.r.t . the map @xmath462 with @xmath463 . similar @xmath464 . + first , we check continuity of @xmath465^{-\frac{1}{n}}=:h_t(\gamma)$ ] on @xmath466 . by the condition @xmath181 and theorem [ nonbranching ] we have @xmath467^{-\frac{1}{n } } & \geq \sigma_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-s)}(|\dot{\gamma}|)\tilde{\rho}_{0}(\gamma_{0})^{-\frac{1}{n}}+\sigma_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(s)}(|\dot{\gamma}|)\left[e^{-\phi_1(\gamma)}\tilde{\rho}_1(\gamma_1)\right]^{-\frac{1}{n } } \ ] ] for @xmath468 $ ] and @xmath460-almost every @xmath225 . that is @xmath469^{-\frac{1}{n}}\leq -\frac{k}{n}|\dot{\gamma}|^2\left[\rho_t(\gamma_t)e^{-\phi_{{s}}(\gamma)}\right]^{-\frac{1}{n}}\ \ \mbox{on } ( 0,1)\end{aligned}\ ] ] for @xmath470-almost every @xmath225 . in particular , @xmath471 is semi - convex and therefore continuous on @xmath466 for @xmath25-almost every @xmath29 . + now , we choose @xmath472 and @xmath473 , and by the condition @xmath31 and again theorem [ nonbranching ] we have @xmath467^{-\frac{1}{n } } & \geq \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(1-{s})}(|\dot{\gamma}|)\tilde{\rho}_{0}(\gamma_{0})^{-\frac{1}{n}}+\tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(s)}(|\dot{\gamma}|)\left[e^{-\phi_1(\gamma)}\tilde{\rho}_1(\gamma_1)\right]^{-\frac{1}{n}}\\ & \geq \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(s)}(|\dot{\gamma}|)\left[e^{-\phi_1(\gamma)}\rho_1(\gamma_1)\right]^{-\frac{1}{n}}\ \ \mbox{for } \tilde{\pi}\mbox{-a.e . } \gamma\in\mathcal{g}(x),\end{aligned}\ ] ] or equivalently , for @xmath470-almost every @xmath225 : @xmath474^{-\frac{1}{n}}\geq \tau_{{\scriptscriptstyle{k}},{\scriptscriptstyle{n}}}^{(s)}((1-\epsilon)|\dot{\gamma}|)\left[e^{-\phi_1(\gamma)}\rho_1(\gamma_1)\right]^{-\frac{1}{n}}.\end{aligned}\ ] ] by continuity of @xmath471 , @xmath475 in @xmath466 , it follows for @xmath476 @xmath477 now , we choose @xmath478 for @xmath479 . integration with respect to @xmath25 yields @xmath480 where we choose @xmath481 in @xmath482 if @xmath483 , and @xmath484 otherwise . + + note that the left hand side of the previous inequality can be rewritten as follows @xmath485 where @xmath486 is the optimal plan between @xmath98 and @xmath487 . @xmath486 arises as the pushforward w.r.t . the map @xmath488 . since @xmath489 for @xmath25-almost every @xmath29 , the left hand side in ( [ lele ] ) is dominated by @xmath490 , and we obtain the following inequality : @xmath491 by construction , @xmath492 is monoton non - decreasing and right continuous . hence , it has only countably many discontinuities , and we can pick an arbitrarily small continuity point @xmath429 . then , for arbitrarily small @xmath393 we can apply ( [ differences ] ) to deduce that @xmath493 is continuous at any @xmath494 . hence @xmath493 is continuous on @xmath495 . in particular , for any @xmath496 and @xmath95 it follow that @xmath497 and @xmath498 for any @xmath499 . hence , @xmath500 for any @xmath429 and @xmath95 . moreover , if @xmath501 in ( [ differences ] ) , we obtain ( [ spheres ] ) . + now , for @xmath429 and @xmath393 given , one considers @xmath235 points @xmath502 in @xmath503 . by ( [ differences ] ) again it follows @xmath504 this implies @xmath493 is locally lipschitz , and therefore weakly differentiable , and its weak derivative is given by @xmath505 @xmath506-almost everywhere . then we can apply gromov s integration trick ( * ? ? ? * lemma 3.1 ) and obtain ( [ balls ] ) . + it remains to consider the case @xmath507 . assume @xmath508 . since @xmath509 is a length space , there is a least one point @xmath510 such that @xmath511 . we then apply apply the previous steps . this yields @xmath512 for every @xmath513 which is a contradiction . otherwise @xmath514 and @xmath515 . assume that @xmath298 satisfies the curvature - dimension @xmath31 for @xmath441 , @xmath156 and a @xmath0-integrable vectorfield @xmath125 . then the support of @xmath49 is compact and its diameter @xmath8 satifies @xmath516 in particular , if @xmath160 , then @xmath509 consist of one point . we argue by contradiction . assume there are points @xmath517 with @xmath518 since @xmath519 , we can assume that @xmath451 . choose @xmath520 such that @xmath521 , and set @xmath522 . let @xmath523 be the optimal transport between @xmath98 and @xmath36 . as in the proof of the bishop - gromov comparison @xmath451 implies that @xmath524 is @xmath49-absolutely continuous . hence @xmath525 the left hand side is dominated by @xmath526 . hence , @xmath527 . on the other hand , theorem [ bg ] implies that @xmath430 for every @xmath429 . for @xmath22 and @xmath125 we define @xmath528\end{aligned}\ ] ] where the supremum is w.r.t . families @xmath529 of disjoint @xmath530-balls . we also define @xmath531.\end{aligned}\ ] ] we have @xmath532 , and if @xmath125 is essentially bounded , and if @xmath533 , then @xmath534 and @xmath535 . + + let @xmath536 be the family of smooth metric spaces @xmath48 such that there exits a measure @xmath49 and an @xmath0-integrable vector field @xmath125 such that @xmath298 satisfies the condition @xmath31 . let @xmath537 be the subset of metric spaces @xmath48 in @xmath536 with @xmath538 . note that @xmath536 usually denotes the class of _ metric measure spaces _ that satisfy a condition @xmath31 . the family @xmath539 of smooth metric spaces @xmath48 such that @xmath298 satisfies @xmath31 for a measure @xmath49 and an @xmath0-integrable vector field @xmath125 with @xmath540 is precompact with respect to gromov - hausdorff convergence . let @xmath541 , and let @xmath49 be a measure and let @xmath125 be an admissible vector field such that @xmath542 and @xmath298 satifies @xmath31 . fix @xmath543 and a family @xmath529 of disjoint @xmath530-balls in @xmath69 . we replace every @xmath530-ball by a slightly smaller @xmath544-ball that is contained in the first one and centered in @xmath7 for some @xmath545 . this is possible since @xmath7 is dense in @xmath69 . the generalized bishop - gromov inequality yields for every @xmath546 with @xmath547 . @xmath548 hence @xmath549 , and the family @xmath539 is uniformily totally bounded , and the claim follows from gromov s ( pre)compactness theorem . let us emphasize that the measures @xmath49 do not play a role explicitly in proving that the family @xmath539 is uniformily totally bounded . the total masses do not need to be uniformly bounded , but the quantity @xmath550 . therefore , a sequence of measures has not to be precompact , and consequently the statement is not formulated in terms of measured gromov - hausdorff convergence ( or measured gromov convergence @xcite ) . this viewpoint becomes even more apparent if one considers just vector fields @xmath125 that are uniformily essentially bounded . then , in @xmath551 the mass cancels , and @xmath552 suggests that one can extract a `` weakly converging '' subseqence of @xmath553 . [ [ riemannian - manifolds - with - boundary ] ] * riemannian manifolds with boundary * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath5 be a geodesically convex , compact riemannian manifold with non - empty boundary , and @xmath554 . for instance , one can consider a convex domain in @xmath555 . condition ( [ transportcondition ] ) is obviously satisfied . note that any wasserstein geodesic in @xmath305 between absolutely continuous probability measures with bounded densities has midpoint measures with bounded densities since @xmath556 for some @xmath557 . let @xmath125 be any smooth vector field on @xmath7 that is @xmath0-integrable w.r.t . assume that @xmath558 for some @xmath171 and @xmath156 . then @xmath559 satisfies the condition @xmath560 if @xmath561 . in particular , if @xmath444 , @xmath562 and @xmath563 , then @xmath564 and @xmath559 satisfies the condition @xmath565 . [ [ n - warped - products ] ] * @xmath2-warped products * + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + let @xmath566 be a @xmath567-dimensional riemannian manifold , and lef @xmath568 be a smooth function . we assume * @xmath569 * @xmath570 * @xmath571 . let @xmath572 be weighted riemannian manifold where @xmath573 is a smooth reference measure and @xmath574 is the riemannian volume with respect to @xmath575 . the riemannian @xmath2-warped product @xmath576 between @xmath577 , @xmath578 and @xmath141 is given by the metric completion @xmath48 of the riemannian metric @xmath579 on @xmath580 , and the measure @xmath581 on @xmath69 ( see also @xcite ) . @xmath582 and @xmath583 are the projections maps . + the @xmath2-ricci tensor of @xmath584 and @xmath585 is computed in @xcite : @xmath586|v|_{{\scriptscriptstyle{c}}}^2\end{aligned}\ ] ] and the previous assumptions together with @xmath587 imply @xmath588 for every @xmath589 and for every @xmath590 respectively . + let @xmath591 be a riemannian manifold and @xmath125 a smooth vector field on @xmath592 such that @xmath593 satisfies the condition @xmath594 . by theorem [ maintheorem ] the condition is equivalent to @xmath595 . let @xmath141 , @xmath596 , and @xmath125 be as before , with constants @xmath597 and @xmath156 . assume @xmath577 has alexandrov curvature bounded from below by @xmath186 , and @xmath598 . if @xmath160 and @xmath599 , we assume @xmath600 . + then the @xmath2-warped product @xmath601 is a smooth metric measure space , and @xmath602 satisfies the condition @xmath603 where @xmath604 . first , by similar computations as in the proof of proposition 3.2 in @xcite we obtain that the @xmath605-ricci tensor of @xmath606 is given by @xmath607|v|_{\scriptscriptstyle{c}}^2\end{aligned}\ ] ] and under the previous assumptions together with @xmath608 , this implies @xmath609 moreover , if @xmath599 , then by the generalized bonnet - myers theorem we know that @xmath610 . note , by theorem 1.2 in @xcite condition ( iii ) from the beginning - provided condition ( ii ) - is equivalent to * @xmath611 if @xmath612 , * @xmath599 and @xmath613 if @xmath614 . then , exactly like in the proof of theorem 3.4 in @xcite we can prove if @xmath25 is an optimal dynamical transference plan in @xmath615 such that @xmath616 is absolutely continuous with respect to @xmath617 , then @xmath618 hence , the @xmath2-warped product is indeed a generalized smooth metric measure space in the sense of our definition . for the proof we need a maximal diameter theorem for @xmath619 that satisfies the condition @xmath620 with @xmath599 . for smooth @xmath125 this is provided by kuwada in @xcite ( we also refer to @xcite for closely related results ) . + finally , following the lines of the proof of the main theorem in @xcite in combination with theorem [ maintheorem ] , we obtain the condition @xmath31 with admissible vector field @xmath621 . * in particular , the previous theorem covers the case of @xmath2-cones and @xmath2-suspensions that play an important role for the study of spaces with generalized lower ricci curvature bounds . * @xmath622 is @xmath0-integrable w.r.t . @xmath617 if @xmath164 . let @xmath623 $ ] and @xmath624 , and let @xmath125 be any smooth vector field on @xmath625 . since @xmath125 is smooth and @xmath625 is compact , we can find a constant @xmath626 such that @xmath627 now , we consider @xmath628 with @xmath629 such that @xmath630 and choose @xmath631 such that @xmath632 . hence @xmath633\geq -\alpha \kappa|v|^2.\end{aligned}\ ] ] and therefore @xmath634 now , we set @xmath635 and @xmath636\rightarrow [ 0,\infty)$ ] with @xmath637 . the previous theorem yields that @xmath601 with @xmath638 satisfies the condition @xmath639 . note @xmath640 one only can achieve if we set @xmath641 . then , any @xmath642 is admissible , @xmath427 and - for @xmath643 - @xmath644 . otherwise , the underlying metric space has singularities at the vanishing points of @xmath141 . also note that @xmath645 is an invariant measure of the corresponding nonsymmetric diffusion operator @xmath646 on @xmath601 where @xmath647 . we observe that the diameter bound @xmath64 of the generalized bonnet - myers theorem is attained but the @xmath2-warped product is not a smooth manifold unless @xmath641 and @xmath643 ! in @xcite kuwada proves that for @xmath435 ( with @xmath648 a smooth manifold , @xmath286 , and @xmath125 a smooth vector field ) that satisfies @xmath649 with @xmath265 and @xmath329 the maximal diameter is attained if and only if @xmath650 and @xmath427 . we also observe that the generalized laplace operator splits into a nonsymmetric part @xmath651 and a symmetric part @xmath652 where the invariant measure @xmath653 is also the invariant measure for the symmetric part , and @xmath654 for any smooth function @xmath655 . in this section we discuss the link between the curvature - dimension condition and contraction estimates in wasserstein space . let @xmath22 be a compact generalized smooth metric measure space that satisfies the condition @xmath31 for an admissible vector field @xmath125 . we assume that @xmath656 , @xmath286 and @xmath125 is smooth . hence , @xmath657 is induced by a smooth riemannian metric @xmath102 on @xmath7 . theorem [ maintheorem ] yields @xmath658 for @xmath171 and @xmath297 $ ] . we also assume @xmath21 . + + we can consider the diffusion operator @xmath659 and the corresponding nonsymmetric dirichlet form @xmath660 on @xmath661 where @xmath662 . note that @xmath663 is a dirichlet form adapted to the dirichlet energy w.r.t . @xmath102 in the sense of @xcite , and @xmath4 is the laplace - beltrami operator of @xmath5 . let @xmath33 be the associated diffusion semigroup , let @xmath664 be the so - called co - semigroup , and let @xmath35 be the dual flow acting on probability measures : @xmath665 where @xmath666 and @xmath667 . @xmath668 denotes the generator of @xmath664 . note @xmath669 for more details about nonsymmetric dirichlet forms and the relation between @xmath33 , @xmath670 , @xmath8 , @xmath668 and @xmath663 we refer to @xcite . since @xmath671 we have @xmath672 . + we assume that there is a @xmath673-kernel @xmath674 for @xmath33 , i.e. @xmath675 such that @xmath676 is @xmath673 . then @xmath677@xmath678 in particular , @xmath37 is @xmath418-absolutely continuous for any @xmath679 , and its density @xmath680 is smooth . note that we can extend @xmath664 ( and @xmath33 as well ) as semigroup acting on @xmath681 . then , for @xmath418-absolutely continuous probability measures @xmath682 we have @xmath683 . by abuse of notation we denote with @xmath8 the @xmath684-generator of @xmath33 , and similar for @xmath664 . then , for instance ( [ uuuu ] ) extends in a canonical way provided @xmath685-integrability for @xmath686 . this is straightforward , and we omit details . [ kuwadaslemma ] let @xmath687 and @xmath125 be as before . let @xmath679 . then @xmath688 is a locally absolutely continuous curve in @xmath689 , and latexmath:[\ ] ] deviding by @xmath767 , and letting @xmath768 yields @xmath769 since geodesic between absolutely continuous probability measures are unique , and since @xmath764 coincides with @xmath758 up to reverse parametrization , we have for @xmath770 @xmath771 and the last line in ( [ seethelastline ] ) vanishes . hence @xmath772 finally , gromwall s lemma yields the claim . it is rather obvious how to improve the previous estimates in context of the condition @xmath31 for @xmath30 in the sense of @xcite by modifying the computations ( see also @xcite ) , and we omit details . let @xmath5 be a compact smooth riemannian manifold , and let @xmath125 be an smooth vector field . we denote with @xmath42 the corresponding metric measure space , and let @xmath33 and @xmath35 be as above . then , the following statements are equivalent . * @xmath775 , * @xmath559 satisfies the condition @xmath23 , * for every @xmath45 @xmath37 is an @xmath46-flow curve starting in @xmath36 , * @xmath35 satisfies the contraction estimate in corollary [ contractionxxx ] , * @xmath33 satisfies the condition @xmath47 .